Treatments for Social Interaction Impairment in Animal Models of Schizophrenia: A Critical Review and Meta-analysis

Abstract

Background

While pharmacological treatments for positive symptoms of schizophrenia are widely used, their beneficial effect on negative symptoms, particularly social impairment, is insufficiently studied. Therefore, there is an increasing interest in preclinical research of potentially beneficial treatments, with mixed results. The current review aims to evaluate the efficacy of available treatments for social deficits in different animal models of schizophrenia.

Study Design

A systematic literature search generated 145 outcomes for the measures “total time” and “number” of social interactions. Standardized mean differences (SMD) and 95% confidence interval (CI) were calculated, and heterogeneity was tested using Q statistics in a random-effect meta-analytic model. Given the vast heterogeneity in effect sizes, the animal model, treatment group, and sample size were all examined as potential moderators.

Study Results

The results showed that in almost all models, treatment significantly improved social deficit (total time: SMD = 1.24; number: SMD = 1.1). The moderator analyses discovered significant subgroup differences across models and treatment subgroups. Perinatal and adult pharmacological models showed the most substantial influence of treatments on social deficits, reflecting relative pharmacological validity. Furthermore, atypical antipsychotic drugs had the highest SMD within each model subgroup.

Conclusions

Our findings indicate that the improvement in social interaction behaviors is dependent on the animal model and treatment family used. Implications for the preclinical and clinical fields are discussed.

Introduction

Schizophrenia is a severe mental disorder that causes significant distress for the individual. A recent systematic review indicates that about 0.28% of the general population is diagnosed with schizophrenia, with no differences between genders in prevalence.¹ The symptoms of schizophrenia are classified into 3 categories based on their pathological characteristics: positive, negative, and cognitive.^2,3 Social impairment is one of the main features of the negative symptoms,⁴ including deficits in social cognition,^5,6 social functioning,⁷ and emotion perception and expression.⁸ These deficits are expressed in decreased motivation to interact with others.⁹ Moreover, social withdrawal is a risk factor for developing schizophrenia later in life.^10–12 In a study of a high-risk population, it predicted the later occurrence of schizophrenia,¹³ and social skills are an essential component in relapse prevention and the recovery process.^14,15 Thus, social capacities have a crucial role in the development and prognosis of schizophrenia.

Treatments for Negative Symptoms in Schizophrenia

The current therapeutic recommendation for schizophrenia is a combination of psychosocial and psychopharmacological interventions.¹⁶ However, the treatments available today primarily ameliorate the positive symptoms.¹⁷ Because treatments for the negative symptoms are only partially beneficial, there is no clear recommendation for optimal treatment for this aspect of the disorder.^17,18 The most common treatments for social withdrawal in schizophrenia are cognitive behavioral therapy and social skills training,^9,19 although 2 recent meta-analyses indicate that their efficacy is nonsignificant or limited to specific capacities.^20,21

Along with psychosocial therapy, several pharmacotherapeutic drugs were partially beneficial for treating social malfunction. These treatments were chosen based on brain targets found to be dysregulated in schizophrenia. Antipsychotic drugs (AP) are the first-line pharmacological treatment.^22–26 AP treatments were found to have beneficial effects on social capacities in several studies.^27–29 For instance, AP treatment shows beneficial effects on theory of mind; after 2 weeks of treatment, patients with schizophrenia showed significant improvement compared to their own baseline and reached similar scores as a control group.³⁰ Antidepressants (AD) are also widely used but usually as an add-on treatment to AP.^31–34 Since dopaminergic system dysregulation is central in the pathobiology underlying schizophrenia,^35–37 some medications target this system.^38–42 Furthermore, the cholinergic system, emphasizing the muscarinic and nicotinic receptors, also plays a substantial role in the pathogenesis of negative symptoms.^43–46 Therefore it has been recommended as a potential target for novel treatments.^47–50

Additional brain targets have recently been proposed to establish a more comprehensive and beneficial treatment. One approach suggested the endocannabinoid system and phosphodiesterase inhibitors (PDEi).^51–60 Another approach addressed the endocrine disruption and proposed to treat with related hormones such as oxytocin,^61–63 vasopressin,^64–66 and melatonin.^67–70 Furthermore, many new treatments, such as interneuron transplants (IT), are currently being studied.^71–76 Lastly, some studies indicate that combining 2 drugs allows regulating more aspects of neuronal abnormalities and is thus preferable to treating with a sole drug.^77,78 These attempts to improve social impairment through various drugs, targeting different biological systems, highlight the current only partial understanding of the disorder’s complex pathophysiology and the limited capacity to treat all pathologic aspects of the negative symptoms; therefore, more research is required to develop a novel and more efficient treatment approach.²²

Animal Models for Schizophrenia

The use of animal models for human psychiatric disorders is a main scientific technique for providing insight into the underlying physiological and neurological mechanisms and as screening models for novel treatments. In the literature, there are many accepted, well-studied, and validated animal models of psychiatric disorders.^79–81 For schizophrenia, several animal models mimic its main symptoms, although this modeling is complex and incomplete due to the lack of knowledge on the disorder’s pathogenesis.^82–84 Each model targets a different aspect of the brain dysregulation and symptomatic manifestation, and hence the models approach the study of schizophrenia and its treatments from different perspectives.

The rodent models for schizophrenia can be classified into several main groups^81,83,85: (a) Pharmacological models cause a chemical imbalance in the neural systems dysregulated in schizophrenia.^86–88 For example, phencyclidine (PCP), ketamine, and MK-801 are used to induce schizophrenia-like symptoms by creating glutamatergic and dopaminergic imbalance.^89–93; (b) neurodevelopmental models focusing on prenatal or perinatal exposure to different adverse environmental factors.^94–99 This group can be subdivided to: (b1) Lesion models—mainly refers to neonatal lesion of the ventral hippocampus based on brain imaging findings from patients with schizophrenia.^100–103 (b2) Genetic models—based on creating a knockout (KO) for a gene or using a strain that inherits phenotypes that mimic schizophrenia symptoms.^104–110 For instance, Spontaneously Hypertensive Rat (SHR) and Gunn strains display a range of behavioral pathophenotypes and biochemical dysregulations associated with schizophrenia.^111–115 (b3) Perinatal Pharmacological models—cause a chemical imbalance in the perinatal period, such as methylazoxymethanol acetate (MAM) and polyriboinosinic–polyribocytidilic acid (Poly I:C), toxins given during gestational days, targeting the immune system or neuroblast development.^116–119 These neurodevelopmental models are initiated during early, sensitive periods of brain development. Therefore, they modify brain development and may induce a wide spectrum of brain pathologies.

In rodents, the social interaction test is considered a reliable method to test the deficits of negative symptoms in animal models.^120,121 In this test, 2 unfamiliar rats or mice are placed together in an empty testing arena (for more details see supplementary appendix A).¹²² The main recorded measures are total time spent in social interaction (eg, sniffing, following, etc.) and number of social interactions.¹²³ Different models of schizophrenia result in performance deficits in this test, and relevant psychiatric drug treatments improve these behavioral deficits, indicating the existence of predictive validity.^81,124

Research Goals

Based on the extensive research in the field, we performed a meta-analysis investigating the effect of various drug treatments on social impairment in animal models of schizophrenia, as expressed in the social interaction test. This meta-analysis allows: (a) to investigate the pharmacological validity of the different animal models, ie, whether well-established pharmacological treatments will reverse social withdrawal in the different animal models^81,125,126; (b) to examine the efficacy of treatments to ameliorate social withdrawal, a major symptom of schizophrenia that does not have satisfactory treatment response. Since each treatment targets differently the neurobiological dysregulation, we hypothesize that they will also have different effects on social deficit; and (c) to evaluate the overall effect sizes and homogeneity in the research field, and to identify moderators that should be considered while conducting future studies. This meta-analysis summarizes the existing research in the literature to date and can suggest how to perform more accurate research, increasing the likelihood of finding an effective treatment for social deficits in schizophrenia.

Methods

Literature Search

The current meta-analysis followed the Preferred Reporting Items for Systematic Reviews and Meta-Analysis guidelines (PRISMA).¹²⁷ The analysis is based on peer-reviewed studies collected from PubMed, PsycNet, and Scopus and published until January 4, 2021. The search terms were (a) “social interaction”; (b) “rats” or “mice”; (c) “treatment”; (d) “schizophrenia”; and (e) “social.” Aside from the term “social interaction,” which could appear throughout the article, the remaining terms were defined to appear in the title or abstract.

Inclusion/Exclusion Criteria

The following exclusion criteria were applied: (a) review or book chapters; (b) not in English; (c) human subjects; (d) no model was used to generate schizophrenia-like behavior; (e) no treatment was given to treat the social deficit; (f) the model did not yield a deficit in the social interaction test; (g) the treatment was given before or parallel to the creation of the model (this is because our focus is about treating the pathology and not preventing it); and (h) the behavioral measures were not “social interaction time” or “number of social interactions.”

The initial search yielded 781 papers from all 3 databases, with 314 duplicates, leaving 467 original papers for further review. Following this, 2 independent coders screened the studies, and 418 studies were excluded based on the above exclusion criteria (figure 1). Consequently, the meta-analysis was performed on the remaining 49 studies and 145 outcomes. Of those, the estimates for “total social interaction time” were calculated from 47 studies and 129 outcomes, and the estimates for “number of social interactions” were calculated from 5 studies and 16 outcomes. For these outcomes, we recorded the variables: study year, authors, animal, sex, model, sample size, treatment, and effect size. Afterward, we classified the various drug treatments into treatment groups because of their wide range. If they did not fit into an official drug group, they were classified based on their brain targets.

Fig. 1.

PRISMA flow diagram.¹²⁸

Outcomes and Statistical Analysis

Based on the primary outcomes measures, standardized mean differences (SMD) between treatment and placebo groups were calculated (see details in supplementary appendix C). Effects greater than zero indicated that the treatment improved social behavior compared to the control group. Moreover, the standard error for each effect size was calculated from the 95% confidence interval (CI).¹²⁹ The random-effects model was used for the analysis. Egger’s test was calculated, and funnel plot asymmetry was examined to test for publication bias.^130,131 Afterwards, we conducted moderator analyses with model, treatment group, and sample size as potential moderators of the total effect size of each measure.¹³² RevMan software was used for statistical analysis and JASP software for figure making.^133,134 All statistical tests were 2-tailed with a significance level set at P < .05.

Results

Study Characteristics

The literature search yielded 49 studies and 145 outcomes, with 4590 subjects (details of the study characteristics are presented in supplementary tables S2 and S3). Overall, 88 outcomes were performed with rats and 57 with mice. Moreover, only 2% of the studies were conducted on females (1 study only with females and 2 studies with females and males).

Publication Bias

For both total time and number of interactions, visual examination of the funnel plots asymmetry did not identify evidence for publication bias (supplementary figure S1). Furthermore, Egger’s test was not significant for the 2 measures [total time: t(129) = −0.012, P = .991; number of interactions: t(15) = −0.678, P = .509].

Main Analyses

SMD for total time and number of interactions are shown in the forest plots in supplementary figures S2 and S3. Overall, there was a significant effect for treatment on total time (SMD = 1.24, CI = 1.08–1.41, P < .0001) and on number of interactions (SMD = 1.1, CI = 0.65–1.55, P < .0001). However, there was a significant heterogeneity between the effect sizes of total time (Q = 519.75, I² = 75%, df = 128, P < .0001) and number of interactions (Q = 54.01, I² = 72%, df = 15, P < .0001).

Moderator Analyses

Nested moderator analyses were conducted using 3 levels of variables that we suspected might be potential moderators—the animal model of pathology, treatment group, and sample size. Figure 2A and B summarizes the subgroups’ deviation (see details on the classification process in supplementary Table S1).

Fig. 2.

Subgroup deviation in the 3 levels moderators for total time (A) and number of interactions (B). Only subgroups with significant heterogeneity (marked with *) continued to the next level. **Subgroups that we could not proceed to the next level because they could not be further divided into subgroups for heterogeneity analysis. AAP, atypical antipsychotic; AD, antidepressant; AP, antipsychotic; CAN, cannabinoids; CHL, cholinergic; DOP, dopaminergic; IT, interneuron transplant; KO, knockout; MAM, methylazoxymethanol acetate; MT, mixed treatment; NOSi, nitric oxide synthase inhibitors; PCP, phencyclidine; PDEi, phosphodiesterase inhibitors; Poly I:C, polyriboinosinic–polyribocytidilic acid; SHR, spontaneously hypertensive rats; ST, stimulants; TAP, typical antipsychotic; VN, vanilloids; VS, vasodilator.

Moderators Level 1: Animal Model

We hypothesized that the model that created the pathology phenotype could explain the variability in effect sizes. In the measure total time, 10 model subgroups were identified—MK-801, ketamine, PCP (3 NMDA receptor antagonists analyzed separately, see Supplementary Material B2), MAM, Poly I:C, SHR, bilateral neonatal hippocampal lesion (Lesion), KO, Gunn, and nitric oxide synthase inhibitors (NOSi). The remaining 5 outcomes were classified as “other” since they do not match any group. Statistics of heterogeneity tests of the subgroups are summarized in table 1A. In the test for subgroup differences in mean SMD, we found significant differences between the subgroups (Q = 62.58, df = 9, P < .0001). Within 4 subgroups—Poly I:C, KO, Gunn, and NOSi—the Q statistic was nonsignificant, indicating homogeneity in their corresponding effect sizes. However, in the other subgroup, significant heterogeneity was found (details in table 1A). Furthermore, in the measure number of interactions, 2 model subgroups were identified—MK-801 and WIN55,212-2 with 1 outcome as “other.” Testing the differences in mean SMD between subgroups yielded a significant difference between those groups (Q = 5.42, df = 1, P = .02). However, while the WIN55,212-2 subgroup had a homogenous effect size, the MK-801 subgroup was still significantly heterogeneous regarding average effect size (all heterogeneity test statistics are summarized in table 1B).

Table 1.

Total time (A) and number of interactions (B) measures—heterogeneity test and test for overall effect for the moderators’ subgroups model and treatment group

Moderator	Subgroup	k	SMD (95% CI)	P	Q	Q’s P
A. Total social interaction time
Model	MK-801	25	1.47 (1.2 to 1.75)	<.001	47.74	<.001
Other: n = 5	Ketamine	5	2.57 (1.94 to 3.20)	<.001	20.20	<.001
	MAM	10	1.14 (0.8 to 2.00)	.009	54.82	<.001
	PCP	44	1.36 (1.09 to 1.62)	<.001	167.41	<.001
	Poly I:C	3	2.22 (1.27 to 3.18)	<.001	3.57	.17
	SHR	21	0.54 (0.27 to 0.81)	<.001	32.4	.04
	Lesion	8	0.36 (−0.16 to 0.89)	.18	15	.04
	KO	2	0.75 (−0.11 to 1.62)	.09	1.42	.23
	Gunn	3	1.36 (0.64 to 2.09)	<.001	0.38	.83
	NOSi	3	1.74 (0.49 to 2.99)	.006	4.13	.13
Treatment (model)
(MK-801)	Antidepressant	5	0.78 (0.05 to 1.51)	.04	7.08	.13
Other: n = 2	Antipsychotic	9	1.81 (1.28 to 2.34)	<.001	14.93	.06
	Cannabinoids	5	1.18 (0.72 to 1.63)	<.001	0.61	.96
	Mixed treatment	4	1.49 (1.02 to 1.96)	<.001	6.55	.09
(MAM)	Typical antipsychotic	2	−0.75 (−3.16 to 1.67)	.54	9.14	.003
	Atypical antipsychotic	3	2.34 (1.58 to 3.10)	<.001	2.18	.34
	Cannabinoids	3	1.69 (0.58 to 2.80)	.003	5.38	.07
	Interneuron transplant	2	0.37 (−0.80 to 1.54)	.060	3.67	.06
(PCP)	Typical antipsychotic	4	0.94 (0.12 to 1.76)	.02	10.2	.02
Other: n = 5	Atypical antipsychotica	8	2.06 (1.65 to 2.47)	<.001	4.35	.74
	Other antipsychotic	3	1.90 (0.87 to 2.93)	<.001	5.45	.07
	Cholinergic	8	1.01 (0.51 to 1.51)	<.001	22.43	.002
	Dopaminergic	5	1.74 (0.78 to 2.71)	<.001	18.97	.001
	Phosphodiesterase	5	1.20 (0.68 to 1.71)	<.001	9.84	.04
(SHR)	Antipsychotic	3	0.98 (0.22 to 1.74)	.01	2.72	.26
	Cannabinoids	11	0.28 (−0.13 to 0.69)	.18	20.26	.03
	Stimulants	3	0.87 (0.34 to 1.40)	.001	0.65	.72
	Vasodilators	2	1.01 (0.34 to 1.68)	.003	0.001	.98
	Vanilloids	2	0.47 (−0.18 to 1.12)	.160	0.02	.89
(Lesion)	Antipsychotic	6	0.24 (−0.19 to 0.67)	.28	5.43	.37
Other: n = 2
B. Number of social interactions
Model	MK-801	13	1.26 (0.75 to 1.77)	<.001	42.88	<.001
Other: n = 1	WIN55,212-2	2	0.33 (−0.33 to 0.93)	.35	0.91	.34
Treatment (model)
(MK-801)	Antidepressant	4	0.41 (−0.71 to 0.98)	.16	0.7	.87
Other: n = 1	Atypical Antipsychotic	5	1.7 (0.48 to 2.91)	.006	26.01	<.001
	Mixed Treatment	3	1.39 (1.04 to 1.74)	<.001	0.12	.94

Notes: k, number of outcomes; K, knockout; Lesion, bilateral ventral hippocampal lesion; MAM, methylazoxymethanol acetate; NOSi, nitric oxide synthase inhibitors; PCP, phencyclidine; PDEi, phosphodiesterase inhibitors; Poly I:C, polyriboinosinic–polyribocytidilic acid; SHR, spontaneously hypertensive rats; SMD, standardized mean differences.

Atypical Antipsychotic subgroup without the treatment risperidone which was excluded.

Moderators Level 2: Treatment Group

Comparisons Between Treatment Groups Within Each Model

Significant heterogeneity in model subgroup continued to level 2 of the moderators analysis. In total time, the drugs were classified into 12 treatment subgroups—AP, AD, Cannabinoids (CAN), Cholinergic (CHL), Dopaminergic (DOP), PDEi, IT, Stimulants (ST), Vasodilators (VS), Vanilloids (VN) and Mixed treatment (MT; when 2 drugs were given together). If the overall effect sizes in the antipsychotics were significantly heterogeneous, they were further split into atypical and typical antipsychotic groups (AAP and TAP, respectively). The classification is detailed in supplementary tables S2 and S3 and the statistics of heterogeneity tests in table 1A. Within the MK-801 subgroup, the treatment subgroups AP, AD, CAN, and MT had homogeneous effect sizes, obscuring the general group’s significant heterogeneity. However, there were no significant differences between these groups in mean SMD (Q = 6.09, df = 3, P = .11). Within the MAM subgroup, the treatment subgroups CAN, IT, and AAP had homogeneous effect sizes, and only the TAP remained significantly heterogeneous. Moreover, a significant difference was found in mean SMD between those treatment subgroups (Q = 11.58, df = 3, P = .009). Within the PCP subgroup, treatments were classified into 6 groups—CHL, DOP, PEDi, TAP, AAP, and other AP. Within the AAP, the treatment risperidone was excluded into a separate group since it was heterogeneous, which could not be explained by the model we performed. For this treatment, the dose can explain the variation in the effect sizes; thus, except for 1 exceptional outcome, there is a positive correlation between treatment dose and the effect sizes (figure 3). Furthermore, there were significant differences in effect sizes between the treatment subgroups (Q = 15.04, df = 5, P = .01). However, only the AAP and the other AP had homogeneous effect sizes, and the other treatment subgroups were still significantly heterogeneous. Within the SHR subgroup, the treatment subgroups AP, ST, VS, and VN had homogeneous effect sizes, and only the CAN remained significantly heterogenous. However, there were no significant differences between the treatment subgroups’ average effect sizes (Q = 5.87, df = 4, P = .21). Lastly, in the lesion subgroup, only the AP treatment subgroup was found, with 2 outcomes classified as “other” since they did not match any group. This AP group had homogenous effect sizes.

Fig. 3.

Correlation between SMD and dose of risperidone treatment. The outcomes refer only to the risperidone subgroup within the PCP model in total time measure. Pearson coefficient correlation with all the 6 outcomes was not significant (r = 0.57, P = .237), but without the unusual outcome (SMD = −0.714) there is a significant strong positive correlation (r = 0.99, P < .001).

In number of interactions, within the subgroup MK-801, the treatment subgroups AD and MT had homogenous effect sizes, and only AAP remained with significant heterogeneity (Statistics summarized in table 1B). Furthermore, there were significant differences in the mean SMD between the subgroups (Q = 9.10, df = 2, P = .01).

Comparison Between Models Treated With Atypical Antipsychotics

To examine the pharmacological validity, we compared the effect of AAP treatment on the different models of schizophrenia. We chose to compare this treatment group because the antipsychotic drugs are the first-line treatment for schizophrenia,^22,135 and within the antipsychotics, AAP was the most common drug in our meta-analysis (36 of 129 outcomes).

In our meta-analysis, the AAP treatment subgroup was given in 7 different models—MK-801, MAM, PCP, SHR, Lesion, Gunn, and NOSi (see supplementary figure S4). Of these, all subgroups had homogeneous effect sizes except for the MK-801 (Q = 14.52, I² = 52%, df = 7, P = .04). Thus, the mean effect size of each subgroup mostly represents the group. The test for subgroup differences revealed a significant difference between the mean effect sizes of these subgroups (Q = 26.52, df = 6, P = .0002).

Moderators Level 3: Sample Size

Significant heterogeneity in treatment subgroups continued to level 3 of the moderators’ analysis. In total time and number of interactions, all sample size subgroups were homogenous except for 2 outcomes in the number of interactions that could not gather into nonsignificant heterogeneity (Statistics detailed in supplementary tables S4 and S5).

Discussion

While the potential beneficial treatments for the negative symptoms of schizophrenia have been extensively researched, conflicting findings have led to the absence of clear recommendations for the negative symptoms, particularly for social impairment.^17,136 The current meta-analysis included a total of 145 outcomes that evaluated the therapeutic effect of pharmacological treatments on social deficits in animal models for schizophrenia. To the best of our knowledge, this is the first meta-analysis that examined this question.

We found that overall, the available pharmacological treatments have beneficial effects on social impairment in animal models for schizophrenia, with significant and high effect sizes for the main measures in the literature: total time and number of social interactions. Nevertheless, there was vast heterogeneity in the SMD. From the moderator analysis, it appears that there are differences between the models, reflecting relative pharmacological validity.

Some perinatal (Poly I:C and NOSi) and adult (MK-801 and ketamine) pharmacological models resulted in large and significant SMD, indicating that they have substantial pharmacological validity. These models further show external validity as they lead to negative-like symptoms in both zebrafish and healthy humans.^137–140 On the other hand, there are specific models, “lesion,” “KO,” and WIN55,212-2 in which the drugs, in general, do not have a significant therapeutic effect (see table 1). However, AAP specifically were effective in the lesion model. Thus the current complex findings suggest that in the context of social deficiency, neurodevelopmental, perinatal, and adult pharmacological models, may represent the high complexity of the disorder.

The models’ different effects on social behavior were further modulated by the particular treatment. In general, the AAP treatment group (or AP when not separated for AAP and TAP) has the highest mean effect size within each model. Therefore, it seems that among the drugs currently offered, according to rodent model studies, it is the preferred drug group for treating social impairment in schizophrenia. Moreover, AAP treatments have different SMD in the different models (perinatal and adult pharmacological models showed the highest SMD; supplementary figure S4), suggesting that AAP may affect social behavior differently on different biological dysregulation backgrounds. This further suggests the importance of carefully choosing the models for studies in this field of research.

The present meta-analysis, based on animal models studies, suggests that several treatment groups may have potential for treatment of asocial aspects of schizophrenia. While AAP showed the most robust pattern of effects, additional treatments showed relative strength (figure 4). Treatments targeting mainly the dopaminergic and cholinergic systems showed high effects. Moreover, the new drug groups ST, VS, PDEi, and new AP molecules were found in the meta-analysis to yield good results, so it is interesting to further investigate them. MT (generally includes a combination of AP and a new drug) also showed high effects for both measures. In contrast, the new treatment groups, IT, and VN, showed insignificant effects on social behavior. Moreover, AD yielded partial effects, as they significantly improved total interaction time but not number of interactions. Lastly, a TAP group also showed partial results.

Fig. 4.

Forest plot for mean SMD of treatment groups in the different models. Size effects refer for the measure “total time,” except the 4 cases refer to the measure “number of interactions” (marked with *). CI, confidence interval; MAM, methylazoxymethanol acetate; NOSi, nitric oxide synthase inhibitors; PCP, phencyclidine; Poly I:C, polyriboinosinic–polyribocytidilic acid; SHR, spontaneously hypertensive rats; SMD, standardized mean differences.

The CAN treatment group showed a more complicated pattern. First, while CAN has considerable effects within MK-801, Poly I:C, and MAM models, in the SHR model it did not. Moreover, WIN55,212-2, a CAN receptor agonist, is used both for model creation and to treat symptom-like deficits created by other models.^141,142 The endocannabinoid system has relationships with systems relevant to schizophrenia, such as dopaminergic and cholinergic,^143–145 and is found to be dysregulated in schizophrenia.^146,147 However, the role of this system in social behavior and the effects of endocannabinoids ligands in this context is complex.^148–150 Thus, exogenic CAN components can lead to improvement in social deficit but on the other hand they can also produce it.^151,152 Further research is needed to better understand how manipulating the endocannabinoid system can be used to improve psychiatric symptoms.

In addition to the combined effects of model and treatment on social behavior, sample size further interacted with model and treatment. The nested moderator analysis of sample size achieved homogeneity in SMD within each subgroup. Although sample size affects the power and validity of the test, the effect size is independent of this bias.¹⁵³ Accordingly, in our moderator analysis, there is no clear direction for an association between sample size and effect size. Sometimes there was a bigger effect size in subgroups with a larger sample, and sometimes vice versa. This is consistent with the absence of evidence for publication bias in this meta-analysis. These results make it difficult to draw concrete conclusions about sample size. Thus, it appears that keeping the sample size as similar as possible between research groups may be essential to reduce its bias on outcomes.

One of the most significant findings in our meta-analysis is the absence of studies on females. Only 3 of the 150 outcomes had female subjects, 2 of which combined male and female subjects.^154–156 In humans, the prevalence of males and females diagnosed with schizophrenia are similar or have slight gaps that alter with age [while males outnumber women at younger ages (1.4:1), women are modestly dominant in older age groups].^1,157 More importantly, studies discovered sex differences in the neurodevelopmental dysregulations of females and males with schizophrenia.^158–161 Furthermore, sex differences were found in response to pharmacological treatments in patients with schizophrenia.^162–165 These findings raise a major problem in the deduction of clinical implications for females from male-only studies and highlight the importance of further research on females.

From Preclinical to Clinical Studies

In the clinical field, negative symptoms, and specifically social deficits, have far-reaching implications on the well-being and prognosis of people with schizophrenia.^{10,15,166–169} Many studies failed to find promising treatments for negative symptoms.^17,18,21 For example, a meta-analysis that examined the effects of intranasal oxytocin treatment on negative symptoms in general found no beneficial effects.¹⁷⁰ In addition, 2 meta-analyses did not find significant effects for pharmacological treatments, but they both examined the effect on negative symptoms as 1 unit within no specific reference to social impairment.^18,171 In contrast, when focusing on studies that measured social capacities, a different picture emerges. Several studies indicate that AAP treatments had beneficial effects on social functioning.^28,30,172 For instance, quetiapine improved social competence.²⁹ Perospirone had beneficial effects on social cognition but not on negative symptoms in general.¹⁷³ Risperidone had more conflicting effects, with studies showing improvement in social capacities and reduction in hospitalizations,^27,174 but no effects on emotion perception nor social competence (measured with another scale).^175,176 Furthermore, oxytocin and ITI-007, a new monoamine modulator drug, had beneficial effects on social impairment in schizophrenia.^{38,61,177–179} However, a recent meta-analysis reported that the beneficial effects of oxytocin are limited to high-level social cognition capacities.¹⁸⁰ In other psychopathologies such as autism and social phobia, there are also reports on pharmacotherapeutic effects on social functioning.^181–184 In the light of the current and other meta-analyses findings, there is a significant need to conduct clinical trials that will examine the effects of pharmacological treatments on specific measures for social deficits of schizophrenia.

Limitations

It is important to acknowledge the possibility of confounding variables that have not been tested as moderators. For example, due to the wide variety of treatments within each treatment group, it was impossible to examine the dose of the treatment or the duration of the treatment as a variable that may affect effect sizes (as we found in the treatment risperidone). Nevertheless, we have examined whether drug responsiveness was affected by differences in model creation. We found no significant differences between acute and chronic administrations nor between neonatal and adolescent periods of model creations (details in supplementary appendix B). A follow-up review could focus on AAP treatments and further identify whether some particular drug components have more beneficial effects on social deficits and whether it is possible to characterize which brain targets should be focused on in the study of new drugs (eg, whether there is a specific receptor(s) whose activation led to more beneficial effects).

Another limitation concerns the existence of other social behavior tests for rodents that are not included in our meta-analysis.¹⁸⁵ For example, the 3-chamber test variations examine different social motivation and memory patterns (eg, social recognition or social vs nonsocial preference),^186–189 and the resident-intruder test focuses mainly on aggressive behavior.¹⁹⁰ However, in the present meta-analysis, our main focus is to assist in finding a beneficial treatment for social impairment in schizophrenia. Therefore, we had to standardize the behavioral phenotype variable to allow statistical comparison of the other variables. In order to minimize this limitation as possible, we chose the most basic and common social test—the social interaction test, recommended by the MATRICS (Measurement and Treatment Research to Improve Cognition in Schizophrenia).¹⁹¹ This test covers the broadest range of social behaviors (sniffing, following, climbing, etc.), and not only approaching or recognizing. Further research can compare the pharmacological effects on additional social tests while controlling other variables (ie, choose 1 animal model or 1 pharmacological treatment and compare the other variables).

Conclusions

Our findings suggest that animal models and pharmacotherapy have diverse effects on social deficits-like symptoms of schizophrenia. Perinatal and adult pharmacological models may be the most promising for this aspect of schizophrenia. Further research for novel treatments should compare their effects to the well-known atypical antipsychotic drugs. Moreover, there is a significant need for clinical research that will specifically examine the effect of pharmacological treatments on asocial negative symptoms of schizophrenia.

Funding

No funding was provided for this research. R.H. received a Presidential Doctoral Fellowship, from Bar-Ilan University.