Peripheral oxytocin and vasopressin: biomarkers of psychiatric disorders? A comprehensive systematic review and preliminary meta-analysis

A large array of studies has investigated peripheral oxytocin (OT) and vasopressin (ADH) as potential biomarkers of psychiatric disorders, with highly conflicting and heterogenous findings. We searched Web of KnowledgeSM and Scopus® for English original articles investigating OT and/or ADH levels in different biological fluids (plasma/serum, saliva, urine and cerebrospinal fluid) across several psychiatric disorders. Sixty-four studies were included. We conducted 19 preliminary meta-analyses addressing OT alterations in plasma/serum, saliva, urine and cerebrospinal fluid of 7 psychiatric disorders and ADH alterations in plasma/serum, saliva, urine and cerebrospinal fluid of 6 psychiatric disorders compared to controls. Hedge’s g was used as effect size measure, together with heterogeneity analyses, test of publication biases and quality control. None of them (except serum OT in anorexia nervosa) revealed significant differences. There is no convincing evidence that peripheral ADH or OT might be reliable biomarkers in psychiatric disorders. However, the lack of significant results was associated with high methodological heterogeneity, low quality of the studies, small sample size , and scarce reliability of the methods used in previous studies, which need to be validated and standardized. between studies heterogeneity non-significant), (n=2, overall sample=86, Hedges' g=-0.219, 95% CI from -0.755 to 0.318, p unc=0.425, between studies heterogeneity non-significant), MDD (n=2, overall sample=75, Hedges' g=-0.333, 95% CI from -0.781 to 0.114, p unc=0.144, between studies heterogeneity non-significant), Psychosis (n=3, overall sample=147, Hedges' g=0.287, 95% CI from -0.330 to 0.903, p unc=0.362, substantial between studies heterogeneity Q=6.482, I 2 p=0.039), and OCD (n=2, overall sample=99, Hedges' g=0.599, 95% CI from 0.194 to 1.004, p unc=0.004, p corr=ns, between studies heterogeneity non-significant).


INTRODUCTION
Oxytocin (hereafter OT) and vasopressin (hereafter ADH) are neuropeptides mainly synthesized in the brain's hypothalamic paraventricular (PVN) and supraoptic (SON) nuclei (Ludwig and Leng, 2006). They are released in systemic circulation through the posterior pituitary gland, where they act as hormones regulating a range of physiological functions (Gimpl and Fahrenholz, 2001;Leng et al., 2015). They are also released in the central nervous system, acting on multiple brain regions as neuromodulators and influencing a range of neurophysiological processes and behaviours (Stoop, 2012), including feeding, anxiety, aggression, social recognition and the stress/fear response to social stimuli (Hashimoto et al., 2012).
Evidence from animal studies has demonstrated the significant role that OT and ADH play in the regulation of social behaviour and cognition (Chang and Platt, 2014). An increasing number of studies have also began to dissect the roles of OT and ADH in human social behaviour (Heinrichs et al., 2009). These neuropeptides are associated with complex social and emotional processing in healthy people which if impaired may account for some of the symptoms present in psychiatric disorders (Meyer-Lindenberg et al., 2011). Furthermore, there is also growing interest in the potential for synthetic neuropeptides in treatment of psychosis (for a comprehensive review see (Gumley et al., 2014)), autism spectrum disorders (ASD) (Thompson et al., 2006;Uzunova et al., 2015), and affective and anxiety disorders (Griebel et al., 2012).
In animals there are multiple methods that allow to reliably either assess or manipulate central OT and ADH levels and their effects on behaviour (e.g. intracerebral microdialysis (Veenema and Neumann, 2008), targeted delivery of neuropeptide agonists or antagonists (Song et al., 2014), gene knockout (Wersinger et al., 2002), and viral gene transfer (Pagani et al., 2014)). However, these are not available in humans, hence researchers have turned to peripheral assays as proxy measures.
Specifically, plasma/serum (Rubin et al., 2014), saliva (Fujisawa et al., 2014), urine (Hoffman et al., 2012 or cerebrospinal fluid (CSF) (Sasayama et al., 2012) OT and ADH levels have been recently tested as putative biomarkers in ASD (Alabdali et al., 2014;Boso et al., 2007;Modahl et al., 1998), Psychosis (Elman et al., 2003;Goldman et al., 2008;Rubin et al., 2013;Walss-Bass et al., 2013), bipolar disorder (BD) (Rubin et al., 2014;Turan et al., 2013), major depressive disorder (MDD) (Goldstein et al., 2000;Ozsoy et al., 2009;Yuen et al., 2014), as well as in anxiety (Hoge et al., 2012;Hoge et al., 2008), personality (Bertsch et al., 2013) and eating disorders (anorexia nervosa, AN and bulimia nervosa, BN) (Frank et al., 2000;Lawson et al., 2011;Lawson et al., 2012), with highly heterogeneous and conflicting results (Al-Ayadhi, 2005;Alabdali et al., 2014;Emsley et al., 1989;Watson et al., 2007). The first aim of the present systematic review and preliminary metaanalysis was to test if the levels of these neuropeptides across different clinical samples and different biological fluids (plasma/serum, cerebrospinal fluid (CSF), urine and saliva) were consistently altered and could therefore be considered as potential reliable biomarkers for psychiatric disorders. The second aim was to investigate and address moderators and confounding factors impacting the preliminary meta-analytical estimates. The second step involved the implementation of an electronic manual search of the reference lists of the retrieved articles. Articles identified through these two steps were then screened on basis of title or abstract reading. The articles surviving selection were fully downloaded (PDFs) and assessed for eligibility on the basis of full-text reading. Discrepancies were resolved through consensus with a third researcher [MR]. To achieve high quality of reporting we adopted MOOSE guidelines (Stroup et al., 2000) (see supplementary materials).

Inclusion criteria
Articles meeting the inclusion criteria for the current systematic review and preliminary metaanalysis: (a) were original articles, written in English; (b) included subjects with a psychiatric diagnosis defined according to international standard definitions (ICD, DSM); c) included a healthy comparison group; d) reported sufficient data on peripheral (plasma, serum, urine, cerebrospinal fluid, saliva) OT or ADH level differences between groups (See supplementary materials for details).

Exclusion criteria
We excluded (a) abstracts, pilot datasets, reviews, articles in language other than English; (b) articles failing to report enough data to perform a meta-analysis (we also contacted the authors to obtain the missing data); (c) articles with overlapping datasets. In case of multiple publications deriving from the same study population, we selected the articles reporting 1) the largest or 2) the most recent data set. In case of conflict between these two criteria, the sample size was prioritised.

Recorded variables
We recorded the following variables from each article: author, year of publication, quality of reporting criterion (adapted Newcastle-Ottawa Scale http://www.ohri.ca/programs/clinical_epidemiology/oxford.asp, see supplementary material for detail), comparison group type (healthy comparisons), epidemiological data of patient and control samples (baseline sample sizes, mean age, proportion of females) and methodological information (specimen type, measurement method (ELISA, RIA), plasma extraction (yes/no)).
For the peripheral level of OT and ADH measured in plasma/serum, saliva, urine and CSF in patients and in controls, we extracted: mean value, standard deviation (SD), and standard error of the mean (SEM) (when presented instead of SD). We excluded data reporting OT or ADH peripheral levels following any drug administration.

Effect size
The meta-analyses were performed using Comprehensive Meta-Analysis Software version 2 (Borenstein M, 2005). We did not perform analyses on diagnoses where there was only one study available. As a measure of effect size, Hedges' g was adopted. Results were Bonferroni-corrected for multiple testing, by dividing 0.05 by the overall number of meta-analyses conducted.

Heterogeneity, publication biases, sensitivity analysis
Heterogeneity among study point estimates was assessed using Q statistics (Paulson and Bazemore, 2010) and the proportion of the total variability in the effect size estimates evaluated with the I 2 index (Lipsey and Wilson, 2000). As meta-analysis of observational studies is supposed to be characterized by significant heterogeneity, random effect models were used (Cooper et al., 2009).
Because of numerical constraints, meta-regressions were not conducted. Publication biases were tested with the Duval and Tweedie "trim and fill" method (Duval and Tweedie, 2000), when at least 3 studies were included. To further assess the robustness of the results, we performed sensitivity analyses by sequentially removing each study and re-running the analysis, when at least 5 studies were included (Higgins and Green, 2011). We conducted other sensitivity analyses to investigate the putative influence of several categorical moderators: assay technique, pregnancy, alcohol use and phase of the psychotic disorder. In particular, we performed sensitivity analyses to clarify the impact of peptide extraction techniques on the results. Because of numerical constraints, this was possible only in two of the subgroups (assessing plasma/serum ADH levels in MDD and psychosis).
We performed an additional sub-group analysis dividing First Episode Psychosis (FEP) and chronic Psychosis vs healthy comparisons (see supplementary materials for details). The potential influence of methodological weaknesses influencing preliminary meta-analytic estimates, according to the quality rating, was discussed.

3.1.Database
After manual and electronic searches, 192 potential articles (PDFs) were screened to assess eligibility (see PRISMA flow chart eFigure1 and MOOSE checklist eTable1). 52 articles were considered eligible according to our inclusion criteria. 12 of these contributed two individual samples each: six investigating both OT and ADH in the same psychiatric disorder; four investigating the same neuropeptide across two different psychiatric disorders; two conducting the analyses on both plasma and CSF. Three articles delivered four individual samples each: two by measuring OT and ADH in two distinct diagnostic groups, and one measuring ADH both in plasma and CSF in two psychiatric disorders.

OT levels in CSF
Meta-analyses of OT levels in CSF were conducted in AN, BN, MDD, OCD and Psychosis. We found no differences between healthy comparisons and AN (n=2, overall sample=52, Hedges' g=-

OT levels in plasma/serum
Meta-analyses of OT levels in plasma were conducted in ASD, BD, Psychosis and AN.

OT levels in saliva
Meta-analyses of OT levels in saliva were conducted in ADS. OT levels in saliva of ASD were not statistically different from those observed in healthy controls (n=2, overall sample=152, Hedges' g=-0.354, 95% CI from -0.701 to -0.006, p unc=0.046, p corr= ns, between studies heterogeneity non significant).

OT levels in urine
Meta-analyses were not performed as insufficient data were retrieved.
Overall effect sizes are depicted in Figure 1a, forest plots are reported in supplementary materials (eFigure 2 and 3) and eTable 6 report heterogeneity summary.

Publication biases and sensitivity analyses
The Duval and Tweedie's method showed that the results were not affected by publication biases.
Sensitivity analysis removing one study per computation when at least 5 studies were included confirmed the robustness of the results (Higgins and Green, 2011).

ADH levels in plasma/serum
Meta-analyses of ADH levels in plasma were conducted in ASD, BD, AN, MDD and Psychosis.
There were no differences between healthy comparisons and patients diagnosed with ASD (n=3, considerable between studies heterogeneity Q=55.984, I 2 =82.14, p<0.001).

ADH levels in saliva and urine
Meta-analyses were not performed as insufficient data were retrieved.
Overall effect sizes are depicted in Figure 1b, forest plots are reported in supplementary materials (eFigure 4 and 5) and eTable 6 report heterogeneity summary.

3.4.Quality assessment
Table 2 describes the quality of the studies. Study quality scores ranged from 1 to 7 (median score was 4 out of 9 possible points). There were no significant differences in Newcastle-Ottawa Scale (NOS) scores between the subgroups (F=0.503, df=18, p=0.943). Unfortunately, since the number of studies included in each meta-analysis was too small to perform meta-regressions, it was not possible to directly assess the effect of NOS scores on the findings of our preliminary metaanalyses.

DISCUSSION
This is the first systematic review and preliminary meta-analysis to comprehensively summarize the available evidence of altered peripheral levels of OT and ADH in psychiatric disorders and investigate whether there are reliable differences which could serve as biomarkers between patients and healthy controls. We found no robust and convincing evidence for significant alterations in the two neuropeptides in psychiatric disorders.
The lack of significant association was secondary to high heterogeneity across individual studies, low quality and significant methodological limitations. Indeed, our results do not contradict the potential role of the two neuropeptides in mental disorders per se but do clearly suggest that peripheral levels do not reliably distinguish cases from controls. The putative reasons for this will be discussed in the following sections.

Reliability of assays
Both the commercially available immunoassays (RIA and EIA) have high heterogeneity and

Recommendations for methodological improvement
It is of crucial importance that future research in this field aim at method validation and standardization before the results of studies exploring the role of neuropeptides in physiology and disease can be considered meaningful.
The sample obtained by EIA, even following the assay manufacturers' recommendations to extract samples, contain multiple non-OT immunoreactive molecules, still to identify. It is therefore pivotal to clarify whether those are reflecting the turnover of bioactive OT, or are non-related reactants.
Moreover, assays with high degree of sensitivity and specificity, such as two-dimensional liquid chromatography separation with tandem mass spectrometry detection, could be use as reference for validation of techniques as RIA and EIA and to confirm the presence of OT and ADH in other biological fluids, such as saliva and urine, where less knowledge is available.
Furthermore, pre-analytical errors should be avoided and systematic evaluation and standardization of pre-analytical procedures should be conducted. In particular, consideration should be payed to variations in the levels of the hormones related to circadian patterns and other physiological conditions (e.g., fasting or post-prandial state, sexual activity, menstrual cycle, pregnancy), time of day and environmental features during collection, tube additives (e.g., EDTA and citrate), use of protease inhibitors and storage conditions.

Peripheral vs Central levels
Our results provide meta-analytical confirmation for the concern, increasingly expressed by many authors, about the lack of neurophysiological evidence of a direct relationship between peripheral levels of OT and ADH and their central release and activity (Grinevich et al., 2015;Kagerbauer et al., 2013;Leng and Ludwig, 2015;Ludwig and Leng, 2006;McCullough et al., 2013;Neumann, 2007). Originally, ADH and OT have been described as hormones, involved respectively in antidiuresis, osmolality regulation, direct vasoconstriction and hepatic glycogenolysis (Appelgren, 1982;Robertson, 2001) and labor (Husslein, 1984), lactation (Crowley, 2015), and sexual activity (Carmichael et al., 1987). Under physiological conditions, the primary drive for ADH peripheral secretion is plasma osmolality. It is unlikely that this homeostatic constraint can be chronically overridden by erratic triggers, such as socioemotional stimuli (Kagerbauer et al., 2013). On the other hand plasma OT levels are stable, except in pregnancy (Leng and Ludwig, 2015), but may be regulated independently from central OT as suggested by converging evidence from animal studies (Ludwig et al., 1994;Moos et al., 1989). In humans no correlation was found between central and The utility of CSF assays is more controversial. In fact, although the extracellular fluid of the brain interconnects freely with the CSF, the contribution of centrally released neuropeptides to CSF levels depends on several factors, as they eventually reach the CSF after having induced central effects by binding to specific receptors and having been possibly degraded by enzymes. It is plausible that once diffused in the CSF, the neuropeptides may have lost their biological activity, since they are unlikely to pass back into the brain parenchyma against the concentration gradient to act on their receptors. Furthermore, the blood-brain barrier prevents ADH and OT to reflow into the central compartment (i.e. brain parenchyma and CSF) once they have reached the peripheral one (i.e., blood and peripheral organs) (Kagerbauer et al., 2013;Mens et al., 1983). Further, peptides appear

Psychosis
Overall, we found no convincing evidence that either plasmatic or CSF ADH and OT levels are altered in Psychosis. We also investigated whether the peripheral ADH impairments in Psychosis were associated with the illness phase. It has been argued that psychotic exacerbations may be concurrent with a transient reset of the osmostasis for ADH secretion, elevated plasma levels of the neuropeptide and eventually water imbalance, as an epiphenomenon of dysfunctions in hypothalamic pituitary adrenal (HPA) axis response to psychological stimuli(Goldman, 2009).
However, we found no meta-analytical evidence for peripheral ADH differences between first episode and chronic Psychosis.

ASD
We did not find evidence that peripheral ADH and OT levels are impaired in ASD, in contrast with converging evidence from animal models of ASD and human studies which implicate alterations in the function of the OT system within ASD. The results align with concerns that these measures are not reliable proxies of the central release and activity of these neuropeptides. In preclinical animal models of neurodevelopmental disorders, autistic-like symptoms were shown to be accounted by alterations of trafficking and/or release of OT from axonal terminals and to be rescued after OT administration (Grinevich et al., 2015). Clinical trials with exogenous OT, administered either via intravenous infusion or intranasal spray, have produced contradictory results, ranging from positive outcomes in a variety of ASD psychopathology domains, namely emotion recognition (Guastella et al., 2010a), social cognition (Andari et al., 2010, and core autistic repetitive behaviours (e.g. ordering, compulsion to tell/ask and touching) (Hollander et al., 2003), to modest (Anagnostou et al., 2012) or null clinical improvement (Anagnostou et al., 2014). Although ADH is less commercially available, the few reports to date show some promise that ADH may exert genderspecific positive effects on processes related to empathic ability (Thompson et al., 2006).
Nevertheless the findings in human studies should be interpreted cautiously and confirmed in larger trials.

BD and MDD
Plasmatic and CSF neuropeptide levels were not significantly altered in affective disorders (BD and MDD) compared to matched healthy controls.
Again these results indicate that peripheral OT and ADH levels are not reliable biomarkers of BD/MDD. In contrast to the null findings shown here, several preclinical and clinical data provide evidence of the association between ADH and affective disorders (BD and MDD). ADH is involved in the regulation of the hypothalamus-pituitary-adrenal (HPA) axis, which has been found to be overactive in BD (Cervantes et al., 2001;Daban et al., 2005) and MDD (Carroll et al., 1981;Dinan, 1994;Holsboer, 2000). Post-mortem brain tissues of patients with MDD and BD contain a greater number of ADH-immunoreactive neurons in the hypothalamic PVN (Merali et al., 2006).
Therefore, antagonists of V1b receptor have been developed and tested in rodents as potential new strategies for the treatment of affective disorders, whose efficacy in humans does still need clarification (Griebel et al., 2012).

OCD, AN and BN
Our meta-analysis did not detect significant alterations of ADH and OT in OCD and BN. A difference in serum OT levels could be observed between patients suffering from AN and healthy comparisons. The reliability of peripheral ADH and OT levels as biomarkers of AN is weakened by the global hypothalamic-pituitary dysfunction, the impact of fluid and sodium restriction and purging behaviours observed in AN and the putative role of the two hormones in the regulation of the gut brain axis and energy homeostasis (Blevins and Ho, 2013). In fact, both neuropeptide levels end to be restored after body weight recovery (Gold et al., 1983). Therefore, once again our findings should be interpreted considering that peripheral levels of OT and ADH are unreliable proxy measures of their central activity.

4.8.Limitations
There are some significant limitations to our study. The subgroups were very small in sample sizes.However, as recommended in the Cochrane Handbook for Systematic Reviews of taking lithium salts (eTable3). Despite its therapeutic value, lithium exerts adverse effects on the kidney, leading to nephrogenic diabetes insipidus (NDI) in up to 10% of the patients on long term treatment (Bendz and Aurell, 1999;Bendz et al., 1996), which may be expected to be associated with a compensatory increase in ADH release (Watson et al., 2007). As regard OT, preclinical studies in rats found that a systemic injection of lithium leads to a significant elevation of peripheral OT levels (Cui et al., 2001;You et al., 2001). Furthermore, 42 studies out of 62 (68%) failed to control for alcohol intake (not applicable to 2 studies which had recruited only children), which was reported to induce degeneration in numerous neurons in the hypothalamic magnocellular system in rats (Silva et al., 2002). The majority of studies did not ascertain the pregnancy state of female participants (70%, 35 out of 50, condition not applicable to 14 studies because of the characteristics of the sample, see eTable3 for further details). During pregnancy dramatic structural and functional changes occur in the OT system of the mother's brain (Hillerer et al., 2014;Rocchetti et al., 2014).
Moreover, high levels of oxytocinase secreted by the placenta might further contribute to the final serum/plasma OT levels (Nomura et al., 2005). Finally, only 18% of the studies investigating OT and none of those investigating ADH analyzed the fluctuations of OT levels across the menstrual cycle phases. Studies investigating plasma OT and ADH levels during the menstrual cycle in different species, including humans, found that these fluctuate across the ovarian cycle (Wathes and Swann, 1982). (Spruce et al., 1985). As a consequence, our meta-analytical estimates are affected by high uncertainty and should be considered as a preliminary synthesis of the available literature.

4.9.Conclusions
The current meta-analysis does not provide convincing evidence that peripheral OT or ADH levels are altered in psychiatric disorders as compared to healthy controls. These findings are characterized by high heterogeneity, various methodological limitations, and poor quality. For more meaningful results to be obtained, methods need to be validated and standardized. Future studies are requested to better address with robust methods and adequate sample sizes the actual role of ADH and OT as biomarkers of mental illnesses.