Clinical Value of Inflammatory and Neurotrophic Biomarkers in Bipolar Disorder: A Systematic Review and Meta-Analysis

Bipolar disorder (BD) is a multifactorial chronic psychiatric disease highly defined by genetic, clinical, environmental and social risk factors. The present systematic review and meta-analysis aimed to examine the relationship between inflammatory and neurotrophic factors and clinical, social and environmental factors involved in the development and the characterization of BD. Web of Science, PubMed, PsycINFO, Scopus and Science Direct were searched by two independent reviewers. The systematic review was registered in PROSPERO (CRD42020180626). A total of 51 studies with 4547 patients with a diagnosis of BD were selected for systematic review. Among them, 18 articles were included for meta-analysis. The study found some evidence of associations between BDNF and/or inflammatory factors and different stressors and functional and cognitive impairment, but limitations prevented firm conclusions. The main finding of the meta-analysis was a negative correlation between circulating levels of BDNF and depression severity score (standardized mean difference = −0.22, Confidence Interval 95% = −0.38, −0.05, p = 0.01). Evidence indicates that BDNF has a role in the depressive component of BD. However, the poor consistency found for other inflammatory mediators clearly indicates that highly controlled studies are needed to identity precise biomarkers of this disorder.


Introduction
Bipolar affective disorder (BD) is a relatively common but serious mental illness, with a lifetime prevalence of 0.6% for bipolar disorder type I (BD-I), 0.4% for bipolar disorder type II (BD-II), 1.4% for subthreshold bipolar disorder (SBD), and 2.4% for bipolar disorder spectrum (BDS) [1]. It is one of the leading causes of disability worldwide [2], presenting high rates of psychiatric and medical morbidity and mortality [3][4][5][6], due to the co-occurrence with medical conditions (such as cardiovascular disorders, diabetes, and obesity), and psychiatric disorders (e.g., anxiety, personality disorders, attentiondeficit/hyperactivity disorder (ADHD), substance use, and suicide) [7][8][9]. Furthermore, an early age of onset is associated with a more chronic, severe, and recurrent course of the disease [10]. In terms of gender, BD-I is equally distributed in men and women, whereas

Search Strategy
This systematic review and meta-analysis were carried out following the recommendations of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement [77]. The protocol was registered in the International Prospective Register of Systematic Reviews (PROSPERO CRD42020180626). An initial review was conducted to identify biomarkers involved in bipolar disorder to generate a list of search terms.
Bibliographic searches were performed in the electronic databases Web of Science, PubMed, PsycINFO, Scopus and Sciencedirect from their inception to May 2020. The database searches were complemented with manual screening of the reference list of the included studies.
Two independent searches were achieved on each database (conducted by AV-N and CG-S-L). Detailed information about the search strategy can be found in the Annex S1 of the Supplementary Materials.
The inclusion criteria applied were the following: (a) type of study: clinical trial, cohort and case-control studies; (b) age: adults subjects (patients over 18 years old); (c) diagnosis: any bipolar disorder (BD-I, BD-II, Cyclothymic Disorder and not otherwise specified BD (NOS-BD)) in any of its phases (mania, depression, mixed state or euthymia) diagnosed according to international classification systems, including clinical interviews that applied the criteria of the DSM (Diagnostic and Statistical Manual of Mental Disorders) or the ICD (International Classification of Diseases); (d) studies that include exposure to clinical, social or environmental factors; (e) studies investigating some type of inflammatory factors (biomarkers indicative of a biological state in the person, in this case the degree of brain inflammation), neurotrophic factors (or neurotrophins, a family of proteins that promote the survival, growth and differentiation of neurons; among them are nerve growth factor (NGF), the insulin-like growth factor (IGF-1), the BDNF and neurotrophin-3 (NT-3)), and cytokines (a set of IL that regulate the immune and inflammatory response) in patients with bipolar disorder; (f) language: English or Spanish only.
The following exclusion criteria were applied: (a) narrative reviews, qualitative studies and case studies; (b) children and adolescents (under 18 years old); (c) subjects diagnosed with another mood disorder (e.g., major depressive disorder, dysthymic disorder, unspecified depressive disorder); (d) subjects diagnosed with any other psychiatric disorder; (e) subjects with a diagnosis of neurological disorder or genetic syndromes; (f) studies that not include exposure to clinical, social or environmental factors related to BD; (g) studies in which the only exposure factor was a pharmacological treatment or psychological intervention; (h) studies that have not investigated neurotrophic factors and inflammatory cytokines; (i) studies with duplicated or unoriginal data.

Study Selection
Title and abstract were screened independently by two reviewers (AV-N and CG-S-L). If the presence of the biomarkers, risk factors or study design could not be ascertained from title and abstract screening, full texts of publications selected were then reviewed by the same investigators to determine final eligibility. Team meetings were held to discuss and resolve any discrepancies and reach a consensus on all inclusion decisions with a third reviewer (JG-P). If the data needed to include the study in the meta-analysis were not available, authors were contacted up to two times (two weeks apart). If no response was received, these studies were not included in the meta-analysis but were discussed narratively.

Quantitative Analysis
A meta-analysis was performed for all factors reported on by at least three studies. Mean and standard deviation for each biomarker, separated into mood phase or subtype of bipolar, were entered into an electronic database and were analyzed with a quantitative meta-analytical approach using R 3.6.3 and Rstudio. The mean differences and 95% confidence intervals between cases and controls were calculated and displayed in forest plots. Correlations between severity scales and circulating levels of biomarkers were transformed to Fisher's Z-score to obtain accurate weight for each study. Montgomery-Asberg Depression Rating Scale (MADRS) scores were translated to Hamilton Depression Rating Scale (HDRS) scores using the equipercentile linking method developed by Leucht and colleagues [78]. The inverse variance method and Random-effects models by Hartung-Knapp-Sidik-Jonkman were used. This method produces more robust estimates and is preferred when the number of studies is small and when there is substantial heterogeneity, because DerSimonian-Laird is prone to produce false positives under this condition [79,80]. Heterogeneity was determined by means of Q statistic tests and the I2 index. Publication bias was evaluated by use of funnel plots in each meta-analysis including ten or more studies, as is recommended for interpretation [81]. Studies included in the meta-analysis were appraised using the Risk of Bias in a Non-randomised Studies (ROBINS-I) tool [82].

Results
A total of 4396 studies were retrieved from the databases. Finally, 51 articles (4 of which had repeated samples with other biomarkers) that analyzed 33 inflammatory mediators were eligible for review and 18 were included in the meta-analysis (Table S2). More detailed information is shown in the PRISMA flow chart (Figure 1). Risk of bias assessment of the studies included in the meta-analysis is summarized in a traffic light plot ( Figure S1) and the general risk of bias of all studies included in the meta-analysis is shown in a summary plot ( Figure S2). The overall risk of bias of the studies included was between moderate and serious. Among the included studies, the most commonly reported outcomes were BDNF and mood state, BDNF and bipolar subtype, correlation between BDNF and severity of mania or depression, TNF-α, and mood state, correlation between TNF-α and severity of mania, correlation between IL-6 and severity of mania or depression. We selected these outcomes and performed 9 meta-analyses. [79,80]. Heterogeneity was determined by means of Q statistic tests and the I2 index. Publication bias was evaluated by use of funnel plots in each meta-analysis including ten or more studies, as is recommended for interpretation [81]. Studies included in the metaanalysis were appraised using the Risk of Bias in a Non-randomised Studies (ROBINS-I) tool [82].

Results
A total of 4396 studies were retrieved from the databases. Finally, 51 articles (4 of which had repeated samples with other biomarkers) that analyzed 33 inflammatory mediators were eligible for review and 18 were included in the meta-analysis (Table S2). More detailed information is shown in the PRISMA flow chart (Figure 1). Risk of bias assessment of the studies included in the meta-analysis is summarized in a traffic light plot ( Figure S1) and the general risk of bias of all studies included in the meta-analysis is shown in a summary plot ( Figure S2). The overall risk of bias of the studies included was between moderate and serious. Among the included studies, the most commonly reported outcomes were BDNF and mood state, BDNF and bipolar subtype, correlation between BDNF and severity of mania or depression, TNF-α, and mood state, correlation between TNF-α and severity of mania, correlation between IL-6 and severity of mania or depression. We selected these outcomes and performed 9 meta-analyses. Following exclusion criteria, a total of 51 studies were selected for systematic review. Among them, 18 articles were included for meta-analysis.
The characteristics and main results of the 51 studies included in the systematic review are summarized in Table 1. The studies were conducted across several countries. The design of the studies was mostly cross-sectional, although clinical trials, post-hoc analysis of a clinical trial, longitudinal studies, retrospective and prospective studies, a historical cohort study, and a naturalistic cohort study were also included. More information on the studies can be found in the Supplementary Materials (Table S1). Following exclusion criteria, a total of 51 studies were selected for systematic review. Among them, 18 articles were included for meta-analysis.
The characteristics and main results of the 51 studies included in the systematic review are summarized in Table 1. The studies were conducted across several countries. The design of the studies was mostly cross-sectional, although clinical trials, post-hoc analysis of a clinical trial, longitudinal studies, retrospective and prospective studies, a historical cohort study, and a naturalistic cohort study were also included. More information on the studies can be found in the Supplementary Materials (Table S1).
Levels of IL-1RA, IL-6 and TNF-α were higher in chronic than early-stage patients. No differences in IL-1RA, IL-6 or TNF-α levels between chronic and early-stage patients in terms of mood state (euthymic, depressive and manic). Correlation between TNF-α levels and MADRS scores in chronic patients. Correlation between TNF-α levels and CGI scores in early-stage patients. No correlation between IL-1RA, IL-6 or TNF-α levels and YMRS nor PANNS scores among chronic and early-stage patients. No significant correlations between biomarkers levels and length of illness and the presence of psychotic symptoms. The changes in IL-6 and IL-1β levels were significantly associated with the changes in HDRS scores (before and after pharmacological treatment). The changes in TNF-α levels were significantly associated with the changes in YMRS scores (before and after pharmacological treatment). The changes in IL-8 levels were not associated with the changes in HDRS scores, nor with the changes in YMRS scores (before and after pharmacological treatment).             Table S3 for abbreviations).  Table 1), included in the present systematic review, evaluated the association between BD subtypes I and II and selected biomarkers, 7 studied BDNF and 9 measured inflammatory biomarkers. Only one article found differences in BDNF levels between BD-I and BD-II subtypes (40 in Table 1), in which patients with the BD-I subtype had significantly lower BDNF levels than patients with the BD-II subtype. Only one article found differences in the circulating levels of inflammatory mediators (sIL-2R and sTNF-R1), which were higher in BD-I patients than in BD-II patients (1 in Table 1). There were two more articles that compared the subtypes of BD with the SBD group (subthreshold bipolar disorder: more than 2-days but less than 4-days duration of hypomania) and found differences in the levels of inflammatory mediators (IL-1ß, IL-6, IL-8) and BDNF (45, 46 in Table 1).
In the present study, we performed a meta-analysis of BNDF between BD subtypes in three selected studies (21, 40, 45 in Table 1). These studies highlighted lower plasma BDNF levels in BDI (40 in Table 1) and SBD (45 in Table 1) patients, and in BD women with higher MADRS scores (21 in Table 1). In the meta-analysis, we found that circulating BDNF levels did not differ between BD subtypes I and II (standardized mean difference  Table S3 for abbreviations).

Subtypes of the Disorder
A total of 11 articles (1, 17, 21, 25, 31, 32, 39a, 39b, 40, 44a, 44b in Table 1), included in the present systematic review, evaluated the association between BD subtypes I and II and selected biomarkers, 7 studied BDNF and 9 measured inflammatory biomarkers. Only one article found differences in BDNF levels between BD-I and BD-II subtypes (40 in Table 1), in which patients with the BD-I subtype had significantly lower BDNF levels than patients with the BD-II subtype. Only one article found differences in the circulating levels of inflammatory mediators (sIL-2R and sTNF-R1), which were higher in BD-I patients than in BD-II patients (1 in Table 1). There were two more articles that compared the subtypes of BD with the SBD group (subthreshold bipolar disorder: more than 2-days but less than 4days duration of hypomania) and found differences in the levels of inflammatory mediators (IL-1ß, IL-6, IL-8) and BDNF (45, 46 in Table 1).
In the present study, we performed a meta-analysis of BNDF between BD subtypes in three selected studies (21, 40, 45 in Table 1). These studies highlighted lower plasma BDNF levels in BDI (40 in Table 1) and SBD (45 in Table 1) patients, and in BD women with higher MADRS scores (21 in Table 1). In the meta-analysis, we found that circulating BDNF levels did not differ between BD subtypes I and II (standardized mean difference  Forest plot of the meta-analysis of BDNF in patients with BD1 subtype compared to those with BD2 subtype in three selected studies (21, 40, 45 in Table 1).

Phases of the Disorder
A total of 16 articles (1, 2, 3, 5, 8, 18, 19, 32, 33a, 33b, 34, 39a, 42, 44a, 44b, 47 in Table  1), included in the present systematic review, studied the association between biomarkers and different mood phases in BD patients, 8 of them assessed inflammatory factors, 5 assessed BDNF and other trophic factors, and 3 both biomarkers. Among the studies in which inflammatory mediators were measured, 5 of them found significant differences Figure 2. Forest plot of the meta-analysis of BDNF in patients with BD1 subtype compared to those with BD2 subtype in three selected studies (21, 40, 45 in Table 1).
Age at onset was assessed in five articles (31,35,39a,42,44a in Table 1). Two of them found an association between age of onset and inflammatory (sVCAM-1) and trophic (GDNF) factors (35, 42 in Table 1). No relationship was found between biomarkers and the number of hospitalizations in any study (3, 44a, 44b in Table 1). No interaction was also found between BDNF levels and chronicity and time [7]. Levels of sVCAM-1 correlated negatively with the bipolarity index (BPx) in the acute and remission phases (35 in Table 1). Another inflammatory mediator (TNF-α) was associated with the duration of untreated BD only in one study (35 in Table 1). Among the 3 studies (15, 19, 41 in Table 1), in which they classify BD patients into early or late stage, all measured inflammatory mediators, 2 of them evaluated BDNF and the other analyzed additional trophic factors. These three studies found significant differences in circulating levels of inflammatory mediators (IL-6, IL-10, IL-1RA, TNF-α) when comparing early-and late-stage patient groups, but not when analyzing circulating levels of trophic factors.

Number of Previous Episodes, Duration of Episodes and Polarity
Regarding the number of previous episodes, the results are heterogeneous. Some studies evaluated the number of lifetime mood episodes, but no association with biomarkers was found (39a, 44b in Table 1). Only one article associated the number of depressive episodes with inflammatory factors (IL-6) (37 in Table 1), but not with BDNF (31 in Table 1) or with the presence of previous episodes of depression (4 in Table 1). In one study, a longer duration of mood episodes (mixed, manic, depressive, euthymic) was associated with inflammatory mediators (IL-6, IL-18) in comparison with episodes of less than one-week duration (33b in Table 1). However, no association was found between mood duration and BDNF or other trophic factors (GDNF) (42 in Table 1). Another study found an association between the duration of current mood state and inflammatory and trophic factors (TNF-α, BDNF, VEGF and sFlt-1) (44a in Table 1), but this result was not confirmed (44b in Table 1). The duration of depressive episodes was associated with inflammatory mediators (TNF-α) in only one study (35 in Table 1).
No associations were found between any biomarker and rapid cycling bipolar patients or atypical depression (20, 39a in Table 1). Furthermore, no differences in circulating levels of inflammatory mediators (sTNF-R1, sIL-6R) were found between patients with subsyndromal BD and without subsyndromal symptoms (9 in Table 1). Similarly, no differences in mean levels of sICAM-1, sVCAM-1, IL-6 and TNF-α were observed when comparing patients in an acute phase and patients in a remission phase of the bipolar disorder (35 in Table 1). Regarding polarity, BD patients with manic polarity had significantly lower levels of sVCAM-1 than BD patients with depressive polarity (35 in Table 1). No differences were found in terms of acute episode polarity and levels of sICAM-1, IL-6 or TNF-α (35 in Table 1).

Cognition and Functionality
Four articles evaluated general cognitive abilities in BD patients, including cognitive and executive functioning, 1 with inflammatory mediators, 1 with BDNF and 2 with both types of factors. Among these reports, significant associations were found between cognitive abilities and inflammatory mediators (IL-1RA, sCD4DL, TNF-α) (4, 16 in Table 1) and BDNF (24, 32 in Table 1). On the other hand, two studies evaluated the degree of functionality. In one, functional impairment was associated with inflammatory mediators (IL-6, IL-10) (30b in Table 1), and in the other was reported a significant interaction between psychosocial functioning and time on BDNF levels (7 in Table 1). In one study, BDNF was not associated with quality of life (24 in Table 1). Finally, a positive correlation between BDNF levels and Global Assessment of Functioning (GAF) scores was found in one study (42 in Table 1), but not when the trophic factor GDNF was analyzed.

Other Clinical Parameters and Comorbidity
Nine articles evaluated age in BD patients, which was associated with inflammatory mediators (IL-6, IL-8, IL-18, IL-18BP, TNF-α, sTNFR1, sTNFR2) in 3 of these studies (10, 17, 44b in Table 1). Only two of these articles included sex as a study variable, but it was not correlated with any biomarker (17,28 in Table 1).
Only one article found an association between the number of previous psychotic episodes and inflammatory mediators (sVCAM-1) (35), and others did not find any association between biomarkers and lifetime features of psychosis (4, 17, 44a, 44b in Table 1). The presence of psychotic symptoms was not associated with inflammatory mediators or BDNF in any study (19, 25, 39a, 44a, 44b in Table 1).
Psychiatric comorbidities were generally not associated with inflammatory mediators or BDNF (2, 3, 4 in Table 1), with the exception of one study (46 in Table 1) that found a relationship with TGF-ß1. Melancholia was associated with inflammatory mediators (sTNFR80, IL-1α) in the two selected studied (39a, 39b in Table 1). Comorbid anxiety symptoms were associated with inflammatory mediators (IL-6, IL-10, TNF-α) in one study (14 in Table 1), but another study found no association between comorbid anxiety syndrome and inflammatory mediators (17 in Table 1). Post-traumatic stress disorder was not associated with the biomarkers assessed (20). The history of suicide attempts was not associated with BDNF (20 in Table 1), nor was suicidal ideation with inflammatory factors (39a in Table 1). ADHD was not associated with the biomarkers evaluated (17 in Table 1). Tobacco use, substance abuse or dependence, and alcohol abuse or dependence, were not associated with any biomarker in the studies evaluated (3, 4, 17, 20, 25, 28 in Table 1).
Childhood abuse and neglect and childhood trauma were associated with BDNF in the two studies that assessed this biomarker (6, 20 in Table 1), but in two other studies inflammatory factors were not related to these variables (28, 36 in Table 1). No association was found between resilience and BDNF (29 in Table 1). An inflammatory factor (sVCAM-1) was associated with family history of BD in one study (35 in Table 1).
Chronotype was associated with an inflammatory biomarker (IL-6) in one study (30a in Table 1). As for sleep disorders, they were associated with the inflammatory factor IL-6 in one study (13 in Table 1). Dietary intake was associated with inflammatory factors (sTNFR1, sTNFR2) in one study (10 in Table 1), and weight cycling was associated with inflammatory factors (IL-6) in another (37 in Table 1). Two studies included BMI, but no correlations with inflammatory factors were found (17, 28 in Table 1).

Discussion
This systematic review and a series of exploratory meta-analyses represent an effort to synthesize reliable evidence on the association of inflammatory and trophic factors with subtypes, phases and severity of the disorder, in addition to factors related to clinical staying, episodes, duration, polarity, among other clinical parameters and comorbidities related to BD. We were able to include 51 studies, comprising 4547 patients with BD. The main finding of this systematic review and meta-analysis is the negative correlation found between BDNF levels and severity of depressive symptoms in BD patients. This evidence suggests that BDNF may be an eventual biomarker for BD.
This result is consistent with one of the most replicated findings in the literature, the decreased peripheral levels of BDNF and its association with depressive symptoms [83,84]. Besides, a positive correlation has recently been found between severity of manic symptoms and BDNF levels in BD patients [85]. Moreover, other studies find lower BDNF levels in BD patients in manic and depressive phases compared to controls [86], but these differences were not significant across affective states in general [87]. Although evidence suggests that BDNF has an important role in the psychopathology and progression of BD [88], there are inconsistent findings due to a number of confounding factors in uncontrolled studies that highlight the need for specificity as a biomarker in the diagnosis of BD. BDNF plays an increasingly important role as a biomarker not only in affective disorders, but also in cognitive impairment, such as that associated with alcohol abuse. It will be important to identify whether the decrease in BDNF observed in BD could be associated with present or ulterior cognitive impairment in BD [89].
The present review and meta-analysis do not show consistent associations between the inflammatory biomarkers studied and the main psychiatric and clinical variables of the disorder. Persistent pathogens likely support a role of an altered immune system in the etiology of mood disorders [90]. In particular cases such as TNF-α, the previously proposed association with BD is not supported by the present meta-analysis, and this finding is supported by recent studies with monoclonal antibodies against this cytokine that showed no effects in the prefrontal neurochemistry of patients with BD [91]. In a systematic review and exploratory meta-analyses in which levels of neurotrophic, inflammatory and oxidative stress biomarkers were analyzed in combination as a function of an affective state (euthymia, mania, depression), no biomarkers were found that could be individually discriminative of each mood phase, with the combination of hsCRP/IL-6, sTNFR1 with BDNF/TNF-α being significant [58]. However, later studies suggest that elevated levels of CRP and TNF-α in manic, depressive, and mixed episodes would point to these substances as biomarkers of mood episode, whereas elevated IL-6 levels during euthymic could refer to a euthymic role as a trait biomarker in BD [92].
Overall, heterogeneity and inconsistency in previous studies are supporting the search for new potential targets with a clinical value in BD. Several additional inflammatory factors, such as specific chemokines (MCP-1, fractalkine), glial factors (GFAP, S100B), or lipidderived mediators such as endocannabinoids, and neurodegeneration-related molecules such as neurofilament light chain, neurogranin and β-amyloid peptide isoforms (Aβ42, Aβ40, Aβ38) and their fragments could be promising biomarkers in patients with BD in relation to prospective clinical outcomes, such as those of neuronal injury, cognitive deficits and risk of dementia [93]. Recent studies have also evaluated non-coding RNA expression, sexual hormones, oxidative stress (uric acid, bilirubin, albumin) and metabolic biomarkers (metabolic syndrome, overweight/obesity, thyroid and liver function) in predicting BD and suicidality in individuals with bipolar disorder [94][95][96][97][98]. In addition, big data of cerebral blood flow and structural and functional connectivity using magnetic resonance imaging are becoming useful tools for actual practice in psychiatry [99,100]. By gathering all this information and using multivariable logistic regression, it could be possible to construct a model to predict BD.
Other results, such as the correlation between the biomarkers and bipolar subtypes, stage of disease, severity of manic symptoms and global severity of illness, were heterogeneous and should be analyzed with caution. Some studies found an association between biomarkers and other clinical variables included in the study, such as stressful life events, childhood trauma, cognitive abilities and functionality, but further studies are needed to confirm the relationship between inflammatory and neurotrophic factors and these variables. In general, more studies are necessary to understand precisely the neurobiology of BD and their interactions with stress and clinical factors. This improvement will presumably translate to more sophisticated treatments.
Taken together, although biomarkers and clinical, social and environmental factors have been shown to play an important role in understanding the etiopathology and development of bipolar disorder, these variables are not clinically replicable [101]. The clinical value of BDNF and other inflammatory biomarkers has yet to be demonstrated by highly controlled studies that follow a precise phenotyping (type, phase, severity, and comor-bidities) of BD patients. Moreover, other promising biomarkers related to mitochondrial dysfunction, oxidative stress, metalloproteinases, HPA axis function, autoimmunity and urinary metabolites, among others, need to be studied [102,103].

Strengths and Limitations
This comprehensive systematic review includes a series of exploratory meta-analyses of neurotrophic and inflammatory factors in bipolar disorder. The results for each biomarker have been separated by clinical and social factors, attempting to provide information on the physiological changes that occur with the illness, and reducing the risk of type 2 error. The use of random effects models adds further validity to the results.
However, this review has several limitations and the results should be interpreted with caution. First, most of the studies were cross-sectional, and changes in trophic/inflammatory biomarkers throughout the development of BD in an individual patient cannot be described. Most of the studies reviewed were observational and at significant risk of bias and confounding, so causality cannot be established. Individual characteristics varied greatly between studies. Many studies included small samples and the aims between them were different. Some factors influencing biomarker levels, such as body mass index, physical activity, drug use, or blood pressure, were not addressed in many studies [104,105]. Some studies measured several cytokines, but the results were not reported due to a lack of sensitivity in assays that detected valid levels. Dilution and storage of blood samples and measurement factors in plasma or serum may also influence the variability and reproducibility of biomarker levels [106][107][108]. The influence of medication is not analyzed in most studies, while others include drug-naïve patients. The inclusion of patients followed different criteria for defining bipolar disorder (i.e., 2 days of hypomania vs. 4 days) and the variation in YMRS and depression scale scores for defining mood phase probably contributed to the heterogeneity of the results. Several clinical variables were assessed from clinical interviews only, which may compromise the reliability and validity of the results. Because of the need to separate results by clinical and social factors, a significant number of meta-analyses were performed, which increases the risk of type 1 error. The decision to perform the meta-analysis with only three studies is arbitrary, but we have taken this decision to increase our statistical power by reducing the standard error of the weighted average effect size. The purpose of our review was exploratory, so multiple testing was not corrected for, and the results should be considered with caution. In addition, it is likely that some meta-analyses were underpowered as a result of a small sample size and the small number of studies included. To determine the validity and precision of a potential biomarker, future methodologically consistent and repeatable studies are needed to allow homogeneity and comparability of results.

Conclusions
The present study shed light on the evidence supporting inflammatory and neurotrophic factors as reliable biomarkers that can characterize clinical features for BD. At the moment, the mixed and contradictory evidence regarding the association of inflammatory and neurotrophic factors with clinical features suggest that highly controlled studies with wide sample sizes need to be conducted. Probably, heterogeneous and small samples, selection bias, different methodologies and bias in the measurement of outcomes, lack of control regarding confounders and bias in the selection of the reported results could explain the mixed evidence found in this review. The meta-analysis clearly indicates that BDNF has a role in the depressive component of BD and more studies need to be conducted to establish the clinical value of this biomarker. The systematic review suggests a possible association between stressors, functional and cognitive impairment, and BDNF and/or inflammatory factors but the small number of studies do not permit to establish solid conclusions. Moreover, the poor consistency of promising inflammatory mediators such as IL-6 and TNF-α indicates the need for homogeneous studies that help identify additional biomarkers of this disorder.

Supplementary Materials:
The following supporting information can be downloaded at: https:// www.mdpi.com/article/10.3390/biomedicines10061368/s1, Table S1: Additional information on the 51 selected studies; Table S2: Summary of the number of studies included in the systematic review and meta-analysis; Table S3: Abbreviations of inflammatory and trophic factors; Annex 1: Details of the search strategy conducted in the present systematic review; Figure S1: Traffic light plot of the domain-level judgments for risk-of-bias assessment; Figure S2: Traffic light plot of the domain-level judgments for risk-of-bias assessment.