Bipolar disorder (BD) is a severe chronic psychiatric disorder characterized by episodes of mania or hypomania, alternating with depression. Because of its early onset, prevalence of more than 1% of the global population, high rate of psychiatric and medical comorbidities and increased premature mortality, BD represents one of the main causes of disability among young people [1]. Both genetic and environmental factors are known to contribute to the onset of BD, and heritability of this disorder has been estimated at 60–85%. Genome-wide association studies (GWAS) have successfully identified a number of genetic loci implicated in this disorder ^[2][3][2,3]. However, the causes of BD, as well as the biological networks involved in this disorder, are still largely unknown. In addition, an accurate and timely diagnosis is difficult, as no biomarker is available and the clinical presentation of BD is often a depressive episode similar to unipolar depression [1].

Pharmacological treatment represents the mainstay in the long-term management of BD. Among mood stabilizers, lithium represents a first-line option because of its effectiveness in the acute phase of the disorder, in the prevention of recurrences and in the reduction of suicide risk. However, clinical response to lithium presents a high interindividual variability, with approximately 70% of patients showing partial or nonresponse [4]. Lithium response is heritable, and initiatives such as the International Consortium on Lithium Genetics (ConLiGen) are contributing to knowledge on the molecular determinants underlying this trait [5]. The first GWAS conducted by ConLiGen suggested the involvement of long noncoding RNAs (lncRNAs) in lithium response [5], and subsequent secondary analyses pointed to different markers and pathways potentially playing a role in this trait ^[6][7][8][6,7,8]. However, the molecular players underlying lithium’s complex mechanism of action are still elusive, and no reliable biomarkers are available to identify patients who might respond to this drug [9]. Even less information is available regarding molecular markers involved in the response to other mood stabilizers, such as the anticonvulsants carbamazepine, valproate and lamotrigine. The identification of reliable biomarkers that respond to mood stabilizers is a priority, since demonstration of proven clinical efficacy is one of the most important factors required for the successful implementation of genomic medicine in the health-care system, together with cost-effectiveness, appropriate knowledge and education and appropriate policy and legislation [10].

Transcriptomic studies provide information complementary to GWAS, allowing for the study of the dynamic evolution of different types of RNA markers based on specific tissues or biofluids, cell types and developmental stage. Among the most investigated RNA biomarkers in BD and in response to mood stabilizers, there are messenger RNA (mRNA) and different types of noncoding RNAs, such as microRNAs (miRNAs), circular RNAs (circRNAs) and lncRNAs.

miRNAs are single-stranded RNA molecules, approximately 20–22 nucleotides in length, that play a substantial role in the regulation of gene expression. Their main mechanism of action consists in promoting gene silencing by guiding argonaute proteins to the 3’ region of a target mRNA, thus allowing for the recruitment of factors that promote translational repression or mRNA decay [11]. Because the majority of protein-coding genes have been shown to present at least one miRNA-binding site, miRNAs are involved in a wide range of cellular functions and biological processes, and their dysregulation has been linked to several human diseases, including psychiatric disorders ^[11][12][11,12]

CircRNAs are single-stranded RNA molecules produced by pre-mRNAs through a process called back-splicing [13]. Unlike linear RNAs, circRNAs are characterized by a covalently closed-loop structure. This conformation makes circRNAs resistant to degradation by exonucleases and more stable compared with other transcripts. CircRNAs were originally considered a result of “splicing noise” with no relevant biological significance. However, in the last few years it has been shown that circRNA molecules are conserved, tissue-specific and involved in relevant biological functions [14]. While their roles are still largely elusive, circRNAs have been suggested to be involved in the regulation of transcription, protein transport and protein–protein interactions. In particular, some circRNAs present miRNA-binding sites and may act as miRNA sponges, sequestering miRNAs and, thus, preventing their interaction with mRNA targets. The prevalence of circRNAs has been largely underestimated until recently because their identification presents technical challenges. In fact, microarrays do not allow the detection of circRNAs or to distinguish the expression of circRNAs from their linear host genes. In addition, since most early RNA sequencing studies focused on measuring levels of RNA with polyA tails, they involved a polyA RNA enrichment step that led to the depletion of circRNAs. Modern RNA sequencing and bioinformatic pipelines are able to identify these markers, which allowed for the realization that circRNAs are much more abundant then first hypothesized. Most circRNAs show specific expression patterns based on tissue, cell type and/or developmental stage and, intriguingly, circRNAs are enriched in brain tissues compared with other tissues [14]. Specifically, in the brain, a significantly greater number of genes, particularly synaptic genes, host circRNAs [14]. In addition, their stability and low turnover rate suggest they might accumulate in postmitotic cells, such as neurons [14]. The abundance of circRNAs in the brain, as well as the peculiar characteristics of these markers, contributes to the increased research interest on these transcripts as potential biomarkers of brain disorders.

LncRNAs are a family of noncoding RNA molecules longer than 200 nucleotides and characterized by substantial differences in terms of length, expression patterns and longevity. Several lncRNAs have been shown to be able to regulate the expression of nearby or distant genes. In addition, lncRNAs can modulate chromatin structure, response to DNA damage and different signaling pathways [15]. As in the case of circRNAs, the measurement of lncRNAs is associated with specific technical challenges because of their generally low abundance. However, while the number of studies exploring these markers is still limited compared with other transcripts, over the last few years there has been increased interest in lncRNAs as disease biomarkers and potential therapeutic targets due to their high specificity in tissue expression patterns, fast turnover and regulation of cellular networks [15].

2. RNA Biomarkers and Bipolar Disorder

The large majority of the studies presented in Table 1 were conducted in whole blood or peripheral blood mononuclear cells (PBMCs), with only a minority of studies exploring RNA levels in plasma/serum ^{[16][17][18][19][20]}[16,20,21,22,23] or plasma-derived extracellular vesicles (EVs) ^[21][22][24,25]. Among EVs, exosomes are membrane vesicles released by different cells into the extracellular matrix that play a pivotal role in intercellular communication and signal transmission through the transfer of bioactive molecules to adjacent or distant recipient cells ^[23][26]. Exosomes carry a variety of molecules, including metabolites, lipids and nucleic acids, and are enriched in miRNAs. Intriguingly, neural exosomes can cross the blood–brain barrier and can be detected peripherally. Therefore, the change of peripheral exosomal content in patients with BD might, at least, partly reflect central changes, thus potentially allowing to identify brain-relevant biosignatures of disease and drug response in a noninvasive way. The few available studies that explored miRNA levels in EVs and exosomes reported promising results. Ceylan and colleagues measured genome-wide levels of miRNAs in plasma exosomes from 69 patients with BD (15 depressed, 27 manic and 27 euthymic) and 41 HCs. After multiple testing correction, three miRNAs showed lower levels (miR-484, miR-652-3p and miR-142-3p) and one miRNA higher level (miR-185-5p) in patients with BD compared with HCs ^[22][25]. The predicted targets of the four miRNAs were enriched for different pathways, including PI3K/Akt signaling, fatty acid biosynthesis/metabolism, extracellular matrix and adhesion pathways. No miRNA was significantly altered among the different states of BD ^[22][25]. Conversely, other studies suggested the potential utility of miRNAs as disease state markers. Namely, Camkurt and colleagues measured the levels of eight candidate miRNAs (selected based on previous evidence of their potential involvement in psychiatric disorders) in whole blood from 58 patients with BD (19 manic and 39 euthymic) and 51 HCs ^[24][27]. The levels of miR-07 were found to be significantly higher in patients with BD compared with HCs but also in patients in a manic episode compared with euthymic patients. Another study conducted by Banach and colleagues observed the downregulation of three miRNAs (miR-499, miR-798 and miR-1908) in patients with BD during a depression episode compared with a euthymic state ^[25][28]. Some studies identified significant differences in the RNA levels based on BD subtype. D’Addario and colleagues measured the mRNA levels of six candidate genes interacting with the brain-derived neurotrophic factor (BDNF) in PBMCs from 54 patients with BD type 1 (BD I), 45 with BD type 2 (BD II) and 42 controls. The authors reported lower levels of the prodynorphin (PDYN) gene in patients with BD II but not BD I compared with HCs. In addition, this study observed increased methylation at the PDYN promoter, as well as higher levels of genes involved in methylation, such as DNA methyltransferase 3 beta (DNMT3b) and methyl-CpG-binding protein 2 (MECP2) in patients with BD II compared with HCs. Other studies included a sample of patients with different psychiatric disorders, such as MDD ^[16][26][16,29] or SZ ^[27][30], aiming to distinguish among RNA markers specifically associated with BD or shared among different psychiatric disorders. Among these studies, Maffioletti and colleagues measured genome-wide miRNA levels in whole blood from 20 patients with BD, 20 with MDD and 20 HCs ^[26][29]. The study reported levels of five miRNAs to be increased specifically in patients with BD compared with HCs (hsa-miR-140-3p, hsa-miR-30d-5p, hsa-miR-330-5p, hsa-miR-378a-5p and hsa-miR-21-3p), while hsa-miR-330-3p and hsa-miR-345-5p showed higher levels in patients with either BD or MDD. However, one the miRNAs specifically associated with BD was found to be altered in MDD patients after treatment with antidepressants in a previous study conducted by the same authors ^[28][31]. Another study aimed at identifying the biosignatures of bipolar from unipolar depression measured genome-wide plasma miRNA levels in a discovery cohort of seven patients with BD, seven with MDD and six HCs ^[17][20]. The study reported higher levels of miR-19b-3p in patients with BD compared with patients with MDD, a result that was validated in a cohort of 27 patients with BD and 32 with MDD. In silico analyses suggested this miRNA to be involved in inflammatory dysregulation associated with experiencing early childhood trauma ^[17][20]. As shown in Table 1, a number of studies provided evidence of a good performance of the investigated RNAs in the discrimination of patients with BD from HCs, based, for example, on the area under the curve (AUC) ^[29][32]. However, it must be considered that several of these studies did not include a replication cohort.