Major depressive disorder (MDD) is a complex, relatively common, costly, and recurrent mental disorder that, according to the World Health Organization, is the leading cause of disability worldwide and is among the leading causes of disease burden globally (Centers for Disease Control Prevention (CDC), 2010; Greenberg et al., 2015)¹. One of the major risks of MDD is suicide, which has been reportedly linked to psychiatric disorders in general and to MDD specifically (Zweig and Hinrichsen, 1993; Lesage et al., 1994). Improving the treatment for MDD would have huge financial consequences and would greatly improve the quality of life for millions of patients globally. Selective serotonin reuptake inhibitor (SSRI) antidepressant drugs, which block serotonin uptake via binding directly to the serotonin transporter (SERT) (Sangkuhl et al., 2009), have remained for the past three decades as the first-line treatment for MDD (Thaler et al., 2012). However, about 30–40% of MDD patients fail to reach sufficient remission with SSRI treatment (Souery et al., 2007). In such cases, clinicians often increase the dosage of the same SSRI drug, switch to another antidepressant of the same or a different class [such as serotonin-norepinephrine reuptake inhibitors (Girardi et al., 2009), tricyclic antidepressants, or serotonin ligands (Ruhé et al., 2006)], or augment the antidepressant by adding the mood-stabilizing drug lithium (Price et al., 1990; Bauer et al., 2003).

Clinical guidelines recommend waiting at least 4–6 weeks before switching to an alternative drug (Kato and Serretti, 2010). Meanwhile, patients may experience long periods of depressive symptoms and an increased risk of suicide with no benefit from their first-line SSRI treatment. However, to date no diagnostic tools for predicting patient response to specific antidepressants are available (Dale et al., 2015).

Biomarkers for predicting antidepressant drug response, in particular for SSRIs as the first-line antidepressants, are therefore needed for aiding clinicians in drug and dosage choice in order to decrease the time from diagnosis to remission for the SSRI non-responder patient population. Additionally, reliable diagnostics are needed for early identification of treatment-resistant depression (TRD, patients who do not respond to any of the approved antidepressants). Several genome-wide association studies (GWAS) have searched for common single nucleotide polymorphisms (SNPs) associated with SSRI drug response (Ising et al., 2009; Garriock et al., 2010; Uher et al., 2010; Ji et al., 2013; Biernacka et al., 2015). However, none of these GWAS findings could be replicated, and several meta-analysis studies have been published (Tansey et al., 2012; GENDEP Investigators, MARS Investigators, STAR^*D Investigators, 2013; Biernacka et al., 2015), which concluded that none of the identified SNPs had genome-wide significance.

A major landmark in mRNA regulation and protein expression levels came about with the discovery of microRNAs (miRNAs). miRNAs are short (22 nucleotides on average) endogenous non-coding RNAs that down-regulate gene expression at the post-transcriptional level (Bartel, 2009; Friedman et al., 2009; Shomron, 2009; Rukov and Shomron, 2011). Thousands of miRNAs are encoded within the human genome; they are prevalent in all cells, tissues (Liang et al., 2007) and body fluids (Weber et al., 2010; Gurwitz, 2015). At least half of all human gene transcripts are estimated to be targets of evolutionarily conserved miRNA regulation (Lewis et al., 2005; Friedman et al., 2009). Many miRNAs were implicated in various diseases including those involved in brain disorders (Mor et al., 2013; Serafini et al., 2014; Modai and Shomron, 2016).

Given that miRNA can potentially target dozens of genes, they have been recognized as master regulators of gene expression in multicellular organisms. Thus, changes in miRNA levels can affect and might even predict changes in global gene expression (Lim et al., 2005). Consequently, miRNAs are being studied as diagnostics, prognostics, therapeutics, or as pharmacogenomic biomarkers. Indeed, in recognition of their important role in health, disease, and drug response, a new trend in molecular medicine, termed “miRNA pharmacogenomics,” has emerged (Shomron, 2010; Rukov and Shomron, 2011; Rukov et al., 2011, 2014).

Accumulating evidence suggests that the expression of miRNAs is altered during chronic stress, and that dysregulation of miRNAs during such stress may contribute to the etiology and pathophysiology of MDD (Dwivedi, 2011; Mouillet-Richard et al., 2012; Issler et al., 2014; Lopez et al., 2014; Garbett et al., 2015). Moreover, miRNAs may be involved in antidepressant response and may serve as novel antidepressant targets, since different studies have demonstrated altered miRNA expression levels following antidepressant treatment (Bocchio-Chiavetto et al., 2013; O'Connor et al., 2013). Owing to the above considerations and because a single miRNA can potentially regulate scores of mRNAs, miRNAs have the potential to serve as pharmacogenomic biomarkers for SSRI response (Rukov and Shomron, 2011; Labermaier et al., 2013).

Affective disorders, including MDD, which seem to uniquely affect humans (or at least only primates) and are not easily modeled by rodent studies, are related to defects in higher human cognitive processes and to the complexity of the human brain compared with the rodent brain (Geschwind and Rakic, 2013; Czéh et al., 2016). MDD is a complex mood disorder with high comorbidity with anxiety disorders, characterized by psychiatric diagnoses of human behaviors such as sadness, guilt, and feeling of worthlessness, among others. Regulation of gene expression (e.g., by miRNAs) has been suggested to mediate stable adaptations in the brain (Tsankova et al., 2007; Labermaier et al., 2013). Thus, the high level of miRNA conservation among primates observed for our three candidate miRNAs and their respective target gene miRNA-binding sites is intriguing and suggests that an important (and thus evolutionary conserved) role exists for the regulation of ITGB3 and CHL1 by miR-221/222 and miR-151a-3p, respectively, in primates (especially in humans) vs. non-primate vertebrates. Moreover, miR-221/222 (to which we will refer together owing to their very close similarity) and miR-151a-3p have a canonical 8-mer site on the 3′UTR region of their targeted genes. Having an 8-mer site was shown to be important for recognition and down-regulation of the target gene (Brennecke et al., 2005; Grimson et al., 2007).

ITGB3, coding for the integrin beta-3 subunit, together with integrin alpha-v or with the integrin alpha-IIb, form the vitronectin receptor (αvβ3), and the fibrinogen receptor (αIIbβ3), respectively. Human ITGB3 has been shown to be regulated by different human miRNAs including miR-98 (Ni et al., 2015), miR-320a (Sun et al., 2015), let-7c (Zhao et al., 2014), and let-7a (Müller and Bosserhoff, 2008). However, in a previous study we did not observe changes in the expression levels of the latter miRNAs following chronic in vitro SSRI treatment of human cells (Oved et al., 2013), suggesting that miR-221/222 play a role in the mode of action of SSRI antidepressants.

Using luciferase assays, our present study confirms the direct regulation by miR-221/222 of their potential target-gene, ITGB3. In addition, the expression of ITGB3 was dramatically increased (3.5-fold) following the 21-day growth of human LCLs in SF media, whereas under these conditions, miR-221 was down-regulated in the same cells by 2.2-fold (Figure 7A).

miR-221/222 and their target-gene, ITGB3, are expressed in the human brain and are implicated in synaptogenesis/neurogenesis (Cingolani and Goda, 2008; Cingolani et al., 2008; Terasawa et al., 2009; Shao et al., 2010; Hamada et al., 2012; Pozo et al., 2012; Cheng et al., 2014). Using Ilumina sequencing, miR-221 and miR-222 were identified among the top 40 highly expressed miRNAs in post-mortem human brains (Shao et al., 2010); both were induced by nerve growth factor in rat pheochromocytoma PC12 cells (Terasawa et al., 2009; Hamada et al., 2012) and were shown to play a role in neurite guidance (Cheng et al., 2014). Moreover, miR-221 potentiated the formation of neurite networks in these cells (Hamada et al., 2012). ITGB3 is involved in synaptic plasticity in mouse neuronal hippocampal cultures (Cingolani and Goda, 2008) and regulates excitatory synaptic strength (Cingolani et al., 2008) and GluA2 AMPA receptor expression in mouse hippocampi (Pozo et al., 2012).

Importantly, ITGB3 is crucial for maintaining the activity of the serotonin transporter (SERT; encoded in humans by SLC6A4), the well-established drug target of SSRI drugs that block serotonin uptake via binding directly to SERT (Sangkuhl et al., 2009). This role of ITGB3 is evident from the diminished SERT activity in platelets (Carneiro et al., 2008), as well as in the midbrain (Whyte et al., 2014) and raphe nuclei synaptosomes (Mazalouskas et al., 2015) of Itgb3-deficient mice, either knockout (Itgb3^−/−) (Carneiro et al., 2008) or heterozygous (Itgb3^−/−) (Whyte et al., 2014; Mazalouskas et al., 2015). Itgb3-knockout mice exhibit altered social and repetitive behavior, e.g., behavior relevant for autism spectrum disorder (Carter et al., 2011). Neuroanatomical assessment of these mice indicated significantly different relative tissue volumes in several brain regions, among them reduced volume of the lateral wings of the dorsal raphe nuclei—a brain region important for the development of the CNS serotonergic system (Ellegood et al., 2012). Moreover, integrins, including ITGB3, interact with CHL1 in the plasma membrane and promote CHL1-induced neuronal migration and neurite outgrowth (Buhusi et al., 2003; Demyanenko et al., 2004; Schlatter et al., 2008; Katic et al., 2014). Considering the above, these findings suggest that ITGB3 plays an important role in correct neuroanatomical development of the CNS, a property it shares with CHL1.

Interestingly, in a pilot clinical study, miR-221 was found to be down-regulated in the plasma of MDD patients treated with the SSRI antidepressant escitalopram (Enatescu et al., 2016). In another study, miR-221 was found to be up-regulated in CSF and serum samples of MDD patients vs. control subjects (Wan et al., 2015). These observations correspond to the decreased expression of miR-221/222 we observed following chronic SSRI treatment in human LCLs (Oved et al., 2013); thus they demonstrate that human LCLs can serve as a legitimate research tool for searching biomarkers for SSRI response in neurons. Whereas in a study that examined blood mononuclear cell proteomes, several members of the integrin signaling pathway, including ITGB3, were found to be differentially expressed between responder and non-responder MDD patients (Martins-de-Souza et al., 2014). In the latter study, ITGB3 expression was found to be increased in responder vs. the non-responder MDD patients.

The identification of ITGB3 as a potential SSRI response biomarker was further supported by our recent study in which lower ITGB3 expression levels (~50% vs. healthy controls) were observed in peripheral blood mononuclear cells (PBMCs) obtained from MDD patients (Rzezniczek et al., 2016). The lower ITGB3 expression levels observed in the PBMCs obtained from the MDD patients is noteworthy and suggests that increased ITGB3 expression upon chronic SSRI treatment plays a role in the therapeutic action of these drugs in MDD.

Human CHL1 (close homolog of L1; also known as CALL or L1CAM2) has also been shown to be regulated by additional miRNAs such as miR-10a, miR-590, miR-182, and miR-21 (Long et al., 2012; Chu et al., 2014; Zhu et al., 2014; Li et al., 2016). However, in a previous study we did not observe changes in the expression levels of the latter miRNAs (Oved et al., 2012), suggesting that miR-151a-3p plays a role in the mode of action of SSRI antidepressants.

Using luciferase assays in cells transfected with miR-151a-3p, we confirmed the direct regulation of CHL1 transcription by this human miRNA, for which the expression levels were observed to be significantly correlated with the in vitro paroxetine sensitivity phenotype of LCLs from unrelated healthy individuals (Oved et al., 2012). In addition, the expression of CHL1 was dramatically increased (3.6-fold) following the 21-day growth of human LCLs in SF media, whereas under these conditions miR-151a-3p, which targets CHL1, was down-regulated in the same cells by ~2-fold (Figure 7A). Moreover, following over-expression of miR-151a-3p in human LCLs, we observed a strong reduction of CHL1 expression (Figure 5B), along with higher in vitro sensitivity to paroxetine (Figure 6), as expected from our previous genome-wide transcriptomic study reporting lower CHL1 expression in human LCLs exhibiting higher in vitro paroxetine sensitivity (Morag et al., 2011). Notably, plasma miR-151a-3p was down-regulated in escitalopram-medicated MDD patients (Enatescu et al., 2016).

CHL1 codes for a cell adhesion protein which plays central roles in neural cell proliferation, migration, differentiation, signal transduction and axon guidance (Maness and Schachner, 2007; Huang et al., 2011). It is implicated in correct brain circuitry (Montag-Sallaz et al., 2002, 2003) and in mental disorders (Frints et al., 2003; Chen et al., 2005). Moreover, CHL1 expression was downregulated in the hippocampus of mice exposed to early post-natal stress, a known aggravator of mood disorders (Desarnaud et al., 2008). Furthermore, CHL1/L1 double knockout (CHL1/L1^−/−) mice have misguided neuronal circuitry from the limbic system to the cerebral cortex (Demyanenko et al., 2010, 2011). In addition, CHL1 was reported to interact with the serotonin 2c receptor and thereby act as a modulator of the serotonergic system (Kleene et al., 2015).

MDD and bipolar disorder are related affective disorders, with BD-II typified mostly by depressive episodes (Baldessarini et al., 2013; Carvalho et al., 2014). Lithium, which is used as a first-line treatment for bipolar disorder, is also employed for augmenting antidepressant therapy in treatment-resistant MDD (Price et al., 1990; Bauer et al., 2003). Notably, lithium was reported to down-regulate miR-221 in the hippocampi of chronically treated rats (Zhou et al., 2009). Moreover, miR-221/222 were down-regulated in the hippocampi of juvenile rats following the induction of lithium-pilocarpine status epilepticus with lithium chloride injections (Ashhab et al., 2013). Taken together, these findings suggest that lithium augmentation of SSRI efficacy in MDD (Price et al., 1990; Bauer et al., 2003) could be related to ITGB3 up-regulation secondary to miR-221/222 down-regulation.

A recent study by Milanesi et al. reported that the in vitro lithium sensitivity of LCLs derived from bipolar disorder patients is affected by IGF-1, an effect that could be observed only in the absence of serum (Milanesi et al., 2015). Notably, IGF-1 was also implicated in MDD (Kopczak et al., 2015). IGF-1 binds directly to integrin αvβ3, and this interaction was found to be essential for IGF-1 signaling through the IGF1R receptor (Saegusa et al., 2009). In addition, IGF-1 assembles the formation of a heterocomplex between IGF1R and the integrin β3 subunit (Tahimic et al., 2016). We therefore compared the transcription of ITGB3 and CHL1 in LCLs maintained in SF vs. serum-containing media. In addition, we examined the transcription of ITGAV, coding for the integrin alpha-v subunit, which together with integrin beta-3, forms the vitronectin receptor (αvβ3), in SF vs. serum-containing media. The expression levels of ITGAV, ITGB3, and CHL1 were all elevated following 21-day growth of human LCLs in SF media (Figure 7A) and a large inter-individual variation in the extent of up-regulated expression was observed for ITGB3 and CHL1 (Figure 7B). A correlation (R² = 0.46; P = 0.031) was observed between the fold-change of the elevated expression levels of these two genes (Figure 7C). This finding is intriguing as ITGB3 interacts with CHL1 at the cell membrane (Katic et al., 2014), and both ITGB3 and CHL1 were implicated in our hypothesized model regarding the mode of action of SSRI drugs (Oved et al., 2013). This model depicts the cell membrane proteins encoded by CHL1 and SLC6A4 (coding for the serotonin transporter), competing on a limited cell membrane protein reservoir of integrin beta-3 (encoded by ITGB3) (Oved et al., 2013).

Das et al. reported that cultivation of cells in serum-free medium in the presence of fibronectin up-regulates the activity of MMP-2 and MMP-9, two matrix metaloproteinases implicated in cell migration and invasion (Das et al., 2008). Expression of αvβ3 correlates with activation of MT1-MMP and MMP-2 in human melanoma cells (Hofmann et al., 2000). Improved cell adhesion of non-adherent cells in serum-free media has been demonstrated, including for Jurkat lymphoblastoid cells, an immortalized cell line of human T-lymphocytes (Thirumala et al., 2007; Audiffred et al., 2010; Nakayama et al., 2014). These findings correspond to our observations of higher expression levels of the cell adhesion genes ITGAV, ITGB3, and CHL1 in human LCLs maintained in serum-free, compared with serum-supplemented media.

The present study has several limitations. Although we have previously observed inverse expression patterns between miRNAs and the target gene mRNAs levels in human LCLs and have now experimentally validated miRNA-targets using luciferase reporter assays, one has to keep in mind that our experiments were carried out in vitro in human cell lines. We used LCLs from healthy unrelated donors, while the expression of the following miRNAs and target genes may differ in brain tissues of MDD patients. In addition, in order to measure the effect of miR-151a-3p upon SSRI response we used an in vitro assay of LCL growth inhibition by the SSRI drug paroxetine as a surrogate for clinical SSRI drug response. Our findings should thus be considered as tentative miRNA/target gene biomarkers for SSRI response until studies with MDD patient blood samples and/or studies with brain tissues of animal MDD models validate our results.

Taken together, our current observations lend further support for the putative role of miR-151a-3p and miR-221/222 and their (here confirmed) respective target-genes, CHL1 and ITGB3, in SSRI responsiveness, and possibly in the pathology of MDD. We therefore propose that miR-151a-3p and miR-221/222 could tentatively serve as SSRI response biomarkers, as well as potential novel antidepressant therapeutic targets, following additional experimental validation in animal models as well as in clinical trials.

NS, DG, and KO were responsible for the study design and interpretation of the data. KO, LF, AG, II, and DH were responsible for acquisition of the data.