Bipolar Disorder (BD) is a severe and chronic disorder characterized by emotional and humoral dysregulation that leads to increased frequency of cognitive dysfunction and interpersonal problems (Chang et al., 2012). BD type I is characterized by one mania episode or rapid (daily) cycling episodes of mania and depression, preceded or followed by hypomanic or major depressive episodes, presenting euphoria and exaggerated behaviors such as glibness, greatness, flight of ideas, leading to social problems. In type II, at least one hypomanic episode alternates with depression phases, occurring less externalizing problems compared to mania episodes regarding to mood switching, interpersonal, and occupational problems (American Psychiatric Association, 2013). It has estimated that BD rank is the 17th cause of years lived with disability in Europe, accounting for 1.6% of the total (World Health Organization, 2016). In the United States 2.6% of adults are affected by BD (National Institute of Mental Health, 2011) and in developing countries like Brazil ~1% of the population are diagnosed with BD type I and II (Lima et al., 2005).

In addition to mood changes, patients with BD tend to show deficits in executive functions, verbal and visuospatial episodic memory, working memory (WM), verbal learning, information processing speed, sustained attention, and sensorimotor performance (Bora et al., 2009; Latalova et al., 2011; Lage et al., 2013; Loschiavo et al., 2013; Lee et al., 2014; Neves et al., 2014). Some studies have suggested more severe cognitive deficits in episodes of manic/mixed mood compared with depressive and euthymic episodes (Sweeney et al., 2000). Latalova et al. (2011) also observed that cognitive deficits tend to persist even in euthymic episodes, whereas, lower scores for verbal short-term memory and executive functions were found in both types of BD mood episodes, in tasks of WM, executive control, and verbal fluency. Some studies have shown that cognitive deficits observed in acute manic and depression episodes (Rubinsztein et al., 2000; Clark et al., 2001; Murphy, 2001), are also present in euthymic periods (van Gorp et al., 1998; Martinez-Aran et al., 2007; Wingo et al., 2009). Moreover, lower cognitive scores persist in visuospatial memory, verbal short and long-term memory, executive function, attention/processing speed, and WM, even after a long period of symptoms remission (van Gorp et al., 1998; Rubinsztein et al., 2000; Robinson et al., 2006; Martinez-Aran et al., 2007; Torres et al., 2007; McKenna et al., 2013; Ajilore et al., 2015). Besides that, first-episode BD is associated with widespread cognitive dysfunction (Lee et al., 2014), this result indicates cognitive deficits even in the earliest stages of disease and reinforces them as primary characters of BD. Symptoms of BD as anxiety and depression seem to reflect the malfunctioning of systems that provide a basis for action within a complex emotionally valenced world (Baddeley, 2013).

Among the cognitive deficits reported, WM seems to be relevant to understand the symptoms of BD. According to Miyake and Shah (1999) there are at least 10 diverse frameworks pointing different nature, structure, and functions of WM, which could diverge in terms of explanations of WM deficits. Then, it is important specifies some issues regarding this cognitive skill and its assessment that will be taken into account in the present paper. For instance, different tasks are considered measures of WM components: simple span tasks as measures of verbal and visuospatial storages, whereas, complex span tasks combine storage and attentional control (Baddeley, 1986). Currently, complex span tasks and n-back tasks are the most used WM measures (Miyake and Shah, 1999), since both require storing and maintain the information of a set of stimuli, and inhibiting interference of the signals recently presented (St Clair-Thompson and Gathercole, 2006). But there are dissimilarities, complex span tasks request keeping a serial order in mind at the same time of an ongoing cognitive task (Daneman and Merikle, 1996; Unsworth and Engle, 2007; Conway and Kovacs, 2013), while n-back tasks demand a constant updating of the relevant information (Kane et al., 2008; Wilhelm et al., 2013; Remoli and Santos, 2017). Apart from that, there are low correlations between n-back tasks and complex span tasks leading a conclusion that both evaluate different processes of WM (Kane et al., 2008; Redick and Lindsey, 2013), particularly concerning to the influence of emotional states (Ribeiro et al., submitted). A systematic review showed that different verbal WM capacities could be observed in studies that used mood induction before the completion of the task. While positive and negative stimuli seem to decrease the span task performance, no effects were found for n-back tasks (Ribeiro et al., submitted). In fact, negative mood induction can decrease WM performance during simple span and complex tasks (Spies et al., 1996; Santos et al., 2015; Soares, 2015).

In this review, WM is assumed to be a temporary information storage system that underpins the capacity for active thought by providing a cognitive workspace in which information may be temporarily maintained and manipulated (Baddeley, 2013). It is composed of multiple components, as the central executive, responsible for attentional control, and two subsidiary systems: the phonological loop and the visuospatial sketchpad, responsible for storing auditory-verbal and visual-spatial information, respectively (Baddeley et al., 2012). Subsequently, the episodic buffer was added to the initial WM model for integrating information from its multiple components with long-term memory (Baddeley, 2000). Previous studies indicated that when emotions and valenced thoughts are brought to WM they elicit somatic states (Bechara and Damasio, 2005), and the activation in brain emotional circuits, as conscious experience (LeDoux, 2000).

In fact, the connection of emotional states and working memory goes beyond the measurement instruments but in theoretical framework as well. For instance, the somatic marker hypothesis presented by Damasio et al. (1991) describes a mechanism for interaction between emotion and cognition in trial and decision-making processes (Bechara and Damasio, 2002, 2005; Bechara et al., 2005; Verdejo-García and Bechara, 2009). Damasio goes further by stating that the central executive component of working memory could be relevant in this process. As far as we know Baddeley's is the only model of WM explicitly addressing the capacity to temporary manipulate stored hedonic information. Baddeley proposed a new component, the Hedonic Detector which is understood as a neutral point that varies between positive and negative valences in response to environmental stimuli by establishing a mean value between stimulus and information retained in the WM to enable choices of future actions (Baddeley, 2007). It means that processing “hot” information (emotional content) disrupt the functioning of “cold” cognitive process, like core WM processes. Accounting this assumption, Baddeley et al. (2012) demonstrated that hedonic judgment of stimuli, such as words, pictures, and faces indeed could be influenced by the valence of an induced mood. Recently, different studies observed that mood induction procedures, e.g., listening valenced instrumental music while retrieving valenced autobiographical memories affect scores in WM tasks of health controls (e.g., Spachtholz et al., 2014; Allen et al., 2015). In this perspective, improper adjustment of the neutral point could be expressed by deficits to maintain, manipulate, and/or update information in WM. In the case of pathological affective episodes likewise in Bipolar Disorder (BD), the episodes itself could uncalibrated the hedonic detector.

Deficits in WM processing have been consistently reported in euthymic patients (MacQueen et al., 2005; Thompson et al., 2007; Daglas et al., 2015). Another study also observed that euthymic BD type I had worse performance on visuospatial tasks compared to healthy subjects (Farahmand et al., 2015). Although, there is no consensus in the literature, these findings suggest that deficits in WM could result from the intensification of emotional valences. In other words, the presence of WM deficits, could be considered a primary trace of BD, beyond state variables (Gruber et al., 2009; Kurtz and Gerraty, 2009). However, this particular issue remains controversial due to the diversity of methodologies used across the studies. In addition to the prevalence of within-subjects design, some authors rarely made the connection between theoretical models and cognitive tasks, limiting the comprehension about WM and their particular components.

In line with this argument, a review by Baddeley (2013) showed that in depressive patients negative mood influences hedonic judgment, explaining the trend of the negative perception of the situations in this clinical population. However, there is no significant evidence in respect to the influence of positive mood in hedonic detection system. Hypothetically, assuming that the neutral point in BD patients corresponds to euthymic phase, the exacerbated positive and negative valences would account for euphoric or depressed mood, respectively (Baddeley, 2007, 2013). As emotion in BD is deregulated, the study of this disorder seems to be necessary to understand the influence of improper adjustment of the hedonic detector neutral point in WM.

There are many meta-analyses which state significant neuropsychological deficits in euthymic bipolar patients (Robinson et al., 2006; Torres et al., 2007; Bora et al., 2009; Kurtz and Gerraty, 2009; Mann-Wrobel et al., 2011) and their relatives (Balanzá-Martínez et al., 2008; Arts et al., 2009; Bora et al., 2009). Most of these studies including working memory performance in bipolar patients Robinson et al. (2006), Torres et al. (2007), and Kurtz and Gerraty (2009) found that BD patients present deficits in measures of verbal working memory. In a recent review, Cullen et al. (2016) found similar results pointing that BD patients performed worse than controls in attention/working memory tasks. Bora et al. (2011) found evidences of verbal working memory deficits both in Type I and II bipolar patients. These deficits are found even in BD's relatives suggesting that these deficits are potential endophenotypes in Bipolar disorder; (Balanzá-Martínez et al., 2008; Arts et al., 2009) found evidences of worse performance in verbal working memory tasks in both BD's patients and relatives. Nonetheless these WM deficits are heterogeneous and seems to be affected by both demographic (e.g., educational level) and clinical aspects of the disorder (e.g., number of episodes) (Arts et al., 2009; Kurtz and Gerraty, 2009).

More recently, some studies using meta-analytical methodology addressed specific cognitive deficit in the bipolar patient as in theory of mind (Bora et al., 2016), social cognition (Samamé et al., 2015), and verbal fluency (Raucher-Chéné et al., 2017). Nonetheless, we did not found studies focusing in working memory deficits in bipolar disorder patients. Most of systematic review consider working memory in comparison to another cognitive functioning and some of than considered these deficits together with another executive functioning component (Robinson et al., 2006; Torres et al., 2007) and even attentional functioning (Cullen et al., 2016). Therefore, the purposes of this study are (i) to update the knowledge of working memory deficits including recent published data about working memory deficits in bipolar patients; (ii) analyze if BD patients show WM deficits, contrasting studies that used complex span tasks and n-back tasks; (iii) to investigate if different BD mood episodes predict different patterns of WM functioning, analyzing its components in each mood phase. The primary hypothesis of the present review is that working memory deficits are worsened during active phases of the disorder but remains in euthymic patients.

This study is a descriptive and informative systematic review of the literature about the association between BD patient's states and the functioning of WM components. We used the instructions of Cochrane Foundation (Higgins and Green, 2011) to ensure the presentation of comprehensive and unbiased data. The following questions were asked to guide the review: Do patients diagnosed with BD have deficits in working memory? Are specific states of BD (mania, hypomania, depression, and euthymic) related to changes in working memory processes, such as updating and serial recall?

After the definition of the guiding questions, the following steps were taken: setting and collection of studies in databases; critical evaluation of studies, data selection and analysis, presentation, and interpretation of results (Bento, 2014).

All articles indexed in MEDLINE, Scopus, SciELO, and web of Science search databases that used, at least, one WM task in patients with BD, during a specified mood episode were included. Papers with available abstracts and written in Portuguese, English, or Spanish were considered.

Selected articles evaluated the WM in adult patients with BD, aged from 18 to 65 years old, and published until 15st July 2016. Literature reviews, case studies, and studies that referred to psychological or pharmacological interventions were excluded. Research and selection of manuscripts were done by the first author (CSF) and reviewed by the second author (FHS), with agreement between them.

For searches in databases, the terms “transtorno bipolar,” “memória operatória,” and “emoção”; “trastorno bipolar,” “memoria operativa,” and “emoción”; and “bipolar disorder,” “working memory,” and “emotion” were used. We found 55 papers in MEDLINE, 63 in Scopus, 1 in Scielo, and 95 in Web of Science in the first search until 15th July 2016. No studies were found in Portuguese and only one study was written in Spanish.

Among the 214 papers identified, 34 were repeated, and 20 were found in more than two databases. Based on the abstract reading, from the 160 remaining studies 25 papers were selected and 135 were excluded, due to the presented criteria: (1) 16 did not use WM tasks; (2) 42 referred to literature reviews; (3) 27 had samples whose participants were under 18- or over 65-years old; (4) eight evaluated the effectiveness of psychological or pharmacological interventions; (5) three were animal sample studies; (6) 39 had undiagnosed BD participants.

Twenty-five articles were thoroughly analyzed, being excluded two studies for the following reasons: (1) the lack of contrast between WM measures and non-social domains, which were used only as demographic control measures; (2) absence of description about the measures of WM and other cognitive domains in this session. As a result, 23 papers were discussed in this review (Figure 1).

Since most studies were methodologically unrelated to each other, carry out a meta-analysis in the present state of art would be premature, therefore we opted to systematically organize the selected articles hoping that this effort will provide new insights for researchers in order to advance in this field. Therefore, the current systematic review structured the analyses focused on the three main working memory processes (maintenance, manipulation, and updating), in respective modalities (verbal or non-verbal). The fundamental aspects of the papers were organized using the PICOS strategy, an acronym that represents the initials of the items: Population, Intervention (procedure), Comparison, Outcome, and Study Design (Bento, 2014). In this way, we used the following topics: population, type of intervention, methodological design, evaluated variables or expected results, and study design (Table 1). Based on these items we could discuss them into three sessions: “Cognitive deficits and neurobiological correlates in BD patients,” “Different patterns of cognitive processes in each mood phase,” and “Relationship among WM and other clinical features.” The first two items addressed clearly the objectives that motivated the present study and the final session account for other findings observed in the selected articles.

The SCImago Journal & Country Rank was used as an indicator of visibility in scientific domains by ranking the journals in which papers were published. The ranking for psychology, neuroscience, and medicine areas in the last 7 years were considered. Twenty-one studies were published in Q1 journals (Fleck et al., 2005; Malhi et al., 2007; Thompson et al., 2007; Barrett et al., 2008; Deckersbach et al., 2008; Drapier et al., 2008; Roiser et al., 2009; Liu et al., 2010; Thermenos et al., 2010; Bertocci et al., 2011; Levy et al., 2011; Mullin et al., 2012; Gruber et al., 2013; Lee et al., 2013; Pomarol-Clotet et al., 2014; Russo et al., 2014; Bauer et al., 2015; Gvirts et al., 2015; McCormack et al., 2015; Muhtadie and Johnson, 2015; Sabater et al., 2016), one in Q2 journal (Miguélez-Pan et al., 2014), and one in Q3 journal (Dittman et al., 2008). These quartiles also indicated that the majority of selected studies were published in higher impact factor journals.

According to the publication period of the selected articles, the first research of a direct association between emotions and WM in BD patients was performed by Fleck et al. (2005). The other studies were published in the following periods. The research studies have been mostly carried out in America, with eleven studies in the United States, eight in Europe, four of them in Asia.

Considering the first topic of analysis, most studies participants were subjects of both gender, except articles by Bertocci et al. (2011), Deckersbach et al. (2008), Malhi et al. (2007), and Thermenos et al. (2010) that only studied women. There were higher prevalence of studies with euthymic phase of BD samples, but three studies considered only depressive episodes in BD subjects (Deckersbach et al., 2008; Roiser et al., 2009; Bertocci et al., 2011). Furthermore, one paper compared a manic/psychotic symptoms group with a euthymic population (Fleck et al., 2005) and another one assessed three groups: mania, depression, and euthymic (Pomarol-Clotet et al., 2014). Articles by Liu et al. (2010) and McCormack et al. (2015) did not characterize explicitly the sample mood despite presented scores of symptoms scales and the study by Gvirts et al. (2015) comprised two subgroups with regard to affective symptoms: asymptomatic and mildly symptomatic patients. Regarding the subtypes of BD, 13 studies used BD I and II mixed groups, nine studies analyzed only BD type I subjects, and one paper compared BD type I, type II, and control groups (Liu et al., 2010).

Most studies evaluated the sample in just one mood episode in BD. Euthymic patients were predominant in the samples (Malhi et al., 2007; Barrett et al., 2008; Dittman et al., 2008; Drapier et al., 2008; Thermenos et al., 2010; Levy et al., 2011; Mullin et al., 2012; Gruber et al., 2013; Lee et al., 2013; Miguélez-Pan et al., 2014; Russo et al., 2014; Bauer et al., 2015; Muhtadie and Johnson, 2015; Sabater et al., 2016), few studies evaluated depressions phases of BD (Deckersbach et al., 2008; Roiser et al., 2009; Bertocci et al., 2011) and only one compared the performance in manic and euthymic episodes in relation to health controls (Fleck et al., 2005). All studies have used between-groups comparison, and only article by Pomarol-Clotet et al. (2014) also had a within-participants design and evaluated the same patients in three characteristics mood phases of BD.

Regarding the intervention, all studies used cognitive tests and self-reported mood scales. Seven studies (Malhi et al., 2007; Deckersbach et al., 2008; Drapier et al., 2008; Thermenos et al., 2010; Bertocci et al., 2011; Mullin et al., 2012; Pomarol-Clotet et al., 2014) also applied functional magnetic resonance imaging (fMRI) to measure brain activity. In relation to the cognitive instruments, 16 studies used span tasks (Malhi et al., 2007; Thompson et al., 2007; Barrett et al., 2008; Dittman et al., 2008; Roiser et al., 2009; Liu et al., 2010; Levy et al., 2011; Gruber et al., 2013; Lee et al., 2013; Miguélez-Pan et al., 2014; Russo et al., 2014; Bauer et al., 2015; Gvirts et al., 2015; McCormack et al., 2015; Muhtadie and Johnson, 2015; Sabater et al., 2016), and six studies chose n-back tasks (Deckersbach et al., 2008; Drapier et al., 2008; Thermenos et al., 2010; Bertocci et al., 2011; Mullin et al., 2012; Pomarol-Clotet et al., 2014), with two of them using the modified EFNBACK instrument. Only one article used a supra span task, the yes/no recognition memory test, based in verbal directed-forgetting paradigm. For the evaluation of mood symptoms, the most used instruments were: Young Mania Rating Scale (YMRS) for mania symptoms (Fleck et al., 2005; Malhi et al., 2007; Thompson et al., 2007; Barrett et al., 2008; Deckersbach et al., 2008; Dittman et al., 2008; Liu et al., 2010; Bertocci et al., 2011; Levy et al., 2011; Mullin et al., 2012; Gruber et al., 2013; Lee et al., 2013; Miguélez-Pan et al., 2014; Pomarol-Clotet et al., 2014; Bauer et al., 2015; Gvirts et al., 2015; McCormack et al., 2015; Sabater et al., 2016); Hamilton Depression Rating Scales (Fleck et al., 2005; Malhi et al., 2007; Thompson et al., 2007; Barrett et al., 2008; Deckersbach et al., 2008; Dittman et al., 2008; Liu et al., 2010; Bertocci et al., 2011; Mullin et al., 2012; Lee et al., 2013; Miguélez-Pan et al., 2014; Pomarol-Clotet et al., 2014; Russo et al., 2014; Gvirts et al., 2015; Sabater et al., 2016), Montgomery Åsberg Depression Rating Scale (Malhi et al., 2007; Roiser et al., 2009; Liu et al., 2010; Bauer et al., 2015; McCormack et al., 2015), and Beck Depression Inventory (Malhi et al., 2007; Thompson et al., 2007; Drapier et al., 2008; Muhtadie and Johnson, 2015) for depressive symptoms (Table 2).

Methodologies of selected studies were characterized by the comparison of cognitive variables and mood episodes between clinical and control groups. Seven articles carried just a comparison of cognitive profile between groups (Fleck et al., 2005; Thermenos et al., 2010; Levy et al., 2011; Miguélez-Pan et al., 2014; Pomarol-Clotet et al., 2014; McCormack et al., 2015; Sabater et al., 2016). In addition, eight studies evaluated correlation between variables, such as cognitive pattern and brain activation pattern (Liu et al., 2010), gender and cognitive pattern (Barrett et al., 2008), functionality, and cognition (Bauer et al., 2015), homocysteine levels and cognitive performance (Dittman et al., 2008), reaction time and perceptual sensitivity (Fleck et al., 2005), humor and WM (Gruber et al., 2013; Russo et al., 2014), and cognitive pattern (Gvirts et al., 2015). Two of the seven fMRI studies stand out, one described brain activation in different mood phases of BD (Pomarol-Clotet et al., 2014) and the other assessed brain effect of mood induction (Malhi et al., 2007). The article by Muhtadie and Johnson (2015) differ from others because combined evaluation of autonomic variables during WM performance (Table 1).

The results showed different brain patterns activity during the performance of WM tests. Most of the neuroimaging studies used n-back tasks to evaluate WM processing. Studies that chose the n-back paradigm found low left dorsal anterior midcingulate cortex (dAMCC) activation in depressive phase of BD type I group (Bertocci et al., 2011), low dorsolateral prefrontal and parietal cortex activation in depression and manic phases (Pomarol-Clotet et al., 2014), and low activity of dorsolateral prefrontal cortex, dorsal anterior cingulate cortex (dACC) and parietal cortex in euthymic group (Mullin et al., 2012). However, some studies have shown conflicting results, as greater prefrontal cortex activation in BD patients and their relatives (Drapier et al., 2008) as well as increased dorsolateral prefrontal cortex and dAMCC activity in episodes of depressed mood (Deckersbach et al., 2008). Low activations in prefrontal cortex in span tasks studies were also confirmed (Malhi et al., 2007; Liu et al., 2010; Thermenos et al., 2010). Article by Liu et al. (2010) indicated that BD type I subjects tend to show lateralized alterations in the right brain hemisphere, while BD type II presented more distributed deficits. In addition, BD patients and their relatives showed frontopolar cortex and insula brain alterations (Thermenos et al., 2010) and euthymic BD subjects exposed to positive and negative valences reported less activation in medial prefrontal cortex, medial frontal cortex, and right parahippocampal gyros (Malhi et al., 2007).

Regarding the cognitive performance of participants in WM tasks, 11 studies demonstrated lower performance in BD patients compared to control groups. Three studies involving processing of visuospatial information (Fleck et al., 2005; Barrett et al., 2008; Gvirts et al., 2015); six papers concerning auditory verbal processing (Dittman et al., 2008; Levy et al., 2011; Gruber et al., 2013; Lee et al., 2013; Pomarol-Clotet et al., 2014; Sabater et al., 2016); and one article studying updating of visuospatial information (Drapier et al., 2008). Concerning effect sizes reported on such articles we decided to analyze mean effect size on studies that produced lower scores for working memory performance. This descriptive analysis showed an average effect size of 0.39 to studies that involved processing of visuospatial information; and 0.41 to the papers concerning auditory verbal processing. The only paper that used measures of updating visuospatial information presented a effect size of 0.33. We can conclude that all the average effect sizes are low (Cohen, 1988). There are also nine papers (Malhi et al., 2007; Deckersbach et al., 2008; Drapier et al., 2008; Liu et al., 2010; Thermenos et al., 2010; Bertocci et al., 2011; Mullin et al., 2012; Pomarol-Clotet et al., 2014; Muhtadie and Johnson, 2015) that reported a variety of neurobiological changes in BD patients (e.g., prefrontal cortex activation or dorsal anterior cingulate cortex; see Table 3).

However, four studies found no difference between groups (Roiser et al., 2009; Miguélez-Pan et al., 2014; Bauer et al., 2015; McCormack et al., 2015). Article by Pomarol-Clotet et al. (2014) distinguished cognitive profiles according to the mood phases of patients, showing that manic or depressive episodes in BD reported worse performance in WM tasks compared with euthymic and controls groups. However, WM deficits (Thompson et al., 2007; Drapier et al., 2008) and other cognitive low scores can also be observed in euthymic phase, for example, visuospatial WM (Barrett et al., 2008), short-term non-affective memory (Bauer et al., 2015), unsocial cognition (Lee et al., 2013), flexibility, plan implementing, set-shifting (Miguélez-Pan et al., 2014), processing speed, resistance to interference and emotion recognition (Sabater et al., 2016), and verbal fluency (Barrett et al., 2008; Miguélez-Pan et al., 2014; Bauer et al., 2015). In addition, WM performance was associated with cyclothymic levels and irritability in euthymic episodes (Russo et al., 2014; Table 3).

Euthymic patients during conducted span tasks showed deficits in several WM components, such as auditory information updating (Thompson et al., 2007), simultaneous processing of visuospatial information (Barrett et al., 2008), and processing of auditory-verbal information (Thompson et al., 2007; Dittman et al., 2008; Levy et al., 2011; Gruber et al., 2013; Lee et al., 2013). BD type I patients also reported lower scores for updating visuospatial information during the n-back task (Drapier et al., 2008).

Some researches of euthymic samples used auditory-verbal complex span tasks (Miguélez-Pan et al., 2014; Bauer et al., 2015; Sabater et al., 2016). Nevertheless, clinical groups performed similarly to controls in processing auditory information (Miguélez-Pan et al., 2014; Bauer et al., 2015). Article by Roiser et al. (2009) also did not found deficits to perform a visuospatial span task in depressive episodes in BD patients during washout of psychotropic medications.

A prevalence of quantitative cross-sectional methodology experiment designs were characterized in selected articles (Fleck et al., 2005; Malhi et al., 2007; Thompson et al., 2007; Barrett et al., 2008; Dittman et al., 2008; Drapier et al., 2008; Roiser et al., 2009; Liu et al., 2010; Thermenos et al., 2010; Bertocci et al., 2011; Mullin et al., 2012; Gruber et al., 2013; Lee et al., 2013; Pomarol-Clotet et al., 2014; Russo et al., 2014; Bauer et al., 2015; Gvirts et al., 2015; McCormack et al., 2015; Muhtadie and Johnson, 2015; Sabater et al., 2016). Only three studies used other experimental designs, such as cohort study (Levy et al., 2011), descriptive study (Miguélez-Pan et al., 2014), and prospective study (Deckersbach et al., 2008) (Table 1).

The aims of the present systematic review were (i) update the knowledge about working memory deficits in BD; (ii) to analyze if BD patients show WM low scores, contrasting studies that used complex span tasks and n-back tasks, and (iii) to investigate if different BD mood episodes could predict different patterns of cognitive processes based on the WM components of the tasks in each mood phase. Considering the first objective our study found similar results to previous systematic review and meta-analyses pointing to working memory deficits in Bipolar patients (Robinson et al., 2006; Torres et al., 2007; Kurtz and Gerraty, 2009; Bora et al., 2011; Mann-Wrobel et al., 2011). Most of these reviews pointed to verbal working memory deficits in BD (Robinson et al., 2006; Torres et al., 2007; Kurtz and Gerraty, 2009). These findings could be explained by the fact that those reviews considered only digit span measures. In our review, we included both verbal and visuospatial working memory measures. In agreement with previous reviews, we found that both verbal and visuospatial working memory are impaired in BD (Bora et al., 2011; Mann-Wrobel et al., 2011).

Considering objective (ii) and (iii) they were respectively presented in Sections Introduction and Methods. This review seems to be necessary because individual results of studies are confusing, due to the fact that there are three emotional phases in BD (euthymic, euphoric, and depressed) and different WM mental operations (serial recall and update), which are expressed in at least two different modalities (phonological and visuospatial) (Table 4). For this purpose, after the selection of the articles according to the inclusion and exclusion criteria, a systematic review was carried out. Twenty-three studies in the restricted period 2005–2016 were found using different approaches as cognitive, behavioral, neuroimaging, and measures of autonomic responses. The study also convey the possibility of Bipolar Disorder as model for the study of the new component of WM, the Hedonic Detector, for this reason the relationship among WM and other clinical features in the final section.

In conclusion, BD mood episodes were associated with both WM processes (updating and serial recall), cognitive lower performance persist even in remission of symptoms. This evidence suggests that BD patients have deficits in monitoring content in WM. But the data at this point is so different across studies that it does not seem prudent to generalize conclusions. Considering that WM deficit apparently is stage-independent in BD patients (Gruber et al., 2009; Kurtz and Gerraty, 2009), future studies should evaluate the convergence of Damasio's Somatic Marker hypotheses and Baddeley's Hedonic Detector in BD. In other words, how the neutral point is deregulated by the interaction between environmental stimuli and mood episodes and affect WM processes.

Conception or design of the work: CS and FS. Data collection: CS. Data analysis and interpretation: CS and FS. Drafting the article: CS, FS, PA, and LM. Critical revision of the article: FS, PA, and LM. Final approval of the version to be published: CS, FS, PA, and LM. Agreement to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved: CS, FS, PA, and LM.