Effect of lithium on circadian rhythm in bipolar disorder: A systematic review and meta‐analysis

Circadian rhythm disruption is commonly reported in patients with bipolar disorder. Lithium has been suggested to have effects on the circadian clock, the biological basis of the circadian rhythm. The objective of the current review was to review systematically the existing studies on the effect of lithium on circadian rhythm in patients with bipolar disorder.

questionnaires, core body temperature, dim light melatonin onset and cortisol secretion profiles are the most commonly used measurements in practice. 2 Circadian rhythm disruption has been well-documented in bipolar disorder. 3 Studies using actigraphy, using a sensor worn on the nondominant wrist of participants, have shown that patients with bipolar disorder have a more variable and less stable circadian rest-activity rhythm. 4,5 Chronotype is the individual preference for the daily timing of activity, a behavioural manifestation of circadian rhythm. 6 Morningness-Eveningness, also known as chronotype, was considered to be trait-like construct. 7 However, a recent study showed its state-like aspects. 8 Individuals who show more morningness have more alertness in the morning and a preference for morning activity.
Eveningness, a general preference for evening activity and delayed sleep-wake phase were commonly noted in patients with bipolar disorder, 3,9,10 which is consistent with findings in studies that patients with bipolar disorder showed delayed evening melatonin secretion. 11,12 Lithium is a first-line treatment for bipolar disorder, and it is widely used worldwide. 13,14 However, its mechanism of action remains unknown. A systematic review examined the impact of lithium on circadian rhythm in cells, animal models and human, suggested that the chronobiological effects of lithium might be an inherent component of its therapeutic effect. 15 The majority of included studies recruited BD patients are candidate gene association studies. We were interested in whether the circadian effects of lithium in laboratory settings translated to clinical context. In this systematic review we sought to update the previous review and focus on the effect of lithium on circadian rhythm only in people with bipolar disorder.

| ME THODS
This review is reported in accordance with PRISMA guideline 16 and the study protocol was registered on PROSPERO (CRD42018109790).

| Search strategy
We searched electronic databases including Cochrane Library, EMBASE, MEDLINE, PsycINFO, clinicaltrials.gov website and PubMed for published and unpublished trials from the inception of the databases until Sep 17st 2020 (see Appendix for details). No restrictions of language were imposed. The reference lists of included studies and systematic reviews in the field were screened for additional eligible studies.

| Participants
We included patients with a primary diagnosis of bipolar disorder according to standardized diagnostic criteria from the International Classification of Disease (ICD)-9, ICD-10 or the Diagnostic and Statistical Manual of Mental Disorders (DSM)-III, DSM-III-R,   DSM-IV, DSM-IV-TR and DSM-5. There were no restrictions by age,   gender, country, phase of illness, subtype (type I, type II, rapid cy-cling, not otherwise specified) and clinical setting.

| Interventions
Studies were included if lithium was administered orally within the therapeutic dose range 0.4 −1.2 mmol/L, either as monotherapy or adjunct therapy. 17

| Comparators
A placebo or comparator intervention (either pharmacological or non-pharmacological) was required for inclusion.

| Types of studies
We included both experimental studies (whether randomized or non-randomized) and observational studies where the primary outcome was a measure of circadian rhythm (see the section below for details).

| Primary outcomes
1. Circadian rest-activity was measured by actigraphy, considering the following ten measures: (i) Intradaily variability (IV), (ii) Interdaily stability (IS), (iii) the average activity during the most active 10-h period (M10), (iv) the average activity during the least active 5-h period (L5), (v) M10 onset time, (vi) L5 onset time, (vii) Relative amplitude. The following three parameters were generated by Cosinor rhythmometry analyses, an approach that fit a sinusoidal wave with a period of 24 h to the actigraphy 18 data. (viii) Acrophase, time at which the maximal activity occurs (ix) Mesor, rhythm-adjusted mean of the sinusoidal wave (x) Amplitude, the difference between the peak and the mean value of a wave. 19 2.2.6 | Secondary outcomes Study quality was also assessed using the AXIS tool for crosssectional studies, 23 the Cochrane Risk of Bias Tool for randomized controlled trials 24 and the Newcastle-Ottawa Quality Assessment Scale (NOS) 25 for cohort studies.

| Data synthesis and statistical analysis
If two or more studies reported about the same outcome, we conducted meta-analyses using Review Manager (RevMan) Version 5.3 (Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration, 2014). For continuous outcomes, standardized mean differences (SMD) were calculated with 95% confidence interval (CI), which enabled comparison to be made among studies using different scales to measure the same outcome. Heterogeneity among studies was tested using I 2 statistic. 26 Expecting substantial heterogeneity among studies, we decided to pool the data using random-effects model.

| Description of included studies
All included studies were published from 2014 to 2018. Four of the included studies were observational studies (three cross-sectional studies [29][30][31] and one cohort study 32 ) and the fifth was an 8-week, non-blind randomized controlled trial. 28 Full description of the characteristics of included studies is reported in Table 1.
All studies included mixed samples of bipolar I disorder (BD I, 10%-57%) and bipolar II disorder (BD II, 42-90%). In three studies all patients were euthymic [30][31][32] ; the other two studies recruited patients who were experiencing a current mood episode. 28,29 Four studies recruited from outpatient settings and a fifth from both outpatient and inpatient settings. 28 Five studies had between 29 and 525 participants, and only one study recruited more than 100 participants. 30 Hwang et al and Benizri et al compared lithium monotherapy with another mood stabilizer, 28,32 whereas the other three studies compared two groups with multiple medications, one took lithium, the other did not. 29-31

| Quality assessment
The quality of the included studies ranged between moderate and low. The randomized control trial had a high risk of bias for blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data and other sources of bias like sponsorship bias (see Table 2 in the Appendix). The cohort study reported unclear information about the follow-up (Appendix Table 3) and in all three cross-sectional studies there were limited measures undertaken to address and categorize non-responders which were susceptible to reporting bias (see Appendix Table 4).

| Circadian rest-activity
Only two studies evaluated the impact of lithium on circadian restactivity using actigraphy. In the randomized controlled trial comparing lithium and quetiapine, acrophase, the authors reported that amplitude and mesor did not exhibit significant changes from baseline in the lithium group after 8 weeks treatment, 28 but they did not report the data, so their findings are not replicable and could not be incorporated in the meta-analysis. By contrast, the second study, a cohort study with 21 days of follow-up, found patients treated with lithium had significantly larger amplitude (SMD 0.68, 95% CI: However, no differences on M10 onset time, L5 onset time were detected between two groups. 32

| Sleep quality
Polysomnography (PSG) is widely considered as the gold standard of measuring sleep objectively. 20 All included studies adopted the Pittsburgh Sleep Quality Index (PSQI) rather than PSG. 27,30 The randomized controlled trial did not detect any significant changes on PSQI scores at week 1, 2, 4, 6 and 8 compared with baseline. In addition, no significant changes were found on actigraphy measured sleep parameters including sleep latency, sleep efficiency, and wakefulness after sleep onset (WASO) in the lithium treatment group which was consistent with the subjective measurements. 27 Similarly, there were no significant differences in PSQI scores found between patients received lithium monotherapy and anticonvulsant monotherapy (SMD 0.06, 95% CI: −0.60 to 0.71). 32 In contrast, a cross-sectional study with more than 100 patients in each arm found that lithium use was significantly associated with the lower PSQI scores in women who had BDI (−23% [−37% to −7%]), whereas such association was not noted in BD II or men who had BD I. 30

| Chronotype
Four observational studies reported data about chronotype and we meta-analysed the results. [29][30][31][32] Three of included studies adopted CSM scale, 30-32 the other one used the MEQ questionnaire. 29 Benizri et al., a cohort study with 21-day actigraphy recording, measured chronotype using CSM questionnaire at one point in time. 32 Thus, we included it in the pooled analysis. Kanagarajan et al. recruited both patients with current mood episodes and euthymic patients, whereas the other included studies recruited only euthymic patients. No differences in CSM total scores were detected between these two groups of patients with different mood states. 29 The meta-analysis provided non-significant evidence (p = 0.08) that lithium is associated with higher level of morning-   (BALM) for chronotype analysis, found that lithium-responsive BD patients showed higher levels of morningness compared with lithium non-responders. 33 Fibroblasts cells from lithium-responsive bipolar patients showed a short period of the cellular circadian rhythm. 33,34 A higher level of morningness was found to be associated with a shorter period. 35 With a similarly long period, phase advance was observed in lithium-responsive cells in comparison with lithium-nonresponsive cells. 33 These results also suggested that chronotype at baseline can be a potential candidate for the predictors of lithium responsiveness.
The second interpretation of this meta-analysis is that the phase advance or an increased level of morningness might be an early sign of the treatment effect of lithium. Morning chronotype was associated with fewer mood symptoms and past suicide attempt. 33 Another study showed that shifts towards morningness were associated with reductions in depression, increases in positive affect and improvements in sleep quality in military veterans receiving brief behavioural treatment for insomnia. 8 It is unclear if lithium had an impact on chronotype independent of changes in mood symptoms.
Finally, the meta-analysis result indicates that increasing the level of morningness may be part of the mechanism of action of lithium.
Lithium showed a period-shortening effect in fibroblast cells from lithium-responsive BD patients. 33 The shorter period was associated with an increased level of morningness. 35 However, the translation of the preliminary findings in bipolar patient fibroblast cells to bipolar patients was incomplete.
All these interpretations of the meta-analysis prompt further investigation in research studies. To elucidate the role of chronotype or the phase changes of the circadian rhythm in the treatment of lithium, whether it serves as part of the mechanism of action or an early signal of a good response or an underlying circadian phenotype associated with lithium responsiveness. We need further investigation with rigorous study design. Therefore, randomized controlled trials with larger sample of BD patients treated with lithium, better control for mood symptoms and sleep quality are required to allow causal inferences to be made.
Sleep dysregulation is well-documented in bipolar disorder. 36 We found that the effects of lithium on sleep quality were mixed. In one included study, Geoffroy et al. found significantly lower PSQI scores in patients with BD I women, but not men, receiving lithium treatment. 30 We are not confident about the validity of these subgroup analysis which were not pre-planned. Any subgroup difference arises from post-hoc rather than a priori hypothesis undermines its credibility. 37 In Kim et al. study, 27 PSQI scores in lithium treatment group was significantly lower than quetiapine group at baseline. We do not know whether the failure of detection of any significant changes in sleep quality in lithium group is caused by their comparatively better sleep quality at baseline.
There are limited studies explored the effect of lithium on circadian rest-activity pattern by using actigraphy. Benziri et al. showed that BD patients received lithium treatment was associated with a larger amplitude of circadian rest activity rhythm compared with anticonvulsants treatment. 32 The effect of lithium on amplitude has been reported in skin fibroblast cultures from healthy participants and BD patients.
A rhythmically expressed, bioluminescent circadian rhythm reporter gene, Per2::luc, was used to indicate circadian rhythm changes. 38 Lithium had an impact on increasing the amplitude of circadian rhythm in fibroblast cultures from healthy participants but failed in BD patients, which may cause by differences in calcium channel and extracellular regulated kinase (ERK) signalling. [38][39][40] Three of five included studies only adopted subjective measures like validated scales. Compared with other physiological markers of circadian rhythm, self-reported scales are more practical and costeffective in large sample studies. However, its external validity has F I G U R E 2 Morningness as a continuous measure on a validated chronotype rating scale in observational studies comparing lithium with other mood stabilizers except for lithium. CI, confidence interval; I 2 , percentage of the variability in effect estimates due to heterogeneity; SD, standard deviation been questioned. Some studies reported that BD participants underestimated their sleep durations using questionnaires and diaries. 41,42 Thus, it is advisable to use both subjective and objective measurements to assess the circadian rhythm in BD patients. 43 Actigraphy, as one of the objective measurements, has increasingly been used in assessing circadian rhythm in mood disorder. 36,44 Actigraphy provides continuous objective measurement of movement over time, which could provide more details of circadian rhythm changes, improve assessment by reducing recall bias. 45 Thus, actigraphy might be a useful tool to further explore the effect of lithium on circadian rhythm.
Mounting evidence supports that lithium has impact on circadian clock. 15 participants prospectively for two years after the initiation of lithium treatment, using actigraphy to monitor daily circadian rest-activity variability. It will also monitor core BD symptoms and illness activity daily by using ecological momentary assessment (EMA) and assess medication adherence over time. 50

| Limitations
The main limitation in our systematic review is the lack of the available high-quality evidence to inform the question whether lithium has an impact on the circadian rhythm. Only five studies published between 2014 and 2018 were included in this study.

| CON CLUS IONS
There is a growing interest in the effect of lithium on circadian rhythm.
Our meta-analysis suggests a possible association between lithium and shifts towards morningness in BD patients. Chronotype could be a potential target of further exploration of biomarkers or biosignatures of lithium treatment. Our systematic review also highlights the paucity of high-quality evidence to elucidate the effect of lithium on circadian rhythm in patients with bipolar disorder. The field requires more high-quality studies where the primary outcome was a measure of circadian rhythm adopting prospective, longitudinal study design, using actigraphy to monitor daily changes of circadian rest activity.

ACK N OWLED G EM ENTS
We thank the following authors of studies for sharing the additional

E TH I C A L A PPROVA L
Ethical approval was not needed for this project because it is a systematic review of aggregate data from existing primary studies.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available from the corresponding author upon reasonable request.