This study was carried out in accordance with the approval and recommendations of the Hamilton Integrated Research Ethics Board with written informed consent from all subjects. All subjects gave written informed consent in accordance with the Declaration of Helsinki. Participants were recruited through community advertisements in the Hamilton and Halton Regions, ON, Canada.

Eighty-five women between 16 and 45 years of age, with regular menstrual cycles (25–32 days) were enrolled. Participants were split into four groups based on their history of BD and PMDD: (1) healthy controls with no history of PMDD (CTRL); (2) women with PMDD but no other psychiatric diagnosis (PMDD); (3) BD with no history of PMDD (BD); and (4) BD with comorbid PMDD (BDPMDD).

Exclusion criteria included: (i) current or recent (previous 3 months) use of systemic hormonal treatment; (ii) pregnancy; (iii) contraindications for MRI; (iv) history of head trauma with loss of consciousness; (v) neurological disorders; (vi) current or recent (previous 6 months) history of alcohol or drug abuse or dependence; (vii) unstable general medical conditions. Regularly cycling women using levonorgestrel intrauterine device were allowed in the study due to its primarily localized hormonal effect. All women performed at least 2 months of prospective symptom charting using the Daily Record of Severity of Problems (DRSP) in order to confirm PMDD diagnosis, as per DSM-5 criteria (36). Study participants were informed that recreational drug use was not allowed during the course of study participation and all participants agreed to comply with this request. We did not perform drug testing on the day of the MRI scans.

Inclusion criteria for participants with BD (both BD and BDPMDD groups) included: (i) a diagnosis of BD according to the SCID-I (SCID-IV-TR) (37); (ii) no current depressive, manic, or hypomanic episodes according to the SCID-I; (iii) no changes in psychotropic medications or mood state within at least 2 months prior to and during the study. Due to the exceptionally high rates of comorbid psychiatric conditions in BD, lifetime but not current history of psychiatric comorbidities was allowed to provide a true reflection of most individuals with BD. Similarly, due to the high rates of comorbid psychiatric conditions in PMDD, a lifetime history of a single major depressive episode, past history of generalized anxiety disorder, or posttraumatic disorder was allowed so long as the individual was fully remitted for a minimum of 6 months prior to study entry.

Exclusion criteria for participants in the CTRL group included: (i) a lifetime history of any psychiatric disorder according to the SCID-I and (ii) greater than a 30% change in the four core symptoms of PMDD in their late luteal phase from their mid-follicular phase according to the DRSP (36). Women with BD without PMDD (BD group) also did not display greater than a 30% change in the four core symptoms of PMDD in their late luteal phase from their mid-follicular phase according to the DRSP. A diagnosis of PMDD (both in PMDD and BDPMDD groups) was established by two independent licensed psychiatrists blinded by subjects’ group status (Benicio N. Frey, Luciano Minuzzi) using 2 months of prospective daily symptom charting as confirmed by the DRSP (36).

Study participation was comprised of three visits to St. Joseph’s Healthcare Hamilton. The first visit consisted of administration of the SCID-I by a psychiatrist or trained Ph.D. candidate, followed by a psychiatric and gynecological history. The second and third visits took place during the mid-follicular phase (days 5–10 of the menstrual cycle) or the late luteal phase of the menstrual cycle (7 days preceding menses). Approximately half of the study participants began with their mid-follicular visit and half with their late luteal visit. Menstrual cycle phase was confirmed through prospective charting and hormonal analysis. Visits two and three included an MRI scan, collection of a blood sample for hormone analysis, and completion of validated clinical questionnaires as detailed below.

The Mongomery-Asberg Depression Rating Scale (MADRS) (38) and Hamilton Depression Rating Scale (HAMD) (39) were administered to assess severity of depressive symptoms. The Young Mania Rating Scale (YMRS) (40) was administered to assess the severity of manic/hypomanic symptoms through the menstrual cycle. Self-reported disruptions in biological rhythms were investigated using the Biological Rhythms Interview of Assessment in Neuropsychiatry (BRIAN) (41). The Pittsburgh Sleep Quality Index (PSQI) was used to asses sleep quality (42) and the State Trait Anxiety Inventory (STAI) was used to identify state and trait-based anxiety (43).

Statistical analysis of demographic and clinical variables was completed using R (Version 3.1.2¹). Shapiro–Wilks tests and Bartlett’s test were used to evaluate normal distribution of clinical scales and homogeneity of variances between groups, respectively. Differences in demographic variables and clinical scales were analyzed using Kruskal–Wallis or Analysis of Variance (ANOVA). Pairwise between-group differences were ascertained using parametric and nonparametric Dunnett Tests using BDPMDD as the reference group. A p-value of < 0.05 was considered significant.

Immediately following both MRI scans, 10 ml of whole blood was collected in serum tubes to confirm menstrual phase. The blood was clotted at room temperature for 45 min, and centrifuged at 20°C for 15 min at 3,000 rpm. Four serum aliquots were obtained and frozen at −80°C until assayed. Serum was assayed for progesterone (P4), 17-β-estradiol (E2), and DHEAS using commercially prepared solid phase enzyme-linked immunosorbent assay (ELISA) kits from ALPCO Diagnostics, Salem, NH, USA. Samples were also assayed for allopregnanolone (ALLO) using ELISA kit from Kiamiya Biomedical Company, Seattle, WA, USA. All serum samples were assayed in duplicate following the manufacturer’s protocol, with a fresh aliquot for each analyte. The inter-assay variations for P4, E2, DHEAS, and ALLO were 11.3, 8.7, 9.2, and 6.0%, respectively. The intra-assay variations were 10.4, 7.7, 9.3, and 11.7% and the sensitivities were 0.1 ng/ml, 10 pg/ml, 0.005 µg/ml, and 0.52 ng/ml, respectively. A licensed gynecologist (Dustin Costescu) confirmed that hormone levels were within physiological range for each menstrual phase.

Participants in the BDPMDD group had a significantly higher BMI than CTRL (p = 0.001) and PMDD (p = 0.009) groups. As a result, BMI was added as a covariate in all imaging analyses. Groups were matched by age (p = 0.054) and years of education (p = 0.09). Neither medication load (p = 0.86), age of onset (p = 0.81), or number of comorbid conditions (p = 0.94) were different between the BD and the BDPMDD groups (Table 1).

Consistent with research suggesting that comorbidity with PMDD is associated with an increased illness burden for women with BD (12, 18), we found a stepwise progression of severity of symptoms from the CTRL group to the BDPMDD group (Table 2). Notably, even though participants in both the BD and BDPMDD groups were clinically euthymic for a minimum of 2 months prior to entering and during the study, the BDPMDD group had significant higher MADRS (p = 0.037) and BRIAN scores (p = 0.007) during the follicular phase of the menstrual cycle. Hormonal levels were all within normal range for each menstrual phase, and there no between-group differences except for higher ALLO levels in the PMDD group (Table 3).

We investigated whole brain differences in Rs-FC using a seed-to-voxel approach. We found significant group differences in Rs-FC particularly in the late luteal phase using the right and left hippocampus as seed points (Table 4, Figure 1). Specifically, we found differences in functional connectivity between the right hippocampus and the right premotor cortex (BA6) (k = 166; x = +64, y = +02, z = +06; p = 0.044, FDR-corrected). In addition, we found the greatest number of clusters when using the left hippocampus as a seed point, which yielded three significant clusters: (i) right somatosensory association cortex (BA5) (p = 0.037, FDR-corrected); (ii) right frontal cortex (BA8) (p = 0.037, FDR-corrected); (iii) right somatosensory cortex (BA1) (p = 0.033, FDR-corrected). Pairwise analyses revealed decreased connectivity between the right hippocampus and the left premotor cortex (p = 0.029, FDR-corrected) in BDPMDD compared to BD, and increased connectivity in BDPMDD compared to BD between the left hippocampus and the right frontal cortex (p = 0.048, FDR-corrected) in the late luteal phase (Figure 1).

We found that MADRS scores were positively correlated with Rs-FC between the right hippocampus and left premotor cortex (r = 0.46, p = 0.037) in BD but not BDPMDD. Further, STAI-state scores were positively correlated with clusters between the left hippocampus and right dlPFC [BA9] (r = 0.54, p = 0.017) and right dorsal frontal cortex [BA8] (r = 0.46, p = 0.048) in BDPMDD but not PMDD.

Moreover, we found increased functional connectivity between the right hippocampus and clusters in the left frontal cortex, right dorsolateral prefrontal cortex, and left dorsolateral prefrontal cortex in BDPMDD compared to PMDD. Further, using the same seed point decreased functional connectivity was found with bilateral clusters in the right and left primary motor cortex in BDPMDD compared to PMDD. Decreased functional connectivity between the left hippocampus and a cluster in the left somatosensory cortex was also found in BDPMDD compared to PMDD (Table 4).

No differences in Rs-FC were found using the following seed points: (i) right and left vlPFC; (ii) right and left vmPFC; (iii) right and left amygdala; and (iv) right and left postcentral gyrus. We tested and we did not find any significant effect of medication load, psychosis, or BMI on any of the Rs-FC results.

We identified six clusters representing significant differences in cortical thickness between BD and BDPMDD groups (Table 5, Figure 2). Compared to BD, BDPMDD women displayed cortical thinning in the following regions: (i) left pericalcarine gyrus (p_CORR = 0.007); (ii) left superior parietal gyrus (p_CORR = 0.0024); (iii) right middle temporal gyrus (p_CORR = 0.0002); (iv) right rostral middle frontal gyrus (p_CORR = 0.0006); and (v) left superior frontal gyrus (p_CORR = 0.04). Further, we found increased thickness in BDPMDD compared to BD in the left superior temporal gyrus (p_CORR = 0.045).

We found seven regions of cortical thinning in BDPMDD compared to controls: (i) left insula; (ii) right middle temporal gyrus; (iii) right medial orbitofrontal gyrus; (iv) left pars triangularis; (v) left rostral middle frontal gyrus; (vi) right cuneus; and (vii) right superior frontal gyrus (all p_CORR < 0.05). Compared to PMDD the BDPMDD group had increased cortical thickness in the (i) left superior temporal gyrus; (ii) right pars orbitalis; (iii) left lingual gyrus; and (iv) right superior parietal gyrus and decreased cortical thickness in the (i) right medial orbitofrontal gyrus and (ii) right inferior parietal gyrus (all p_CORR < 0.05) (Table 5, Figure 2). There were no within-group differences in cortical thickness across the mid-follicular and late luteal menstrual phases.

We investigated the volume of subcortical regions of interest between groups in each menstrual phase, including right and left caudate; left and right putamen; left and right hippocampus; left and right amygdala; left and right thalamus; left and right accumbens area; and left and right ventral diencephalon. We found significant between-group differences in the left caudate during the late luteal menstrual phase. Post hoc analyses revealed that these results were driven by increased volume in the left caudate in BDPMDD compared to BD (p_CORR = 0.007). We did not find any other differences in the volume of other regions of interest in either menstrual phase between groups.

We found that the BDPMDD group reported significantly higher scores than all of the other groups in most of the clinical questionnaires. This is of particular importance as in our study all women with a diagnosis of BD were clinically euthymic for a minimum of 2 months before study entry. The presence of subthreshold depressive and anxious symptoms, as well as disturbances in sleep and biological rhythms during the mid-follicular phase may render these women more vulnerable to relapse. This is supported by literature that highlights that the severity of circadian rhythm disruptions and subthreshold symptoms are associated with higher risk of relapse (57–62). For instance, a systematic review by Jackson et al. found that disruptions in sleep were the most common prodrome of mania and sixth of bipolar depression (63). In addition, a greater severity of subthreshold symptoms has also been associated with poorer psychosocial functioning (58). This may highlight susceptibility of women with BD and comorbid PMDD to develop more frequent mood episodes and contribute to their burden of illness. These findings support a study by Dias et al., which found that women with BD and premenstrual mood worsening represent a phenotype of BD that is more symptomatic and relapse prone (12). This is also consistent with our recent study showing that women with BD and comorbid PMDD had a larger number psychiatric comorbidities, higher rates of rapid cycling, and increased number of hypo/manic and depressive episodes (18).

We identified differences in Rs-FC between BD and BDPMDD groups using a seed-to-voxel analysis using the right and left hippocampus as seed points in the late luteal phase. Aberrant prefrontal-limbic circuitry is well documented in BD; many of the brain regions in this circuitry are also influenced by hormonal fluctuations and implicated in the pathophysiology of PMDD—the hippocampus as a primary example. The hippocampus is central to the trait-based pathology of BD and to the regulation of memory (encoding and retrieval) and emotional processes (21, 24, 64, 65). It also is a prominent site of E2 and 5HT receptors (31, 66), has high expression of BDNF (67) and plays an inhibitory role in stress response (68) and HPA axis regulation (20). Interestingly, it has been postulated that sex hormones interact with the HPA axis to facilitate the response to chronic stress (69). On the other hand, stress hormones may downregulate the action of sex hormones, by exerting modulatory effects on the reproductive neuroendocrine axis (19). Abnormalities in the HPA axis are also present in patients with BD and may have important consequences on the neurobiology of BD (20). The activity of the HPA axis is modulated by the hippocampi, prefrontal, and orbitofrontal cortices, ACC and amygdala—all of which have been implicated in the pathophysiology of BD (20, 24). Thus, considering that the women enrolled in our study were clinically stable, it is conceivable that the differences functional connectivity that we observed in the right and left hippocampi of women with comorbid BD and PMDD may be either an attempt from the brain to regulate the network involved in stress response, or a trait abnormality that may render these women more vulnerable to develop future episodes. Longitudinal studies are required to test these hypotheses. Literature also suggest that progesterone-derived neurosteroids may play an important role in mediating neuronal plasticity neural survival and neurogenesis in the hippocampus (67). Future neuroimaging studies correlating sex hormones with brain structure and function in BD and PMDD would be useful.

Structural changes have been observed in the hippocampus across the menstrual cycle as studies in healthy women report both decreased gray matter in the anterior hippocampus and increased volume in the bilateral hippocampi during the follicular phase (70, 71). fMRI studies in healthy women highlight increased functional coupling at rest between the bilateral hippocampi and bilateral superior parietal lobe in the late vs. early follicular phase, and decreased activity of the hippocampus in the luteal phase with response to emotional faces (72). In this context, the increased functional coupling between the left hippocampus and right frontal cortex in BDPMDD compared to controls may suggest (i) compensatory changes in functional connectivity resulting from the late luteal menstrual phase in women during the “hormonal sensitivity” period (premenstrual) and/or (ii) a functional trait marker of the added weight of PMDD on the pathophysiology of BD.

Interestingly, we found decreased coupling between the right hippocampus and the left premotor cortex in the late luteal phase in BDPMDD compared to BD. Previous studies using PET imaging have reported increased activity of the premotor cortex during “sadness” inducing paradigms in remitted but not depressed patients with BD (73), but decreased activity in patients with major depressive disorder (74). Further, a recent study reported that decreased connectivity of the somatomotor network, including the premotor cortex (as seen in the BDPMDD group), is present in bipolar depression and may represent aberrantly slow flow of inner time—a common feature of depression (75). Therefore, we hypothesize that this pattern of functional decoupling in BD and PMDD vs. BDPMDD may highlight the impact of comorbid PMDD in the late luteal phase in women with BD. Together, these findings suggest that, premenstrually, women with BD and comorbid PMDD may display changes in Rs-FC that mirror the ones observed in bipolar depression.

Compared to PMDD, increased functional connectivity between the right hippocampus and left frontal and bilateral dlPFC, and decreased functional connectivity between the right hippocampus and bilateral primary motor cortex, and left hippocampus and left somatosensory cortex in BDPMDD, further highlights involvement of somatomotor and executive control networks in the pathology of BD. These patterns of functional connectivity may reflect activity that is abnormally enhanced between limbic and frontal regions, and impaired between limbic and somatosensory and motor regions which play a role in mediating affective regulation and risk-taking behavior, both of which are affected in BD (76). Thus, aberrant patterns of functional connectivity between BDPMDD and PMDD in the late luteal phase may reflect functional differences related to the pathology of BD, and/or the pathology of comorbid PMDD and BD on brain functioning compared to PMDD. These findings also emphasize the need for more research exploring the impact of the endogenous hormones on women with BD.

We found decreased thickness in several frontal and temporal gyri central to the default mode and cognitive networks in BDPMDD compared to CTRL, PMDD and BD. The only exceptions were the left superior temporal gyrus, which was thicker in individuals with BDPMDD compared to BD and PMDD, and the right pars orbitalis, left lingual, right superior parietal gyri compared to the PMDD group. A systematic review of cortical thickness studies in BD reported decreases in thickness in the bilateral superior frontal gyri, superior parietal gyrus, middle temporal gyrus and pericalcarine gyri compared to controls (77). As these regions are thinner in individuals in the BDPMDD group than BD and CTRL groups, this suggests, again, that comorbid PMDD may impose a worse impact on brain structure in women with BD. This also may help explain the affective lability reported in our clinical data and in literature in women with BD and comorbid PMDD. This was further reinforced as women in the BDPMDD group had increased gray matter volume of the caudate nucleus compared to the BD group. The caudate nucleus facilitates cross talk between prefrontal–cortical networks and subserves behavioral adaptations required for achievement of complex goals (23, 78). Studies investigating caudate volume between BD and controls have reported mixed findings with no differences (79, 80) and decreased volume compared to controls (81). Decreased volume of the caudate has also been reported in women with major depressive disorder (82), this may suggest that women with comorbid BD and PMDD may display structural and functional brain changes that mirror aspects of depression. Moreover, our seed-to-voxel analysis results show increased prefrontal-limbic activity, which neurobiological models of BD postulate may be mediated by both the dorsal and ventral prefrontal–cortical networks involving the striatum. When taken in conjunction with our Rs-FC findings, increased caudate volume in BDPMDD could also potentially reflect abnormal activity of prefrontal–cortical circuitry central to the trait-based pathology of BD, which seems to be exacerbated in the late luteal phase.

The limitations of our study deserve attention. First, the DRSP is a self-administered tool used to chart symptoms of PMS across the menstrual cycle (36). It is possible that women may provide an inaccurate account of their premenstrual symptoms or that use of the DRSP may be confounded by stressful life events. In both groups with a diagnosis of BD (BD and BDPMDD groups), symptoms reported on the DRSP may also be confounded by exacerbation of their bipolar illness. However, the use of mood, sleep, and biological rhythms questionnaires in both menstrual phases increased our confidence in self-reported results.

Second, Rs-fMRI provides an indirect measure of spontaneous neuronal activity in the ultralow frequency range (0.01–0.10 Hz) (83). The fMRI techniques used in this paper, seed-to-voxel and ROI-ROI are based on an oversimplification that BOLD activation measured is static through the duration of the scanning paradigm (84). Further, results may be confounded by the participant’s ability to remain awake for the duration of the scan, and inability to control the participant’s memory in the scanner. Although we advised participants not to think about anything in particular, and remain awake with their eyes fixed at the fixation point throughout the entire resting-state brain scan, no objective measures, such as simultaneous electroencephalogram or eye tracking, were in place to confirm that participants followed our instructions. Third, the effects of psychotropic medications on Rs-FC are a common limitation of fMRI research in BD (85). Studies investigating Rs-FC in BD, including our current and previous work (86), have largely ruled out the influence of medication following correlation analysis with BOLD activation (87–89). It would be almost impossible to recruit a sizeable sample of BD subjects in sustained clinical remission without treatment. The primary goal of this study was to investigate brain structure and function of comorbid BD and PMDD. We believe this goal could have only been achieved with a primarily medicated sample, by ensuring that all participants with BD were euthymic throughout the study (37). Another limitation of our sample was that it comprised both women with BD type-I and BD type-II. We encourage future work on differences in Rs-FC between women with BD type-I and BD type-II. Finally, it is important to note that our study was cross-sectional and as a result this study only reflects a current picture of disease state and not disease progression. Longitudinal studies in this area are required to provide a picture of disease burden and progression in women with BD and comorbid PMDD.

In conclusion, we found differences in the thickness of the cerebral cortex in regions critical to emotional regulation and cognition, as well as volumetric enlargement of the left caudate in the late luteal phase of participants in the BDPMDD group compared to BD group. Further, results from Rs-FC analysis highlight differences in brain regions dense in E2 and 5HT receptors, such as the right and left hippocampus and the temporal cortex, which may be liable to hormonal influence. Such differences in brain structure and function may help explaining the worse course of illness and clinical profile in women with these comorbid conditions. When taken in the context of other literature on this population, our results suggest that women with BD and comorbid PMDD display a distinct clinical and neurobiological phenotype of BD involving sensitivity to endogenous hormones.