We re-analyzed data from a study including social and non-social aversive emotional pictures . The original paradigm included exposure to emotional images of negative or positive valence. All emotional-image slides were proceeded either by a red cue, signaling negative affect, or a green, associated with positive pictures. The timing was such that the color cue was displayed 5 s before a 2 s exposure of the social slide, and followed by a 2.5 to 3.5 s black screen with a inter-trial interval of 9 to 11 seconds.
The emotional stimuli, 15 negative and 15 positive pictures, were selected from the International Affective Pictures System (IAPS) . For an example of the paradigm see Additional file 1. We analyzed BOLD responses to socially relevant and irrelevant negative emotional stimuli. Negative stimuli were chosen because PMDD mainly comprises negative emotional symptoms .
Seventeen women with PMDD and 16 asymptomatic controls were recruited through a newspaper advertisement and from women with a PMDD diagnosis.
PMDD was diagnosed according to the definitions in the Diagnostic and Statistical Manual of Mental Disorders IV . Details of the diagnostic procedure have been described previously . Briefly, prospective ratings of daily symptoms using the Cyclicity Diagnoser (CD-scale) were completed to confirm the presence of PMDD and to estimate the severity of PMDD symptoms. The number of days during the 10 days before menses when participants reported a score of 2 or more on the CD-scale for each of the the four core symptoms of PMDD (irritability, depression, anxiety and mood swings) (i.e. a scale 0–40) , and number of days when social interaction was avoided (0 to 10) were used as measures of PMDD severity. The asymptomatic controls were physically healthy women with regular menstrual cycles and no history of premenstrual dysphoric symptoms. None of the controls reported premenstrual dysphoric symptoms on daily ratings. The study was approved by the Ethical Review Board of Uppsala, Sweden, and all participants gave written informed consent.
Exclusion criteria were pregnancy; treatment with hormonal compounds or psychotropic drugs; or presence of any ongoing psychiatric disorder. Absence of other psychiatric disorders was confirmed using the structured psychiatric interview, Mini International Neuropsychiatry Interview . Furthermore, participants with pacemakers, cardiac defibrillators, aneurysm clips, cochlear implants or other implants including magnets, batteries or wires were excluded. One woman with PMDD and one healthy control dropped out after the first scanning session due to personal reasons, and two healthy controls and three women with PMDD were excluded due to movement artifacts (peaks of movement in the x/y/z-axis of more than 3 mm or more than 2 degrees of rotation), or incomplete scanning sessions due to hardware problems. There were no significant differences in demographic or behavioral data between excluded and remaining participants. Fourteen women with PMDD and 13 healthy controls were analyzed.
Timing according to the menstrual cycle
fMRI scanning was performed twice, once in the mid-follicular phase (6 to 12 days after the onset of menstrual bleeding) and once to coincide with the late luteal phase (postovulatory day 8 to 13), according to a positive luteinizing hormone assay (Clearplan, Unipath, Bedford, UK). Monitoring of the luteal phase was confirmed by progesterone serum concentrations and records of the next menstrual bleeding. The study was counterbalanced across the menstrual cycle with half of the participants scanned first in the follicular phase and then in the luteal phase, and the other half scanned in the reverse order.
Blood samples were drawn before each scanning. Estradiol and progesterone serum concentrations were determined by competitive immunometric electrochemistry luminescence detection at the Department of Clinical Chemistry, Uppsala University Hospital. The samples were run on a Roche Cobas e601 with Cobas Elecsys reagent kits (Roche Diagnostics, Bromma, Sweden). The measurement interval was 0.1 to 191 nmol/l for progesterone and 18.4 to 15,781 pmol/l for estradiol. The progesterone intra-assay coefficient of variation was 2.21% at 2.39 nmol/l and 2.82% at 31.56 nmol/l. The estradiol intra-assay coefficient of variation was 6.8% at 85.5 pmol/l and 2.8% at 1,640 pmol/l.
Mood and anxiety scales
Prior to each fMRI scan, participants completed the self-rated version of the Montgomery-Åsberg Depression Rating Scale (MADRS-S)  and the state portion of the Spielberger State-Trait Anxiety Inventory (STAI-S) .
Functional magnetic resonance imaging - scans and paradigm
fMRI was performed using a 3 T whole body scanner (Achieva 3 T X Philips scanner Philips Medical Systems, Best, The Netherlands) equipped with an eight-channel head coil. At the beginning of each scanning session, an anatomical T1-weighted reference data set to a voxel size of 0.8 × 1.0 × 2.0 mm3 and 60 slices were acquired. During stimulus presentations, BOLD imaging was performed using a single shot echo-planar imaging sequence with parameters echo time/repetition time 35/3,000 ms, flip angle 90°, acquisition matrix 76 × 77, acquired voxel size 3.0 × 3.0 × 3.0 mm3 and 30 slices.
The participants lay facing upwards in the scanner with their heads lightly fixated. Visual stimuli were presented through goggles mounted on the head coil (VisualSystem, NordicNeuroLab, Bergen, Norway). The stimulus paradigm was implemented using the commercial software package E-prime (Psychology Software Tools, Sharpsburg, PA, USA). To synchronize the paradigm and the MR sequence, a SyncBox (NordicNeuroLab) was used. The paradigm included 15 negative pictures selected from the IAPS  preceded by a color cue indicating the valence. We compared the eight slides displaying negative social situations (for example, injured humans, abduction of a young female; IAPS: 3320, 2710, 3051, 3160, 6312, 6570, 8230, 9042) with the seven pictures containing negative, but non-social stimuli (for example, snakes, threatening dogs; IAPS: 1050, 1052, 1111, 1201, 1274, 1525, 9620). After scanning, participants again viewed and rated pictures for valence and arousal using the Self-Assessment Manikin used in the IAPS material . Arousal ratings are available in Additional file 2 but are not included here, as we did not test any arousal-related hypotheses. The valence ratings for social and non-social stimuli were analyzed in a Group by Phase analysis of variance, with additional follow-up t-tests.
Functional magnetic resonance imaging - preprocessing and analysis
The Digital Imaging and Communications in Medicine images from the scanner were converted to Neuroimaging Informatics Technology Initiative files using the freeware package MRicron . The data were then analyzed in MatLab (MathWorks, Natick, MA, USA) using SPM5 . The individual BOLD images were realigned to a mean image for the session, slice timed to the middle slice of each whole brain volume, co-registered with the individual anatomic scan, normalized into Montreal Neurological Institute (MNI) coordinates space using normalization parameters obtained from a segmentation into the white matter, grey matter and cerebrospinal fluid of the individual anatomical scan, and smoothing was performed using an 8 mm kernel.
For each individual, BOLD signal changes in the fMRI time series were regressed on to social and non-social negative images. Onsets and durations for stimuli included in the paradigm but not analyzed in the present study (that is, anticipatory periods, positive emotional stimuli) and the six movement parameters obtained in the realignment step were included in the model. Contrast maps were calculated for each individual of the contrast between social and non-social negative images. These contrast maps were then used for group comparisons. Analyses of group differences were first performed to compare women with PMDD and healthy controls during the luteal phase. Regions of interest (ROIs) were generated using the automatic anatomical labeling definitions in the Wake Forest University School of Medicine PickAtlas [41–43] and included the bilateral amygdala, insula and ACC. Then, a ROI defined by the group differences observed in the luteal phase was used for between-group comparisons in the follicular phase and for within-group comparisons between phases. To test the a priori hypothesis of increased reactivity in the amygdala and insula as well as attenuated reactivity in the ACC in PMDD during the luteal phase, an uncorrected p-value of 0.05 with k ≥5, corrected for the search volume of each ROI, was used. Functional couplings during the luteal phase between the amygdala and the insula, respectively, to the ACC, were evaluated with extracted data from the significant clusters, as defined by the between-participants effects in the luteal phase, used as seeds for correlations. These analyses were performed in each group separately. The relatively lenient statistical threshold was deliberately chosen as we restricted analyses only to ROIs where specific hypotheses were advanced. This approach does not only focus on type I errors but also gives a balance between type I and type II errors [44, 45].
Self-reports and affective picture ratings were compared by paired and independent t-tests, respectively. Estradiol and progesterone levels were compared using Mann-Whitney U test and Wilcoxon signed rank tests, respectively. Symptom severity and number of days when social interaction was avoided were evaluated using Student’s t-tests. In addition, partial correlations adjusted for affective ratings were performed between alterations in brain reactivity and change in ovarian steroid hormone levels (follicular to luteal phase) to evaluate if brain activity was tied primarily to changes in hormonal activity or subjective ratings.