All participants gave written informed consent after complete description of the study and prior to their inclusion. The study protocol was approved by the ethical committee of the University of Ulm and was conducted according to the principles expressed in the Declaration of Helsinki.

Eighteen unmedicated patients with recurrent major depressive disorder and seventeen healthy control participants took part in the study (demographics in Table 1). Patients were recruited from the outpatient department of a psychotherapeutic institute and diagnosed by two trained clinicians (ST and HeK) using the Structured Clinical Interviews I and II for DMS-IV Diagnosis (German version; [15]). Patients reported between 1 and 15 depressive episodes (M[SD] = 5.6[5.5]) and their age at first occurrence of depression was between 8 and 40 years (M[SD] = 19.3[8.2]). Patients had received various types of medication and psychotherapy prior to consulting the aforementioned institute. However, none of the patients had received treatment within at least 6 months prior to inclusion in the study. Exclusion criteria were other psychiatric conditions as main diagnosis, substance abuse, significant medical or neurological conditions (including medical causes of depression), psychotropic medication, and eye problems. Control participants were recruited from the community, matched for age, sex and education and had no history of previous depressive episodes or other psychiatric conditions (SCID). All participants were right-handed. In both groups, depression severity and general symptoms of psychopathology were assessed using the Beck Depression Inventory (BDI, [16]) and the revised Symptom Check List (SCL-90-R, [17]), respectively.

To assemble individualized and personally relevant stimuli that related to depressive symptoms, an OPD interview (Operationalized Psychodynamic Diagnosis) [14] was conducted with each patient and each control participant. OPD is a multiaxial system assessing psychopathology on several levels [18]. Beyond a pure description of symptoms (Axis V), it includes experience of illness (Axis I), dysfunctional interpersonal relations (Axis II), psychodynamic conflicts (Axis III) and psychological structure (Axis IV) [14]. Although OPD is at its core a psychodynamic approach, dysfunctional relations (Axis II) are considered by most therapeutic schools to be important in the development and maintenance of depression [19], [20]. The OPD interviews were conducted by a trained clinician (HeK) and videotaped. Dysfunctional relations were rated independently by 2–3 expert raters blind to the status of the interviewees. Although not suffering from depression, control participants also experienced dysfunctional relations which reflected a major interpersonal source of distress. From the systematic and item-based diagnosis [14] four sentences were identified representing the core dysfunctional relationship theme of each person (e.g. “You wish to be accepted by others.”, “Therefore you do a lot for them.”, “That is often too close for them, so they retreat.”, “Then you feel empty and lonely.”). It is important to notice that these sentences do not represent “depression” or “negative mood” in general, but intentionally point to a significant and specific aspect of each individual's depression – its development and/or maintenance. These individual sentences served as stimuli during the fMRI-session (OPD condition). Word count and semantic structure of the stimulus sentences (i.e. distribution of the thirty-two items assigned) did not differ between patients and controls.

The control condition was termed “traffic” and comprised four sentences describing stressful traffic situations. Participants were instructed to recall a stressful traffic situation they had experienced whilst reading the “traffic” sentences. The rationale behind this control condition was to induce negative emotions and recall autobiographical memories with a personally relevant situation including human interactions, but without engaging in material that might interfere with participants' depression or interpersonal distress.

In order to separate the two conditions (OPD and traffic), and allow subjects the opportunity to recover after emotionally demanding sentences, “relaxation” sentences were inserted between conditions. These sentences instructed participants to relax by thinking of a safe place. Subjects were prepared for the “relaxation” condition before the experiment.

Whereas the OPD sentences were derived individually for each person, “relaxation” and “traffic” were the same sentences across all subjects. OPD sentences were slightly but significantly longer (M[SD] = 49.8[9.1] characters) than “traffic” sentences (43.5 characters, p<.001). There was, however, no significant difference in length between the OPD sentences for patients and controls. All sentences were presented in German.

The four sentences of each condition (OPD, traffic, relaxation) were individually presented for 7.5 seconds while subjects were in the scanner. During the OPD block participants were asked to mentally engage in situations with significant others, as described by the OPD sentences. Subjects received no instruction to regulate their emotions, but were instructed to allow spontaneous thoughts, emotions and memories come to mind. According to the logic of the OPD Axis II, the four sentences comprising the dysfunctional interpersonal relation form one complex that should activate a specific and disorder-related mental representation [13]. Therefore, the four sentences were modelled as a whole in fMRI analyses. “Traffic” and “relaxation” conditions also comprised four sentences each lasting 7.5 seconds. The instructions were to mentally engage either in the recalled traffic situation or to relax. In total, 12 “relaxation”, 6 “traffic” and 6 “OPD” blocks were presented. Blocks were separated by a 5-seconds fixation cross. The entire experiment lasted approximately 15 minutes.

Four to six weeks prior to fMRI assessment, participants were interviewed (SCID I+II, OPD), completed questionnaires (BDI, SCL-90-R) and gave written consent to participation. At the beginning of the fMRI session and prior to scanning, subjects were presented with their individual OPD sentences and asked whether the sentences adequately represented their problematic relations. To control for state affectivity, all participants filled out the German version of the Positive and Negative Affect Schedule (PANAS; [21], [22]) before entering the scanner. After scanning, a second PANAS was completed together with a questionnaire assessing on a 7-point Likert scale the extent to which the OPD sentences were correct and caused emotional arousal.

MRI data were recorded (DW and PE) using a 3-T SIEMENS Magnetom Allegra head scanner (Siemens, Erlangen, Germany). Subjects were positioned on the scanner couch and wore foam earplugs to reduce scanner noise. An experienced psychotherapist (ST or HeK) assisted with the setup procedure and spoke to the patients both prior to and after the experiment. Further, the therapist explicitly asked the subjects whether they were fully awake and ready to continue in the break between the scanning sessions. Data acquisition started with anatomical images (3D high resolution T1-weighted isotropic volume, MPRAGE-sequence (MPRAGE = Magnetization Prepared Rapid Gradient Echo; [23]); TR = 2.3 s, FOV = 256×256×176 mm, TE = 4.38 ms, TI = 900 ms, flip angle = 8°, 1 mm isovoxel, total acquisition time 14.45 min). Functional scans were performed using a single shot echo planar imaging sequence (EPI). A total of 365 T2*-weighted whole brain volumes were acquired (EPI-sequence; TR 2500 ms, TE 30 ms, flip angle 90°, FOV 192 mm, matrix 64×64, 44 slices, slice thickness 3 mm, interleaved acquisition order, AC-PC- Orientation, total acquisition time: 15.18 min).

Data were analyzed and visualized using Brain Voyager QX 1.10 (Brain Innovation, Maastricht, Netherlands). Preprocessing: Functional data were slice-time corrected, motion parameters were estimated, and motion was corrected relative to the first volume of the run. To remove low frequency drifts, data were high-pass filtered (3 cycles, three sine waves fall within the extent of the data). Structural and functional data were transformed into the standard space of Talairach and Tournoux [24], data points were labeled using Talairach Daemon [25]. The design matrix was modeled using the two gamma hemodynamic response function. Functional data were smoothed using an 8 mm full width at half maximum (FWHM) isotropic Gaussian Kernel.

Statistics: Group data were analyzed using random effects analyses based on z-transformed functional data. An ANOVA, including the within-factor CONDITION (OPD vs. traffic sentences) and between-factor GROUP (patient vs. control) was performed to identify differences in hemodynamic response. Separate brain maps were generated for the main effect CONDITION and GROUP and for the interaction CONDITION x GROUP. The main effect of CONDITON is displayed as a t-statistic, which yields the same results as the F-statistic, but allows color-coding the direction of changes. Motion-correction parameters were included in the GLM-Model. Maps are shown with a threshold of p<0.001. Correction for multiple comparisons for the within-factor CONDITION was based on False Discovery Rate (FDR) [26]. However, literature suggests differences between controls and patients in relatively small cortical and subcortical regions [3], but FDR is very strict for small active areas. Thus, the between-factor GROUP and the interaction are reported on p<.001 (uncorrected). For all reported comparisons, the likelihood of Type I error was reduced based on cluster size threshold estimation [27], [28] involving a Monte Carlo simulation calculating the likelihood to obtain different cluster sizes. Calculations resulted in a cluster size threshold of 16 voxels. Active voxels are displayed in native resolution without interpolation and plotted on the Talairach-transformed brain.

Table 1 shows behavioral data for patients and controls. Patients had significantly higher depression scores (BDI, Table 1; Figure 1) and general symptoms of psychopathology (GSI-scale of the SCL 90-R). Both groups judged the OPD sentences to be adequate descriptions of their dysfunctional interpersonal relations. After the fMRI session, all participants reported that the OPD sentences caused emotional arousal in the scanner. There were no significant differences between groups in terms of adequacy or arousal induced by the OPD sentences (Table 1; Figure 1). Patients with depression had significantly higher levels of negative affect as measured by the PANAS, and there was a tendency toward reduced negative affect after completion of the fMRI session. This tendency was seen in both groups (main effect GROUP: F(1,29) = 16.38; p <.001; main effect Pre/Post fMRI; F(1,29) = 3.35; p<.077, interaction n.s.). There were no differences between groups in positive affect as measured by the PANAS prior and after the fMRI session (PANAS positive affect, main effect GROUP, Pre/Post fMRI and interaction n.s.). See Table 1 for mean values. PANAS ratings in both groups were comparable to normative data obtained from a large group of healthy subjects under stress-free conditions [21].

The main effect CONDITION, displayed as a t-contrast, identified regions with a stronger signal for OPD relative to traffic sentences. These regions were located in the occipital cortex, in the superior parietal lobe, the superior frontal gyrus, in the anterior cingulate cortex (ACC), and in the medial frontal gyrus. Conversely, a stronger signal for Traffic relative to OPD sentences was observed in a cluster including parts of the superior and middle frontal gyrus (Table 2; Figure 2).

A significant group by condition interaction was found in a variety of regions, including the inferior frontal gyrus, the postcentral gyrus, the amygdala, the precentral and middle frontal gyrus, and the basal ganglia. In general, the hemodynamic response pattern can be described as a signal increase for patients when confronted with OPD relative to Traffic sentences. In contrast, controls show a signal decrease for this comparison (Table 2; Figure 3).

Central conclusions are derived from the GROUP X CONDITION interaction reported above. In order to establish whether the traffic condition itself might yield differences between groups, a “TRAFFIC minus RELAX” comparison was performed between groups. This comparison addressed the potential confound that would be introduced should patients with depression be shown to exhibit stronger responses in limbic areas in the traffic condition - potentially due to a negative bias in handling emotional stimuli in this condition. However, the comparison revealed no active clusters, showing no differences between patients and controls in the traffic situation. In a similar vein, groups were compared based on the difference “OPD minus RELAX”. OPD relative to RELAX sentences led to stronger BOLD responses in seven clusters, including the amygdala, the inferior/middle frontal gyrus, the right and left postcentral gyrus, showing a large overlap with areas that have been reported for the interaction GROUP X CONDITION.

Several limitations of this study deserve mention. Firstly, while the use of OPD sentences as the main stimuli is the genuine strength of this study, it is also the major source of potential confounds. It is impossible to know what subjects are actually thinking of when instructed to mentally engage in the problematic interpersonal relation depicted by the OPD sentences. If one follows the logic of the OPD system and interpersonal theories of depression, the stimuli are highly specific and directly related to a significant factor contributing to the development and maintenance of depression. On the other hand, they are less controllable in terms of what reactions they produce in subjects than standardized and widely-used stimuli such as IAPS pictures [41]. A further possible confound lies in the traffic-related sentences as a control condition. It has been reported that patients with depression show a negative bias in the evaluation of emotional stimuli, which could lead them to react to unspecific traffic-related stress with enhanced hemodynamic responses in brain areas involved in the processing of (negative) emotions (e.g. limbic structures). In order to ensure that the limbic responses found in the interaction GROUP X CONDITION were caused by the OPD sentences themselves we conducted additional comparisons demonstrating that the contrast “TRAFFIC minus RELAX” yielded no differences between patients and controls and further, that the contrast “OPD minus RELAX” indicated hemodynamic activity in amygdala and inferior/middle frontal gyrus in the patient group. This suggests that the interaction effect GROUP X CONDITION is not driven by differences between groups in the TRAFFIC condition. Finally, another potential problem may lie in the standardized style of the TRAFFIC sentences. Thus, it could be the case that brain responses reflect the difference between personalized and general stimuli. This limitation lies in the study design and could not be ruled out by additional analyses.

The present fMRI study describes clear differences in hemodynamic responses between patients with depression and non-depressed control participants using personalized stimuli in a highly standardized fashion, thus supporting the model of limbic hyperactivity in depression. The stronger response in the amygdala and basal ganglia found for OPD sentences in patients could indicate particular involvement of these structures in the processing of clinically derived and personally relevant material. Increased responses in cortical midline structures when confronted with problematic interpersonal sentences suggests that our novel experimental design engaged both groups of participants in self-referential and emotional processing.