This study was a part of the ‘Neurobiology of Personality’ project at the University of Turku and University of Helsinki (Finland). The same dataset has been used also previously in Tuominen’s et al. studies [22,23,24]. The participants for this study (n = 39) were selected from the prospective population-based Young Finns Study (YFS) that started in 1980 (n = 3596 in the baseline study, born in 1962, 1965, 1968, 1971, 1974, 1977). The original sampling of YFS was designed to include a population-based sample of non-institutionalized Finnish children, representative with regard to Eastern vs. Western regions in Finland, sex (female vs. male), and rural vs. urban environment.

For the present sub-study, we utilized data that were originally collected to examine the role of brain opioid system in the temperament trait Harm Avoidance [22]. The participants were selected from the YFS sample on the basis of their Harm Avoidance scores (HA, a scale of the Temperament and Character Inventory). In the current study, we invited all the participants with low/high HA who could be matched with each other with regard to age, sex, and educational level. Further details about the sampling can be found elsewhere [22, 23, 25].

All participants were screened to be healthy on the basis of blood and urine screening, medical examinations and interviews, MRI, and ECG examination. An extensive urine drug screen was conducted, and alcohol use was assessed with the AUDIT questionnaire and blood laboratory tests. Participants were allowed to have past affective disorders, but none of them fulfilled diagnostic criteria at the time of PET imaging (six participants had mild/moderate depressive or anxiety symptoms). According to the Hamilton Depression Rating Scale (HAM-D-17), all participants had scores <10. Participants with regular smoking were excluded because smoking is known to affect [11C]carfentanil binding potential [26].

The study was conducted in accordance with the Declaration of Helsinki. The original design of the YFS was approved by the ethical committees of all the Finnish universities with medical schools. Further, the current study protocol was approved by the Joint Ethical Committee of the University of Turku and the Turku University Central Hospital. All the participants gave a written informed consent before participation.

Childhood family environment was assessed with three scores: (1) stress-prone childhood events, (2) disadvantageous emotional family atmosphere, and (3) adverse socioeconomic environment. The risk scores have been used and validated previously [27, 28]. All the childhood environmental characteristics were assessed with questionnaires presented for the parents in 1980.

The score of adverse socioeconomic environment included parents’ low occupational status (1 = highest parental occupational status was manual worker, 0 = at least one parent had lower- or higher-level non-manual occupation), parents’ low educational level (1 = parents’ highest completed education was comprehensive school, 0 = at least one parent had completed high school, occupational school, or academic level), low family income in relation to family size (1 = at least 1 SD below the sample average, 0 = other values), unstable employment situation (1 = at least one parent was unemployed or in a long-term sick leave, 0 = other employment situations), and over-crowded apartment (1 = number of rooms at home in relation to family size was at least 1 SD below the sample average, 0 = other values). We calculated a total score of adverse socioeconomic environment and dichotomized it (1 = at least one socioeconomic risk factor, 0 = no socioeconomic risk factors).

The score of stress-prone childhood events included the following factors: change of residence (at least once), change of school (at least once), parental divorce, mother’s or father’s death, mother’s or father’s hospitalization (for at least one day), and child’s absence from school due to sickness (at least 11 days during past 12 months). Each event was dichotomously encoded (0 = not occurred, 1 = occurred). We calculated a sum score of the stressful life events and classified it as 0 (no stress-prone life events) or 1 (at least one stress-prone life event).

The score of disadvantageous emotional family atmosphere included mother’s or father’s mental disorder (0 = none of the parents had mental disorder, 1 = at least one parent had mental disorder), mother’s or father’s frequent alcohol intoxication (0 = parents reported alcohol intoxication at most once a week, 1 = at least one parent reported alcohol intoxication at least two times a week), emotional distance between the child and parent (0 = emotional closeness between the parent and child, 1 = emotional detachment between the parent and child), parental intolerance toward the child (0 = no parental intolerance toward the child, 1 = parental intolerance toward the child), and parental life dissatisfaction (0 = no parental dissatisfaction, 1 = parental dissatisfaction at least one life sector). We calculated a total score of disadvantageous emotional family atmosphere in childhood and dichotomized it (0 = no emotional risk factors, 1 = at least one emotional risk factor).

More specifically, emotional distance between the parent and child was evaluated with a four-item questionnaire (e.g., “The child is emotionally important for me”, ”I can fulfill myself with the child”). The items were responded with a 5-point scale (e.g., 1 = little, 5 = much). Emotional detachment between the parent and child was defined to be present if the parent responded to at least one item with either of the two most unfavorable response alternatives. Parental intolerance toward the child was evaluated with a three-item scale (“I get nervous when spending time with the child”, “The child is a burden in challenging situations”, “The child consumes my time too much”). The items were responded with a 5-point scale (1 = frequently, 5 = never). Parental intolerance toward the child was defined to be present if the parent responded to at least one item with “frequently” or “quite frequently”. The items measuring parenting attitudes have been used also previously [29, 30].

Parental life satisfaction was assessed with a three-item questionnaire measuring parent’s satisfaction in three life sectors: as a parent, spouse, and employee. The items were responded with a 5-point scale (1 = satisfied, 5 = dissatisfied). This questionnaire has been adapted from the Operation Family Study questionnaire [31] and has been used also previously [32, 33]. Parental dissatisfaction was defined to be present if parent reported being “dissatisfied” or “quite dissatisfied” in at least one of the life sectors.

We conducted additional sensitivity analyses using Harm Avoidance and adult attachment style as covariates since they are previously found to associate with μ-opioid receptor availability [22, 24].

Harm Avoidance was measured in 2012 with the Harm Avoidance scale of the Temperament and Character Inventory (TCI) [34]. The scale includes 35 items that are responded with a 5-point scale (1 = totally disagree; 5 = totally agree). The internal consistency of the scale was good (Cronbach’s alpha = 0.96). We calculated a sum score of the items for all the participants who had responded to at least 50% of the items.

Adult attachment style was assessed using the Adult Attachment Interview (AAI) [35] that was further encoded using the Dynamic Maturational Model (DMM) [36]. The DMM has been widely used also previously [24, 37, 38]. The assessment of adult attachment style was conducted by an experienced AAI interpreter (A. H.) with qualified training on both AAI and DMM. Adult attachment style was categorized into three categories: avoidant (n = 15), ambivalent (n = 19), secure attachment (n = 15). More details about the assessment of adult attachment can be found elsewhere [24].

All participants underwent a PET scan with μ–opioid receptor tracer [11C]carfentanil as described in detail in [22]. [11C]carfentanil (dose 423.6 ± 73.9 MBq; mass 1.08 ± 0.84 μg) was injected as an intravenous bolus. A brain‐dedicated high‐resolution PET scanner (ECAT HRRT, Siemens Medical Solutions) was used for PET imaging collect emission data for 69 min using 16 frames (3 × 1 min, 4 × 3 min, and 9 × 6 min). During the PET scans, head of the subject was fixed using an individually molded thermoplastic mask. A T1‐weighted MRI scan with 1 × 1 × 1 mm3 resolution voxel size was obtained from each subject using Philips Gyroscan Intera 1.5 T CV Nova Dual MRI scanner to exclude structural abnormalities and for anatomical reference.

PET images were preprocessed using the automated PET data processing pipeline Magia [39] (https://github.com/tkkarjal/magia) running on MATLAB (The MathWorks, Inc., Natick, Massachusetts, United States). PET data was first corrected for motion by realigning the frames of each scan. Radiotracer binding was quantified using non-displaceable binding potential (BPND), which is the ratio of specific binding to non-displaceable binding in the tissue [40]. The BPND is taken here as an estimate for number of target receptor/transporter available for tracer binding (receptor availability). Binding potential was calculated applying basis function method for each voxel using the simplified reference tissue model [41], with occipital cortex serving as the reference region. The parametric images were spatially normalized to MNI-space via segmentation and normalization of T1-weighted anatomical images, and finally smoothed with an 8-mm full-width half maximum Gaussian kernel.

The data were analyzed by averaging BPND’s within regions of interest (ROIs). Atlas-based ROIs were generated in the brain regions rich with μ–opioid receptors (amygdala, hippocampus, ventral striatum, dorsal caudate, thalamus, insula, prefrontal cortex, orbitofrontal cortex, and anterior cingulate cortex using AAL [42] and Anatomy [43] toolboxes.

Data were analyzed with STATA MP 16.0 statistical software with general linear models. Mean regional [11C]carfentanil was extracted for each region and was predicted separately by three dichotomous variables: 1) stress-prone life events (0 = no stressful life events, 1 = at least one stressful life event), 2) disadvantageous emotional family atmosphere (0 = no disadvantageous emotional factors, 1 = at least one disadvantageous emotional factor in childhood environment), and 3) adverse socioeconomic environment (0 = no unfavorable socioeconomic factors, 1 = at least one unfavorable socioeconomic factor). Since age and sex affect [11C]carfentanil BPND [44], they were used as covariates. In further sensitivity analyses, we used Harm Avoidance and adult attachment style as covariates because Harm Avoidance and adult attachment style are found to correlate with μ-opioid receptor availability in the same dataset [22, 24]. We reported both uncorrected p-values and false discovery (FDR) corrected p-values (Benjamini-Hochberg procedure) [45].