Data from two independent cohorts of the Study of Health in Pomerania (SHIP) was used (51). SHIP comprises adult, general population-based samples which are drawn from local registries in the northeast of Germany. The assessments and examinations of the SHIP studies were approved by the institutional ethics review board of the University Greifswald. The data collection of the SHIP studies and all analyses of the present manuscript were performed according to the declaration of Helsinki including written informed consent of all participants.

The first SHIP cohort (SHIP-START-0: N = 4,308) was recruited between 1997 and 2001. Interview and medical examination data as well as laboratory measurements in blood samples of the 10-years-follow-up (SHIP-START-2: 2008–2012; N = 2,333) are included in the present analyses. All participants of SHIP-START-0 still alive in 2006 were re-invited to participate in the Study of Life Events and Gene-Environment-Interaction in Depression (SHIP-LEGEND: 2007–2010; N = 2,400). SHIP-LEGEND was used to assess life events and multiple psychiatric phenotypes in the SHIP-START sample with 1,944 participants assessed in both SHIP-START-2 and SHIP-LEGEND. Psychometric data of SHIP-LEGEND are used in the present analyses. Psychometric data (SHIP-LEGEND) was assessed about 19 months (SD = 14.3 months) before the interviews, medical examinations and blood samples (SHIP-START-2).

A second, independent cohort (SHIP-TREND-0; N = 4,420) was recruited between 2008 and 2012 in the same local area as SHIP-START. Participants of SHIP-START and SHIP-TREND are not overlapping. Interview, medical examination and questionnaire data of SHIP-TREND-0 were used in the present analyses as a second, independent sample.

During a computer-assisted face-to-face interview, sociodemographic variables were assessed including sex and age. Cigarette smoking was assessed as either non-smoker, ex-smoker or current smoker; current and ex-smokers were summarized as ever-smokers for the present analyses. For ever-smokers, pack years were calculated according to Baumeister et al. (52). Alcohol consumption was quantified by the mean intake of ethanol in grams per day, averaged over the past 30 days (53). According to Baumeister et al. (53), risky alcohol consumption was defined as alcohol consumption of >20 g/d for women and >30 g/d for men. Participants were asked to bring their drug prescription sheet or medication containers and to report all medication used during the past week. Compounds were classified according to the Anatomical Therapeutic Chemical (ATC) classification (54).

Body height was measured to the nearest 1 cm. Body weight was measured in light close to the nearest 100 g. Waist circumference was measured to the nearest 1 mm with an inelastic tape midway between the lower rib margin and the iliac crest while the participant was standing plain on both feet. As a proxy for central body fat, the waist-height ratio was calculated (55). The body mass index (BMI) was used to define obesity with a BMI ≥30 classified as obese. Blood pressure (BP) was measured three times via a BP cuff on the upper arm. While the participant was sitting calmly, the first measurement was conducted after a five-minute rest; the second and third measurement was each conducted with a three-minute delay. The mean of the second and third BP measurements was used for statistical purposes. Hypertension was defined if the participant reported the intake of antihypertensive medication (ATC C02, ATC C03, ATC C07, ATC C08, ATC C09 or self-report) or if a systolic BP ≥140 mmHg or a diastolic BP ≥90 mmHg was measured.

Single-point blood samples were taken from the cubital vein and stored at −80°C in the Integrated Research Biobank of the University Medicine Greifswald (56). The exact time of blood sampling was recorded. Participants of SHIP-TREND-0 but not participants of SHIP-START-2 were asked to fast before blood sampling. Fasting time was calculated in both cohorts by quantifying the time difference between the time of blood sampling and the time of the last reported caloric intake.

Processing of the serum samples to measure the cortisol concentrations is described elsewhere (57). Briefly, serum samples were prepared and frozen at −80°C directly after blood sampling. Samples were thawed and processed before the measurements. To quantify the serum cortisol concentrations, an immunoassay with low-cross reactivity was used on the AVIDA Centaur XP System (Siemens Healthcare Diagnostics, Eschborn, Germany). The coefficients of variation observed during the course of the measurements were low for both analyzed cohorts (30, 57).

White blood cell count (WBC) was measured either on the XT2000, XE 5000 or SE9000 analyzers from Sysmex (Sysmex Deutschland GmbH, Norderstedt, Germany) or on the Advia 2120i (Siemens Healthcare Diagnostics, Eschborn, Germany). Glycated hemoglobin (HbA1c) concentrations were quantified by high-performance liquid chromatography (Bio-Rad Diamat, Munich, Germany). Triglycerides were determined enzymatically (Dimension VISTA, Siemens Healthcare Diagnostics GmbH, Eschborn, Germany). High-density lipoprotein cholesterol (HDL-C) was measured enzymatically after the preparation with phosphotungstic acid/MgCl2.

All assays were performed by skilled personnel according to the manufacturer's instructions.

CM was assessed via the Childhood Trauma Questionnaire (CTQ) (58). Based on 28 items, five subscales are assessed: emotional, physical and sexual abuse as well as emotional and physical neglect. Each item is answered on a five-point scale ranging from “never true” to “very often true”. A summary score operationalizes the severity of CM overall (range: 25–125) with higher values indicating more severe CM. Summary scores of the subscales can be categorized into “none”, “mild”, “moderate” or “severe”, often used to compare “none/mild” vs. “moderate/severe” CM (58, 59). Accordingly, dichotomous abuse and neglect scores were calculated as indicators of the absence or presence of CM for the present manuscript. The dichotomous scores thus extend the severity information given by the CTQ summary score.

To assess subjective health complaints, the Subjective Health Complaints (SHC) questionnaire was used (60). All symptoms are rated on a four-point scale. Based on the results of a factor analysis, 30 symptoms are assigned to eight clusters (60). According to Klinger-König et al. (61), the clusters anxiety/depression and exhaustion were subsumed as mental complaints and the remaining six clusters were subsumed as somatic complaints. For both categories the mental (range: 13–52) and the somatic SHC (range: 17–68) a summary score was calculated. For both scores, higher values indicate more severe health complaints.

Depressive symptoms were differently assessed in SHIP-TREND-0 and SHIP-START-2. Both questionnaires assessed the severity of depressive symptoms during the past two weeks. In SHIP-TREND-0, the depression module of the Patient Health Questionnaire was used (PHQ-9) (62). The PHQ-9 uses nine items that mimic the symptoms used as MDD diagnostic criteria in the Diagnostic and Statistical Manual of Mental Disorder - 4th Version (DSM-IV) (63). The experience of each symptom is rated on a four-point scale from “not at all” to “almost every day”. A summary score is calculated (range: 0–27) with higher values indicating more severe depressive symptoms. In SHIP-START-2, the Beck Depression Inventory-II (BDI-II) was used (64). The BDI-II uses 21 items to assess depressive symptoms. Each item is rated on a four-point scale formulated individually for each item. A summary score is calculated (range: 0–63) with higher values indicating more severe depressive symptoms.

The selection of the analytic sample is presented in Supplementary Figure S1. Participants were excluded from analyses if data sets were incomplete for serum cortisol concentrations, the CTQ summary score, abuse and neglect, age, sex, fasting time, time of blood sampling, HbA1c and WBC (SHIP-TREND-0: N = 326; SHIP-START-2: N = 469). Further, participants were excluded if reporting the intake of corticosteroid medication (ATC H01, ATC H02, ATC R03; SHIP-TREND-0: N = 211; SHIP-START-2: N = 115) or sexual hormones or hormonal contraception (ATC G03, ATC G02B; SHIP-TREND-0: N = 366; SHIP-START-2: N = 109). Thus, the analytic sample comprised 3,517 participants for SHIP-TREND-0 and 1,640 participants for SHIP-START-2. Descriptive statistics of these analytic samples are provided in Table 1 and the results section. Descriptive statistics of covariates comparing the initial study samples with the analytic samples of both cohorts are presented in Supplementary Table S1.

Statistical analyses were calculated with R 4.2.1 (65). For descriptive purposes, continuous variables are reported in mean (M) and standard deviation (SD); dichotomous variables are reported in percentage (%). All analyses were calculated separately for both cohorts in the first step and afterwards integrated via meta-analyses. Only meta-analytic results were interpreted.

In mediation analyses, serum cortisol concentrations were used as the outcome variable and the CTQ summary score, abuse and neglect were used as predictors. As potential mediators, the following variables were used: systolic BP, hypertension, WHtR, obesity, triglycerides, HDL-C, depressive symptoms, mental and somatic SHC, alcohol consumption, risky alcohol consumption, pack years and ever smoking. Association matrices of the mediators in SHIP-TREND-0 and SHIP-START-2 are presented in Supplementary Table S2. Associations between CM and serum cortisol concentrations, between CM and the potential mediator as well as between the potential mediators and serum cortisol concentrations are preconditioned to analyze statistical mediation effects. Hence, these associations were checked before calculating the mediation models. Highly skewed variables (skewness >|1|: CTQ summary score, systolic BP, triglycerides, depressive symptoms, alcohol consumption, pack years) were log-transformed (66). Log-transformation resulted in a distribution closer to normal. Specific sets of covariates were considered for each analysis as described below.

Associations between the CM variables and all potential mediators are provided in Table 2. Significant associations with the CTQ summary score were observed for WHtR, HDL-C, depressive symptoms, both SHC scores, alcohol consumption and pack years (β = −.037 – .260, all p_FDR< .05). All associations were positive, except for HDL-C and alcohol consumption. For both abuse and neglect, associations were observed with depressive symptoms, both SHC scores and alcohol consumption (abuse: β = −.125 – .650, all p_FDR< .05; neglect: β = −.153 – .439, all p_FDR< .001). A positive association was observed between abuse and pack years (β = .195, p_FDR= .012). For neglect, a positive association was found with WHtR (β = .094, p_FDR= .045).

All potential mediators were significantly associated with serum cortisol concentrations (β = −.180 – .273, all p_FDR < .05), except for systolic BP. For test statistics and FDR-adjusted p-values see Table 3. Positive associations were observed with hypertension, triglycerides, HDL-C and both alcohol consumption variables. Inverse associations were observed with all CM variables, WHtR, obesity, depressive symptoms, both SHC scores, pack years and ever smoking.

Accordingly, the following variables were included as potential mediators of the associations between CM and serum cortisol concentrations: (1) depressive symptoms, mental SHC, somatic SHC and alcohol consumption were tested for all CM variables, (2) WHtR for the CTQ summary score and neglect, (3) pack years for the CTQ summary score and abuse and (4) HDL-C for the CTQ summary score.

All 30 symptoms of the SHC were associated with the CTQ summary score (β = .065 – .219, all p_FDR< .001) and abuse (β = .130 – .590, all p_FDR< .05). For neglect, a significant association was observed for all symptoms (β = .092 – .419, all p_FDR< .05), except for suffocated feeling (β = .088, p_FDR= .066). Nevertheless, only eleven symptoms were associated with serum cortisol concentrations. These items covered four symptom clusters: exhaustion, difficulty breathing, pain and disturbances of sensations. The results of the associations between the SHC symptoms and serum cortisol concentrations are presented in Figure 3. As suffocated feeling was not associated with serum cortisol concentrations, all eleven symptoms significantly associated with serum cortisol concentrations were tested as potential mediators of the associations with the CTQ summary score, abuse and neglect, respectively.

The results of the separate mediation analyses are presented in Supplementary Table S3. The impact of the CTQ summary score on serum cortisol concentrations was partly mediated by all symptoms (5.86–13.72% mediated, all p_FDR< .05). The impact of abuse on serum cortisol concentrations was partly mediated by the three symptoms of exhaustion (13.49–14.69% mediated, all p_FDR= .012) and shortness of breath (8.65% mediated, p_FDR= .028). However, no indirect effect was observed for headache/facial pain, joint pain/pain in the limbs and numbness. A similar structure of the results was found for neglect: The association between neglect and serum cortisol concentrations was partly mediated by the three symptoms of exhaustion (8.65%–11.01% mediated, all p_FDR= .038) and shortness of breath (6.53% mediated, p_FDR= .049). No indirect effect was observed for neck/shoulder pain, headache/facial pain, joint pain/pain in the limbs, heaviness/tiredness in the legs and numbness.

Conclusively, our analyses were based on a general population sample with a rather low CM and symptom severity. We used two large and independent cohorts and integrated the results via meta-analysis which makes the overall results statistically more robust. The meta-analytic approach also enabled us to overcome differences between the samples, e.g., different fasting times. Nevertheless, the tested associations might be more prone in patient samples. Additionally, future studies should aim at longitudinal instead of cross-sectional data to support the mediation direction not only by theory but also via data structure.

Our analyses used basal serum cortisol concentrations as the outcome. Thus, the results are focused rather on the basic functioning of the HPA axis than on stress reactivity. Moreover, we used single-point cortisol measurements. Repeated individual cortisol measurements would allow to assess diurnal slopes or calculate the area under the curve and thus enable the use of a more robust outcomes. Nevertheless, the measured serum cortisol concentrations were validated in a multi-omics study including cortisol-associated transcriptome concentrations (82). Here, the exact time of blood sampling was included as a covariate and modeled non-linear to adjust for the circadian rhythm of cortisol secretion. Further, we excluded participants with drug use potentially inferring with the cortisol secretion.

The mediators included in the present analyses interact with each other. Thus, a more complex mediation model might be needed to display the pathways. We aimed to fit a wide range of behaviors, disease symptoms and diseases potentially mediating the association between CM and basal cortisol concentrations. We focused on previously well-described variables and our analyses provide a first overview integrating these potential mediators in one study. Although the selection of potential mediators was theory driven, the description of suitable mediators was based on p-values and rather exploratory. This approach enabled to extend previous research by integrating and comparing a wide set of potential mediators as well as to narrow this set down. However, focusing on a narrower set would ease the integration of multiple mediators in more complex models.

Mental symptoms were represented by mental SHC and depressive symptoms. Anxiety symptoms or anxiety disorders were not included in the present study as these were insufficiently assessed. Besides, depressive symptoms and SHC were assessed by self-report questionnaires. Interviews and, validated diagnoses particularly for mental disorders, could improve the data validity. Further, similar mediation effects should be addressed with other mental symptoms and disorders in future studies. As Ju et al. (35) reported a mediation by dysfunctional attitudes, integrating personality traits as a mediator might also be promising.

Finally, although the used questionnaire limits CM to childhood and adolescents, no exact time of CM was assessed. Thus, a potential difference between early or late CM could not be evaluated. However, including CM-related timing information in future studies and differentiating between types of CM could extend our presented results.

The present study investigated a wide range of potential mediators for the association between CM and basal cortisol concentrations. Our analyses assessed several mediators described in multiple studies before, but followed a more integrative approach. In our two large general population-based samples, associations between the potential mediators and CM were largely absent. Thus, CM seem to add up to the effects of both physical and mental symptoms and should be addressed in both psychiatric and somatic medicine. Yet, mediation effects of depressive symptoms and somatic SHC were robust. Hence, especially in psychiatric medicine, the physiological impact of mental diseases should be addressed in the light of childhood maltreatment. Finally, our results are a first step towards more complex models representing the nuanced structure between the tested mediators.