Participants

Patients were recruited from the specialised out-patient clinic for risk-taking and self-harming behaviour (AtR!Sk; Ambulanz für Risikoverhaltensweisen und Selbstschädigung ^{Reference Kaess, Ghinea, Fischer-Waldschmidt and Resch24} at the Clinic for Child and Adolescent Psychiatry, Center for Psychosocial Medicine, University of Heidelberg, Germany). Healthy controls were recruited via public advertisement (e.g. recruitment at public places) and matched to the patient group by age. As self-harming behaviours differ between genders,^{Reference Andover, Primack, Gibb and Pepper25} only females were included to this analysis. Previous data from the study, focusing on behavioural and psychophysiological outcomes, were published earlier.^{Reference Koenig, Lischke, Bardtke, Heinze, Kröller and Pahnke26} The recruitment period lasted from July 2016 until May 2018. Inclusion criteria for the patient group were at least five incidents of NSSI during the past 12 months, age between 13 and 17 years, and female gender. Inclusion criteria for the healthy control group were age between 13 and 17 years and female gender. General exclusion criteria were deficient language skills or clinically relevant impairments in intelligence, glucocorticoid medication intake, pregnancy, any underlying neurological or endocrinological diseases, acute psychosis, acute suicidality, substance dependency and a body mass index (BMI) <17.5 kg/m² or >30 kg/m². An additional exclusion criterion for the patient group was current BPD drug treatment. A further exclusion criterion for participants of the control group was any current and former psychiatric disorder or lifetime NSSI. After completion of the study, every participant received an allowance of €40 for study participation.

For the NSSI group, 180 (100%) adolescents were screened and 37 (20.56%) were included in the study. Reasons for study exclusion are provided (see Supplementary Table 1 available at https://doi.org/10.1192/bjo.2024.728). During study participation, seven participants in the NSSI group dropped out of the study: one (0.56%) reported acute headaches, two (1.11%) showed dissociative symptoms that required interruption of experimental procedures, three (1.67%) did not show up for the second appointment and one (0.56%) dropped out because of further reasons not documented in detail. This resulted in a final sample of n = 30 (16.67%) for the NSSI group. A total of 66 (100%) female adolescents were screened as healthy controls for study participation. Of those, 31 (46.97%) were included in the study. One (1.52%) refused to participate in the stress task and hence dropped out of the study. Finally, one (1.52%) had to be excluded from NIRS analyses because of technical issues during the assessment. This resulted in a sample of n = 29 healthy controls (43.94%).

The study comprised two appointments. The first appointment included an extensive clinical characterisation of participants via interviews and self-reports. Appointment two comprised the actual stress induction experiment. Written informed consent was provided by participants and their caregivers before inclusion in the study. The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008. All procedures involving human patients were approved by the ethical committee of the University of Heidelberg (study identifier: S-685/2015).

Procedures

At the first appointment, patients completed a diagnostic interview and answered self-report questionnaires. Diagnostic tools that are relevant for the present manuscript are presented here. Further instruments are reported elsewhere.^{Reference Koenig, Lischke, Bardtke, Heinze, Kröller and Pahnke26} For all diagnostic tools, a validated German translation was used. During the interview, demographic data (e.g. school form, age, height, weight) were queried. Suicidality and NSSI behaviour were assessed with the Self-Injurious Thoughts and Behaviors Interview (SITBI-G),^{Reference Fischer, Ameis, Parzer, Plener, Groschwitz and Vonderlin27} BPD criteria were assessed with the Structured Clinical Interview for the DSM-IV (SCID-II).^{Reference Wittchen, Zaudig and Fydrich28} The semi-structured Mini-International Neuropsychiatric Interview for children and adolescents (M.I.N.I.-Kid)^{Reference Sheehan, Lecrubier, Sheehan, Amorim, Janavs and Weiller29} was conducted with all participants to check for common Axis 1 disorders. In self-report questionnaires, dimensional characteristics of BPD criteria were determined with the Borderline Symptom List Short Form (BSL-23).^{Reference Bohus, Kleindienst, Limberger, Stieglitz, Domsalla and Chapman30} Moreover, traumatic childhood experiences were determined with the Childhood Trauma Questionnaire (CTQ).^{Reference Klinitzke, Romppel, Häuser, Brähler and Glaesmer31} In addition to the diagnostic interview, possible influencing factors of the physiological stress reaction were assessed. Therefore, somatisation, depression and anxiety were assessed with the Brief Symptom Inventory (BSI-18).^{Reference Spitzer, Freyberger, Stieglitz, Carlson, Kuhn and Magdeburg32} Healthy controls also underwent a diagnostic interview and questionnaires, which included a demographic assessment, the SITBI-G, BSL-23, BSI-18 and CTQ. To make sure that healthy control participants did not fulfil criteria for any psychiatric disorder, the SCID-II Non-Patient (SCID-N/P)^{Reference First, Spitzer, Gibbon and Williams33} was conducted. All interviews and questionnaires were digitised with LimeSurvey for Windows (LimeSurvey GmbH, Hamburg, Germany; www.limesurvey.org/).

The second appointment comprised the actual experiment, including the stress induction paradigm. The detailed procedures are described elsewhere.^{Reference Koenig, Lischke, Bardtke, Heinze, Kröller and Pahnke26} The Trier Social Stress Test (TSST) for stress induction in children and adolescents was used as the stress task.^{Reference Kirschbaum, Pirke and Hellhammer34} The rationale for using the TSST over other paradigms is that the TSST is one of the best validated stress paradigms to reliably induce interpersonal stress in adolescents.^{Reference Seddon, Rodriguez, Provencher, Raftery-Helmer, Hersh and Labelle35} The aim was not to specifically induce academic stress. The respective cover-story has been suggested for adolescents and has previously been widely used. As adolescents with NSSI or BPD are known to show difficulties in handling interpersonal stress, examining stress during real-world interpersonal interaction is important in this patient group. Thus, it was intended to present the participants with a real-world scenario in a controlled stress-provoking setting. Before the TSST, participants performed the Color Detection Task (CDT) as a cognitive low-demand baseline task.^{Reference Jennings, Kamarck, Stewart, Eddy and Johnson36} They were asked to count the occurrence of a specific colour of a rectangle over 5 min. The TSST consisted of a preparation phase and two stress tasks. During the preparation phase, participants had 5 min to prepare themselves for an interview for their dream school. During phase 1 of the TSST, they were asked to present themselves in front of two interviewers. They were asked to elaborate on why they were the perfect candidate for a new school. Participants believed that their speech was videotaped. The interviewers did not show any reaction to the participants' speech and did not answer any questions. After 5 min they interrupted the participant and continued with a mental arithmetic test during which the participant had to subtract mentally. Every time the participant made a mistake, they had to start subtracting from the beginning. Before and after the stress task, the participants answered two short questionnaires regarding their momentary mental state (Positive and Negative Affect Schedule (PANAS);^{Reference Watson, Anna and Tellegen37} Dissociation-Tension Scale (DSS-4)^{Reference Stiglmayr, Schimke, Wagner, Braakmann, Schweiger and Sipos38}), and rated their stress levels on a visual analogue scale ranging from 0 to 100. At the end, all participants were debriefed.

fNIRS measurement

An eight-channel continuous-wave fNIRS system was used for continuous recordings of PFC oxygenation (OctaMon, Artinis Medical Systems, The Netherlands). It is an optical neuroimaging device that contains eight LED light sources and two PIN diode receivers that are placed with a headband onto the forehead of the participant. The light sources emit light in the near-infrared spectrum, which passes the skull cap and is absorbed by the O₂Hb and the HbR. The attenuation of light by different tissues is described by the modified Beer-Lambert Law. As O₂Hb absorbs light with a wavelength >800 nm and HbR absorbs light with a wavelength <800 nm, the OctaMon device emits light in two wavelengths, 760 nm and 850 nm. Receivers and transmitters are summed up as optodes. The inter-optode distance is 35 mm, which results in an estimated cortical penetration depth of 17 mm. The optodes were placed onto the forehead according to the international 10–20 system for electroencephalography electrodes placement.^{Reference Jaspers39} The optode configuration and their estimated coordinates according to the Montreal Neurological Institute brain template are displayed in Fig. 1. When placing the headband to the forehead, the instructor made sure that there was no hair between optodes and skin, and that the received light was between 3 and 97% of the emitted light. Values close to 0% indicate that almost no light reaches the receiver, whereas values close to 100% indicate that environmental light is also received. According to the general equation for the differential path length factor and in regard to current study protocols, the differential path length factor was set to 6 cm.^{Reference Scholkmann and Wolf40} The sampling rate was set to 50 Hz per channel.

Fig. 1 Optode placement on the forehead.^{Reference Koenig, Höper, van der Venne, Mürner-Lavanchy, Resch and Kaess23} R indicates receiver and S indicates source.

fNIRS data preprocessing

Raw haemoglobin density values (O₂Hb and HbR) were recorded by the fNIRS device and sent to a laptop via Bluetooth, where it was stored as *.oxy3-data using the Oxysoft software, version 3.1.103 for Windows (ArtinisMedical Systems, Elst, The Netherlands).⁴¹ All fNIRS data were segmented according to the start and end times of the different blocks during the study. In preparation for the preprocessing of the data, they were imported to MATLAB⁴² for Windows (The Math Works Inc., Natick, Massachusetts, USA) with the oxysoft2matlab function provided by Artinis Medical Systems (The Netherlands). Preprocessing was conducted with the HOMER2 toolbox.^{Reference Huppert, Diamond, Franceschini and Boas43} First, the raw optical density measures were converted to optical density values (hmrIntensity2OD). Next, we corrected for motion artefacts as recommended.^{Reference Cooper, Selb, Gagnon, Phillip, Schytz and Iversen44} We applied a two-step correction process. The first step was the application of a wavelet-based motion correction with a probability threshold of α = 0.01 (hmrMotionCorrectWavelet). In the second step, we corrected for motion artifacts by using the hmrMotionArtifact function. This function removes signal changes where the standard deviation exceeds the factor 20 (SDThresh) or where the peak-to-peak amplitude exceeded 0.5 (AMPThresh) within 0.5 s (tMotion). High-frequency noise was removed with a third-order Butterworth filter (LowFrequencyPass), which only allowed frequencies lower than 0.5 Hz so that high-frequency physiological noise (e.g. cardiac) was removed. Finally, optical density rates were converted to haemoglobin concentrations in μmol/L (10⁻⁶ mol/L) for O₂Hb, HbR and total haemoglobin (O₂Hb + HbR), and were exported for segmentation and further analyses to Stata/SE software, version 16.0 for Windows (StataCorp LLC, College Station, Texas, USA).⁴⁵ After preparing the fNIRS data for analysis, recordings from channel five of one healthy control were removed from the analysis during the CDT and the preparation of the TSST, as the recorded values were unrealistically high, suggesting that external light was absorbed. Visualisations were prepared with BrainNet Viewer, version 1.6 for Windows (www.nitrc.org/projects/bnv/).^{Reference Xia, Wang and He46}

Statistical analysis

All statistical analyses were conducted in Stata/SE software version 16.0.⁴⁵ In line with previous research,^{Reference Alderliesten, De Vis, Lemmers, van Bel, Benders and Hendrikse21,Reference Artemenko, Soltanlou, Ehlis, Nuerk and Dresler47–Reference Shi, Sakatani, Okamoto, Yamaguchi and Zuo51} we calculated mean values and standard deviations for O₂Hb for all relevant time blocks (four time blocks: CDT; Preparation TSST; TSST, divided into the free speech and the mental arithmetic test). For O₂Hb, mean values per channel and a grand mean value per block were calculated. First, to check for differences between groups on demographical data and on the clinical interviews, t-tests (continuous variables) and χ ²-tests (categorical variables) were calculated. Second, a linear mixed-effects model was calculated for O₂Hb to test the hypothesis that adolescent patients engaging in NSSI would show an increased PFC activation in response to an acute stress task. Predictors were time (CDT, preparation TSST, free speech, arithmetic task) and group (patients versus healthy controls), as well as their interaction included as fixed effects. The participants’ identifiers were addressed as random effect. Where applicable, planned contrasts were used to investigate effects of group at single events of time, to address when effects emerged and how long they lasted. To account for the severity of BPD symptoms and to test the hypothesis that severity of BPD pathology would be positively correlated with the increase in PFC activation in the NSSI group, models were repeated using a dimensional approach of BPD severity based on BSL-23 and the number of BPD criteria (SCID-II). The interaction of time × severity was addressed. Consistency between the two measurement modalities (BSL-23 and SCID-II) was inspected. For consistent interactions between the modalities, margin plots at fixed levels of BPD symptoms (BSL-23: 0, 1, 2, 3, 4; BPD criteria: 0, 3, 6, 9) were composed. Next, effects of self-reported levels of dissociation, stress, and positive as well as negative affect on PFC oxygenation were addressed. Each measure was included separately as a predictor to the mixed linear-effects model on O₂Hb, and it was checked whether the inclusion of self-reports improved the overall model fit. Group differences on self-reports in response to stress are reported elsewhere.^{Reference Koenig, Lischke, Bardtke, Heinze, Kröller and Pahnke26} Finally, connectivity between fNIRS channels across the PFC was analysed; cross correlation coefficients were determined for each time block and analysed in mixed-effects models with time and group as predictors. Contrast analyses were used to address changes in connectivity strength by time and group. Furthermore, BPD dimensionality (BSL-23 and SCID-II) was also included in additional mixed-linear effects models as indicated above. Because of technical issues, one healthy control had to be excluded from the connectivity analysis. Hence, the sample resulted in n = 30 patients engaging in NSSI and n = 28 healthy control participants. For simplicity and in line with previous research, only O₂Hb was included in all analyses, as it is the most informative haemoglobin variable and changes in haemoglobin are easier to detect.^{Reference Koenig, Höper, van der Venne, Mürner-Lavanchy, Resch and Kaess23,Reference Condy, Friedman and Gandjbakhche52}