EEG source functional connectivity in patients after a recent suicide attempt

(cid:1) Depressed patients after a recent suicide attempt exhibit increased nonlinear alpha functional connectivity. (cid:1) Nonlinear alpha functional connectivity is positively linked to depressive symptom severity in suicidal patients. (cid:1) Referencing the alpha range based on the individual peak frequency does not seem to improve validity of connectivity metrics.


Introduction
Despite a global decline in annual suicide rates over the last 20 years, the number of people suffering from a suicidal crisis remains at a dauntingly high level (WHO, 2019).A history of suicide attempts (SA) is a consistently identified risk factor for nonfatal and fatal suicidal behavior (Berman, 2018;Large et al., 2018;Park et al., 2018;Voss et al., 2019).While suicidal thoughts and behaviors frequently occur in the context of various comorbidly presenting psychiatric and neurodevelopmental disorders, depression is the most common among suicidal individuals (Bachmann, 2018;Wiebenga et al., 2021).Despite their robustness, these risk factors have failed to reliably predict suicide-related outcomes (Franklin et al., 2017).Additionally, established measures of suicide risk exclusively rely on subjective momentary appraisals that are inadequate in gauging the multi-facetted nature of suicidality (Bhatt et al., 2018;Cochrane-Brink et al., 2000;Klonsky et al., 2016).Consequently, there is a strong need for more objective and comprehensive suicide risk assessments that adequately reflect these complex pathomechanisms.In this context, the search for biophysiological patterns that are associated with suicidal thoughts and behaviors has gained momentum within recent years.Not only would the detection of such patterns advance approaches in personalized psychiatry, but they could also pave the way towards novel treatments for suicidal crises (Arns et al., 2022;Johnston et al., 2022).While some candidate neurophysiological features have been identified, none of these have been integrated yet into clinical practice.An electroencephalogram (EEG), on the other hand, is often already part of the standard routine examination during in-patient psychiatric treatment.Thus, EEG features show great potential to generate information on neurophysiological patterns in suicidal patients in a cost-and time efficient manner and could expand existing suicide risk assessments (Olbrich and Arns, 2013).
Among EEG-derived oscillatory waves, the alpha frequency range (AFR) between 8 and 12 Hz is the visually most prominent rhythm of human brain activity and can be best observed during eyes-closed resting-state recordings (Clayton et al., 2018).Further, the AFR shows substantial intra-and interindividual variance (Newson and Thiagarajan, 2019).Current evidence suggests an involvement of the AFR in memory, attention, and visual processing.Hence, an association between AFR alterations and psychopathological manifestations is highly probable (Mazaheri et al., 2018).Unsurprisingly, EEG based correlates of suicidality, such as frontal asymmetry or differences in frequency power, are most consistently detected within the AFR (Dolsen et al., 2017;Iznak et al., 2021;Park et al., 2019;Roh et al., 2020).
The vigilance regulation model (Hegerl et al., 2012;Hegerl and Hensch, 2014) highlights the promising potential of resting-state paradigms to shed light on pathomechanisms in affective disorders and suicidality.Additionally, resting-state paradigms exhibit high feasibility in vulnerable populations, including severely depressed or suicidal patients who often experience difficulties engaging in cognitive tasks (Rosazza and Minati, 2011).An EEG based resting-state measure yet to be explored in suicidal populations is functional connectivity, defined as a temporal dependence between spatially distinct neural events (Pascual-Marqui et al., 2011;van den Heuvel and Hulshoff Pol, 2010).A number of studies that relied on functional magnetic resonance imaging (fMRI) reported altered patterns of functional connectivity in suicidal patients (e.g., Gosnell et al., 2019;Ho et al., 2021;Kim et al., 2017;Malhi et al., 2020;Ordaz et al., 2018;Zhang et al., 2016).However, in contrast to EEG, fMRI exhibits a rather low temporal resolution and does not allow for a direct measurement of neurophysiological processes.To the best of our knowledge, no studies were published yet that investigated the association between AFR connectivity and suicidality.Promising findings have, however, been reported for patients with MDD.Olbrich et al. (2014) tested for group differences in source-based delta, theta, alpha, and beta connectivity between MDD patients and healthy controls.At baseline, MDD patients exhibited increased connectivity within the AFR compared to the control group, while no differences were found for the remaining frequency bands.Similarly, Leuchter et al. (2012) found the increase in sensor-level coherence in patients with MDD to be most robust within the AFR and beta frequency and argued that alterations in beta connectivity -that are also at a higher risk of contamination due to muscular artefacts (Krepel et al., 2021) -might be augmented by the AFR.Furthermore, several studies show that functional connectivity within the AFR is predictive of treatment response to pharmacological treatment (Iseger et al., 2017;Lee et al., 2011) and electroconvulsive therapy (Kirsten et al., 2020).A less clear pattern seems to emerge regarding focality of functional connectivity changes.Whereas Olbrich et al. (2014) focused more on frontal and prefrontal brain regions, Kirsten et al. (2020) found that widespread patterns of connectivity were predictive of treatment response.Still, these findings provide consistent support that alterations in EEG based functional connectivity within the AFR might relate to depression.Considering, again, the high prevalence of depression among suicidal individuals, we suspect that altered patterns of functional connectivity within the AFR could also be linked to suicidal thoughts and behaviors.
Following the provided evidence, we therefore hypothesized that patients after a recent SA exhibit altered EEG based wholebrain functional connectivity compared to HC. Considering that altered patterns of connectivity in MDD patients seemed to be most robust within the AFR, while focality appeared to be more ambiguous, we limited our analyses to source-level AFR connectivity; thus, aiming at a higher spatial resolution.A higher spatial resolution would also provide the possibility to compare our results to subsequent fMRI-based analyses.Due to a lacking consensus regarding the discriminatory power of existing EEG based functional connectivity measures (Wendling et al., 2009), we mapped both linear (i.e., covariance of frequency power) and nonlinear (i.e., phase synchronization) measures.Further, considering the large interindividual variance in the AFR, we applied an individually referenced AFR (iAFR) based on the individual alpha peak frequency ([IAPF]; i.e., the individual spectral peak within the AFR) in addition to the standard AFR ([sAFR], 8-12 Hz), hoping to increase validity.We therefore expected group differences to be more pronounced within the iAFR compared to the sAFR.Further, we hypothesized that EEG based linear and nonlinear functional connectivity within the sAFR and iAFR are associated with suicidal ideation and depressive symptom severity.We also performed exploratory subgroup analyses for suicidal patients with a current depressive disorder or episode (DD) and those without DD to allow a discussion of potential confounding effects of depression.Considering the current scarcity of neurophysiological patterns that are specific to suicidality, distinct patterns of EEG based functional connectivity could, on the one hand, yield valuable insight into the pathomechanisms of suicidal behavior and ideation within a transdiagnostic sample.On the other hand, identification of functional connectivity patterns that could be explored further in biomarker-based approaches would promote time-efficient and more personalized strategies for the prevention and treatment of suicidal crises.

Subjects
We used baseline data collected within the framework of a larger clinical randomized controlled trial (registered at ClinicalTrials.gov,NCT03732300).For a more detailed description of the study design, see Section A1 in Appendix A. Fig. 1 details the acquisition of the analysis samples.The final sample consisted of 70 patients after a recent SA (mean age = 31.3,SD = 12.9) who received treatment at the Psychiatric University Hospital Zurich, Switzerland, and 70 age-and gender-matched HC (mean age = 31.8,SD = 12.9).For details on age-and gender-matching see Table B1 in Appendix B. Recruitment for this sample took place between August of 2019 and February of 2022.For the exploratory subgroup analysis, patients were further divided into two subsamples: patients with a recent SA who either suffered from a current depressive episode (F32.X), a recurring depressive disorder (F33.X), or a bipolar disorder with a current severe depressive episode (F31.4)(=subsample with DD, n = 53) and patients with a recent SA without DD (n = 17).Diagnostic criteria were based on the 10th revision of the international statistical classification of diseases and related health problems ([ICD-10]; WHO, 1993).Age-and gender-matched HC were also divided into subsamples, analogous to the patient subsamples with and without DD.Demographic characteristics for all patients and the resulting subsamples are presented in Table 1.Adult patients who had attempted suicide within 6 months prior to study participation were considered eligible.Exclusion criteria for patients and HC were checked prior to study participation and can be found in Section A2 and A3 in Appendix A, respectively.Additionally, two clinical interviews, the Hamilton depression rating scale ([HDRS]; Hamilton, 1960;Strauß and Schumacher, 2004) and the mini-international neuropsychiatric interview ([MINI]; Sheehan et al., 1998) were administered during the baseline assessment to ensure the absence of current or past mental, behavioral, or neurodevelopmental disorders in HC.HC were recruited via mailing lists, online platforms, and flyers and received a compensation of 90 Swiss francs for study participation.Patients and HC signed the informed consent prior to study participation.The study was carried out in accordance with the declaration of Helsinki 2008 and was approved by the ethics commission of the canton of Fig. 1.Flow chart for our analysis sample.Initially, we recorded 96 electroencephalograms (EEG) in patients and 75 in healthy controls (HC).However, only 71 of each group matched in terms of age and gender.These groups were further divided into subsamples: A patient subsample with a current depressive disorder or episode (DD) and a patient subsample without DD.HC were divided analogously.

Clinical measures
Current suicidal ideation was assessed using the German version of the Beck scale for suicide ideation ([BSS]; Beck and Steer, 2015).The self-rating scale includes 19 items on suicidal ideation and preparatory acts within the last 7 days as well as two additional items on past suicidal behavior.For the total BSS score, only items 1 to 19 are used.Current severity of depressive symptoms was assessed with the German version of the 21-item HDRS.The HDRS is a standardized clinical interview that covers a wide range of depressive symptoms, including depressive mood, anxiety, sleep disturbances, suicidal ideation, psychosomatic, and psychotic symptoms (Hamilton, 1960;Strauß and Schumacher, 2004).Investigators were trained to guarantee a standardized administration of the HDRS and sufficient interrater accordance.For a detailed description of the psychometric properties of the BSS and HDRS, see Section A4 in Appendix A. Patients' ICD-10 diagnoses at baseline were based on the agreement of two clinicians' clinical judgement of core and additional symptoms according to ICD-10 criteria.Additionally, we administered the MINI to confirm patients' diagnosis and ensure absence of any psychiatric diagnoses according to ICD-10 in HC.For the patient sample, diagnoses were checked again after the baseline assessment to affirm diagnostic stability.

EEG acquisition and preprocessing
EEGs were recorded between 9 a.m. and 3 p.m.During the recording, participants were in a reclining position in a lightattenuated room.Room temperature was kept between 20 and 23 °C.The 15-minute eyes-closed resting-state recordings followed standardized instructions according to Hegerl et al. (2017).EEGs were recorded with an eego TM amplifier (ANT Neuro; Hengelo, Netherlands).The 64 electrodes (Ag/AgCI) including an electrooculogram for vertical eye movements were placed according to an extended 10-20 system and referenced against CPz.Two additional bipolar electrodes for horizontal eye movements, placed near the canthus, and two bipolar electrodes at the wrists for electrocardiogram recording were applied as well.Data was sampled at 4 kHz while impedances were kept under 50 kX.
Preprocessing of the EEG recordings was done using BrainVision Analyzer 2 (''BrainVision Analyzer 2," 2019).Raw data was filtered at 70 Hz (low-pass), 0.5 Hz (high-pass) and 50 Hz (notch-filter).1second intervals were inspected for movement artefacts and removed manually if necessary.No more than 6% of segments were excluded during manual artifact rejection.Additionally, an independent component analysis (ICA) was conducted.For this, data was down-sampled to 500 Hz.Components were chosen carefully based on visual inspection of the ICA topographic maps and comparisons with the original EEG time series signal.Accordingly, 2 to 5 components that were likely based on eye movements or muscle artefacts were removed.For exact low resolution electromag-   Marqui et al., 2006) analysis, the preprocessed data was segmented into 4-second intervals to achieve a sufficient frequency resolution (i.e., 0.25 Hz) and increase the number of epochs to reach high validity in the functional connectivity analysis.

eLORETA based functional connectivity mapping
Linear (i.e., covariance of frequency power) and nonlinear (i.e., phase synchronization) EEG-based source functional connectivity was mapped using eLORETA.eLORETA represents a unique inverse solution with zero localization error under ideal noise-free conditions and allows for an enhanced spatial resolution, compared to sensor-level connectivity (Pascual-Marqui, 2007).For eLORETA, the two mastoid electrodes (M1 and M2), the two sets of bipolar electrodes for the electrocardiogram and horizontal eye movements as well as the single electrode for vertical eye movements were not considered.A complete list of electrodes for eLORETA analysis and corresponding coordinates according to Towle et al. (1993) are listed in Table B2 of Appendix B. To avoid volume conduction bias, we applied lagged measures of functional connectivity that omit phase-lags of zero.These were calculated for the 84 Brodmann areas (i.e., 42 at each hemisphere without Brodmann areas 12, 14, 15, 16, 26, and 48-52) that are predefined within the LORETA software which resulted in (83x84)/2 = 3486 edges.Based on the findings of Olbrich et al. (2014) and Kirsten et al. (2020) analysis was limited to the AFR.The sAFR was set to 8-12 Hz, while the iAFR was defined as IAPF Ç 1 Hz.We used the vigilance algorithm Leipzig ([VIGALL]; Hegerl et al., 2017) to determine the IAPF.VIGALL calculates the IAPF as an average of two occipital channels (O1 and O2) within a 2 Hz window that contains the greatest mean power in a spectrum of 8-14 Hz.For VIGALL analysis, EEG data was average-referenced.

Network-based statistics
Connectivity matrices for lagged linear and lagged nonlinear measures within the sAFR and iAFR were exported to networkbased statistics ([NBS]; Zalesky et al., 2010).NBS is a two-step non-parametric approach that relies on permutation testing to account for the multiple comparisons problem in large-scale network analyses.An exact description of the method can be found in Zalesky et al. (2010).Group differences between patients after a recent SA and HC were assessed for each modality separately, applying an F-test with primary thresholds between 0.5 and 8, in intervals of 0.5.The same thresholds were applied in Kirsten et al. (2020).The secondary threshold for significance was set to p 0.05.In case of significant networks, we performed two onesided t-tests (i.e., patients > HC and HC > patients) in a post-hoc manner, using the equivalent t-statistical primary threshold (i.e., approximately 0.5 to 3) to determine the directionality of group differences.Since a total of 4 group comparisons were performed for each connectivity modality, we also applied a corrected pvalue of p 0.0125.We conducted further statistical analysis in R studio, R version 4.0.5 (R Core Team, 2021).To examine for potential confounding effects of psychotropic drugs in the group comparisons, we tested for significant (p 0.05) correlations according to Kendall's s between medication intake and global connectivity -defined as mean functional connectivity across all possible connections for each distinct modality (i.e., lagged linear and lagged nonlinear connectivity within the sAFR and lagged linear and lagged nonlinear connectivity within the iAFR).Further, Mann-Whitney U-and t-tests were applied where applicable to test for differences in sociodemographic characteristics between patients and HC as well as across subsamples.Graphs were either created in R studio or BrainNet Viewer (Xia et al., 2013).

Multiple regression analysis
The association between EEG functional connectivity and both, BSS and HDRS scores, was investigated by means of multiple regression models.To reduce the number of tests, we used global connectivity (i.e., mean lagged connectivity across all edges) for each modality.We assumed that subgroup allocation (i.e., either HC, patient subsample with DD, or patient subsample without DD) would explain most of the variance in the outcome variable (i.e., either BSS or HDRS).Hence, a simple regression for outcome and subgroup was defined as the basic model.Next, we fitted an interaction model, including a main effect for subgroup, a main effect for global connectivity and the interaction between global connectivity and subgroup.Finally, a covariate model was fitted with added main effects for age and sex.Accordingly, three nested models were fitted for each modality (i.e., lagged linear and lagged nonlinear connectivity within the sAFR and iAFR) and outcome (i.e., BSS or HDRS).F-tests were then applied to test whether any of the more complex models would significantly outperform the basic model.Additionally, we calculated the Akaike information criterion (AIC) for each model to allow comparisons across nonnested models.A p-value 0.05 was deemed significant.Prior to regression analysis, BSS, HDRS, global connectivity, and age were centered and scaled to account for floor effects in clinical measures and differences in scaling across these variables.

Group differences in sociodemographic and clinical measures
According to Mann-Whitney U tests, there was no significant difference in age between patients and HC (p =.64) as well as no significant age differences among patient (p =.40) and HC subsamples (p =.38).Table 2 shows mean scores and standard deviations for the BSS and HDRS.Patients consistently showed significantly higher BSS and HDRS scores compared to HC (p <.001).Further, patient subsamples did not differ in BSS (p =.21) or HDRS scores (p =.70).

NBS: Group differences in functional whole-brain connectivity
A correlation analysis revealed no significant (i.e., p 0.05) correlations according to Kendall's s across all subjects (N = 138 due to missing data on medication intake in two patients) between medication and any of the global functional connectivity metrics.Accordingly, we deemed confounding effects of medication unlikely.F-tests across first-degree thresholds 0.5 to 8 revealed no significant differences in network-based lagged linear or lagged nonlinear functional connectivity within neither the sAFR nor the iAFR (see Fig. 2).Features and p-values for the identified, but insignificant networks can be found in Tables B3 to B6 of Appendix B.
To investigate effects that might relate to depression, we also performed subgroup analyses.First, patient subsample with DD (n = 53) was compared to the matched HC subsample 1. F-tests with first-degree thresholds of 0.5 to 8 revealed significantly distinct patterns of lagged nonlinear functional connectivity within the sAFR (see Table B8 in Appendix B).Accordingly, we performed two equivalent one-sided t-tests.Patient subsample with DD consistently exhibited increased lagged nonlinear functional connectivity within the sAFR compared to HC subsample 1.These networks are visualized in Fig. 3. Patterns in posterior regions, particularly within the right hemisphere, revealed the most robust and pronounced differences across varying first-degree thresholds.However, these did not survive the corrected threshold of p 0.0125.There were no significant networks for which patients exhibited a decrease in lagged nonlinear functional connectivity within the sAFR compared to HC. Measures of lagged linear functional connectivity within the sAFR revealed trend-level (p <.1) significant networks for first-degree thresholds 1.5 to 8.However, since these did not exceed the a priori defined significance level of p 0.05, these were not considered for further analysis (see Table B7 in Appendix B).Group comparisons for lagged linear (see Table B9 in Appendix B) and lagged nonlinear functional connectivity (see Table B10 in Appendix B) within the iAFR did not render significant differences.
Second, the same analyses were applied to the patient subsample without DD (n = 17) and the matched HC subsample 2. However, no significant group differences were detected for any of the functional connectivity metrics (see Tables B11 to B14 in Appendix B).Finally, patient subsample with DD (n = 53) was compared to patient subsample without DD (n = 17).Since these subsamples were not age-and gender-matched, an additional correlation analysis across patients was performed to check for potential associations between age, gender, medication intake and functional connectivity metrics.Kendall's s showed no significant correlations.Accordingly, we considered confounding effects of these factors unlikely.Group comparisons between patients with and without DD revealed no significant differences across   any of the functional connectivity metrics (see Tables B15 to B18 in Appendix B).

Association between functional connectivity and clinical measures
The 3 nested regression models were fitted for each outcome and global connectivity modality.The parameters for these regression models and the corresponding F-tests that were applied to compare the nested models can be found in Tables B19 to B34 in Appendix B. For BSS scores, there was a significant main effect of global linear functional connectivity within the sAFR (b ¼ À0:16; CI ¼ À0:030; À0:01; p ¼ :039Þ in the interaction model (see Table B19 in Appendix B).However, none of the interaction or covariate models for the BSS outperformed the basic model (see Tables B27 to B30 in Appendix B).For HRDS scores, there was a significant main effect of sex across all covariate models, in that male participants scored lower on the HDRS (see Tables B23 to B26 in Appendix B).None of the interactions between global connectivity and subgroup were significant, nor was there a significant effect of age.The covariate model outperformed the basic model for both linear connectivity modalities (see Tables B31 and B33 in Appendix B) with a significant main effect of linear connectivity within the iAFR (b ¼ 0:14; CI ¼ 0:01; 0:28; p ¼ :037Þ.The interaction model outperformed the basic and covariate model for the nonlinear connectivity modalities (see Tables B32 and B34 in Appendix B) with significant main effects for nonlinear connectivity within the sAFR (b ¼ 0:14; CI ¼ 0:01; 0:27; p ¼ :036Þ and iAFR (b ¼ 0:13; CI ¼ 0:00; 0:26; p ¼ :048Þ.Across all HDRS-models, the interaction model, including the significant main effect for nonlinear connectivity within the sAFR performed best, based on the F-test and AIC (AIC = 227.162,adjusted R 2 = 0.703), with an additional explained variance of 1% compared to the basic model (see Table B24 in Appendix B).Despite prior scaling, the assumption of homoscedasticity was not met for the models which likely indicates a floor effect in clinical measures across subgroups.Fig. 3. Glass brain view and binary adjacent matrices for distinct patterns of lagged nonlinear functional connectivity within the standard alpha frequency range (sAFR) between patients after a recent suicide attempt (SA) with a depressive disorder or episode (DD) and healthy controls (HC).t-statistical thresholds from 0.5 to 3 all reached significance (i.e., p 0.05).Network patterns in posterior brain regions showed the highest robustness across thresholds.

Discussion
The primary objective of this analysis was to explore the association between EEG based whole-brain functional connectivity within the AFR and suicidal thoughts and behaviors in patients after a recent SA and related subpopulations.Contrary to our hypotheses, we did not observe significant group differences in functional connectivity within the AFR between the entire sample of patients after a recent SA (N = 70) and matched HC (N = 70).Further, none of the AFR connectivity metrics significantly improved variance explanation in suicide ideation, measured with the BSS, beyond the grouping factor.However, the patient subsample with DD showed significantly increased nonlinear functional connectivity within the sAFR compared to their matched HC.This was most evident in right-hemispheric posterior brain regions.Additionally, the interaction model for nonlinear functional connectivity within the sAFR outperformed the remaining models for the HDRS and indicated that increased phase synchronization within the AFR is associated with higher depressive symptom severity.
These findings are in line with Olbrich et al. (2014) who reported increased phase synchronization within the AFR in MDD patients compared to HC.Initially, we assumed that validity of the frequency-dependent functional connectivity measures would increase as contamination through neighboring frequency bands (i.e., beta and theta for the AFR) decreases.However, similar to Debnath et al. (2021), we did not provide support for improved validity in phase-based connectivity through IAPF referencing.One possible explanation is that the location from which the power was derived (i.e., occipital) was not the optimal reference for IAPF and subsequent iAFR detection.Following the approach from Voetterl et al. (2023), different electrode sites could have been compared.However, this would have been beyond the scope of this paper.Furthermore, VIGALL calculates the IAPF based on a 2 Hzwindow that contains the highest frequency power across all segments.Hence, the IAPF was detected at varying stages of the resting-state recording for the different subject.This might have in fact increased intra-group variance.Finally, while individually referencing the AFR based on power could be valuable in linear measures of connectivity that rely on amplitude, this might not be necessary or potentially even counterproductive in phasebased indices, as these narrow-band frequency ranges could reflect distinct functional patterns of phase synchronization (Bazanova and Vernon, 2014).The association between phase-based connectivity and the IAPF should be explored in more detail.Additionally, it could be interesting to explore various operationalizations of IAPF detection and iAFR referencing.
While increases in AFR power and the related linear connectivity metrics are likely associated with inhibition, phase synchronization within the AFR is thought to be associated with an active top-down regulation of neuronal processing across largescale cortical networks (Palva and Palva, 2011).In a graphtheoretical sense, this phase-based hyperconnectivity, might thus imply reduced efficiency in the regulation of intrinsic restingstate brain networks (Bullmore and Bassett, 2011;Palva and Palva, 2011).Similarly, a growing body of evidence suggests that the down-regulation of brain arousal is impaired in MDD patients (Hegerl et al., 2012(Hegerl et al., , 2012;;Hegerl and Hensch, 2014;Ip et al., 2021;Olbrich et al., 2012).Additionally, in a previously published paper, we could show that a dysregulation of the autonomic nervous system, measured via heart rate variability, could discriminate patients after a recent SA from HC (Rüesch et al., 2023).Taken together, individuals who suffer from depressive or suicidal crises seem to exhibit neurophysiological hyperarousal that might be associated with phase-based hyperconnectivity.Symptoms that are common among these individuals, such as sleep disturbances, agitation, or rumination could also relate to this neurophysiological hyperarousal (Joiner et al., 2018).However, whether hyperarousal is specific to a depressive-suicidal syndrome or whether it is a transdiagnostic marker of general psychopathology remains unclear.More research should focus on the combination of various neurophysiological or multi-modal features to potentially increase specificity.Further, it would be interesting to assess the relation between functional connectivity, EEG wakefulness, and heart rate.This could be further investigated by comparing groups of individuals who suffer from varying psychopathologies.
In this context, a possible viewpoint is that the observed increase in phase synchronization in the patient subsample with DD could be primarily attributed to depression rather than suicidal behavior.However, it is important to highlight that although our patient subsample with DD shares a common feature of an ongoing depressive disorder or episode, these patients still represent a heterogeneous group of individuals who have recently engaged in suicidal behavior.This diversity is evident in the high percentage of psychiatric comorbidity ($70%) within this group, as illustrated in Table 1.Additionally, there were no significant group differences between the patient subsamples.Therefore, attributing the observed group differences solely to depression would likely oversimplify the intricate psychopathology associated with suicidal behavior.Our results rather implicate a potential interplay between suicidal behavior, depression, and altered regulation of intrinsic brain networks through AFR phase synchronization.Again, this interplay needs to be investigated further by comparing various clinical subpopulations, including suicidal and non-suicidal depressed patients.
This would be of particular interest since our results could also point to the existence of distinct subpopulations among individuals who suffer from suicidal crises.These subpopulations might be characterized by both, distinct psychopathological (e.g., depressed vs. non-depressed) and neurophysiological patterns (e.g., phase synchronization).It is thus likely that no group differences were observed between non-depressed suicidal patients and their matched HC as the patient subsample without DD exhibited high pathophysiological variance because it encompassed multiple distinct subpopulations.Additionally, the sample size of 17 was relatively small.The proportion between patients with and without DD within our sample reflects a real-world clinical sample since the vast majority of individuals who experience suicidal crises also suffer from depression (Bachmann, 2018).This highlights the scientific and clinical value of our findings.A prospective design that investigates the possibility to stratify a larger sample of transdiagnostic individuals who suffer from suicidal thoughts or behaviors based on patterns of AFR phase synchronization could be promising.Exploring the association between these connectivity-derived subgroups and treatment outcome might be a particular beneficial approach and could pave the way for less established treatments that affect AFR phase synchronization, such as transcranial alternating current stimulation (Rostami et al., 2021).
While our study has valid implications, some important limitations ought to be mentioned.A major limitation is its exploratory nature which reduces statistical power and limits the generalizability of our results.Further, the significant group effects did not survive a corrected p-value threshold.No correction for multiple testing was applied across the multiple regression models and the assumption of homoscedasticity was not met.Hence, these results should be interpreted with caution and rather used as a gateway to subsequent hypothesis-driven and replicatory analyses.Despite non-significant correlations between medication intake and global connectivity measures, we cannot completely rule out a confounding effect of medication intake.Nevertheless, the majority of suicidal in-patients are medicated by which our sample relates to a real-world clinical setting.Controlling for each individual drug and combined intake would therefore likely impair the external validity and clinical feasibility.Additionally, Olbrich et al. (2014) reported no significant effects of antidepressant treatment on AFR connectivity.The application of ICA during preprocessing may have further affected the functional connectivity measures (Castellanos and Makarov, 2006), although excluded components were chosen carefully.Finally, the frequent comorbid occurrence of some depressive symptoms among the nondepressed patient subsample can be considered as a limitation since it might indicate an arbitrary distinction between the two patient subsamples.However, the HDRS assesses a variety of symptoms that are also common among other mental disorders such as sleep disturbances, anxiety, or suicidality.Hence, high depressive symptom severity according to the HDRS does not necessarily justify the diagnosis of a depressive disorder or episode (Gibbons et al., 1993).Finally, while an additional explained variance of 1% in HDRS scores was statistically significant, this cannot be considered clinically relevant.We therefore second a shift from classification to treatment stratification since we believe this approach to be more beneficial for clinical translation (Arns et al., 2022).

Conclusion
Our results generally provide further support that depressed suicidal patients exhibit increased functional connectivity within the AFR.We extend existing literature by applying various EEG based functional connectivity modalities out of which phase synchronization within the sAFR has shown the highest potential for further investigation in prospective designs.While exploring the predictive value of AFR phase synchronization for suicidal thoughts and behaviors could offer valuable insights, assessing its potential for treatment stratification might be even more impactful.Unfortunately, evidence-based and specialized interventions for suicidality are scarce (Belsher et al., 2019).Identifying candidate neurophysiological markers that predict treatment outcome in a transdiagnostic population of individuals who exhibit suicidal thoughts and behaviors could advance the implementation of targeted interventions for suicidal crises.
= 70 (all patients), n = 53 (subsample with DD), n = 17 (subsample without DD).Abbreviations.DD, depressive disorder or episode; ICD, International classification of diseases: SA: suicide attempts.a Refers to the simultaneous prescription of at least two types of psychotropic drugs from different drug classes.b Refers to diagnoses at baseline according to ICD-10 criteria which were based on two clinicians' clinical judgement.Diagnoses were further confirmed by the study team through the mini international neuropsychiatric interview (MINI).F1 = Mental and behavioral disorders due to psychoactive substance use; F2 = Schizophrenia, schizotypal, delusional, and other non-mood psychotic disorders; F3 = Affective disorders; F4 = Anxiety, dissociative, stress-related, somatoform and other nonpsychotic mental disorders; F5 = Behavioral syndromes associated with physiological disturbances and physical factors; F6 = Disorders of adult personality and behavior; F9 = Behavioral and emotional disorders with onset usually occurring in childhood and adolescence.c Refers to the simultaneous presence of at least two mental health disorders according to ICD-10 criteria.netic tomography ([eLORETA] Pascual-Marqui, 2007; Pascual- Note.N = 70 (all patients), N = 70 (all HC), n = 53 (subsample with DD), n = 17 (subsample without DD).BSS and HDRS were administered at baseline.Abbreviations.BSS, Beck scale for suicide ideation; DD, Depressive disorder or episode; HC, Healthy controls; HDRS, Hamilton depression rating scale.a Refers to BSS scores.Data was missing for one patient and one HC.Accordingly, sample sizes for mean BSS scores and SD are as follows: n = 69 (all patients), n = 70 (all HC), n = 52 (subsample with DD), n = 53 (HC subsample 1), n = 17 (subsample without DD), n = 17 (HC subsample 2).b Refers to HDRS scores.Data was missing for three patients and two HC.Accordingly, sample sizes for mean HDRS scores and SD are as follows: n = 64 (all patients), n = 69 (all HC), n = 48 (subsample with DD), n = 53 (HC subsample 1), n = 16 (subsample without DD), n = 16 (HC subsample 2).

Fig. 2 .
Fig. 2. Results of the network-based statistics (NBS) analysis.Plots show the number of edges (left y-axis) and corresponding p-Values (right y-axis) as a function of firstdegree F-statistical threshold (x-axis).No statistically significant (i.e., p 0.05) differences in connectivity were detected between patients after a recent suicide attempt (SA) and healthy controls (HC) for first-degree F-thresholds 0.5 to 8 in any of the connectivity modalities.These group comparisons were performed for linear and nonlinear functional connectivity within the standard alpha frequency range ([sAFR], Fig.2A and 2B) and the individually referenced alpha frequency range ([iAFR], Fig.2C and 2D).

Table 1
Sociodemographic characteristics of all patients and patient subsamples at baseline.

Table 2
Means (M) and standard deviations (SD) of clinical measures at baseline.