Wei Zhang

Eijndhoven, Schene, Beckmann, & Tendolkar, 2015). In general, a breakdown of the SN gives rise to impaired detection and signaling of salient stimuli or events, resulting in significant consequences for cognition and emotion. The characterization of the SN and its interactions with other large-scale brain networks such as the DMN and CEN has been proposed as an important way to understand dysfunction in a variety of neuropsychiatric disorders.

Chapter 2 elaborates on my observations concerning increased connectivity within the SN following acute stress induction and the correlation between the magnitude of this increase and individual cortisol stress-responsiveness. In parallel, decreased DMN connectivity was found after stress induction, which also exhibited an association with higher cortisol stress-responsiveness. These findings may suggest potential interactions between these large-scale networks and the HPA axis that give rise to individual variability in cortisol responsivity. This result is generally in line with previous studies suggesting the critical roles played by the amygdala, hippocampus and medial PFC – also the key regions from the SN and DMN – in regulating cortisol secretion in response to acute stress. With these findings, I identified neural markers at the network level that exhibit sensitivity to individual differences and that could potentially be relevant to stress-related psychopathology. In Chapter 4, I discussed the link between these connectivity changes and subsequent PTSD symptom development after trauma and reported the finding that SN connectivity patterns with posterior DMN prospectively predicted perceived stress levels, while changes in SN connectivity (i.e., SN-cerebellum connectivity) over time also explained symptoms at follow-up. Interestingly, the specific DMN connectivity changes in Chapter 2 showed neither predictive nor acquired effects of PTSD symptoms after trauma in the study reported in Chapter 4. However, DMN was involved in predicting post-trauma stress levels via its interaction with the overall SN. These results align well with the current literature reporting the SN-DMN interactions, predominantly based on connectivity between key regions of these networks, in relation to stress-related processing and psychopathology (Veer et al., 2010; Sripada et al., 2012; Van Der Werff et al., 2013; Lei et al., 2015; Koch et al., 2016; Akiki et al., 2017). This involvement of SN-DMN interaction in stress-related processing was further supported by the findings reported in Chapter 3, where I directly investigated inter-regional interactions within and across these large-scale networks (i.e., connectivity patterns) in relation to acute stress induction. Using a novel analytic approach, I identified critical interactions of key regions from the SN (i.e., dACC and amygdala) and DMN (i.e., PCC and PCu) that, together, substantiated the discrimination between the stressed and non-stressed brain states. Again, these results suggest that connectivity patterns of the SN and DMN might be the most relevant neural signature for stress sensitivity and can thus be informative for studying long-term consequences for mental health. In conclusion, these findings demonstrate that neural responses to challenging situations can be useful 1) for understanding inter-individual variability in stress reactivity and possibly in stress adaptation, and 2) for linking that variance to long-term consequences.

Strengths, Implications and Limitations

Strengths

From an evolutionary point of view, stress responses evolved as adaptive processes. However, severe and prolonged stress responses may lead to undesired consequences (Selye, 1956). How these presumably adaptive responses can become destructive is a key to understanding individual differences in stress resilience versus vulnerability. Furthermore, such insight can potentially facilitate precise assessments and alternative classifications beyond the DSM categories of PTSD by identifying varying dysfunctions in general biological and psychological systems (Cuthbert and Insel, 2013; Howlett and Stein, 2016; McFarlane et al., 2017). In the studies reported in this thesis, based on indications that reactivity to challenges might provide useful biomarkers for stress vulnerability (see review for Michopoulos, Norrholm, & Jovanovic, 2015), I used a formal acute stress induction to experimentally probe stress responses – particularly at the neural-network level. Importantly, having used a longitudinal design, the work described in this thesis also fits well with a recently proposed framework for resilience research in which prospective longitudinal studies are urged to investigate resilience factors and thereby complement traditional pathophysiological research on stress (Kalisch et al., 2017). Hence, the findings reported in this thesis can also be considered relevant for studying resilience factors.

Implications

At the clinical level, the majority of investigations into stress-related network-level connectivity have used a cross-sectional design, leaving the predisposed and acquired neural effects unclear. Results reported in this thesis show that acute stress-induced reconfiguration of large-scale networks, particularly the SN reconfiguration, can be highly relevant for further clinical studies. More specifically, insufficient SN-DMN interactions in response to stress may lead to unwanted high stress levels after trauma exposure, whereas SN-cerebellum interactions under stress may become sensitized with increased symptom levels (as shown in Chapter 4). Although the studies in the thesis focused on a relatively resilient sample with sub-clinical symptoms, the findings reported here are, to a certain degree, in line with the existing theories and hypotheses about the neurocircuitry of PTSD. For example, based on neuroimaging findings, Admon et al. (2013) proposed a causal model for PTSD with abnormalities within the amygdala and dACC – the core regions of the SN – being predisposing risk factors of PTSD. In this research, instead of within-SN abnormalities, we observed weakened SN synchronization with other brain circuits (i.e., the DMN) being predictive of higher post-trauma stress levels. The discrepancy here might result from the different experimental designs used in the current research and in the prospective studies reviewed by Admon and colleagues. Nevertheless, these findings all point toward the SN as a potential biomarker for PTSD vulnerability. Interestingly, in contrast to the abnormalities in the DMN (i.e., the hippocampal-vmPFC) that were proposed as the acquired deficits (Admon et al., 2013), this research observed intensified SN-cerebellum synchronization in participants with higher symptom levels after trauma. Again, the inconsistency may arise from different experimental designs. Yet, the alterations in SN connectivity pattern match well the existing observations for discrete structures of the SN connectivity in PTSD (see review for Koch et al., 2016), as well as the findings from large-scale network-based studies on PTSD (see review for Akiki et al., 2017). In short, the current research provides empirical evidence supporting aberrant SN function as a predisposing risk factor for PTSD. In line with the existing models for PTSD biomarkers, this thesis highlights the critical role of SN connectivity in stress-related processing and potentially relevant psychopathology – not only at the local regional level, but also at a cohesive and unified network level.

Limitations

This thesis is not without limitations. For example, the experimental setup for inducing acute stress only allowed for capturing of the neural processing that was plausibly a mixture of stress reactivity and stress recovery. Specifically, although at the group level, cortisol level in the current sample peaked 20 minutes after the onset of stress induction and remained high for another 10 minutes; some participants showed a decline from the peak level, while others exhibited a sustained elevation even 30 minutes after the onset. As the acquisition of rs-fMRI took place in the timeframe between 20 and 30 minutes after the stress onset, the observed neural responses for some participants may be more relevant to stress reactivity than to stress recovery and for others vice versa. Consequently, disentangling these interrelated, yet conceptually distinct processes may help identify neural circuits associated with and the factors that can influence individual variances in these processes. Teasing these processes apart with a higher temporal resolution of neuroimaging measures may also provide better insight into the relationship between the short-term adaptation to a stressor and the long-term consequences for mental health.

Additionally, the sample studied in this research was relatively resilient and demonstrated low overall levels of PTSD symptoms after trauma. Therefore, the observed neural-network responses to acute stress and their association with subsequent symptom development may well involve processes relevant to stress resilience as well. To delineate the underpinning processes for resilience versus vulnerability, future investigations may consider the use of much higher sampling frequency during and after trauma for studying trajectories of stress responses, assuming that stress resilience is a dynamic process of successful adaptation to adversity that is presumably time-varying and individually variable (Kalisch et al., 2017). It may also help to take into account genetic make-up and early-life experiences that, together, are suggested to program individuals with improved resilience for later-life challenges when adverse experiences in early and later life are similar or with enhanced vulnerability when there is a mismatch (Daskalakis et al., 2013).

Lastly, the network-level focus in this research inevitably missed neural processing at a more refined regional or sub-regional level, which may direct one to more specific neural circuits pertinent to stress-related processing. However, in acknowledgment of this drawback, I applied a more data-driven method to identify sub-regions from these large-scale networks in relation to acute stress responses; whether these functional units can offer more information about long-term stress effects remains untested.

Open Questions and Future Directions

The studies reported in this thesis aimed to identify predictive biomarkers of stress-related symptom development. I focused primarily on large-scale network connectivity but also made an effort to explore inter-regional interaction within and across those networks. As mentioned earlier, it remains unknown whether the observed interactions at more local level (i.e., between sub-regions) that significantly responded to acute stress can provide us more information on long-term stress effects. This question is relevant because, if we can break down acute stress-induced brain reconfiguration from the network level to regional and sub-regional level and demonstrate the association between these more local-level functional characteristics and long-term stress consequences, we may be able to specify neural mechanisms for distinct yet highly overlapped processes, such as stress reactivity and stress recovery. In fact, previous investigations into prolonged stress effects provided a hint of different connection patterns involved in stress recovery as opposed to reactivity, which appeared to be associated with cortisol stress response (Veer et al., 2011; Vaisvaser et al., 2013; Quaedflieg et al., 2015; Dimitrov et al., 2018).

Another relevant question concerns the outcome measures for stress symptoms used in this research. I used three outcome measures and found different results (i.e., results in PSS and CAPS but not in PCL), which may be due to different aspects of the measures (i.e., PCL was less sensitive to the current sub-clinical sample). Alternatively, these differences may have arisen from different administration methods (i.e., the CAPS being administered via telephone interview). One possible way to reconcile all these outcome measures is to combine the information from all the stress measurements. In relation to this aspect, a number of recent studies set out to sub-type tested samples for a better understanding of complex disorders, such as PTSD. One approach used in these studies is latent variable modeling, which allows for the discovery of a set of invisible (thus, latent) variables based on the observed variables. This data-driven approach appears to better characterize individuals with reported symptom scores as well as with information about trauma experience and socio-demographic characteristics. Further investigations can then follow to identify the associated neurophysiological mechanisms for characterized symptom groups (Bondjers, Willebrand, & Arnberg, 2018; Galatzer-Levy, Nickerson, Litz, & Marmar, 2013; Murphy, Ross, Busuttil, Greenberg, & Armour, 2019; Rahman et al., 2018). As PTSD is highly heterogeneous and comorbid with several other psychiatric disorders (Spinhoven, Penninx, van Hemert, de Rooij, & Elzinga, 2014), it is important to find efficient ways to better describe and characterize the tested samples (i.e., phenotyping), as this may not only advance our understanding of this complex disorder but also ultimately benefit clinical populations through the use of targeted treatment strategies. For example, with the identification of neural and behavioral signatures for a specific PTSD subgroup, non-invasive brain stimulation approaches are now suggested for treating patients with PTSD (Etkin et al., 2019). Interventions using a real-time fMRI-based neurofeedback technique are also emerging and showing promising outcomes for targeting specific brain regions known to be involved in the pathophysiology of PTSD (Nicholson et al., 2017, 2018).

Concluding remarks

“It’s not stress that kills us; it is our reaction to it.” – Hans Selye

Exposure to stressful situations can lead to undesirable consequences for mental health. Such exposure can also potentially prepare us for challenges later in life, thus facilitating resilience. It is, therefore, important to uncover brain processes that result in divergent pathways of long-term stress effects. This research investigated potential biomarkers for stress vulnerability at the neural, endocrine and subjective levels. The findings suggest that focusing on acute stress responses at a large-scale network level is a promising way to investigate long-term stress effects. In particular, SN-based reconfiguration upon acute stress may serve as a potential marker for stress vulnerability.