Adaptive contextualization: A new role for the default mode network in affective learning

INTRODUCTION

Learning that a particular sensory cue signals danger is fundamental to successful behavioural adaptation in humans and other animals and is governed by similar neural networks across species [Bechara et al., 1995; Büchel et al., 1998; Fanselow and LeDoux, 1999; Freese and Amaral, 2009; McHugh et al., 2014; Sah et al., 2003]. However, learning that a cue that was previously associated with danger is now signalling safety is equally important because it reduces the physiological consequences of stress and enables the adaptation of behaviour to changing circumstances [Lupien et al., 2009]. Safety learning can be assessed through Pavlovian fear conditioning and extinction experiments, in which participants are first trained to predict an electric shock that follows an otherwise neutral tone (conditioning) and then learn not to respond to the tone with fear when it is no longer followed by the electric shock [extinction; Greco and Liberzon, 2016; Phelps et al., 2004; Schiller et al., 2008]. Animal studies have shown that extinction and hence safety learning are facilitated by activity in the ventromedial prefrontal cortex (vmPFC), which inhibits the fear response previously associated with a danger-signalling sensory cue in the basolateral amygdala [Greco and Liberzon, 2016; Do-Monte et al., 2015; Quirk et al., 2003; for a review, see Tovote et al., 2015].

In humans, neuroimaging studies have provided evidence for a similar neural mechanism of safety learning by demonstrating increased vmPFC activity during extinction [Kalisch et al., 2006; Milad and Quirk, 2010]. However, a recent meta-analysis of neuroimaging studies showed that regions overlapping with the default mode network (DMN)—including the vmPFC—are de-activated during the processing of danger cues and activated during the processing of safety cues [Fullana et al., 2016]. This finding suggests that in addition to the vmPFC, a large-scale network, and specifically the DMN may play a functional role in safety learning. The DMN is known to be activated during internally-driven cognitive operations, for example, self-referential processing, long-term memory retrieval, and mind-wandering, which are especially likely to occur during the presentation of safety cues that do not direct attention to the environment [Anticevic et al., 2012; Buckner et al., 2008; Burianová and Grady, 2007; Burianová et al., 2010]. As such, Fullana et al.'s [2016] finding of DMN activation during the presentation of safety cues could simply reflect a corollary rather than functional role during safety learning. That is, the DMN might be an epiphenomenon of safety learning—allowing the individual to engage in mind-wandering whenever they consider themselves to be safe—rather than an integral part of the learning process that leads to the impression of safety associated with a reduced fear response. However, a number of recent studies demonstrate reduced or aberrant connectivity within the DMN associated with an impairment in safety learning in patients with post-traumatic stress and anxiety disorders, thereby suggesting that in healthy individuals, activity within the DMN may play a functional role and actively contribute to successful safety learning [DiGangi et al., 2016; Jovanovic et al., 2012; Lissek and Grillon, 2010; Lissek et al., 2014b; Patriat et al., 2016].

The objective of this study was to investigate whether the DMN has a functional rather than corollary role in safety learning in healthy individuals. Learning is associated with changes in physiological and neural responses to altered cue-reinforcement contingencies, for example, if a cue were repeatedly associated with a shock, the observation of an increased BOLD response in the amygdala together with an increased physiological fear response to the cue would be interpreted as fear learning. Therefore, if the DMN were functionally related to safety learning then activity in the main nodes of the DMN would increase significantly in response to modulations of cue-reinforcement contingencies and in parallel with physiological fear responses. In contrast, if DMN activity were a consequence of safety learning, no such increases over time would be observed and the DMN would be activated to the same extent whenever safety signals are presented. To test this hypothesis, we assessed the skin conductance response (SCR) and functional brain activity during danger and safety learning with functional magnetic resonance imaging (fMRI) in a fear conditioning and extinction paradigm, in which the cue-reinforcement contingencies shift from a cue predicting danger (conditioning) to the cue predicting safety (extinction). To increase the number of changes in cue-reinforcement contingencies, a standard differential delay-conditioning paradigm was modified to include repeated conditioning and extinction phases, resulting in an A-B-A-B fear-learning paradigm. Based on previous human neuroimaging studies [Fullana et al., 2016], we hypothesized that (i) during danger cue processing, SCRs would be stronger and functional brain activity would significantly increase in the amygdala, thalamus, somatosensory cortex, anterior insula, and anterior cingulate cortex, and (ii) during safety cue processing, SCRs would be weaker and functional brain activity would significantly increase in the main nodes of the DMN, such as the vmPFC, hippocampus, inferior parietal lobule, and posterior cingulate cortex. Finally, positing that DMN activity fulfils a function in safety learning, we hypothesized that (iii) the difference between the two activation patterns associated with safety and danger learning would increase over time and that activity in DMN regions would increase with repeated extinction phases, whereas SCRs would decrease.

MATERIALS AND METHODS

Procedure

Participants took part in a partially reinforced, differential fear conditioning experiment with repeated conditioning and extinction phases (A–B–A–B paradigm; see Fig. 1). During each phase, two visual stimuli (a triangle and a circle) were repeatedly presented in randomized order. The stimuli were presented in either of two contexts (blue or orange background) associated with the different experimental phases to reduce interference from context-conditioning effects [Lang et al., 2009]. Participants were asked to identify the stimuli by pressing one of two buttons with the second and third digits of their right hand. One of the two stimuli (CS+) was paired with electro-dermal stimulation (unconditioned stimulus, UCS) during the first and third phases (conditioning) but not the second and fourth phases (extinction), while the other stimulus (CS−) was never paired with stimulation. The CS+ constitutes a danger signal during the conditioning phases and a safety signal during the extinction phases, whereas the CS− constitutes a control safety stimulus throughout the experiment. Stimuli and contexts were randomly assigned and assignments were counterbalanced across individuals. Each block started with 15 s of background presentation to allow for the electro-dermal response to settle and for the participants to habituate. During each experimental block, following 1 s of background, 20 stimuli (10 CS+, 10 CS−) were presented for 3 s and followed by 14 s of background in a randomized order. All stimuli were presented using Presentation software (Neurobehavioral Systems, Inc.) and projected onto a screen, which could be viewed with a mirror attached to the head coil.

Sixty percent of CS+ presentations co-terminated with a 50 ms electro-dermal stimulation using two pre-gelled carbon snap electrodes attached to the right wrist (EL508, Biopac Systems, Inc.). Prior to scanning, stimulation strength was adjusted to individual tolerances following established procedures to ensure that stimulation was highly uncomfortable, but not painful [LaBar et al., 1998]. Stimulation was administered using a STIMISOC isolator connected to a STM100C stimulator, which was controlled by a MP150 (Biopac Systems, Inc.).

Skin conductance responses (SCRs) were sampled at 1 kHz using pre-gelled carbon snap electrodes (EL508, Biopac Systems, Inc.) attached to the medial phalanges of the second and third digits of the left hand and connected to an EDA100C module attached to a MP150 (Biopac Systems, Inc.). SCRs were defined as the peak response of the low-pass filtered (0.1 Hz) electro-dermal activity above 0.02 µS that occurred within 1–4 s after the onset of the conditioned stimuli [Lockhardt, 1966].

RESULTS

There were no significant differences in reaction times in response to CS+ and CS− presentations. However, a trend suggested slower reaction times in responses to CS+ than to CS− in the initial conditioning phase (t(29) = 1.9, P = 0.07, d = 0.39; see Fig. 2a). Psychophysiological evidence of successful differential fear conditioning was provided by a 2 × 2 × 2 analysis of variance of the SCRs with factors stimulus (CS+, CS−), context (conditioning, extinction), and time (first, second presentations) that yielded significant main effects for the factors stimulus (F(1,1) = 5.4, P < 0.05) and context (F(1,1) = 17.2, P < 0.001). Repeated, two-sided t-tests demonstrated significantly larger SCRs to CS+ than CS− presentations as well as to CS presentations during the conditioning than during the extinction phases (all t(29) > 2.1, P < 0.05, d > 0.27; see Figure 2b).

Whole-brain fMRI analysis resulted in one significant LV (P < 0.001) accounting for 51.4% of the variance in the data. The LV differentiated activity related to danger cues from activity related to safety cues. In contrast to safety cues, the presentation of danger cues (i.e., CS+ during the first and second conditioning phase) engaged dorsal anterior cingulate cortex, bilateral insula, middle temporal gyrus, temporal pole, ventrolateral prefrontal and inferior parietal cortex, thalamus, brainstem, and medial cerebellum (see cool coloured regions in Figure 3). In contrast to danger cues, the presentation of safety cues (i.e., CS+ and CS− during the first and second extinction phase as well as CS− during the second conditioning phase) engaged orbitofrontal cortex, ventromedial prefrontal cortex, subgenual anterior cingulate cortex, bilateral hippocampus, precuneus, superior parietal gyrus, right fusiform gyrus, medial superior frontal gyrus, posterior cingulate gyrus, ventral striatum, and lateral cerebellum (see warm coloured regions in Fig. 3). The results further show that this pattern increased over time and repeated measures, two-tailed t-tests revealed significant differences between presentations of CS− during the first and the second conditioning and extinction phases (for both t(29) > 2.8, P < 0.008, d > 0.57) but not between presentations of CS+ during the first and second extinction phase (t(29) = 0.7, P = 0.5, d = 0.16).

The comparison between the DMN mask derived from 820 resting-state data sets and our results demonstrate an overlap of 42% of the pattern associated with safety learning and regions considered to form the core of the DMN [Andrews-Hanna et al., 2014] including the ventromedial prefrontal cortex, anterior and posterior cingulate cortex, superior frontal gyrus, and angular gyrus (see Fig. 4).

DISCUSSION

The aim of this study was to investigate whether the activation of the DMN that has been observed incidentally in previous fear conditioning studies [Fullana et al., 2016] is functionally related to safety learning. Based on patient studies, which link impaired safety learning to a reduction in DMN activity, we hypothesized that in healthy adults, the DMN would be activated in response to safety cues and that DMN activity would increase as learning progresses with repeated presentations of safety cues, whereas physiological fear measures would decrease [DiGangi et al., 2016; Jovanovic et al., 2012; Lissek and Grillon, 2010; Lissek et al., 2014b; Patriat et al., 2016]. Our results demonstrate activity within the core nodes of the DMN (ventromedial prefrontal cortex, anterior and posterior cingulate cortex, superior frontal gyrus, and angular gyrus) in response to safety cues. Importantly and in contrast to previous studies [Fullana et al., 2016], the multivariate, data-driven method used in our analysis was able to identify the hidden factors or latent variables underlying the co-variation within brain regions, thereby yielding two activation patterns that are negatively correlated with each other [Abdi and Williams, 2010]. As a consequence, our results extend previous findings by showing that DMN activity associated with safety cues is directly anti-correlated with activity in regions associated with fear learning. In other words, these findings strongly suggest that the DMN is deactivated whenever the fear-learning network is activated and vice versa. Furthermore, we show increases in DMN activity over time in response to safety cues during extinction (CS+ and CS−) and also, importantly, during the conditioning phases (CS−), while the physiological indicator of fear (skin conductance response) decreases over time in response to safety cues during extinction (CS+ and CS−) as well as during the conditioning phases (CS−). These parallel changes over time not only demonstrate that the DMN is related to safety learning but also suggest that activation of the DMN is not an epiphenomenon of safety learning but that it performs the function of safety learning. Taken together, the findings of this study show that the DMN is activated during safety learning, directly anti-correlated with fear learning and, its activity increases over time in parallel to decreasing physiological indicators of fear. Thus, our study provides strong evidence for a functional rather than a corollary role of the DMN.

Based on these findings, we propose that it may be the DMN's function during affective learning to contextualize safety memories and that such contextualization of safety memories inhibits fear responses and thereby contributes to the adaptive generalization of fear associations. At the core of our proposal is the idea that the engagement of the DMN during safety learning reflects the contextualization of safety memories. The DMN is mainly associated with perceptually decoupled and internally-driven cognitive processes, such as mind-wandering, self-referential processing, and autobiographical long-term memory retrieval [Andrews-Hanna et al., 2010; Binder et al., 1999; Herold et al., 2016; Mantini and Vanduffel, 2013; Mason et al., 2007; Smallwood et al., 2013; Spreng and Grady, 2010]. However, a number of studies have reported DMN activity related to behavioural performance in stimulus-directed tasks [Chen et al., 2013; Elton and Gao, 2015; Gilbert et al., 2007; Hahn et al., 2007; Smallwood et al., 2012; Vatansever et al., 2015], suggesting that the DMN is also involved in unfocused attention and broad monitoring of the external environment for unexpected events [Buckner et al., 2008; Leech and Sharp, 2014; Pearson et al., 2011; Mantini and Vanduffel, 2013]. The convergence of these two different aspects of the DMN is plausible because stimulus-independent processes (such as self-generated thoughts) often occur in conjunction with unfocused monitoring of the environment (as when one travels on the bus, for instance). In a recent proposal, Hasson et al. [2015] provide evidence for the view that these two aspects—internally-driven processing and unfocused environmental monitoring—are even more integrated and argue that the DMN is engaged in the formation of a ‘process memory’ that constitutes the meaningful continuous context of our subjective experience over time on the scale of several minutes. As such, the DMN seems to be responsible for the maintenance of a cohesive sense of self across time that forms a reference point for the interpretation of novel information [Andrews-Hanna et al., 2014; Kajimura et al., 2016; Northoff and Bermpohl, 2004; Roy et al., 2012]. Our finding that the DMN is engaged during safety learning might thus reflect the process of embedding safety cues that predict the absence of shock within the context of self-generated thought.

The contextualisation of safety memories is important for fear learning in general, as adaptive fear learning depends not only on the generation of appropriate fear responses but also on the inhibition of inappropriate fear responses [Rescorla and Wagner, 1972]. Fear associations readily generalize to similar stimuli, whereas the inhibition of fear responses is highly context-dependent, which suggests that context-dependent inhibition plays a crucial role in preventing the maladaptive overgeneralization of fear associations [Bouton, 2004; Quirk et al., 2003]. Recent evidence implicates a role of the DMN in the regulation of fear generalization by showing that the generalization of fear is counterbalanced by activity within the DMN [Lissek et al., 2014a]. Such a regulatory function is commonly associated with the ventromedial prefrontal cortex (vmPFC)—the frontal node of the DMN, which shows the strongest overlap with the areas activated during safety learning in our results. Animal and human studies have demonstrated that the vmPFC is engaged in the inhibition of fear responses [Todd et al., 2014; Quirk and Beer, 2006] and that it plays a major role in regulating emotions and decision-making [Bechara et al., 2000; Etkin et al., 2011]. The engagement of the vmPFC and the DMN during safety learning suggests that the contextualisation of safety memories contributes to the inhibition of inappropriate fear responses and consequently to the adaptive generalization of fear.

Importantly, our findings demonstrate that the DMN is engaged in safety learning during both extinction and differential conditioning, that is, learning that both a previously conditioned stimulus (CS+) after a period of non-reinforcement and a non-reinforced conditioned stimulus (CS−) within the context of threat signal safety. It is yet unclear how these different forms of safety learning are related to each other and more research is necessary to address this question. Based on our results, we suggest that the DMN might constitute a high-level form of safety learning that integrates other cognitive functions involved in threat and safety processing, such as perception, attention, cognitive inhibition, and higher-order reasoning, as well as working and long-term memory [Anderson et al., 2016; Dayan et al., 2016; Elton and Gao, 2015]. The DMN's functional profile, the rich-club properties of its hub regions (especially the medial prefrontal and medial temporal cortices), as well as its effective, global structural connectivity make it well suited for such an integrated, contextual form of safety learning [Van den Heuvel and Sporns, 2011; Vatansever et al., 2015].

Our proposal that the DMN's function in adaptive, affective learning may be the contextualization of safety memories has important implications for pathologies that are characterized by an over-generalization of fear associations, more specifically for anxiety and stress disorders [Dymond et al., 2015; Jovanovic et al., 2012]. The findings of this study suggest that these patient populations might be impaired in their ability to appropriately contextualize and integrate safety cues into their current experience as a result of insufficient recruitment of the DMN or less sustained activation within the DMN. Our proposal might help improve our understanding of aberrant DMN activity and connectivity commonly found in patients suffering from anxiety disorders [Zhao et al., 2007], post-traumatic stress disorder [Bluhm et al., 2009; Reuveni et al., 2016], or bipolar disorder [Vargas et al., 2013], as these disorders are also characterized by an impairment in safety learning [Grillon, 2002; Jovanovic et al., 2012; Kim et al., 2011; Woody and Rachman, 1994] and pathological over-activation of limbic regions and over-generalization of fear associations [Dymond et al., 2015; Grillon and Morgan, 1999; Lissek and Grillon, 2010; Lissek et al., 2014b]. Based on the findings of this study, we suggest that disruptions in functional and structural connectivity within the DMN might be a potential biomarker for the early detection of insufficient contextualization of safety memories in affective, mood, and stress disorders, informing effective therapeutic and pharmacological interventions. In addition, our results support emerging evidence that patients with affective disorders might benefit from therapeutic interventions, which target the DMN through mindfulness techniques or brain stimulation [Dichter et al., 2014; Fox et al., 2014; Hölzel et al., 2011; King et al., 2016; Liston et al., 2014; Taylor et al., 2013].