Nearly half of all individuals with major depressive disorder (MDD) also have an anxiety disorder (Kessler et al., 2015). Thus, comorbid MDD and anxiety disorders (e.g., generalized anxiety disorder, social anxiety disorder, panic disorder, or specific phobia) are among the most common presentations of affective psychopathology in psychiatric settings. Anxious depression is associated with greater treatment resistance (Ionescu et al., 2014), worse side effect profiles (Gaspersz et al., 2017), poorer treatment outcomes (Penninx et al., 2011), quicker symptom relapse (Fava et al., 2008) and higher levels of suicidal ideation (Kessler et al., 2015) than non-anxious depression. However, treatment engagement appears to be higher in this comorbid group (Kessler et al., 2015). Despite the pervasiveness of anxious depression, the neural and behavioral processes characterizing this form of psychopathology remain incompletely understood. A better understanding of the underlying process dysfunctions may be an important first step in the development of more mechanistically oriented interventions.

Depression and anxiety share abnormalities across a number of different neurocognitive processes, such as negative affect (Clark and Watson, 1991), life stress (Syed and Nemeroff, 2017), and cognitive function (Millan et al., 2012). However, studies comparing the mechanisms of anxious vs. non-anxious depression are rare, which may be one factor impeding improvements in treatment development. Hyperarousal, defined as an abnormal state of increased responsiveness to stimuli that is marked by various physiological and psychological symptoms, is a feature of both anxiety and anxious depression (Clark and Watson, 1991) that is not present in non-anxious depression. It may therefore be a useful intermediary target mechanism. Since anxiety disorders are also commonly characterized by abnormalities in the domain of interoception (Khalsa et al., 2018), the mechanistic probing of interoception in anxious depression might inform a deeper understanding of the underlying pathophysiology that could help to tease apart the differences between anxious and non-anxious depression.

Interoception refers to the process by which the nervous system senses, interprets, and integrates signals originating from within the body (Khalsa et al., 2018). Interoceptive awareness has many features, including: (1) interoceptive attention (i.e., cued or spontaneous focus on internal physiological sensations); (2) interoceptive accuracy (i.e., the match between true and consciously perceived internal physiological sensations); (3) interoceptive magnitude (i.e., the perceived intensity of internal physiological sensations (Khalsa and Lapidus, 2016; Khalsa et al., 2018); (4) interoceptive sensibility (i.e., self-perceived tendency to focus on interoceptive stimuli (Garfinkel et al., 2015); and (5) interoceptive insight (i.e., the metacognitive correspondence between self-perceived and objective measures of interoceptive accuracy). Whether, and how often interoceptive processes enter awareness varies according to the organ system and level of physiological arousal, and therefore multimodal assessments involving self-report, behavioral, and physiological forms of measurement across multiple domains may be important for gaining an informed perspective on this process. The insular cortex is a key hub within the neural circuitry implicated in interoception (Berntson and Khalsa, 2021; Nord et al., 2021). Related to this, models of embodied emotion (Craig, 2008; Smith and Lane, 2015; Barrett, 2017) have been proposed that integrate interoception with wider emotional processing and motivation. In this view, the co-activation of regions such as the anterior insula and anterior cingulate cortex (ACC) during emotion processing (Thielscher and Pessoa, 2007) and across psychiatric disorders (McTeague et al., 2020) is consistent with the view that an emotional experience involves body states detected via interoceptive processes that may be altered in psychopathology. It follows from this that functional neuroimaging tasks which specifically perturb interoception can elucidate the neural circuits underlying interoceptive psychopathology (Smith et al., 2021; Teed et al., 2022). Furthermore, the anatomical overlap between circuits associated with fear conditioning (Maren and Quirk, 2004) and the central autonomic network (Thayer and Lane, 2000; Smith et al., 2017) includes the ventromedial prefrontal cortex (vmPFC), cingulate cortex, insula, and amygdala. This overlap forms the basis of a neurovisceral model of fear conditioning (Battaglia and Thayer, 2022) and makes the study of aversive interoception particularly relevant.

Aberrant interoception has been strongly linked to many forms of psychopathology beyond anxious depression, including MDD (Avery et al., 2014; Wiebking et al., 2015), substance use disorders (Paulus and Stewart, 2014; Stewart et al., 2020), eating disorders (Khalsa et al., 2015; Kerr et al., 2016; Berner et al., 2018), suicidal behaviors (DeVille et al., 2020, 2021), and schizophrenia (Yao and Thakkar, 2022). Yet findings in depression are mixed. For example, in one study (Dunn et al., 2007), a severely depressed group did not differ from healthy comparators in heartbeat perception accuracy, whereas a more moderate group showed significantly lower accuracy, as also seen in other studies (Terhaar et al., 2012). This could suggest a non-linear effect of depression on interoception (Eggart et al., 2019). Alternatively, it might be explained by the heterogeneity of depression and/or the inclusion of co-morbid anxiety in the patient groups under investigation. Depression has been generally associated with blunted cardiac interoceptive awareness (Pollatos et al., 2009; Furman et al., 2013), whereas anxiety has often been conceptually associated with increased sensitivity (Van der Does et al., 2000; Domschke et al., 2010), perhaps driven by abnormally heightened interoceptive prediction signals of a somatic threat (Paulus and Stein, 2006; Khalsa and Feinstein, 2018). Interactions in the effect of depressive and anxiety symptoms on interoception have been suggested dimensionally in healthy individuals (Pollatos et al., 2009) and those with mood disorder symptoms (Dunn et al., 2010), although firm conclusions from such studies are limited by reliance on a measure of cardiac interoception with several drawbacks (Desmedt et al., 2018; Ring and Brener, 2018; Corneille et al., 2020; Ferentzi et al., 2022). Despite the available evidence of transdiagnostic interoceptive dysfunction, there is little work comparing interoception in individuals with MDD vs. those with MDD and comorbid anxiety.

The goal of the current cross-sectional investigation was to determine whether individuals with anxious depression show different sensitivity to interoceptive sensations of a threat compared to non-anxious depressed individuals. Our approach was to compare a sample of depressed individuals who were propensity-matched based on demographic characteristics to individuals with comorbid anxiety and depression using an interoceptive battery of tasks developed to measure different aspects of cardiorespiratory interoceptive and nociceptive processing. Our overarching hypothesis was that individuals with anxious depression would show increased interoceptive reactivity to aversive sensations relative to their depressed counterparts.

This study examined whether individuals with anxious depression exhibit increased sensitivity to aversive interoceptive signals relative to demographically matched individuals with non-anxious depression. There were two main results: first, the Dep+Anx group reported greater fear of suffocation during a breath-hold task, despite similar breath-hold durations compared to the Dep group. Second, the Dep+Anx group removed their hand from painful cold water more quickly than the Dep group, despite having similar self-reported pain ratings over time. However, there were no differences between the groups on the heartbeat-perception task or the VIA task. This could be explained by the Dep+Anx group (relative to Dep only and possibly relative to healthy comparisons) perceiving body signals that accompany the breath-hold task as more threatening and reporting this subjectively. Then, in the cold-pressor task this Dep+Anx group reacted in such a way as to avoid threatening nociceptive signals by terminating the trial earlier. Together, these results support the hypothesis that anxious depression is characterized by an exaggerated response to aversive interoceptive signals relative to those with non-anxious depression and reduced tolerance of nociceptive threat. To our knowledge, this is the first time that differences in interoceptive reactivity between anxious vs. non-anxious depression have been identified, which was specifically enabled by our propensity-matching approach.

Previously, we showed increased emotional reactivity to threat in a startle EMG task in anxious depression compared to non-anxious depression, using the same propensity-matched sample (Ironside et al., 2022). An association between emotional reactivity and increased interoceptive awareness has also been previously shown (Pollatos et al., 2007; McTeague et al., 2011), which may suggest that a chronically increased outflow of sympathetic signals might be one variable contributing to the establishment of high interoceptive sensitivity/fearfulness, though potentially to the detriment of interoceptive accuracy (Paulus and Stein, 2010). Interestingly, in the cold pressor challenge, both groups had similar pain ratings over time, but the Dep+Anx group turned these feelings (pain) into action (hand removal) more quickly than the Dep group. This could be explained by reduced cognitive control or stronger emotional experiences in anxious depression for the same level of pain. Another possibility is a potential limbic-motor dysfunction in non-anxious depression, which prevents the interoceptive signal from activating motivational resources. In contrast, we see at least a preservation, if not an exaggeration of motivational activation to bodily sensations of threat in anxious depression. One explanation of altered interoception in psychiatric illnesses suggests a combination of rigid and exaggerated expectations of aversive interoceptive experiences and low signal-to-noise ratio of visceral input to the insula, resulting in erroneous evaluation of benign signal changes as significant, with accordingly motivated action (Paulus et al., 2019). It is unclear if the group difference seen here represents limbic motor dysfunction in non-anxious depression or erroneous evaluation in anxious depression, or (likely) a combination of the two.

Previous work based on symptom patterns supports a tripartite model (Clark and Watson, 1991), i.e., a bifactor model consisting of a general negative affect factor and specific depression and anxiety factors. The anxiety-specific factor in the tripartite model is hyperarousal. In the breath-hold and cold pressor tasks, the aversive interoceptive signal activates a defensive system. As expected from the tripartite model, the Dep+Anx group show hyperarousal, represented by higher fear of suffocation and quicker hand removal, which is not observed in the Dep group, suggesting a key difference between these sub-groups. In terms of behavior, the approach-withdrawal model helps tease apart the different factors present in anxious depression, with depression being associated with reduced approach and anxiety being associated with increased avoidance (Davidson, 1998). In the present study, the non-anxious depressed group showed a reduced degree of defensive action during a nociceptive threat (i.e., hand withdrawal), which was preserved in anxious depression. Future research should examine how those with anxious vs. non-anxious depression respond to treatments targeting the defensive action system. Compared to a group with anxiety only, it seems possible that an anxious depressed group might show both reduced approach and increased avoidance (e.g., using approach/avoidance threat paradigms). Further studies are needed to tease apart the potentially unique disease that the intersection of anxiety and depression represents.

We previously proposed (Ironside et al., 2022) a process model for anxiety and anxious depression to help distinguish disease modifiable processes that could be useful targets in treatment development. Individuals with anxiety may be characterized by an excessive defense system that takes incoming or internally generated stimuli and evaluates them with respect to threat-to-self, with a high prior expectation of aversive outcomes (Paulus et al., 2019). This process may be highly taxing and may result in exhausting affective processing capacities ultimately resulting in depression. This proposition is consistent with the observation that two-thirds of individuals with lifetime comorbid anxiety disorders and MDD reported an earlier age-of-onset of their first anxiety disorder than their MDD (Kessler et al., 2015). It is unclear whether this subsequent depression is characterized by a lack of response to both positive/negative affect, i.e., if anxious depression often becomes non-anxious depression eventually. It is also unlikely that the entire sub-group will show the same phased process. In comparison, depressed-only individuals might be characterized by a primary lack of reactivity to positive and negative stimuli, which results in a lack of anxious responding even when threatening stimuli are encountered. Therefore, whereas the primary disease-modifying process for anxious depression would be to attenuate threat-related processing, the primary disease-modifying process for non-anxious depression would be to enhance valence-independent motivation in general. A lot of focus is given to reward responsivity in depression, but we propose a more general factor of blunted motivation (both approach and avoidance motivation), supported by our prior (Ironside et al., 2022) and current findings. This would modify the tripartite model, transforming the anxiety-specific factor of hyperarousal into a general “arousal/motivation” factor, with depression and anxiety pulling in opposing directions. One consequence of this suggestion would be that treatments targeting hyperarousal in non-anxious depression could actually worsen an already blunted emotional reactivity. Interestingly, the same pattern of blunted aversive interoception was found when comparing the larger T500 study sample in terms of their past suicide attempt history, with the past attempters showing higher pain tolerance in the cold pressor challenge and less suffocation on a breath-hold task (DeVille et al., 2020, 2021). Perhaps this is descriptive of the process model suggested above, with initial anxious depression as a predecessor to the potentially more chronic state of non-anxious depression. However, this is at odds with worse reported outcomes and higher rates of suicidality in anxious depression.

The present study had several limitations. First, case-control designs and cross-sectional studies cannot support causal inferences. Longitudinal designs will be necessary to determine cause and effect. Second, the Dep+Anx group was twice as large, which may have increased the statistical power to detect effects in the Dep+Anx vs. the Dep-only group. However, as there were no group differences in the heartbeat task or the VIA tasks, this concern does not appear to have indiscriminately influenced the current findings. Third, we used a somewhat imprecise, albeit commonly employed, measure of nociceptive stimulation in the cold-pressor challenge. It would be advantageous to clarify whether anxious depression is associated with impaired processing of other pain signals, such as heat (Thompson et al., 2008) or pinprick forms of pain. Furthermore, heartbeat perception tasks have been the subject of criticism, reflecting the difficulty of assessing cardiac interoception at rest (Khalsa and Lapidus, 2016). We attempted to overcome this with our endogenous perturbation approach (i.e., breath-hold incorporation), but other exogenous perturbation approaches (e.g., Teed et al., 2022) are worth utilizing in future investigations. In addition, according to our power calculations our fMRI analyses were underpowered and therefore could only be deemed exploratory. Finally, without including an anxiety-disorder only comparator group it is unclear if the increased reactivity observed in our sample is due to anxiety, or whether the interaction of depression and anxiety has a unique profile. Future studies should assess differences in interoceptive and nociceptive processing between those with anxiety disorders only, depression only, and comorbid anxiety and depression. We did not find significant differences between the groups on a number of measures. However, those that involve aversive sensations such as the Cold-Pressor Challenge and the Breath-Hold Challenge showed group differences. This could be because of the limitations of the tasks outlined above, or, it could be indicative of the group differences stemming from the processing of aversive sensations only and thus driven by differences in threat sensitivity. Additionally, the fact that there were no statistically significant group differences in the activity of insula subregions during interoceptive attention (possibly due to low statistical power) limits our ability to dwell further on the possible neural correlates of differences in aversive interoceptive reactivity in individuals with anxious depression vs. those with depression alone. One possibility is that the observed differences in perception are related to activity in brain regions other than the insular cortex, although further study would be needed to verify this notion.

In sum, these findings provide further support for the view that anxious and non-anxious depression have distinct neurocognitive and neurophysiological profiles. One outcome of this arrangement may be that exaggerated reactivity to aversive interoceptive sensations in anxious depression could promote avoidance behavior, whereas a lack of valence-independent motivation could be a driving factor of similar outcomes in non-anxious depression. Our findings add to the growing body of evidence that interoceptive and nociceptive signals are processed differently in individuals with comorbid depression and anxiety than in individuals with depression alone, providing support for a process model of increased threat sensitivity and hyperarousal in anxious depression.

The raw data supporting the conclusions of this article will be made available by the authors pending the establishment of data sharing agreements between all parties.

The studies involving human participants were reviewed and approved by Western Institutional Review Board. The participants provided their written informed consent to participate in this study prior to study entry, and received compensation for their study involvement.

MP and SK: study conception and design. MI, DD, RK, AT, KB, RS, MP, and SK: analysis and interpretation of results. MI, DD, and SK: draft manuscript preparation. All authors contributed to the article and approved the submitted version.