Our aim with these two meta-analytic reviews was to help advance understanding of how human cognition operates in affectively laden environments, by synthesizing data on the impact of affective information on WM and the neural correlates of this effect. WM is implicated in virtually all day-to-day cognition (Barrett et al., 2004; Engle & Kane, 2004; Miyake & Shah, 1999) and much of its operation takes place in affective contexts, ranging from the overt manipulation of affective information to the performance of relatively neutral tasks in the context of affectively laden goals and plans. The studies reviewed here have tried to measure these forms of interplay by looking at WM in affective versus comparatively affect-neutral contexts within the laboratory and scanner using carefully controlled tasks.10 The challenge inherent in these tasks is to pursue the relatively neutral task goals while dealing with affectively laden contexts of different types as a proxy for the challenges faced in day-to-day cognition.
Our findings show that neural and behavioral data reviews and syntheses can complement each other; in this case with evidence for widespread neural engagement that arguably reflects broader cognitive engagement than the resultant behavioral data reveal (Barrett, 2009). This is a vindication of models such as Pessoa’s (2009) DCF and others (e.g., the conceptual act theory, Barrett, 2014; the model of the cognitive control of emotions, Ochsner et al., 2012) that seek to generate and integrate sets of both behavioral and neural predictions. These complementary insights from the current set of reviews further highlight the importance for future data synthesis endeavors of including, where possible, measures of behavioral performance as well as functional neuroimaging data. While this conclusion appears self-evident there is a surprising lack of meta-analytic reviews that integrate findings in this manner.

The dissociation between behavioral and neural findings in healthy individuals was strongest for the moderating effect of task-relevance. Interestingly, across the behavioral studies affective targets had a negligible-to-small enhancing effect on WM accuracy, whereas the effect of affective distractors was small and dependent on valence. The neuroimaging meta-analysis, however, showed that affective distractors led to more frequent recruitment of the predicted brain regions (including, amygdala, vlPFC) whereas affective targets did not. Moreover, task-irrelevant affective distractors had a greater impairing effect on WM accuracy compared with affective targets in individuals with mental health issues.
The differential behavioral and neural effects of affective stimuli on WM in healthy individuals arguably evidence the efficiency of the cognitive control system in mitigating any impact of affective information on performance. The increased recruitment of the vlPFC may reflect the organism’s effort to inhibit attention and responses toward these distractors and regulate any affective experience elicited by the distractor. This is particularly adaptive in our contemporary environments that are populated with myriads of affective distractors (e.g., phone alerts). This dissociation of behavioral and neural results observed in healthy individuals is in line with research into the interaction between affect and other types of cognition including long-term memory (Erk, von Kalckreuth, & Walter, 2010). This dissociation appears to be maintained across time (Erk et al., 2010), with behavioral memory performance for affective material being unaffected by whether individuals had been instructed to regulate their affective responses to the memoranda at encoding 12 months prior (in line with Dolcos, Labar, & Cabeza, 2005). At the neural level, however, amygdala activation during encoding of affective items that were viewed without attempts to downregulate affective experiences was stronger than amygdala activation to items encoded 12 months prior while individuals were attempting to regulate their affective responses. The reviewed evidence further suggests that it is in particular the connectivity between this vlPFC node and the amygdalo-hippocampal complex that reflects the efficacy of healthy individuals in controlling any potential interference from affective information in WM (Krause-Utz, Elzinga, Oei, Paret et al., 2014; Ladouceur et al., 2013; Ziaei, Salami, & Persson, 2017). Interestingly, Ladouceur et al. (2013) showed reduced downregulation of amygdala reactivity by the vlPFC in response to negative and positive distractors in young people with a parent suffering from bipolar disorder compared with a healthy age-matched sample. This differential pattern of neural activation across groups was observed in the absence of behavioral performance differences. Altered functional connectivity during WM performance in the presence of affective compared with neutral material may therefore constitute a sensitive marker for mental health problems before the behavioral differences that were observed in the current behavioral meta-analysis emerge. Together the studies lend support to the argument that competition for resources from affective information is being routinely resolved in the vlPFC.
In mental ill health, however, maladaptive behavioral responses and involuntary attentional engagement with affective distractors are characteristic of many disorders (e.g., anxiety disorders; Bar-Haim et al., 2007). WM performance and its neural substrates in the presence of affective distractors may therefore constitute a source of individual differences associated with mental health problems. In line with this argument, Menon’s (2011) triple neural network model of mental health proposes that weak mapping from the salience network is involved (among other things) in “[. . .] aberrant bottom-up detection of salient events, [and] aberrant control signals to other large-scale networks that facilitate access to attention and working memory resources, [. . .]” (Menon, 2011, p. 501). That is, mental health problems may be associated with particularly impaired WM performance in the presence of affective distractors due to both aberrant salience attribution to affective information at the perceptual level of competition as well as impaired control at the executive level of competition.

A critical prediction, although somewhat underresearched in the literature, is the impact of stimuli’s degree of affective significance on executive performance. Here we used age and mental health status as proxies for affective significance. In line with our predictions older people were faster to respond to positive material and WM performance in individuals with mental health problems was significantly impaired by affective information.
The age results were in line with the age-related positivity effect shown in the attention and memory literature (for a meta-analytic review, see Reed, Chan, & Mikels, 2014). However, little is known about the development of WM in affective contexts from childhood through into adulthood. Of the included studies fewer than 10% (n = 14) were conducted in children and/or adolescents (Bertocci et al., 2014; Cromheeke & Mueller, 2016; Ladouceur et al., 2005, 2013, 2009; Mueller et al., 2015; Passarotti, Ellis, Wegbreit, Stevens, & Pavuluri, 2012, 2010, 2011; Pavuluri, Passarotti, Fitzgerald, Wegbreit, & Sweeney, 2012; Schenkel, Passarotti, Sweeney, & Pavuluri, 2012; Tavitian et al., 2014; Visu-Petra, Ţincaş, Cheie, & Benga, 2010; Z. Li et al., 2009) and there was no study of the typical development of affective WM. This is particularly surprising given that affective WM in developmental samples may provide evidence for those at risk for emotional disorders by virtue of problems with affective control capacity. Moreover, interventions that augment executive control in affective contexts may constitute efficient forms of prevention, especially when administered early in development (Wass, Porayska-Pomsta, & Johnson, 2011).
There are likely to be variations in the pathways through which the effects of affective significance create perceptual and executive competition across different mental health disorders. Arguably, differences in affective significance may exert their impact on perceptual competition in a similar way across diverse forms of psychopathology, whereas the intersection of affective significance and executive competition may rely upon different mechanisms across disorders. For example, engaging in cognitively costly emotion regulation strategies (including rumination in depression, or suppression in anxiety disorders; Aldao et al., 2010) in response to affective stimuli versus increased executive competition due to resources deployed to disambiguate affective information in schizophrenia (Kohler, Walker, Martin, Healey, & Moberg, 2010). Similarly, there are likely to be variances in the relative affective significance of the stimuli included in standard experimental paradigms across disorders. Despite these potential differences, all of the mental health disorders included in the current behavioral meta-analysis have been shown to be associated with affective dysregulation: alcohol dependence (Cheetham, Allen, Yücel, & Lubman, 2010); anxiety disorders (Cisler, Olatunji, Feldner, & Forsyth, 2010); attention-deficit and hyperactivity disorder (Graziano & Garcia, 2016; Shaw, Stringaris, Nigg, & Leibenluft, 2014); borderline personality disorder (Carpenter & Trull, 2013); mood disorders (Hofmann, Sawyer, Fang, & Asnaani, 2012; Townsend & Altshuler, 2012); obsessive–compulsive disorder (Calkins, Berman, & Wilhelm, 2013); PTSD (Frewen & Lanius, 2006); and schizophrenia (Horan, Kring, & Blanchard, 2006; Trémeau, 2006). Poor WM performance in the presence of affective material then may be a transdiagnostic marker of dysregulated affect across these disorders.
At the neural level the paucity of available studies means that the current analysis cannot speak to finer-grained questions concerning the neural substrates of the effects of affective significance across disorders. As and when further evidence emerges on affective WM from each disorder, future meta-analyses should investigate the interaction between behavioral and neuroimaging findings in these clinical populations. We currently know little about the neural substrates associated with individual differences in affective WM and potentially different pathways to interference from affective information across mental health problems. As with the posited cognitive-level pathways, the neural signatures are argued to be both overlapping and distinct across different types of mental health problems. Interestingly, the networks proposed in Menon’s (2011) triple neural network model of mental health overlap with the neural networks shown in the current meta-analysis to be associated with the effects of affective stimuli on WM performance (i.e., the salience and fronto-parietal control networks). Future research is warranted to explore the neural substrates of affective WM both within and across disorders.
