Affectively Biased Competition: Sustained Attention is Tuned to Rewarding Expressions and is Not Modulated by Norepinephrine Receptor Gene Variant

A total of 310 participants (76 males, 1 non-binary) were recruited from the University of British Columbia and received course credit or $20 for participating in the study. Participants were of European Caucasian descent, had no history of anxiety disorders, traumatic brain injury and epilepsy, and reported normal to corrected-to-normal vision. 113 were excluded due to a programming error in resulting in slow monitor refresh rates, such that no peaks were observable at the driving frequencies, 12 participants were excluded due to insufficient SNR among posterior electrodes, one participant was excluded after visual inspection of frequency amplitudes, and genotyping failed for two participants. There were additional errors in recording accuracy data for four participants, who were omitted from the analysis (due to a bug in the software, accuracy values were uninterpretable – not 0 or 1). Thus, the final N was 182 (46 males, 1 non-binary, mean age = 20.75 years, SD = 2.85, 87 ADRA2B deletion carriers). All procedures were approved by the clinical research ethics review board of the University of British Columbia. The sample size was determined a priori by power analysis based on effect sizes observed forADRA2B effects on emotional bias in previous studies. After discovering our screen refresh error we collected additional data to meet our sample size estimates. To mitigate against spurious false positives found in targeted polymorphisms studies with relatively small sample sizes (Grabitz et al., 2018), we adopted a split-sample approach to the data analysis: Following data collection, equal numbers of ADRA2B deletion carriers and noncarriers were randomly assigned to the discovery and replication samples. There were 90 participants in the discovery sample (19 male, 1 non-binary, mean age = 20.71SD = 2.74, 43 deletion carriers), and 92 participants in the replication sample (26 male, mean age = 20.79, SD = 2.98, 44 deletion carriers). For demographic information see Table 1.

Upon arriving in the lab, participants were asked to provide consent and respond to questions regarding vision, hours of sleep, caffeine and alcohol consumption, and history of traumatic brain injury. Additional questionnaires were also administered as part of a standard battery but were not examined for this study. EEG data was subsequently collected while participants performed a change detection task. Following EEG recording, participants performed a stimulus rating task wherein 72 facial stimuli used in the EEG task were rated on scales for valence and arousal. Finally, participants were asked to provide a (~2 ml) saliva sample, collected using an Oragene OG-500 DNA kit (DNA Genotek, Ottawa, ON; http://dnagenotek.com) at the end of each testing session. Genotyping for ADRA2B was performed at CAMH in Toronto (1^st 125 participants of the final sample) and at the BCCH (BC Children’s Hospital Research Institute) in Vancouver (58 participants of the final sample).

Seventy-two photographs of 14 actors (12 female) posing angry, neutral, and happy facial expressions were selected from the Karolinska Directed Emotional Faces (KDEF; Lundquist, Flykt, & Ohman, 1998; http://www.facialstimuli.com). These images were converted to grayscale. A Gabor patch was overlaid on each image in a change detection task. Gabor patches were 225 × 305 pixels, with 8 cycles across the vertical dimension. Stimuli were 562 × 762 px (6.244 × 8.467 inches) images that were presented against a gray background on a 24-in monitor with a vertical refresh rate of 60 Hz using Presentation software.

Participants performed a change detection task that was adapted from Wieser, McTeague, and Keil (2012) (Figure 1). Frequency tagging was used to flicker overlaid stimuli at opposing frequencies, producing distinct electrocortical signatures, such that power at each driving frequency indexed allocation of cortical resources to that stimulus. In this way we could measure attentional allocation to two distinct stimuli that occupied the same regions of space. Each experimental trial consisted of a face stimulus flickered at one of two frequencies (either 15 or 20 Hz), overlaid by a semi-transparent Gabor patch that flickered at the other frequency (e.g., if the face was flickered at 15 hz, the Gabor patch was flickered at 20 hz and vice versa). Both stimuli were presented foveally for 3000 ms. The driving frequencies of face vs. gabor patch were counterbalanced in blocks within subjects across 144 experimental trials. Every trial consisted of 0, 1, or 2 phase reversals of the transparent grating with the first change occurring between 666 ms and 1333 ms and the second change occurring between 1666 ms and 2333 ms. At the end of each trial, a prompt appeared asking the participant to press the number key indicating how many changes were observed (Figure 1b). Whereas the gabor patches were targets, the faces were task-irrelevant. 16 practice trials providing feedback (correct, wrong, or missed) were provided to ensure participants were performing above chance.

Participants viewed the 72 face stimuli once in random order, rating each one of the Self-Assessment Manikin Scale (SAM; Bradley & Lang, 1994) for affective valence and arousal. This self-reporting tool instructs participants to select one of five figures sitting on each of the two 9-point scales to match their experience while viewing a particular stimulus.

Lower scores on the arousal scale indicate a calm vs anexciting or agitating feeling indicated by higher scores. On the valence scale, lower scores more negative ratings while higher scores indicate more positive ratings. Mean ratings were calculated after grouping each into the three emotions (angry, neutral, happy).

EEG was recorded from 64 active electrodes using the 10–20 Biosemi Active-Two amplifier system. Common mode sense (CMS) and driven right leg (DRL) ground electrodes were placed over medial-parietal cortex. A vertical EOG directly below the right pupil and horizontal EOGs on the left and right outer canthus were attached to record artefactual eye movements, for later correction.

EEGLAB 13.6.5b (Delorme & Makeig, 2004), ERPLAB 6.1 (Lopez-Calderon & Luck, 2014), the PREP pipeline 0.55.2 (Bigdely-Shamlo, Mullen, Kothe, Su, & Robbins, 2015) and MATLAB 2015b (The MathWorks Inc., 2015) were used for offline analysis of data. EEG was predominantly recorded at 256 Hz. EEG was inadvertently recorded at 512 Hz for 4 participants, and 2048 Hz for 2 participants; the data for these participants was downsampled to 256 Hz prior to any pre-processing. The PREP pipeline was used with default parameters to remove line noise and reference all EEG channels to a robust average reference. Bad channels were detected and were replaced with spherical spline interpolated values.

To select electrodes for statistical analysis, signal-to-noise ratio (SNR) was calculated for each electrode as the ratio of the stimulus driving frequencies to the average amplitude of the 20 surrounding frequency bins, excluding those bins immediately adjacent to the driving frequency as well as other frequencies at integer multiples of five to ensure that harmonic frequencies of the driving frequencies were not included in the noise estimate. This SNR calculation follows previous research [e.g., (Liu-Shuang, Norcia, & Rossion, 2014)] with the exception of the additional exclusion of harmonic frequencies. ssVEPs were calculated for each participant as the amplitude at 15 Hz and 20 Hz among the three electrodes with the highest SNR. To obtain 15 Hz and 20 Hz amplitudes for each electrode and trial, the filter-Hilbert method was used with a 12^th order Butterworth filter. Obtained amplitudes were then baseline corrected by subtracting the amplitude present at 400 to 200 ms prior to trial onset. As in previous studies (Wieser & Keil, 2011), ssVEP amplitudes for the face and the Gabor patch were then extracted for three time windows (100 to 700 ms; 800 to 1400 ms; 1700 to 2300 ms). For difference score analyses, at each time window a difference score was calculated (ssVEP_Face–ssVEP_{Gabor patch}) such that positive values indicate higher face ssVEP amplitudes than Gabor patch ssVEP amplitudes, and negative values indicate higher Gabor patch ssVEP amplitudes than face ssVEP amplitudes. For the item analysis, this difference score was extracted for each trial, and then averaged across participants to obtain an average measure of attentional allocation for each trial. For the participant-level analysis, the trial-level data was first averaged within each block, and then across blocks, so that for each participant there was an average ssVEP amplitude for the face, the Gabor patch, and a difference between these two amplitudes, at each specified time window.

Participants were genotyped for the ADRA2B deletion variant. Extraction and genotyping of the DNA was performed at the Neurogenetics Laboratory at the Centre for Addiction and Mental Health in Toronto, Canada for the first 125 participants and at the BC Children’s Hospital Research Institute in Vancouver for the last 58 participants included in the final sample. For each variant, Hardy-Weinberg equilibrium was calculated using an online Hardy-Weinberg Chi Square calculator (Rodriguez, Gaunt, & Day, 2009).

The ADRA2B 9bp insertion/deletion variant was assayed using PCR followed by Sanger sequencing. A total of 50ng genomic DNA was combined with 1 × AmpliTaq Gold 360 buffer, 2.0 mM Magnesium Chloride, 360GC Enhancer 4 ul, 200 uM dNTPs, 0.5 uM Forward primer ACGAAGGTGAAGCGCTTCT and 0.5 uM Reverse primer GGCCAGAAGGAGGGTGTTT, AmpliTaq Gold 360 DNA Polymerse 0.625 U/reaction and total volume 25 ul in a 96 well plate. Initial denaturation was at 95°C for 8 min, followed by 38 cycles at 95°C for 50 sec, 60°C for 30 sec and 72°C for 50 sec and a final extension step of 7 min at 72°C. The PCR products were cleaned up by ExoSAP-IT Express and analyzed by Sanger sequencing (ABI 3130, Applied Biosystems).

ADRA2B genotype frequencies fell within Hardy-Weinberg equilibrium (did not differ from expected frequencies),X² = 0.03, p > .05. Of the 182 individuals, 95 participants did not carry the ADRA2B deletion variant, 72 carried one copy of the allele, and 15 were homozygous for the deletion variant. We were unable to obtain reliable genotyping forADRA2B for two participants. For all analyses, based on previous research (de Quervain et al., 2007; Rasch et al., 2009a;Todd et al., 2015; Todd et al., 2013; Todd et al., 2014), homozygote and heterozygoteADRA2B deletion carriers were treated as a single group due to the low number of homozygotes.

Two-way (ADRA2B group) × 3 (emotion) mixed ANOVAs were performed on arousal and valence ratings for face stimuli in the face ratings tasks and for accuracy in the change detection task. Analyses were first conducted independently in the replication and discovery samples. After results were examined in the two separate samples, data from all participants were combined for increased power and analyses were repeated in the full sample. For all ANOVAs reported contrasts were Bonferroni corrected for multiple comparisons and Greenhouse-Geisser corrections were used when sphericity was violated. We report results of analyses of the full sample below. Results of all behavioral analyses for Replication and Discovery samples can be found in Table 2.

There was a main effect of emotion, F(1.74, 311.81) = 86.57,p < .001, $ηp2=.33$⁠, with higher levels of arousal for happy than neutral faces and higher levels of arousal for angry than happy and neutral faces (ps < .001). There was no main effect ofADRA2B, F(1,179) = 0.55, p = .457, $ηp2=0.00$ and no emotion ×ADRA2B interaction, F(1.742, 311.81) = 0.15, p = .829, $ηp2=0.00$⁠. Thus, participants found angry faces to be more arousing than happy faces which in turn were more arousing than neutral faces (Figure 2). There was no evidence that carrying the ADRA2B deletion variant influenced subjectively rated arousal of emotional faces.

There was a main effect of emotion, F(1.33, 236.15) = 899.42,p < .001, $ηp2=.83$⁠, indicating that, as expected, angry faces were rated as negatively valenced and happy faces were rated as positively valenced (ps < .001) There was no main effect ofADRA2B, F(1,178) = 0.66, p = .416, $ηp2=0.00$ and no emotion ×ADRA2B interaction, F(1.33, 236.15) = 0.63, p = .471, $ηp2=0.00$⁠. These findings confirm that participants subjectively rate angry, neutral and happy faces as appropriately valenced (Figure 2). There was no evidence that carrying the ADRA2B deletion variant influenced subjective experience of stimulus valence.

There was no main effect of emotion, F(2,352) = 2.35,p = .097, $ηp2=.01$⁠, and no main effect of ADRA2B, F(1,176) = .24, p = .622, $ηp2=.001$ (Figure 2). There was an emotion × ADRA2B interaction,F(2,352) = 3.17, p = .043, $ηp2=.02$⁠, with non-deletion carriers demonstrating greater accuracy in trials with angry face distractors than trials with happy face distractors (paired contrast ps < .05). However, this interaction was non-significant in both the Discovery and Replication samples (see Table 2). Here effect sizes were consistent across samples, indicating the power of the full sample may have been required for a small effect showing facilitation of performance in the presence of angry faces for those who did not carry the deletion variant. This effect may be so small as to be trivial, and would require further replication to merit discussion.

In order to ascertain the potential presence of competition effects between targets and distractors, we first conducted an omnibus mixed ANOVA on ssVEP power, with ADRA2B (deletion carrier vs. non-carrier) as a between-subjects factor and stimulus type (target or distractor), emotion (angry, neutral, happy) and time window (early, middle, late) as within subject factors. All results of this ANOVA are reported in Table 3. There was a main effect of stimulus type, F(1,180) = 14.34, p < .001, $ηp2=0.07$ such that there were higher amplitudes in response to the Gabor targets than to the face distractors, indicating biased competition favoring target stimuli. There was also a consistently observed stimulus type × emotion interaction, indicating the presence of affectively-biased competition, F(2,360) = 13.06,p < .001, $ηp2=0.07$⁠. Paired contrasts revealed that whereas amplitudes were greater for targets in the presence of all distractor types (p’s < .05), the differences were smaller in trials with happy face distractors. There was also a time window × stimulus type interaction, F(1.21,218.69) = 14.75,p = .000, $ηp2=.08$⁠. Pairwise contrasts indicated greater power differences between targets and face distractors in window 1 and 3 than in window 2 (ps < .005). Importantly, there was no main effect of ADRA2B, F(1, 180) = .88, p = .350, $ηp2=0.00$ and no hypothesized stimulus type × emotion × ADRA2B interaction,F(2, 360) = 1.49, p = .227, $ηp2=0.01$⁠.

In summary, in all samples there was a pattern of higher amplitudes for target stimuli, with competition effects indicating that participants maintained greater attention to task-relevant stimuli than emotionally salient distractors. We also reliably observed an interaction between stimulus type and distractor expression, indicating that the difference in power between targets and distractors was reduced in the presence of happy faces. We further observed that the differences in ssVEP power varied by time window. Notably, there was no evidence of any effect ofADRA2B genotype.

To follow up on the interaction between facial expression and stimulus type, and thus better characterize the effects of emotion on competition for neurocognitive resources between targets and distractors, we next calculated a difference score between targets and distractors for each emotion expression in each time window. Thus, larger difference scores would indicate greater competition between targets and distractors and vice versa. Moreover, while our previous studies have found robust effects ofADRA2B on emotion/cognition interactions in samples of this size or smaller, suggesting our study was not underpowered to find equivalent effects, we wished to further probe the reliability of the null effects observed for ADRA2B. We thus next conducted a Bayesian ANOVA on the full sample using JASP (JASP Team, 2018) with default priors. One participant who showed difference scores –6.5 standard deviations from the mean was excluded from these analyses. The pattern of results did not differ when the outlier was included.

First, we conducted a mixed ANOVA with ADRA2B (deletion carrier vs. non-carrier) as a between subject factor and emotion (angry, neutral, happy) and time window (early, middle, late) as within subject factors. Consistent with the time window by stimulus type interaction observed in the previous analysis, there was a main effect of time window,F(1.53, 274.06) = 12.55, p = .000, such that there were overall largest difference scores in the third time window than in either of the first 2 (ps < .001). There was also a main effect of emotion, F(2, 358) = 7.59,p = .001, $ηp2=.04$⁠. Contrasts revealed that difference scores were smaller in trials with happy faces, indicating greater competition for neurocognitive resources from happy faces than from neutral (p < .001) or, marginally, angry (p = .06) faces, which in turn did not differ from each other (p > 1) (Figure 3). In this analysis this finding was qualified by a time window × emotion interaction, F(3.76, 672.63) = 3.30,p = .012, $ηp2=.02$⁠, indicating that the greatest happy face bias effect was observed in time window 2 (ps < .05). Crucially, there was no main effect of ADRA2B, F(1, 179) = .08, p = .774, $ηp2=0.00$⁠, no emotion ×ADRA2B interaction, F(2, 358) = .728,p = .483, $ηp2=0.00$ (Figure 3), nor was there a time window × emotion ×ADRA2B interaction, F(3.76, 672.63) = .34, p = .850, $ηp2=0.00$⁠. As a subsequent item analysis (described below) revealed that two happy face stimuli elicited ssVEP difference scores that were greater than 2.5 SD from the mean, we next reran the analysis with the trials using those faces removed. The pattern of results did not change with the happy face outliers removed. Crucially, the main effect of emotion showing greater competition for happy faces (p = .010, $ηp2=.02$⁠) and the emotion by time window interaction (p = .006, $ηp2=.02$⁠), indicating this effect was most pronounced in the middle time window, were still observed. Thus, the happy face distraction effect was not driven by outliers.

Analysis of the effects of a Bayesian Mixed ANOVA across matched models revealed a main effect of time window (BF_Inclusion = 15302.41), a main effect of face emotion (BF_Inclusion = 1064.12) and, in this analysis, no interaction between these two factors (BF_Inclusion = 0.003). There is some evidence against the main effect of ADRA2B (BF_Inclusion = 0.31) and its interaction with face emotion (BF_Inclusion = 0.27) and its interaction with time window (BF_Inclusion = 0.02). There is considerable evidence against a three-way interaction among these factors (BF_Inclusion = 0.002). These results demonstrate that there is evidence against ADRA2B modulating the degree to which emotional stimuli influence biased competition.

In summary, effects that were consistent across all frequentist and Bayesian analyses of ssVEP amplitude indicated that, within each trial, competition for neurocognitive resources favouring target stimuli over face distractors increased over time. Competition was also modulated by the facial expression of the distractor, such that smiling faces captured more neurocognitive resources from targets than angry or neutral faces. Effects of time window on emotionally-biased competition were not reliable. Importantly, there was no evidence for – and some positive evidence against — our hypothesis that putative greater NE availability resulting from carrying the ADRA2B deletion variant influenced emotionally-biased competition between targets and distractors.

While reliable, the observed effects of emotional valence on difference scores were small. In order to examine whether the higher levels of competition for neurocognitive resources was modulated by social anxiety, as has been previously observed (Wieser et al., 2012), we next probed effects of emotion on the difference scores in the 25% of our sample that was highest (n = 45), and the 25% that was lowest (n = 44) in social anxiety based on LSAS scores. Mean scores for social anxiety (LSAS) were (M = 9.68, SD = 7.39) in the low anxiety group and (M = 58.87,SD = 14.20) in the high anxiety group.

A 2 × 3 × 3 (LSAS group × Emotion × Time window) mixed ANOVA was performed on ssVEP difference scores described above. As previously reported, there was no main effect of emotion,F(1.29, 112.56) = 3.26, p = .063, $ηp2=.04$⁠. There was a main effect of time window, F(1.60, 139.57) = 5.84, p = .007, $ηp2=.06$ but no time window × emotion interaction F(4, 348) = 2.13, p = .076, $ηp2=.02$⁠. Finally, there was no main effect of LSAS, F(1, 87) = 2.44, p = .122, $ηp2=.03$ and no emotion × LSAS interaction,F(1.29, 112.56) = 1.06, p = .322, $ηp2=.01$⁠.

To confirm the null effect of social anxiety, analysis of the effects of a Bayesian Mixed ANOVA across matched models again revealed a main effect of time window (BF_Inclusion = 3.45), a main effect of face emotion (BF_Inclusion = 71.60). There was inconclusive evidence for both a main effect of social anxiety and an interaction between social anxiety and face emotion (BF_Inclusion = 0.86;BF_Inclusion = 0.418, respectively). All remaining interactions revealed null effects (allBF_Inclusion < 0.07). Thus there was no evidence suggesting that social anxiety influenced affectively-biased competition in this sample of participants.

An outstanding question concerned whether the effects of emotional valence on biased competition were due to the emotional content of the stimuli or capture of attention by low level features, such as contrast, that might differentiate happy faces from neutral and angry faces. To probe this question, we extracted a variety of image metrics. First, we used the Image Processing Toolbox packaged with MATLAB 8.6.0 (The MathWorks, Natick, MA) to calculate whole image contrast, average local image contrast within distinct square regions of different sizes (8 × 8, 16 × 16, 32 × 32, 64 × 64, and 128 × 128 regions), edges, and average log luminance. Second, we used the SHINE toolbox (Willenbockel et al., 2010) to calculate low and high spatial frequency content. Third, we used the Visual Saliency Toolbox (Walther & Koch, 2006) to calculate visual salience and the Euclidean distance between the first and second predicted fixations. Each of these measures were correlated with ssVEP difference score values. Contrast was calculated by finding the standard deviation of the grayscale pixel values. Edges was calculated with MATLAB’s edge function, using a Canny filter with a threshold of 0.5. Average log luminance was calculated as the exponential of the mean value of the logarithm of the grayscale pixel values. Spatial frequency content was calculated using the sfplot function from the SHINE toolbox, which extracts the energy at each spatial frequency in the image. From the output of this function, low frequency content was calculated as the mean energy of frequencies below the median frequency, and high frequency content was calculated as the mean energy for the remaining frequencies. A single image was removed from the analysis because it was identified visually as an outlier and had a face-Gabor ssVEP difference greater than 2.5 standard deviations from the mean. Results showed that only edges may have marginally influenced the face-Gabor ssVEP differences (r = 0.18, p = .131, 95% CI [–0.05, 0.40]), while all other image properties very clearly did not correlate with this measure (ps > .270) (Supplementary Figure 1). For each trial we also extracted subjective valence and arousal ratings averaged across participants.

Next, we conducted a one-way ANOVA to test whether the number of edges differed between facial emotion categories. If the ANOVA revealed that there was no difference among the face emotions on the amount of edges in the images, then our primary result demonstrating modulation of amplitude differences by face emotion should not be due to differences in the amount of edges in the image. However, the ANOVA revealed that facial emotion category had a significant impact on the number edges in an image,F(2, 68) = 4.80, p = .011, $ηp2=0.12$⁠). To determine whether this difference in number of edges across emotion categories could be responsible for the observed Face-Gabor patch ssVEP differences, above and beyond the influence of emotional arousal and valence, we performed a regression with edges, as well as ratings of arousal, valence, and the interaction of arousal and valence for each face stimulus as predictors of Face-Gabor patch ssVEP difference. The full regression model showed no significant effect of edges, β = 0.13, t(66) = 1.12, p = .265, and did not differ from a more parsimonious model with only arousal, valence, and their interaction as predictors, F(1,66) = 1.26, p = .265, so we will describe the results of the more parsimonious model. Face-Gabor ssVEP difference was significantly predicted by valence, β = 0.47, t(67) = 3.03, p = .003, but not arousal, β = 0.04, t(67) = 0.27, p = .791. There was a trend level interaction, β = –0.45,t(67) = –1.87, p = .065, which we illustrate in Figure 4 via simple slopes of the regression of Face-Gabor ssVEP difference on arousal at three different levels of valence (Mean valence, 1 SD below mean valence, and 1 SD above mean valence.) Follow-up tests of these three simple slopes using bootstrapping with 9999 bootstrap replicates and calculating the BCa confidence interval (DiCiccio & Efron, 1996) reveal that arousal predicts higher Face-Gabor ssVEP difference only among low valence face stimuli (i.e. negative faces), 95% CI [0.12, 0.79]. Face-Gabor patch difference is not significantly predicted by arousal at mean valence (95% CI [–0.25, 0.31]), or high valence, (95% CI [–1.05, 0.15]). In sum, though edges differed across stimulus emotion categories, it did not significantly predict Face-Gabor patch ssVEP differences. Instead, we found that once again valence predicted difference scores, confirming the happy face advantage. In addition, higher stimulus arousal results in higher Face-Gabor ssVEP differences (i.e. a relatively higher ssVEP in response to face stimuli), but only among more negatively valenced stimuli. Thus neither any of the low level features we measured nor subjective ratings of arousal accounted for the happy face advantage.

In the present study, we used ssVEPs to examine effects of a deletion variant inADRA2B, a common genetic variation thought to influence norepinephrine availability, on affectively-biased competition in the visual cortex. We further examined within-subject effects of emotional valence in a large sample of healthy young adults. Counter to our hypotheses, carrying a deletion variant of theADRA2B gene had no influence on competition for population-level neuronal responses between task-relevant stimuli and affectively salient distractors. Thus, the evidence suggests that naturally occurring variation in NE levels putatively associated with this genetic variation neither drive nor modulate such affectively-based visual cortex competition when attention must be sustained over seconds. In the group as a whole, the presence of happy faces captured cortical resources from target stimuli, particularly at middle latencies, indicating biased competition for smiling faces at the population level of neuronal competition. Although photographs of happy faces did differ from those of angry and neutral faces in the number of edges they contained, follow-up analyses showed that the effect of emotional expression on biased competition between happy faces and other facial expressions was not due to differences in any of the low-level features we measured or in subjective ratings of arousal.

Our own previous research has found robust effects for the ADRA2B deletion variant influencing affective biases in attention and memory. Building on findings that carrying the deletion variant influences emotional enhancement of memory (de Quervain et al., 2007), we have found that carrying this variation enhances rapid attentional selection for emotionally relevant words, enhances the link between subjective emotional arousal in encoding and memory accuracy and confidence, and enhances the subjective perception of the relative vividness (signal to noise ratio) of emotionally salient images (Todd et al., 2015; Todd et al., 2013; Todd et al., 2014). It has also been associated with differences in responsiveness of the amygdala and VMPFC to enhanced processing of salient stimuli (Rasch et al., 2009b; Todd et al., 2015; Todd et al., 2013; Todd et al., 2014), which can drive affectively enhanced processing in the visual cortex.

The present study was motivated by the hypothesis that putatively higher levels of NE availability in deletion carriers would facilitate biased visual cortex competition favoring affectively salient stimuli, providing some mechanistic understanding of our previously-observed results. This was not the case. It is worth noting that all of our previous studies on the influence of ADRA2B on attention and subjective perception tapped relatively rapid processes, whereas the effects of biased competition measured here emerged more slowly. Moreover, previous studies measuring effects of the deletion variant on emotional attention and perception engaged subjective judgments of emotional content (Ehlers, Ross, & Todd, 2018), or the relative level of noise overlaid on a visual stimulus (Todd et al., 2015), or rapid engagement of selective attentional prioritization of emotional content (Todd et al., 2013). In contrast, the present study required sustained attention to a neutral stimulus over the course of seconds, and therefore continuous inhibition of distraction by the emotional expression. The LC/NE system is thought to influence emotional effects on perception, attention and memory via amygdala modulation of other brain regions [for review see (Ehlers & Todd, 2017; Markovic et al., 2014; Roozendaal, Luyten, de Voogd, & Hermans, 2016)]. Given electrophysiological evidence that amygdala neurons are phasically activated – showing transient increase in firing at the onset of salient or novel stimuli stimuli (E. J. Kim et al., 2018) — it may be that the sustained pulse of emotional distractors did not elicit amygdala activation that would differentiate deletion carriers and non-carriers. If true, one boundary condition of effects ofADRA2B on subjective perception and attention would it does not influence sustained or ongoing processes. This hypothesis should be tested directly in future research.

It is also notable that our previous research found that, in contrast to findings of emotionally enhanced activation in the ventromedial prefrontal cortex and the amygdala (Rasch et al., 2009b; Todd et al., 2015), deletion carriers did not differ from non-carriers in patterns of enhanced visual cortex activation for emotionally salient stimuli. These findings are convergently consistent with the conclusion that effects of NE on emotionally-biased processing do not function via visual cortex activity.

The results of a Bayesian ANOVA further suggested our null finding was conclusive, despite the relatively small sample size for a targeted polymorphism study. Our power analysis indicated that the study was amply powered based on the weakest effect size of our previous studies examining ADRA2B. Thus, we conclude that commonly observed differences related to NE activity do not play a role in modulating biased competition in the visual cortex, at least not for sustained firing patterns measured here. However, it is worth noting that previous studies investigating effects of ADRA2B have employed more arousing affective stimuli than emotionally expressive faces, using images and words that evoke more physiological arousal [e.g., (A. K. Anderson, Yamaguchi, Grabski, & Lacka, 2006). Thus, it is possible that a higher level of stimulus-evoked arousal may be required for differences in NE functioning to influence biased competition effects.

We did, however, find reliable evidence that the smiling faces we used captured more visual cortex resources, indexing greater attention, from task-relevant stimuli than angry or neutral faces. These findings are inconsistent with a prevalent “standard hypothesis” in affective science, which has proposed that threatening stimuli such as angry faces are universally prioritized because of an evolutionary advantage for rapid threat detection (Ohman, Flykt, & Esteves, 2001; Ohman, Lundqvist, & Esteves, 2001). However, findings of attentional prioritization of threatening relative to rewarding facial expressions have been repeatedly challenged on several fronts. First, there is an extensive literature documenting systematic individual differences in the direction of biases [for review see (Todd, Cunningham, Anderson, & Thompson, 2012)]. Valence preferences can also vary with developmental stage, such that young children show far more sensitivity and prioritization of happy faces than young adults (Picardo, Baron, Anderson, & Todd, 2016; Todd, Evans, Morris, Lewis, & Taylor, 2011). In the present study, this pattern did not reflect the individual differences in ADRA2B genotype and social anxiety we predicted and measured in this sample. This may be because previous studies finding social anxiety effects used participants with less extreme levels of social anxiety and less variability in the high social anxiety participants (Wieser et al., 2012). However, it is likely that given the degree of variability observed (Figure 2), and the modest effect size, this pattern is subject to individual differences linked to other traits.

Second, ERP studies of emotional effects on visual processing have reported different early patterns of response and time courses for positive vs. negative stimuli, potentially reflecting differences in contributions of reward circuitry to valenced responses. There is evidence that the C1 component, which is generated by striate cortex as rapidly as within 50 ms of stimulus onset, responds preferentially to negatively valenced over neutral stimuli in a study where positive valence was not tested (West, Anderson, & Pratt, 2009), and to monetary loss relative to monetary gain (Rossi et al., 2017). In contrast, later time windows have shown enhanced activation for both monetary gain and loss over neutral stimuli. Other studies have also found reward-related responses, putatively mediated by estimated anterior cingulate activity, between 200 and 300 ms following stimulus onset (Hickey, Chelazzi, & Theeuwes, 2010). Thus, whereas very rapid visual cortex activity appears to favor negatively valenced stimuli, activity that may reflect more sustained re-entrant processing may favor salient stimuli in general. Our findings showing that the greater competition for happy faces was greatest in the latest time window, allowing for potential feedback from reward circuitry, are not inconsistent with these findings.

Third, studies looking at overall within-subject effects of valenced stimuli on attention have yielded conflicting results. After controlling for low-level features, some researchers have observed consistent happiness superiority effects as we did (Becker, Anderson, Mortensen, Neufeld, & Neel, 2011). One suggestion has been that, after controlling for low-level perceptual features, it is arousal rather than valence that drives attentional prioritization (Juth, Lundqvist, Karlsson, & Ohman, 2005). However, our present findings are not consistent with this interpretation. In the present study, participants rated angry faces as more arousing than happy faces, and arousal accounted for ssVEP difference scores in angry faces only — there was no relationship between arousal and biased competition favoring happy faces. Previous findings of happy face prioritization have been interpreted as reflecting easier visual discrimination of the unambiguous communication that is afforded by smiling faces (Becker et al., 2011). This interpretation is the most consistent with our results, where rewarding content that was unrelated to arousal captured attention.

In summary, our findings are consistent with robust evidence that in some contexts happy facial expressions, which are experienced as rewarding, are prioritized over angry ones. These findings did not reflect low-level features nor did they reflect patterns of subjectively rated arousal. Nonetheless, such valence effects are subject to numerous boundary conditions that constrain our interpretation. We measured a wide range of low level features, including contrast at a multiple spatial scales, edges, luminance, and a measure of visual saliency that predicts eye movement behaviour, and predicted eye-movement behaviour, and none of these predicted ssVEP difference scores; however, in schematic faces the happy face advantage has been linked to more subtle featural confounds (Purcell & Stewart, 2010), such as the relative orientation of face and mouth contours (Stein & Sterzer, 2012). In the case of photographic faces, prioritization of angry vs happy faces can vary depending on the stimulus set used (Savage, Becker, & Lipp, 2016; Savage, Lipp, Craig, Becker, & Horstmann, 2013). Future research will be required to establish whether the increased biased competition from happy faces we observed generalizes to other affective stimulus sets.

Finally, although examination of social anxiety was not the focus of this study, we failed to replicate previous findings that social anxiety modulated the biased competition effect in a larger sample (Wieser et al., 2012). Unlike the previous study, we did not screen for high and low anxiety, and our high-anxiety group had lower levels of mean anxiety than that of the previous study. Thus, it may be effects would be replicated in samples where anxiety levels are more extreme.

In sum, our findings provide compelling evidence that affectively biased competition in sustained attention is not modulated by individual differences in noradrenergic activity related to alpha2b receptor function. Moreover, although we observed consistent affectively-biased competition for smiling faces, the present findings add to growing evidence indicating multiple boundary conditions for previously observed valence effects. Studies using larger sample sizes for within-subject designs, investigation of individual differences, and increased use of tools such as Bayesian analyses to probe the conclusiveness of null results – along with increased willingness to publish null results – will play an important role refining our understanding of cognition-emotion interactions.

All data and analysis scripts are available on Open Science Framework: https://osf.io/n3g6e/?view_only=458920fd64c048028c651f78f7192f55.

This research was supported by Canadian Institutes for Health Research (CIHR) Operating Grant #491746, National Engineering and Research Council (NSERC) Discovery Grant RGPIN-2014-04202, and Canadian Foundation for Innovation Leaders Opportunity Fund grant F13-03917. Individual researchers were supported by a NSERC doctoral fellowship to K. H. Roberts and CIHR New Investigator Salary awards and Michael Smith Foundation for Health Research Scholar awards to both C.J.D. Ross and R.M. Todd. We would like to thank Drs. Todd Handy and Julia Kam as well as Mana Ehlers and Lena Fürst for their generous contributions to this project.

The authors have no competing interests to declare.

• Contributed to conception and design: RMT & MJW

• Contributed to acquisition of data: MGMM & KHR

• Contributed to analysis and interpretation of data: KHR, MGMM, CJDR, DJM, MJW, & RMT

• Drafted and/or revised the article: KHR, MGMM, & RMT

• Approved the submitted version for publication: KHR, MGMM, CJDR, DJM, MJW, & RMT

The author(s) of this paper chose the Open Review option, and the peer review comments are available at: http://doi.org/10.1525/collabra.202.pr