Evidence for a minimal role of stimulus awareness in reversal of threat learning

Abstract

In an ever-changing environment, survival depends on learning which stimuli represent threat, and also on updating such associations when circumstances shift. It has been claimed that humans can acquire physiological responses to threat-associated stimuli even when they are unaware of them, but the role of awareness in updating threat contingencies remains unknown. This complex process-generating novel responses while suppressing learned ones-relies on distinct neural mechanisms from initial learning, and has only been shown with awareness. Can it occur unconsciously? Here, we present evidence that threat reversal may not require awareness. Participants underwent classical threat conditioning to visual stimuli that were suppressed from awareness. One of two images was paired with an electric shock; halfway through the experiment, contingencies were reversed and the shock was paired with the other image. Despite variations in suppression across participants, we found that physiological responses reflected changes in stimulus-threat pairings independently of stimulus awareness. These findings suggest that unconscious affective processing may be sufficiently flexible to adapt to changing circumstances.

Figure 1.

Schematic description of experimental design and procedure. In each trial of the acquisition phase, participants were presented with one of two stimuli (schematic pictures of spiders, presented monocularly for 6 sec and suppressed from awareness by a CFS mask shown to the other eye). One image (spider A) always terminated with a mild electric shock to the wrist, whereas the other (spider B) never did. Halfway through the experiment, with no warning, the contingencies flipped and the reversal phase began: The formerly safe stimulus (spider B) now predicted the shock, and the old threat-associated one (spider A) was now safe. Each spider was shown eight times in each phase. Trial order was pseudorandomized (see the Materials and Methods) and spider identity (A or B) was counterbalanced across participants. To assess the success of the awareness manipulation, participants answered the questions “Which seen?” (1 = flower, 2 = spider) and “How confident?” (1 = guess to 3 = sure), presented binocularly (1.5–2 sec each), beginning 1 sec after the offset of every CS, and followed by an 8- to 10-sec intertrial interval (the questions are only shown here for the first depicted trial, but were repeated in all trials). Participants who underwent the same procedure without CFS were shown identical CSs, but the CFS mask was absent.

Figure 2.

Physiological reversal learning. (A) Time courses reveal reversal of threat responses with and without continuous flash suppression. Data points represent trial-wise mean responses to spider A (the CS+ during acquisition) and spider B (the CS− during acquisition). Both groups showed reversal learning, as indicated by greater responses to spider A during the acquisition phase and greater responses to Spider B during the reversal phase. Error bars represent standard errors. (B) Mean reversal learning index for each group. Error bars represent 95% confidence intervals, indicating that the interaction of stage and stimulus and thus the magnitude of reversal learning in both groups was significantly greater than zero. (C) Heightened anxiety is associated with impaired reversal learning under CFS. A negative correlation between baseline anxiety measures and the strength of threat reversal learning is evident for state and trait anxiety. Blue lines show linear fits of each score to the reversal index, and ribbons around lines indicate bootstrapped 95% confidence intervals around the estimate. Note that the participant with the highest reversal index provided data for the STAIS and STAIT, but not FSQ. (STAIS/STAIT) state/trait anxiety subscale of the Spielberger State-Trait Anxiety Inventory, (FSQ) Fear of Spider Questionnaire, (∼) P < 0.1, (*) P < 0.05.

Figure 3.

Characterizing the relation between perceptual awareness and reversal learning in the CFS group. (A) Correlation between the awareness index of even and odd-numbered trials. Each data point represents an individual participant. The strong positive correlation between these independent measures of awareness demonstrates that individual participants’ awareness ratings—even those with extreme values of zero or one—are unlikely to be due to measurement noise. For illustrative purposes, the color scheme marks all participants with an awareness index of 0 in even trials in red ([UA] unaware, N = 27) and classifies the rest of the CFS group in three tertiles (T1–T3). Note that some data points overlap. (B) Reversal learning plotted against perceptual awareness for individual participants, for data obtained from even-numbered trials. The color scheme is the same as in A. (C) Reversal learning plotted against perceptual awareness for individual participants, for data obtained from odd-numbered trials. Individual participants are marked with the same color as in the previous panels; the overall distribution of participants is highly similar across panels. (D) Reversal learning as a function of perceptual awareness in the CFS group, using data pooled from all trials. The intercept, indicating the magnitude of reversal learning in the absence of awareness, is positive and significantly different from zero. (E) Reversal Index intercepts and their 95% confidence intervals in a series of regression models. Model 1 depicts the intercept (the value of the reversal index when the awareness index equals zero) shown in D. Model 2 shows the intercept when the regression model includes STAIT scores in addition to the perceptual awareness index. Model 3 regresses the reversal index onto the perceptual awareness index, STAIT and tracking scores. (Excluding the potential outlier in the top left corner of D weakens significance of the intercept in model 1, P = 0.07; the intercepts of models 2 and 3 remain significant after removal of this outlier.) Blue lines show linear fits, and ribbons around lines indicate bootstrapped 95% confidence intervals around the estimate.