The outbreak of COVID-19 altered the relationship between current negative memory biases and future depressive degree in nonclinical depression.

The outbreak of the novel coronavirus disease 2019 (COVID-19) has increased concern about people's mental health under such serious stressful situation, especially depressive symptoms. Cognitive biases have been related to depression degree in previous studies. Here, we used behavioral and brain imaging analysis, to determine if and how the COVID-19 pandemic affects the relationship between current cognitive biases and future depression degree and the underlying neural basis in a nonclinical depressed population. An out-expectation result showed that a more negative memory bias was associated with a greater decrease in future depressive indices in nonclinical depressed participants during the COVID-19 pandemic, which might be due to decreased social stress. These data enhance our understanding of how the depressive degree of nonclinical depressed populations will change during the COVID-19 pandemic and also provide support for social distancing policies from a psychological perspective.

Introduction

The outbreak of the novel coronavirus disease 2019 (COVID-19) is a major emergency that people around the world are facing. Until now, the virus has rapidly spread in most countries, with more than 60 million cases reported (https://covid19.who.int/). This serious stressful situation has increased concern about people's mental health, especially depressive symptoms (Adhanom Ghebreyesus, 2020; Shi et al., 2020). In addition to the possible direct impact of the pandemic on mental health, recent studies have argued that governments' policies for the prevention of viral spread (for example, social distancing policies) may also have an impact on psychological distress and cause mental disorders (Galea et al., 2020; Orben et al., 2020; Venkatesh and Edirappuli, 2020). Nonclinical depressed individuals, who are associated with depressive symptoms but do not meet the criteria for clinical depression, are widely distributed around the world and are at high risk of depression (Ayuso-Mateos et al., 2010; Ronald C. Kessler et al., 1997). Therefore, it is important to investigate how the depressive degree changes during the COVID-19 pandemic in the nonclinical depressed population.

The cognitive model of depression proposed by Beck clarified the cognitive factors that are associated with depressive degree (Beck, 2008). According to this model, biased attention, interpretation, and memory play primary roles in the development and maintenance of depression (Disner et al., 2011). Current negative cognitive biases have been related to future depressive symptoms (Gotlib and Joormann, 2010; Kleim et al., 2014; Romero et al., 2014; Rude et al., 2003; Smith et al., 2018). Therefore, to understand how the depressive degree will change during the COVID-19 pandemic in the nonclinical depressed population, it is important to investigate how the outbreak of COVID-19 may affect the relationship between current cognitive biases and future depression.

Two possible effects of the COVID-19 pandemic on the relationship between current cognitive biases and future depression in the nonclinical depressed population can be put forward. On the one hand, the COVID-19 pandemic is a threat to life. Fear and anxiety about a new disease and what could happen can be a powerful source of stress and increase the depressive degree. Recent studies have also found significant increased symptoms and a high prevalence of depression during the COVID-19 pandemic (Bauerle et al., 2020; Chen et al., 2020; Choi et al., 2020; Corbett et al., 2020). Therefore, in nonclinical depressed individuals, those with more negative biased cognition might become more depressed when facing elevated stress during the COVID-19 pandemic. Thus, the outbreak of COVID-19 may promote a positive correlation between negative cognitive biases and future depressive degree. On the other hand, some measures (for example, social distancing policies) have been carried out by governments to prevent the spread of the virus. These measures have caused significant changes in social habits. Although recent studies have argued that these prevention measures may cause psychological distress, we propose that these measures may result in decreased social stress in the nonclinical depressed population, which has been reported to be a crucial risk factor for depression (Pinquart and Sorensen, 2003; Schuster et al., 1990; Slavich et al., 2009). In addition, negative cognitive biases play a key role in the causing and maintaining of social stress (Clark and McManus, 2002; Heinrichs and Hofmann, 2001; Hirsch and Clark, 2004; Ledley and Heimberg, 2006; Musa and Lepine, 2000). Therefore, in nonclinical depressed individuals, those with more negative biased cognition may experience more social stress before the pandemic and thus show more relief of depression when facing reduced stress during the COVID-19 pandemic. Thus, the outbreak of COVID-19 may change the relationship between current cognitive biases and future depressive degree and promote a negative correlation between them.

According to a previous meta-analysis of functional neuroimaging of depression (Miller et al., 2015), ten brain areas were highlighted in youth with depression. These areas were the subgenual anterior cingulate cortex (ACC), ventrolateral prefrontal cortex (vlPFC), caudate, thalamus, parahippocampal gyrus, cuneus, dorsal cingulate cortex (DCC), insula, dorsolateral prefrontal cortex and superior temporal cortex (Miller et al., 2015). The roles of these areas in cognitive biases have also been discussed in previous studies (Disner et al., 2011; Sakaki et al., 2020; Wiers and Wiers, 2017). For example, the DCC functions in the inhibitory processing and cognitive control of emotion and its greater activation when inhibiting attention to negative stimuli in individuals with depression has been considered to be associated with the cognitive biases of depression (Disner et al., 2011). The cognitive biases associated with depression have been proposed to be particularly prominent for emotional faces (Stuhrmann et al., 2011). Greater brain response to sad faces in the depressed population was interpreted as neurobiological evidence of negative cognitive biases (Stuhrmann et al., 2011; Suslow et al., 2010). Attenuated response to happy faces has also been reported as an evidence of impaired processing of happy expressions (Fu et al., 2007).

Based on what has been discussed above, we hypothesize that the outbreak of COVID-19 may alter the relationship between current cognitive biases and future depression degree in the nonclinical depressed population. In the present study, nonclinical depressed participants' current depressive indices, cognitive biases, brain response to emotional faces, and future depressive indices were assessed. Some of the nonclinical depressed participants' future depressive indices were assessed before the outbreak of COVID-19, and some were assessed during the COVID-19 pandemic. We also included a group of nondepressed participants to ensure the results were specific to the effect of the COVID-19 pandemic. All of the nondepressed participants' future depressive indices were assessed before the COVID-19 pandemic. We compared the correlations between negative cognitive biases and future depressive indices between samples whose future depressive indices were assessed before or during the COVID-19 pandemic. Regions of interest (ROI) analysis was also carried out to test whether the relationship between brain activity in the cognitive bias-related regions and future depressive indices also showed the same pattern under the effect of the COVID-19 pandemic as that between cognitive biases and future depressive indices.

Results

A total of 98 adults participated in our study. Three participants were excluded because two quit the study after the behavior tests and the other slept during the experiment. The participants attended a present depressive indices assessment session, a cognitive biases assessment session (see Figure S1), an fMRI scan session, a future depressive indices assessment session, and a socializing information collection session (Figure 1). The participants' present (measured no more than 1 week before the day of the experiment) and future (measured 3 month later) depressive indices were assessed using the Beck Depression Inventory-II (BDI) (Beck et al., 1996). The participants were classified as the healthy group (HG, n = 34, BDI cutoff range: 0–13) or the nonclinical depressed group (NDG, n = 61, BDI cutoff range: 14–63) based on their present BDI score (Everaert et al., 2014). Members of the NDG whose future depressive indices were assessed before the pandemic were further classified as the before the pandemic group (BG, n = 31), and those whose future depressive indices were assessed during the pandemic were further classified as during the pandemic group (DG, n = 30). For detailed experimental design and analysis method, see Transparent methods.

Figure 1

Procedure. The procedures of the experiments are shown in a flow chart

A word identification task (WI) plus incidental free recall (IFR) task was used to assess memory biases. Interpretation biases were evaluated using the ambiguous scenarios test for depressed mood (AST-D). A visual search task (VST) was used to investigate attentional biases to sad faces. The fMRI experiment was a face viewing task composed of 2 runs, each with 6 blocks. The facial stimuli in each run consisted of grayscale normalized sad, happy, and neutral expressions of 30 men and 30 women. In each block, 5 trials of male faces and 5 trials of female faces were shown. In each trial, a face was shown in the center of the screen for 200 ms, followed by a black screen. Each trial lasted for 2 s. Then, the next face appeared. A fixation cross was presented for 20 s between two blocks and for 10 s before the first block and after the last block of each session. Prior to the experiment, the subjects were instructed to attentively watch the faces and recognize the corresponding expressions.

Demographics and BDI scores

There were no significant differences among the three groups in regard to age, gender, or years of education (p > 0.1, Table 1). Moreover, the present BDI of the HG was significantly lower than that of the BG and DG (F(2, 92) = 69.48, p < 0.001, Table 1), but there was no significant difference between the BG and DG (t(59) = 1.31, p = 0.19). In addition, the future BDI of the HG was also significantly lower than that of the BG and DG (F(2, 92) = 12.32, p < 0.001, Table 1), and there was no significant difference between the BG and DG (t(59) = 0.84, p = 0.40).

Table 1

Sample demographics and BDI scores

Sample group	HG (n = 34)	BG (n = 31)	DG (n = 30)	F/χ2	p Value
Age (mean ± SD, year)	21.59 ± 1.88	21.32 ± 1.81	20.70 ± 1.70	2.01	0.14
Gender (M/F, n)	24/10	16/15	15/15	3.52	0.17
Education (mean ± SD, year)	15.74 ± 1.71	15.26 ± 1.59	15.03 ± 1.16	1.81	0.17
Present BDI (mean ± SD)	6.74 ± 3.71	19.10 ± 6.44	21.13 ± 5.64	69.48	<0.001
Future BDI (mean ± SD)	7.29 ± 5.80	13.65 ± 8.22	15.23 ± 6.31	12.32	<0.001

For continuous variables, one-way analysis of variance was carried out. For categorical variables, χ tests were carried out.

Cognitive biases comparison

For memory biases, no significant difference was found among the three groups in regard to the percentage of negative (F(2, 88) = 2.17, p = 0.12) or positive (F(2, 88) = 1.63, p = 0.20) words recalled. However, for interpretation biases, the pleasantness ratings in the HG were significantly higher than those in the other two groups (F(2, 92) = 9.12, p < 0.001; compared with the BG, t(63) = 3.74, p < 0.001; compared with the DG, t(62) = 3.67, p < 0.001), with no significant difference found between the BG and DG (t(59) = 0.60, p = 0.55). For attention biases, the accuracy and reaction time (RT) of the sad targets among the neutral distractors condition in the visual search task (VST) were analyzed, and no significant differences were found among the three groups (accuracy: F(2, 92) = 1.09, p = 0.34; RT: F(2, 92) = 0.18, p = 0.84).

Comparison of changes in future BDI scores

There was no significant change in future BDI in the HG (t(33) = 0.73, p = 0.47). A significant decrease in the future BDI was found in both BG (t(30) = −3.03, p < 0.01) and DG (t(29) = −4.70, p < 0.0001). However, there was no significant difference in the change in future BDI between the BG and DG (t(59) = 0.20, p = 0.84).

The COVID-19 pandemic altered the ability of negative memory biases to predict future BDI scores

Regarding memory biases, there was no significant difference among the three groups in the correlation between the percentage of negative (for the HG and BG, z = 0.80, p = 0.42; for the HG and DG, z = 0.93, p = 0.35; for the BG and DG, z = 0.13, p = 0.89) or positive (for the HG and BG, z = 0.0077, p = 0.99; for the HG and DG, z = 1.22, p = 0.22; for the BG and DG, z = 1.18, p = 0.24) words recalled and present depressive indices. However, the correlation coefficients between the percentage of negative words recalled and future depressive indices in the DG were significantly different than those in the HG (z = 2.58, p < 0.01) and BG (z = 2.87, p < 0.005), with the percentage of negative words recalled significantly predicting future depressive indices in the DG (r = −0.63, p < 0.001, Figure 2A) but not in the BG (r = 0.057, p = 0.77, Figure 2A) or the HG (r = −0.052, p = 0.77, Figure 2A). The correlation coefficients were not significantly different between the HG and BG (z = 0.41, p = 0.68). Moreover, the correlation coefficients between the negative words recalled and future or present depressive indices in the DG were also significantly different (z = 2.74, p < 0.01), and no significant correlation was found between the negative words recalled and present depressive indices (r = −0.076, p = 0.70). In addition, we also found that the percentage of negative words recalled also significantly predicted changes in depressive indices in the DG (r = −0.50, p < 0.01, Figure 2B) but not in the BG (r = 0.081, p = 0.68, Figure 2B) or HG (r = −0.21, p = 0.24, Figure 2B). The results of the comparisons of correlations were consistent when using analysis of variance (ANOVA) (Table S1). No significant differences were found among the three groups in the correlation between the percentage of positive words recalled and future depressive indices (for the HG and BG, z = 0.73, p = 0.47; for the HG and DG, z = 0.14, p = 0.89; for the BG and DG, z = 0.82, p = 0.41).

Figure 2

Correlations between negative memory biases and future BDI or change in BDI

(A) Correlations between negative memory biases and future BDI in the DG, BG, and HG. Future BDI scores were plotted against the percentage of negative words recalled in the IFR task.

(B) Correlations between negative memory biases and changes in BDI in the DG, BG, and HG. Changes in the BDI scores were plotted against the percentage of negative words recalled in the IFR task.

Regarding interpretation biases, there were no significant differences among the three groups in the correlation coefficients between the pleasantness ratings and the present depressive indices (for the HG and BG, z = 0.067, p = 0.95; for the HG and DG, z = 0.37, p = 0.71; for the BG and DG, z = 0.43, p = 0.67), with the pleasantness ratings significantly or nearly significantly correlated with the present BDI in the DG (r = −0.32, p = 0.083), BG (r = -0.42, p = 0.019), and HG (r = −0.41, p = 0.017). Moreover, the pleasantness ratings also significantly predicted future depressive indices in all three groups (for the DG, r = −0.53, p < 0.005; for the BG, r = −0.37, p = 0.038; for the HG, r = −0.43, p = 0.011).

Regarding attention biases, there were no significant correlations between the present depressive indices and accuracy (for the DG, r = −0.15, p = 0.41; for the BG, r = 0.26, p = 0.16; for the HG, r = −0.23, p = 0.19) or the average RT for correct responses (for the DG, r = 0.17, p = 0.38; for the BG, r = −0.0094, p = 0.96; for the HG, r = −0.15, p = 0.41) in the sad target condition. Moreover, no significant differences were found among the three groups for the correlation coefficients between future depressive indices and accuracy (for the HG and BG, z = 0.72, p = 0.47; for the HG and DG, z = 1.67, p = 0.10; for the BG and DG, z = 0.94, p = 0.35) or the average RT for correct responses (for the HG and BG, z = 0.44, p = 0.66; for the HG and DG, z = 1.06, p = 0.29; for the BG and DG, z = 0.61, p = 0.54) in the sad target condition.

Correlations between the brain response to sad faces and negative memory biases

ROI were defined as the areas proposed in the introduction section (see Figure 3A). We calculated the correlations between the responses of those areas to sad faces and negative memory biases. The results showed that the response to sad faces was significantly correlated with negative memory biases in the bilateral DCC (left: r = 0.29, p = 0.049, FDR adjusted, Figure 3B; right: r = 0.33, p = 0.045, FDR adjusted, Figure 3B), and right vlPFC (right: r = 0.31, p = 0.049, FDR adjusted, Figure 3B). No significant correlation was found in the other areas after the correction for multiple comparisons (all p > 0.1, FDR adjusted). The areas (left insula, left STG, left thalamus, left vlPFC, right cuneus, right dlPFC, right insula) in which the response to sad faces was significant (p < 0.05) or nearly significant (p < 0.1) correlated with negative memory biases but didn't survive the correction for multiple comparisons were shown in Table S2.

Figure 3

Correlations between the brain response to sad faces and negative memory biases

(A) Locations of ROIs. The location of the spheres in the figure represent the average coordinates of the corresponding ROIs.

(B) Correlations between brain activity and negative memory biases during the emotional face viewing task. Percentage of negative words recalled in the IFR task were plotted against the average beta values (activation of sad faces minus neutral faces) of each voxel in the left and right DCC and right vlPFC. Corrected p values were shown.

The COVID-19 pandemic altered the ability of the brain response to sad faces to predict future BDI scores

We further investigated whether the COVID-19 pandemic altered the ability of the response to sad faces in the bilateral DCC, and right vlPFC to predict future depressive indices. The response to sad faces in the left DCC significantly predicted future depressive indices in the DG (r = −0.41, p = 0.026, Figure 4A) but not in the BG (r = 0.035, p = 0.86) or the HG (r = 0.10, p = 0.60). The correlation coefficients in the DG were also significantly more negative than those in the BG (z = 1.68, p = 0.047, one side) and the HG (z = 1.97, p = 0.025, one side). Although the correlations between the future BDI and response to sad faces in the right DCC were not significant in the DG, the trend was the same as that in the contralateral region (r = −0.31, p = 0.10, Figure 4A). Moreover, the response to sad faces in the DCC were also negatively correlated with the change in depressive indices in the DG (left: r = −0.35, p = 0.065, Figure 4B; right: r = −0.40, p = 0.032, Figure 4B), and no significant correlation was found between the response to sad faces in the bilateral DCC and present BDI (left: r = −0.052, p = 0.79; right: r = 0.13, p = 0.51). No significant correlation was found between the response to sad faces in the right vlPFC and future depressive indices (r = −0.16, p = 0.41) or change in future depressive indices in the DG (r = −0.021, p = 0.91). And the correlation coefficients between the response to sad faces in the right vlPFC and future depressive indices (z = 0.79, p = 0.43) or change in depressive indices (z = 0.71, p = 0.48) were not significantly different between the BG and DG. For the left insula, left STG, left thalamus, and right cuneus, although their correlation with negative memory biases did not survive the correction for multiple comparisons, the correlation between their responses to sad faces and future BDI scores were also significantly (or nearly significantly) changed during the pandemic compared with the correlation between their responses to sad faces and present BDI scores, with their responses to sad faces more negatively correlated with future BDI scores (see Table S1).

Figure 4

Correlations between the response to sad faces and future BDI or change in BDI in the DG

(A) Future BDI scores were plotted against the average beta values (activation of sad faces minus neutral faces) of each voxel in the left and right DCC in the DG. Responses to sad faces in the bilateral DCC were significantly (or nearly significantly) negatively correlated with future BDI scores.

(B) Change in future BDI scores were plotted against the average beta values (activation of sad faces minus neutral faces) of each voxel in the left and right DCC in the DG. Responses to sad faces in the bilateral DCC were significantly (or nearly significantly) negatively correlated with change in BDI scores.

Comparison of changes in future BDI scores in high or low memory biases subgroups

We further compared the changes in future BDI scores between the BG and DG in the higher and lower negative memory biases subgroups separately. We calculated the quartiles of the percentage of negative words recalled in the incidental free recall (IFR) task in the SDG and defined those with a memory bias higher than the 75% quantile or lower than the 25% quantile as the high bias group and the low bias group, respectively. Two-way ANOVA was carried out, with changes in future BDI as the response variable and pandemic (the BG or DG) and memory bias (high or low bias) as two factors. Significant interactions were found between pandemic and memory bias factors (F(1, 25) = 8.62, p < 0.01, Figure 5), with changes in future BDI in the high bias subgroup of the DG being significantly lower than those in the high bias subgroup of the BG (t(13) = 3.37, p < 0.005, Figure 5). No significant difference was found between the low bias subgroups of the DG and BG (t(12) = 0.94, p = 0.37, Figure 5).

Figure 5

Comparisons of the changes in future BDI scores between the BG and DG in the higher and lower negative memory bias subgroups

Data are represented as mean +/− standard error. ∗∗ represents p < 0.01. two-way ANOVA was carried out.

Change in socializing during the COVID-19 pandemic and its effect on the relationship between negative memory biases and future BDI scores

To explore the reason underlying the fact that a more negative memory bias was associated with a greater decrease in future depressive degree in nonclinical depressed participants during the COVID-19 pandemic, we further investigated the change in socializing during the COVID-19 pandemic and its effect on the relationship between negative memory biases and future BDI scores. We found that social distance significantly increased in all three groups (for the HG, t(32) = 2.55, p = 0.016; for the BG, t(27) = 2.59, p = 0.015; for the DG, t(29) = 3.47, p < 0.005, Figure 6A). Social frequency (for the HG, t(32) = −7.13, p < 0.001; for the BG, t(27) = −3.20, p < 0.005; for the DG, t(29) = −5.67, p < 0.001, Figure 6A) and time spent on socializing (for the HG, t(32) = −6.18, p < 0.001; for the BG, t(27) = −4.79, p < 0.001; for the DG, t(29) = −10.00, p < 0.001, Figure 6A) significantly decreased. Moreover, we also found that degree of stress from socializing significantly or nearly significantly decreased during the COVID-19 pandemic in the BG (t(27) = −1.77, p = 0.088) and DG (t(29) = −3.75, p < 0.001, Figure 6A) but not in the HG (t(32) = −1.22, p = 0.23). A significant correlation was also found between the degree of change in social frequency and future BDI scores in the DG (r = 0.94, p = 0.016, Figure S2). However, the correlation between the degree of change in the stress from socializing and future BDI scores was not significant (r = 0.04, p = 0.96). The lack of significance might be due to the narrow range and unbalanced distribution of the degree of change in the stress from socializing (see Figure S2).

Figure 6

Change in socializing during the COVID-19 pandemic and its effect on the relationship between negative memory biases and future BDI scores

(A) Change in socializing during the COVID-19 pandemic. The degree of changes in the social distance, social frequency, time spent on socializing, and stress from socializing were shown. Data are represented as mean +/− standard error. ∗ represents p < 0.05, ∗∗ represents p < 0.01, ∗∗∗ represents p < 0.001, (∗) represents p < 0.1, N.S. represents no significance. One sample t-tests were carried out.

(B) Correlations between negative memory biases and future BDI in the higher and lower decrease of socializing subgroups of the DG. The median of change in socializing in the DG was used as the grouping criterion to ensure the balance of sample size in two subgroups. For the stress from socializing, social frequency, and time spent on socializing, subjects with a degree of change in socializing lower (or not higher, depends on the balance of sample size in the two subgroups) than the median was grouped into the high decrease subgroup. For social distance, subjects with a degree of change in social distance higher than the median was grouped into the high decrease subgroup. The subgroup with a higher decrease of socializing always shows a more negative correlation between the negative memory biases and future BDI scores.

We further divided the DG into subgroups according to their degree of change in socializing. The subgroup with a higher decrease of socializing always shows a more negative correlation between the negative memory biases and future BDI scores (Figure 6B).

Discussion

In the present study, we found that a more negative memory bias was associated with a greater decrease in future depressive indices in nonclinical depressed participants during the COVID-19 pandemic but not before the pandemic. The result of the lack of a significant difference between the HG and BG in the correlation between current negative memory biases and future depressive indices provides further evidence that this finding is specific to the effect of the pandemic. We further showed that the responses to sad faces in the DCC, which were significantly correlated with negative memory biases, were also negatively correlated with future depressive indices during the COVID-19 pandemic. A comparison of the change in future depressive indices revealed that the change in future BDI in the high bias subgroup of the DG was significantly lower than that in the high bias subgroup of the BG. Investigation of socializing showed that social stress decreased significantly during the pandemic in the nonclinical depressive population and the subgroup of the DG with a higher decrease of socializing showed a more significant and more negative correlation between the negative memory biases and future BDI scores.

As we explained in the introduction section, the COVID-19 pandemic may have two possible effects on the relationship between current cognitive biases and future depressive indices in the nonclinical depressed population. One is an increased stress effect considering that the COVID-19 pandemic is a threat to life. The other one is a decreased stress effect considering that social distancing may reduce social stress during the COVID-19 pandemic. Our results support the latter effect. We have provided behavioral evidence that the COVID-19 pandemic promotes a negative correlation between negative memory biases and future depressive indices in the nonclinical depressive population, which means that a more negative memory bias was associated with a greater decrease in future depressive indices during the pandemic. This effect might be due to the low degree of social stress during the COVID-19 pandemic. The investigation of the socializing of participants also corroborated this explanation. Decreased social frequency, increased social distance and decreased social stress were all found in the nonclinical depressed population during the pandemic. Social stress has been reported to be a crucial risk factor of depression (Paykel, 2003; Slavich et al., 2009). Thus, reduced social stress during the COVID-19 pandemic may lead to decreased depressive indices. Our finding that the subgroup of the DG with a higher decrease of socializing showed a more significant and more negative correlation between the negative memory biases and future BDI scores provided further evidence for this explanation. However, it is worth noting that although future depressive indices were negatively correlated with negative memory biases in the DG, no significant difference was found in the future depressive indices between the BG and DG. A possible explanation for this result is that reduced social stress only had a significant effect on individuals with a greater negative memory bias. For those with a less negative memory bias, the development of their depressive mood may be due to other events but not social stress. Thus, their depressive indices did not decrease with social stress during the pandemic. As a result, the depressive indices in the DG were not significantly lower than those in the BG. The results showing that the depressive indices in the high memory bias subgroup of the DG were significantly lower than those in the BG also corroborated this explanation. Previous studies have found an association between social stress and negative memory biases (Fan et al., 2017, 2020; Romano et al., 2020; Zhang et al., 2019), which is also in line with our findings.

The other possible effect, that the COVID-19 pandemic may increase survival stress and thus promote a greater increase in future depressive indices in participants with a more negative cognitive bias, was not supported by our data (although a positive correlation was found between negative interpretation biases and future depressive indices in the DG, the correlation coefficient was not significant different from that in the BG). The reason for this finding may be that the most severely affected region in China was Wuhan during the COVID-19 pandemic, and the participants in this study mainly lived in other cities and might not have experienced much survival stress. However, the social distancing policy was carried out all over China, thus a reduced social stress effect was found in our data.

Brain imaging analysis further corroborated the findings of the behavioral data. We tested ten ROIs highlighted by a previous meta-analysis and found that the responses to sad faces in the bilateral DCC, which were found to be correlated with negative memory biases, were also negatively correlated with future depressive indices during the COVID-19 pandemic. The predefined DCC area was mainly in the ACC in this study. The ACC has been related to the function of cognitive control of emotion (Bush et al., 2000; Disner et al., 2011; Li et al., 2020; Ochsner and Gross, 2005). Individuals with a greater activation in the ACC when viewing sad faces may have a deficient inhibition ability and thus require greater cognitive effort to divert attention away from negative stimuli (Disner et al., 2011). A deficient ability to inhibit attention for negative stimuli may contribute to negative memory biases.

Although the response to sad faces in the right vlPFC was also correlated with negative memory biases in our results, their correlation with future depressive indices did not show the same pattern as in the behavioral data. A possible explanation for this finding is that the activity in this area might be related to many kinds of depressive symptoms, not only memory biases. Thus, their relationship with future depressive indices was not altered as in the relationship between memory bias and future depressive indices.

In the present study, only the relationship between current memory biases and future depression degree was affected by the COVID-19 pandemic in the nonclinical depressive population, which might be due to the mechanisms of the development of memory biases in depression. Previous studies have found that stress can suppress hippocampal neurogenesis (Gould and Tanapat, 1999), inhibit dopamine neurons (Tye et al., 2013), and sensitize the amygdala (Roozendaal et al., 2009), and these phenomena impair pattern separation and disrupt the encoding of positive experiences and bias retrieval toward negative events, respectively, thus leading to negative memory biases in depression (Dillon and Pizzagalli, 2018). Therefore, people with greater negative memory biases may experience more social stress during their daily lives, and thus, their depressive indices may have been decreased more when social stress was decreased during COVID-19 pandemic. The relationship between current attention or interpretation biases and future depressive indices was not altered by the COVID-19 pandemic, which might imply that there are some underlying particularities of the mechanism of memory biases compared with other cognitive biases, which requires further investigation.

Our results also support social distancing policies that carried out during the pandemic from a psychological perspective. Increasing social distance can not only prevent the spread of COVID-19 but may also reduce social stress and thereby decrease depressive indices in the nonclinical depressed population, especially in those with a more negative memory bias. Moreover, considering that social stress may re-emerge when social distancing goes back to normal, the resumption of work and school should be carried out gradually, not only to prevent the recurrence of the pandemic but also to provide time for people to adapt to re-emerging social stress in case they experience an increased degree of depression. Previous studies have linked too close interpersonal distances to increased feelings of threat and decreased pleasantness (Ahs et al., 2015; Kroczek et al., 2020), which is also in line with our findings.

In summary, this study revealed that a more negative memory bias was associated with a greater decrease in future depressive indices during the COVID-19 pandemic in the nonclinical depressive population, which might be due to decreased social stress. The responses of the bilateral DCC to sad faces were found to be correlated with negative memory biases, and their negative correlation with future depressive indices converged with the results from the behavioral data. Our findings also support social distancing policies from a psychological perspective. Increasing social distance not only prevents the spread of COVID-19 but may also reduce social stress and thereby decrease depressive indices in the nonclinical depressed population, especially in those with a more negative memory biases.

Limitations of the study

Some limitations of this study must be acknowledged. For the VST, only sad and neutral faces were used. Thus, the reason why no significant correlation between attention biases and future depressive indices was found might be that the nonclinical depressive population may not pay more attention to negative stimuli but may instead pay less attention to positive stimuli. Moreover, all of the nondepressed participants' future BDI were assessed before the pandemic. Thus, our results might not generalize to nondepressed populations.

Resource availability

Lead contact

Further information and requests for resources should be directed to the lead contact, Xiaochu Zhang (zxcustc@ustc.edu.cn).

Materials availability

The sources of stimulus materials are provided in the Supplemental information file.

Data and code availability

The datasets supporting the current study have not been deposited in a public repository but are available from the lead contact on request. The statistics were performed on R version 3.6.1 (64-bit) platform. The code for statistical analysis is available from the Lead Contact on request.

Methods

All methods can be found in the accompanying Transparent methods supplemental file.