Social Network Position Moderates the Relationship between Late-life Depressive Symptoms and Memory Differently in Men and Women

Late-life depression has been considered to be associated with memory deficits and hippocampal volume reduction. Considering that not all depression patients undergo the same amount of cognitive impairment or regional brain volume loss, moderating factors such as complex mental activity and social activity have been examined to assess whether these factors attenuate the detrimental impact of depressive symptoms on cognitive function and regional brain volume. However, the premise that a cognitively stimulating experience may modify the association between depressive symptoms and memory or hippocampal volume has not been investigated using social network data, which would reflect individuals’ concrete characteristic of everyday social activity. In a social network, a brokerage position which connects two otherwise unconnected others demands mental and physical efforts. Using complete social network data in an entire village in South Korea, we examined whether opportunities for brokerage in social networks alter the negative association between depressive symptoms and episodic memory function or hippocampal volume in older adults. Initially, 125 participants were included in the analysis involving episodic memory function. Then, of which 65 participants completed the MRI scan, and were included in the subsequent analysis containing the hippocampal volume. Furthermore, we investigated the gender-specific effect of brokerage based on the previously reported gender difference in the effect of social networks. We found a gender-specific moderating effect of brokerage. For men, a large opportunity for brokerage weakened the negative association between depressive symptoms and memory performance and left hippocampal volume. In contrast, women showed that a large opportunity for brokerage was not beneficial for assuaging the impact of depressive symptoms on memory performance and hippocampal volume. In women, the opportunity for brokerage was positively associated with the detrimental impact of depressive symptoms on memory performance and hippocampal volume. Our findings suggest that occupying a bridging position in a social network may minimize the impact of depressive symptoms on memory function and hippocampal volume among older men, whereas the opposite holds true for older women.

Late-life depression has been considered to be associated with memory deficits and hippocampal volume reduction. Considering that not all depression patients undergo the same amount of cognitive impairment or regional brain volume loss, moderating factors such as complex mental activity and social activity have been examined to assess whether these factors attenuate the detrimental impact of depressive symptoms on cognitive function and regional brain volume. However, the premise that a cognitively stimulating experience may modify the association between depressive symptoms and memory or hippocampal volume has not been investigated using social network data, which would reflect individuals' concrete characteristic of everyday social activity. In a social network, a brokerage position which connects two otherwise unconnected others demands mental and physical efforts. Using complete social network data in an entire village in south Korea, we examined whether opportunities for brokerage in social networks alter the negative association between depressive symptoms and episodic memory function or hippocampal volume in older adults. Initially, 125 participants were included in the analysis involving episodic memory function. Then, of which 65 participants completed the MRI scan, and were included in the subsequent analysis containing the hippocampal volume. Furthermore, we investigated the gender-specific effect of brokerage based on the previously reported gender difference in the effect of social networks. We found a gender-specific moderating effect of brokerage. For men, a large opportunity for brokerage weakened the negative association between depressive symptoms and memory performance and left hippocampal volume. In contrast, women showed that a large opportunity for brokerage was not beneficial for assuaging the impact of depressive symptoms on memory performance and hippocampal volume. In women, the opportunity for brokerage was positively associated with the detrimental impact of depressive symptoms on memory performance and hippocampal volume. Our findings suggest that occupying a bridging position in a social network may minimize the impact of depressive symptoms on memory function and hippocampal volume among older men, whereas the opposite holds true for older women.
Late-life depression, including depressive symptoms, has been associated with both poor cognitive function and regional brain volume loss, particularly in the hippocampus [1][2][3][4][5] . In late-life depression, the memory loss generally occurs with hippocampal volume reduction and it is possibly mediated by hippocampal volume loss [6][7][8][9] . Also, depressive symptoms have predicted subsequent cognitive decline independently of the baseline cognitive impairment, implicating the detrimental influences of depressive symptoms 10 . However, the severity of memory impairment and degree of hippocampal volume atrophy varied across depression patients. Even among those diagnosed with depression, not all patients undergo the same amount of cognitive impairment and regional brain volume changes, including hippocampus 11,12 . The inconsistency of the severity of memory decline and hippocampal atrophy in depression indicates the possibility of the moderating factors in the relationship between depressive symptoms and its detrimental effects on memory function or hippocampal volume. Neuropsychological test sample. Gender-specific regression models were generated to investigate whether the relation between depressive symptoms and memory performance is contingent on the brokerage measure among older adults. The analysis results are presented in Table 3. The two-way interaction term (depressive symptoms × brokerage) predicting LTM Recognition was significant among both older men (p = 0.04) and women (p = 0.01), indicating that the relation between depressive symptoms and memory performance was modified by brokerage role for both genders. The identical two-way interaction term in the model predicting the LTM Recall index was not significant for men or women.
In the subsequent analysis, the three-way interaction term (depressive symptoms × brokerage × gender) was significant in the model predicting the LTM Recognition index (p < 0.01). In an identical model predicting the LTM Recall index, the three-way interaction term was not statistically significant.
To test the robustness of the regression model predicting LTM Recognition, outlier detection was conducted using Cook's distance. For men, after the removal of the data point of one participant who met the Cook's distance outlier criteria, the statistical significance of the two-way interaction term was improved (p = 0.01). For women, no outlier was detected using the Cook's distance criterion. Due to a solid correlation between social network size and the brokerage measure, the social network size covariate was registered as an additional covariate in the above models. The significance of the interaction terms was not undermined by accounting for the effect of social network size (men: B = −0.04, p = 0.04; women: B = 0.05, p < 0.01).
Magnetic resonance imaging subsample. In the MRI subsample, a statistical analysis identical to that of the neuropsychological test sample was conducted to predict bilateral hippocampal volume. The analysis results are presented in Table 3. For men, the two-way interaction term (depressive symptoms × brokerage) predicting adjusted left hippocampal volume was marginally significant (p = 0.10). For women, this two-way interaction term was significant (p = 0.03). For both men and women, the two-way interaction term predicting adjusted right hippocampal volume was not statistically significant. www.nature.com/scientificreports www.nature.com/scientificreports/ Subsequently, the three-way interaction term (depressive symptoms × brokerage × gender) was tested. In line with the analysis in the neuropsychological test sample, the three-way interaction term was significant (p = 0.01) in the model predicting adjusted left hippocampal volume. This result revealed that the moderation pattern differs significantly across gender.
To test the robustness of the regression model, outlier detection was conducted using Cook's distance. For men, two data points were classified as outliers based on Cook's distance. After excluding two outlier data points, the two-way interaction term (depressive symptoms × brokerage) predicting adjusted left hippocampal volume was significant (p = 0.02). For women, one data point was detected as an outlier. After this data point was removed, the effect of two-way interaction was not significant. The significance of the interaction terms was not

Discussion
The present study is the first to examine the gender-specific role of bridging position in a social network in the context of late-life depression. Men with sufficient brokerage opportunities demonstrated an assuaged negative impact of depressive symptoms on memory performance and left hippocampal volume. In contrast, for women, more opportunities for brokerage were associated with greater detrimental impact of depressive symptoms on memory performance and hippocampal volume.  www.nature.com/scientificreports www.nature.com/scientificreports/ Given that a complex and cognitively stimulating life-time experience may play a role in attenuating the detrimental effects of late-life depressive symptoms on cognitive function and brain volume, we inferred that men who had more opportunity for brokerage in their social network were more likely to deal with and integrate diverse or even conflicting demands from their social ties and this opportunity for the brokerage position could weaken the negative association between depressive symptoms and episodic memory performance or hippocampal volume. In contrast, a buffering role of brokerage under the influence of depressive symptoms was not observed among women. Rather, women with more opportunity for brokerage showed greater adverse effects of depressive symptoms on their memory performance and hippocampal volume. Although the results found for women may not be consistent with the previous studies indicating advantages of bridging position 31,42 , the possibility of gender differences in brokerage has been discussed earlier because men tend to receive benefits from the brokerage position while women appear to be burdened by it 37 . The gender difference in brokerage could be attributed to the fact that socialization and affiliative relationship patterns differ between men and women 39,40 . Men tend to be satisfied with large and coarse social network in which they share their activities rather than building close emotional or supportive relationships. In contrast, women tend to rely on their social network for emotional, instrumental, and informational support 43 . Likewise, women are likely to receive more benefit from kin and other women rather than men, indicating that homophily may enhance the positive outcomes of social networks, such as physical and psychological health 39 . Consistent with the hypotheses, Carboni and Gilman 36 found that brokerage was positively associated with social stress and negatively associated with life satisfaction among women. Therefore, in the women of the present study, it seems that the costs of brokerage may have far outweighed the advantages of bridging opportunity.
This study has examined the role of a social network position called "brokerage" in the context of late-life depression. Even though individuals cannot know their social network position, the social network position captures the unique characteristics of an individual's daily social activity. By using a complete social network, we were able to define a social network position involving complex mental activity and found its benefits and costs in late-life depressive symptoms. Being located in a bridging position would require more energy and resources to keep complicated and conflicting relationships 33,34 . Therefore, men with brokerage seem to have benefited from their experience of a demanding social position when they have symptoms of depression. Nevertheless, we admit that we cannot eliminate the possibility that individuals who have better memory function or larger hippocampal volume are more likely to occupy the bridging position 44 and that they would then show better memory performance or larger hippocampal volume under the influence of depression. However, we may still insist on our hypothesis, which is centered on the buffering role of brokerage, in that we could not find a simple correlation between the brokerage opportunity and memory performance or hippocampal volume in the data.
More broadly, our findings provide one plausible explanation for the individual difference in the severity of memory deficits and hippocampal atrophy under a similar severity of depressive symptoms. Given that depressed older adults with cognitive deficits are more likely to progress to dementia than those depressed older adults without severe cognitive deficits 45,46 , finding a moderating factor in the association among depressive symptoms, memory performance, and hippocampal volume is a meaningful achievement. Our results represent an important step forward in examining a moderating factor weakening the influence of depressive symptoms on cognitive impairment by adding a measurement of hippocampal volume. One of the possible neurobiological mechanisms of depression-related memory decline is the dysregulation of the hypothalamic-pituitary-adrenal axis and subsequent hippocampal volume loss 8,47 . However, cognitively stimulating life-time experience has been known to lessen age-related regional brain volume loss, including the hippocampus. In a longitudinal study, older adults who experienced complex mental activity during their lifetime showed slower hippocampal atrophy rate 48 . Moreover, in a cross-sectional study, time spent in cognitively engaging activities was associated with larger www.nature.com/scientificreports www.nature.com/scientificreports/ cortical and subcortical volume, including the hippocampus 49 . Therefore, the fact that older men with larger opportunity for brokerage showed reduced association between depressive symptoms and left hippocampal volume could be attributed to their cognitively stimulating experiences from demanding social position.
We note several limitations in this study. Firstly, the use of cross-sectional data limits the interpretation. The approach outlined in this study should be replicated in a longitudinal study to elucidate the pathway among depressive symptoms, memory, and hippocampal volume and the moderating influence of bridging position on this pathway. Even though it is possible that smaller baseline hippocampal volume may have contributed to the age-related cognitive dysfunction or psychopathologies 50 , animal studies supported the causal direction among stress, hippocampal atrophy, and cognitive decline 51 by showing that aged rats exposed to chronic stress showed hippocampal dendritic atrophy and inhibited neurogenesis with memory decline. Secondly, even if we assessed the social network property of an entire village, the number of participants included in the analysis of neuropsychometry and regional brain volume was relatively small. In particular, the MRI subsample consisted of 65 older adults, with smaller number of men. Hence we cannot rule out the possibility that MRI analysis results may not be robust. Yet, the consistency of the results within each gender may minimize the possibility that the MRI subsample analysis result was unstable due to small sample size. Further, it may be inappropriate to generalize the findings to individuals with clinical depression since our results were based on self-reported late-life depressive symptoms in a community setting. Lastly, we found the moderation effect of brokerage only in the left hippocampus among the bilateral volumes. Although it may limit our finding, this is probably consistent with the previous finding that lateralization of hippocampal atrophy in patients with psychiatric disorders was related to the hypersecretion of glucocorticoids, including depression 52 .
In conclusion, social network characteristics, specifically brokerage position, measured using the entire population of a village, may play a protective role against the negative consequences of psychiatric disorders since there was a decreased impact of depressive symptoms on memory function and hippocampal volume among older men. Even though longitudinal research is necessary to elucidate the exact pathways of brokerage influence, our research has taken a step in the direction of examining the benefits of occupying a bridging position in a social network.
Methods participants and procedure. This study used data of the Korean Social Life, Health, and Aging Project (KSHAP). The KSHAP was conducted in Township K, a rural town in South Korea. Initially, 591 participants aged 60 or older and their spouses were administered a social network survey to collect a complete social network data 53 . Of the participants, 194 participants were assessed with neuropsychological tests, including measurements of episodic memory. Of these participants, those who did not meet the exclusion criteria (n = 125) were included in the analysis as neuropsychological test sample. Among these 125 participants who completed both the social network survey and neuropsychological tests, 65 participants (25 males and 40 females) were available for structural MRI data acquisition. Table 2 describes the characteristics of the participants. As shown in Table 2, demographic variables, depressive symptoms, memory function, and social network characteristics were assessed in the neuropsychological test sample. Subsequently, bilateral hippocampal volume and intracranial volume were measured in the 65 MRI subsample. In both the neuropsychological and the MRI subsamples, gender difference was not observed, except for years of education and intracranial volume. The study was approved by and administered in accordance with the relevant guidelines and regulations by the Institutional Review Board of Yonsei University, and all participants provided written informed consent for the research procedure. exclusion criteria. Participants were excluded if they had any history of psychiatric or neurological disorders, vision or hearing problems, metals in their body that could not be removed, hypertension or diabetes uncontrollable with drugs or insulin, a history of losing consciousness due to head trauma, infarction, or a history of stroke. Participants who scored below the fifth percentile on the Long-Term Memory Recall Index in Elderly Memory Disorder Scale 54 and below 1.5 SD on the Mini-Mental State Examination for Dementia Screening 55 were interviewed based on the Clinical Dementia Rating Scale. They were then excluded from the analysis unless the interview revealed that their low cognitive performance was not due to preclinical symptoms of dementia. Among the participants who completed the entire imaging protocol, those who had either diffuse old infarction or excessive head motion that degraded the quality of an image were excluded. Furthermore, participants who had no or only one social tie were excluded because it was not possible to calculate a social network index indicating the opportunity for brokerage. social network characteristics. Social network variables were created using data from the third wave of the KSHAP. The KSHAP adopted a name generator from the General Social Survey and National Social Life, Health, and Aging Project 53,56 . Individuals' social connections were constructed from a name generator that identifies social network members, including a spouse, if any, and up to five discussion partners, with information on real names, gender, and residence. The questionnaire was as follows: "From time to time, most people discuss things that are important to them with others. For example, good or bad things that happen to you, problems you are having, or important concerns you may have. Looking back over the last 12 months, who are the people with whom you most often discussed things that were important to you?".
To combine each respondent-centered network into a complete network of Township K, the people who appeared in different respondents' networks were identified as identical based on the following criteria: (1) those who were not the spouses of respondents and lived outside of Township K were excluded, (2) at least two out of three Korean characters in their names matched, (3) their gender was the same, 4) their age difference was less than five years, and (5) their addresses belonged to the same Ri (the smallest administration unit in South Korea). Finally, we identified a complete network of 835 township residents and their 1,285 ties.
www.nature.com/scientificreports www.nature.com/scientificreports/ Social network constraint was used to measure the extent of bridging role (or brokerage). It is an inverse measure of brokerage since brokerage is supposed to reduce the network constraint 57,58 . The network constraint that person j exercises on person i comprises two parts: person i is constrained by person j as much as 1) person i invests time and resources into a relationship with person j (p ij ), and 2) person i maintains a relationship with person k who has a connection with person j (p ik p kj ). We used p ij to represent the ratio of the social tie between i and j to the total ties of i and calculated it as follows: where a ij = 1 means i responded that i has a tie with j. Otherwise, a ij = 0. Thus, the formula to obtain an individual i's network constraint is calculated as c p pp in the following analysis. Social network size was measured by the number of social connections reported by each individual. Figure 1 describes more detailed characteristics of brokerage in social network. In the figure, A has ties with people who are connected to each other, and this type of social network is described as a highly constrained network: A is more likely to be constrained by the same strong demands from surrounding people because they are connected to each other and share the same norms and attitudes and exert influences on person A unanimously. In contrast with A, B maintains a relationship with others who are not connected to each other; thus, B occupies a brokerage position and enjoys fewer constraints but has to deal with diverse norms and demands from people with various backgrounds. Even though A and B have the same number of connections in their social network, presumably B has more brokerage potential and benefits from fewer constraints. Figure 4 offers a visualization of the complete social network in our study, indicating participants located in the brokerage position. Different colors indicate different Ri (the smallest administration unit in South Korea) in the village, and the size of each vertex represents the opportunity for bridging in their social network.

Neuropsychological assessment.
In the neuropsychological test sample, the Elderly Memory disorder Scale (EMS) 54 , a standardized neuropsychological battery, was administered to measure episodic memory. The EMS includes the Elderly Verbal Learning Test (EVLT), the Story Recall Test (SRT), and the Simplified Rey Figure  Test  Assessment of depressive symptoms. Depressive symptoms were assessed using the Geriatric Depression Scale (GDS), a self-rating measure of depressive symptoms designed to screen late-life depression 59 . It includes 30 yes or no questions about an individual's depressive mood. The GDS has high test-retest reliability (0.80 to 0.98) 59 , and it is widely used among community-dwelling older adults 60 . The GDS was translated into Korean by two psychiatrists and two clinical psychologists, but the original meaning of the items has been maintained 61 .  www.nature.com/scientificreports www.nature.com/scientificreports/ Magnetic resonance imaging acquisition and regional volumetry. MRI images were acquired using a 3-Tesla MAGNETOM Trio 32 channel coil; 1 mm 224 sagittal slices of a whole brain T1-weighted image were acquired for each participant, with a MPRAGE sequence using the following parameters: TR = 2300 ms, TE = 2.3 ms, FOV = 256 × 256 mm, and FA = 9°. The time gap between social network measurement and MRI acquisition was 16 to 21 months. T1-weighted images were introduced into an automated segmentation procedure using FreeSurfer software 5.3 (http://surfer.nmr.mgh.harvard.edu/). This procedure has high accuracy comparable with manual labeling 62 . Bilateral hippocampal volume was proportionally adjusted with total intracranial volume (ICV) 63 . statistical analysis. Statistical analysis was conducted using the SPSS package (version 22; IBM, USA).
Preliminarily, Pearson correlation analysis was performed not only to describe the associations among the variables of interest, age, and years of education but also to determine whether they should be registered as covariates in the following analysis. In addition, an independent t-test was conducted to investigate possible gender differences in the study variables.
We used PROCESS macro 2.16 implemented in SPSS 64 to generate and test multiple regression models predicting LTM indices and bilateral hippocampal volume using the neuropsychological test sample and the MRI subsample population, respectively. First, we examined whether the effect of depressive symptoms (X) on the LTM indices (Y) is moderated by brokerage (M) among men and women. Then, we examined whether this moderation effect (XM) is additionally moderated by gender (W). Therefore, gender-specific multiple regression models included the two-way interaction term (XM), and the following model included the three-way interaction term (XMW). Both models were tested in the neuropsychological test sample. Age and education were adjusted as covariates of no interest. Subsequent to this analysis, to predict bilateral hippocampal volume, an identical statistical procedure was repeated in the subsampled population who had undergone an MRI scan, adjusting for the effect of age.

Data Availability
The datasets generated during and/or analyzed during the current study are available in the Open Science Framework repository at https://osf.io/56cfm/.