The role of lipoprotein profile in depression and cognitive performance: a network analysis

Lipid profile (total cholesterol and lipoprotein fractions) has been found to correlate with depression and cognitive impairment across the lifespan. However, the role of lipid levels in self-rated depressive state and cognitive impairment remains unclear. In this study, we examined the relationship between lipid profile (total cholesterol, triglycerides, high-density lipoprotein cholesterol and low-density lipoprotein cholesterol) and cognition in adults with and without self-rated depression. Four hundred and thirty-eight healthy participants completed the Repeatable Battery for the Assessment of Neuropsychological Status (RBANS), the Self-Rating Depression Scale (SDS), and a serum lipoprotein test. Using multivariate ANOVA, partial correlation and network analysis, a network linking lipoprotein profile, depressive state and cognition was constructed. A significant difference in serum lipid profile between the high and low depressive groups was detected. Depressive state had a strong negative correlation with cognitive performance. Of the lipid profile, only high-density lipoprotein was positively correlated with depressive symptom severity, whereas the other three indices showed negative correlation with both depressive state and cognitive performance. Our results suggest that serum lipid profile may be directly linked to self-rated depression and cognitive performance. Further studies recruiting larger clinical samples are needed to elucidate the specific effect of lipoprotein on cognitive impairment in mood disorder.


Scientific Reports
| (2020) 10:20704 | https://doi.org/10.1038/s41598-020-77782-9 www.nature.com/scientificreports/ In addition, they suggested that age, education level, smoking habit and body mass index (BMI) are key mediating factors that have significant negative impact on decreased HDL and increased LDL levels 18,24 . Pan et al. 27 also found significant correlations between individual components of the metabolic syndrome (CHO and TG) and depression in their meta-analysis on the bidirectional association between depression and metabolic syndrome. More recently, a meta-analytic review has linked depression with overall lower serum LDL level, though high heterogeneity was also noted 28 . The aim of this study was to examine the relationship between lipoprotein profile (CHO, TG, HDL, and LDL) and cognitive function in adults with and without self-rated depression. Most previous studies that explored the relationship between serum lipid levels and depression used multiple linear regression analysis after adjusting for confounding factors 29 . Here we adopted the network analysis approach, a new visualization tool that takes into account partially controlled correlational links among conceptualized variables 30 . We hypothesized that: (1) in general, significant negative correlations would be present between depressive state and cognitive performance, while serum lipoprotein levels would associate positively with depressive state and negatively with cognitive performance; (2) people with high and low self-rated depressive state would differ in cognitive function and serum lipoprotein levels; and (3) major demographic and lifestyle confounding factors (i.e. age, sex, education level, smoking status and BMI) would have a significant effect on serum lipoprotein levels, cognitive function, depression status and their relationships.

Results
Sample characteristics and group differences. Table 1 summarizes the demographic, lipoprotein and cognitive information of the sample. The Self-Rating Depression Scale (SDS) total score did not differ between male and female participants after controlling for age. There was no significant difference in sex, education level, smoking status, BMI, and lipoprotein levels between the low-depressive (LD) and high-depressive (HD) groups. Although we sought to recruit individuals with and without depressive symptom in the study, all participants with a history of any physical and mental illness including major depressive disorder, alcohol and drug dependence were also excluded. Therefore, only a small part of the participants was classified into the HD group. The HD group was significantly older than the LD group (F = 2.17, p = 0.03). However, all lipoprotein levels except HDL were found to be slightly higher in the LD group compared with the HD group. This pattern was reaffirmed in the subsequent multivariate ANOVA after controlling for confounding factors, in which significant discrepancies in CHO (F = 3.821, p = 0.002), TG (F = 15.649, p < 0.001), LDL (F = 3.654, 0.003) and HDL (F = 7.606, p < 0.001) levels were detected between the two groups. As for cognitive function, the LD group showed significantly higher scores on the Repeatable Battery for the Assessment of Neuropsychological Status (RBANS) except for the "language" subscale compared with the HD group. Moreover, after controlling for confounding factors, all sub-scale scores differed significantly between the two groups (see Table 2). Additionally, in our whole sample, BMI did not show any significant correlations with either depressive state (r = 0.018) or total RBANS score (r = −0.079). www.nature.com/scientificreports/ Network estimation, centrality, network accuracy and network stability. A network of 11 nodes representing depressive state (SDS total score), lipoprotein levels (CHO, HDL, LDL and TG) and cognitive performance (immediate memory, visuospatial/constructional, language, attention, delayed memory, and RBANS total score) was constructed. As seen in Fig. 1, the community detection function identified three major communities, with depressive state (SDS total score) in red, lipoprotein indices (CHO, TG, LDL and HDL) in green, and cognitive performance (RBANS scores) in blue. Edges linking nodes within each community were the strongest, suggesting robust measurement validity. The nodes R_T (standing for "RBANS total score") and TG (triglycerides level) had the largest number of edges, and were situated in the center of the cognitive and serum lipoprotein communities respectively. In general, the node SDS had a prominent negative link with the cognitive community, suggesting adverse effect of depression on adult cognitive performance; whereas edges from the serum lipid community had both positive and negative connections with the other two communities. Firstly, LDL, CHO and TG levels were all negatively connected with the cognitive performance community. Secondly, TG level had a similar negative link to the SDS node, while CHO and LDL levels were irrelevant to depressive state after controlling for all other nodes. Only HDL level was positively connected to both self-rated depressive state and cognitive performance. Intriguingly, TG and CHO levels were positively linked as one "unit" or "axis". Within the lipid community, LDL and HDL each had one positive and one negative edge with this axis. LDL exhibited a strong positive connection to CHO and a weaker negative connection to TG. Inversely, HDL had a strong negative connection to TG and an equally strong but positive edge with CHO. Network inference measures showed that the SDS node had the highest standardized strength centrality in the network, followed by R_T, both of which also had the highest closeness estimate in the network, suggesting that they had the shortest path connecting them to other nodes. CS coefficients were 0.594 for strength, 0.516 for closeness, and below 0.25 for betweenness. Since the main indicator of stability was the CS coefficient of strength, which was above 0.5, it was safe to assume that the current model was stable. The EI was at the maximum obtainable value of 0.75 31 . Additionally, results of the bootstrapped difference test, the edge weight accuracy test, and the edge weight difference test can be found in Supplementary Materials.
Correlations between lipoprotein profile, cognitive performance and depressive symptom. Table 3 shows that for the LD group (n = 379), CHO, TG and LDL all had significant negative correlations with cognitive scores. Specifically, scores on the memory related sub-scales (i.e. "Immediate Memory" and "Delayed Memory") were inversely correlated with CHO and LDL levels, while "Attention" was negatively correlated with TG level. Of the lipid profile indices, the correlation of HDL with cognitive performance was not significant. For the HD group (n = 59), TG had a moderate to strong correlation with score on the "Attention" sub-scale of the RBANS (r = −0.317, p < 0.05), while all other lipoprotein levels' correlation indices were not statistically significant.

Discussion
The present study examined the relationships among serum lipoprotein profile, cognitive impairment and selfrated depression using network analysis. The constructed network demonstrated the negative correlations of serum LDL, CHO and TG levels with cognitive performance. Moreover, TG level was negatively linked to depressive state, while HDL level was positively connected to both self-rated depressive state and cognitive performance. www.nature.com/scientificreports/ Group comparison revealed that the HD group exhibited lower levels of serum LDL, CHO, and TG, higher level of serum HDL, and performed significantly worse in cognitive tests compared with the LD group. Prior research suggests that cognitive impairment may be a core feature of self-rated depression 15 . In general, we found that self-rated depression had a significant negative impact on cognitive performance. Lipoprotein profile also differed significantly between the HD and LD groups after controlling for age, sex, education level and smoking status. Intriguingly, of the four serum lipid indices, only HDL displayed a positive correlation with cognitive function, and this correlation was significantly weakened in the HD group that exhibited higher HDL level, as well as lower CHO, TG and LDL levels (see Fig. 1 and Table 2). This result supports the negative correlation between TG level and cognitive impairment in major depressive disorder 15 . However, the negative relationship between high lipoprotein levels and cognitive performance did not appear in the correlation matrix of the HD group, which may be due to our small sample size (n = 59), or a genuine group-wise discrepancy in serum lipid profile.
Notwithstanding a well-recognized association between lipoprotein and depressive symptoms, the specific roles of CHO, TG, HDL and LDL are not clearly known. Our results are mostly consistent with the previously proposed "unfavourable serum lipid profile" featuring higher HDL, lower LDL and CHO 24 . Previous findings  Table 3. Correlations between lipoprotein levels, TG, and cognitive performance across the LD and HD groups. Partial correlations, controlling for age, gender, smoking and education years, 2-tailed, *p < 0.05, **p < 0.01. www.nature.com/scientificreports/ suggest that depressive symptoms are associated with lower LDL level in adult women 18 and men 24 , which are supported by our network analysis. However, contrary to studies that reported significant correlations of higher TG and CHO levels with depressive symptoms 15,32 , our network found that only HDL showed a positive association with self-rated depressive state. A critical but unresolved issue therefore arises regarding whether HDL is protective against or contributes to depressive symptoms. We note that our finding of higher HDL level in the HD group contradicts with the results from most studies 27,32 . This inconsistency may reflect the complexity of HDL's relationship with depression, although variations in methodology (different self-report scales, covariates or confounders considered) along with demographics and ethnicity of the sample studied, may also account for the discrepancy. Another reason for the inconsistent results might lie in the nature of the depressive spectrum. In this study, group differences in participants with higher SDS scores (above 50 points) and lower scores (below 50 points) were compared, while some other studies investigated the change of HDL between patients with clinically diagnosed major depressive disorder and healthy controls. For example, results from a larger sample in the Netherlands Study of Depression and Anxiety (NESDA) indicated lower HDL level in 761 participants currently with major depressive disorder than 629 healthy controls (p = 0.007) 32 . However, such significance disappeared after controlling for lifestylerelated factors such as BMI 32 . Therefore, it is important to recognize that adverse lifestyle behaviors (e.g. lower dietary fibre intake and smoking) may underlie serum lipid levels which consequently affect cognitive and mood symptoms 33 . Consistent with our hypothesis, major demographic and lifestyle confounding factors (i.e. age, sex, education level and smoking status) exhibited a significant effect on serum lipid levels, cognition, depressive status and their associations.
This study has several limitations. Firstly, despite the present findings suggesting a correlational pattern between self-reported depression and lipoproteins, this study did not empirically test the directionality of effect between depressive symptoms and serum lipid levels 24 . Secondly, due to difficulties in data collection, the comparison between depressive and non-depressive individuals in our samples was not matched for age and intelligence quotient (IQ), which might have biased our results. Further research would benefit from a larger and more demographically-balanced sample. Also, exclusion of participants with diagnosis of major depressive disorder is a major limitation for us to address the association between depressive symptoms, cognition and lipoprotein profile. Thus, future study should recruit participants with established major depressive disorder to examine whether a similar network pattern would be shown. Finally, our sample size was relatively small (n = 438) for network analysis, which limited our network's stability. Future studies should investigate the relationship between increased HDL level and cognitive impairment in people with major depressive disorder and examine how this contributes to the core symptoms of major depressive disorder.
In summary, our results suggest that there may be distinct correlations between serum lipoprotein levels, cognitive performance and self-rated depressive status. The use of network analysis has added to the current knowledge in this field by illustrating the relationships between specific lipoprotein levels, cognitive function and self-rated depression after controlling for all other factors in the conceptual network. Serum lipoprotein levels may play a putative role in cognitive impairment in people with self-rated depression.

Methods
Participants. Participants were recruited from the annual physical examination event of the Wenzhou Kangning Hospital. A total of 575 people attended the event, and 512 people agreed to take part in the study upon invitation. Seventy-four participants were excluded due to physical health problems, taking regular medications and returning incomplete checklists. The final sample consisted of 438 adults, with 312 females and 126 males. The following inclusion criteria were applied: (1) staff of the hospital; (2) 18-70 years of age; (3) no history of mental illness that was diagnosed in the course of the physical and psychological examinations of newly recruited staff; (4) no history of alcohol and drug dependence; and (5) good past physical health. The study protocol was approved by the Ethics Committee of the Wenzhou Kangning Hospital, and all procedures were carried out in accordance with the approved guidelines and regulations. Written informed consent was obtained from each participant. . Participants had to complete all tests to obtain a total cognitive score. The present study adopted the Chinese version of the RBANS that has been shown to have good test-retest reliability and validity in a healthy Chinese population 35 .

Measurements.
The SDS is a 20-item checklist which assesses the depressive state of participants 36 . The original SDS had a sensitivity of 97%, a specificity of 63%, and an 82% correct classification rate in distinguishing between depressed and non-depressed individuals. The responses were rated from 0 to 3 and summed to obtain a total score, which was then transformed into a standard score 37 . Based on a morbidity cutoff index from previous research 38 , participants who had a SDS standard score below 50 points were classified into the LD group, and those with a score above 50 were classified into the HD group in the present study.
Serum lipoprotein levels. The overnight fasting blood sample was collected from all participants by venipuncture, between 7:00 and 9:00 am. The sample of approximately 5 ml venous blood was allowed to clot at room temperature, and then centrifuged at 3500 rpm for four minutes. After collecting the serum, the sample www.nature.com/scientificreports/ was stored at − 80 °C until it was thawed for assay. CHO, HDL, LDL and TG levels were measured by enzymatic colourimetry with a HITACHI 7180 automatic biochemistry analyzer (Hitachi High-Technologies Corporation, Japan) and commercially available kits (Medical System Biotechnology, Ningbo, China).

Network estimation and statistical analysis. Network analysis is a new tool that visualizes partially
controlled relational links among conceptualized factors 39 . In this study, we used the R software (version 3.5.0, available at https ://cran.r-proje ct.org/) to construct a partial correlation network encompassing data from all participants (n = 438). A total of 11 nodes representing: (1) lipid profile encompassing CHO, HDL, LDL, and TG levels; (2) a self-rated depressive state indexed by the SDS total score; and (3) cognitive performance level indexed by the total score and scores on the five sub-scales of the RBANS, were included to explore the relationships among lipoprotein levels, self-rated depressive state and cognitive performance of the participants. Two measures were applied to estimate network inference: Strength Centrality and Expected Influence (EI). Strength Centrality measures the number of direct connections of one node in the whole network, which connotes the importance of a particular node. One step EI is the sum of all edges of a node, while two-step EI measures the accuracy and stability of the constructed network. The bootstrapped difference test, the edge weight accuracy test, and the edge weight difference test were applied to all node centrality indices. The stability of the centrality estimates was also quantified by the correlation-stability-coefficient 40 .
To analyze the differences in lipoprotein levels and cognitive performance between the LD and HD groups, multivariate ANOVA was conducted while controlling for four major confounding factors (i.e. age, sex, education level and smoking status) using SPSS 16.0 (https ://www.ibm.com/analy tics/spss-stati stics -softw are). Due to the small number of participants in the HD group (n = 59), we applied an additional partial correlation analysis to investigate the between-group differences in the influence of serum lipid levels on cognitive performance.