Sera from people with HIV and depression induce commensurate metabolic alterations in astrocytes: toward precision diagnoses and therapies

Abstract Objectives People with HIV (PWH) have high rates of depression and neurocognitive impairment (NCI) despite viral suppression on antiretroviral therapy (ART). Mounting evidence suggests that immunometabolic disruptions may contribute to these conditions in some PWH. We hypothesized that metabolic dysfunction in astrocytes is associated with depressive symptoms and cognitive function in PWH. Methods Human astrocytes were exposed to sera from PWH (n=40) with varying degrees of depressive symptomatology and cognitive function. MitoTrackerTM Deep Red FM (MT) was used to visualize mitochondrial activity and glial fibrillary acidic protein (GFAP) as an indicator of astrocyte reactivity using the high-throughput fluorescent microscopy and image analyses platform, CellInsight CX5 (CX5). The Seahorse platform was used to assess glycolytic and mitochondrial metabolism. Results More severe depression, as indexed by higher Beck's Depression Inventory (BDI-II) scores, was associated with lower MT signal measures. Better cognitive function, as assessed by neuropsychiatric testing t-scores, was associated with increased MT signal measures. GFAP intensity negatively correlated with several cognitive t-scores. Age positively correlated with (higher) MT signal measures and GFAP intensity. Worse depressive symptoms (higher BDI-II scores) were associated with decreased oxygen consumption rate and spare respiratory capacity, concomitant with increased extracellular acidification rate in astrocytes. Conclusions These findings show that factors in the sera of PWH alter mitochondrial activity in cultured human astrocytes, suggesting that mechanisms that alter mitochondrial and astrocyte homeostasis can be detected peripherally. Thus, in vitro cultures may provide a model to identify neuropathogenic mechanisms of depression or neurocognitive impairment in PWH and test personalized therapeutics for neurologic and psychiatric disorders.


Data Summary
There are data for n=40 participants.The following tables and figures summarize variables used for the analyses.Note that measure units were guessed to be mg/dL for cholesterol and triglycerides.The following figure shows Pearson's correlations for associations between outcomes and predictors.The above figure shows a general pattern of negative correlations between the MT predictors and BDI-II scores and mostly positive correlations between the predictors and cognitive T-scores, with exception of Learning domain.GFAP target average was negatively correlated with cognitive outcomes.A more detailed summary is given in section 2.3.

Simple associations with covariates
September 18, 2023

Simple associations with covariates
The tests shown below are not corrected for multiple testing.Due to small group sizes, ethnicity groups were combined into 2: white and non-white.Due to skewness and large number of observations tied at 0, the duration of D'drug exposure was dichotomized: exposed (> 0) and not exposed (0 months duration).Associations with both, numeric and binary, variables are shown, but only the binary variable is considered for inclusion into multivariable models.
Effect of the extreme outlier in ART duration was tested.Associations with numeric covariates were tested for non-linear patterns, where such were suspected.Results of these tests are summarized in section 2.3.
The plots below are clustered by outcomes.

Simple correlations between outcomes and predictors
Learning and Motor T-scores were investigated for effects of outliers (71.5 & 7, respectively).Learning T-score did not have significant associations with predictors when the outlier was removed.Motor T-score did not have significant associations with predictors with or without the outlier.
MT spot average area has non-significant, but consistently negative correlation of moderate effect size with BDI-II total and subscores.It has positive associations with T-scores, approaching significant p-values for Global Mean T-score and Executive Functioning T-score and reaching significance for Working Memory T-score.Its associations with medical outcomes are not statistically significant.
MT spot average intensity has significant negative associations with BDI-II measures, specifically BDI-II Total, Apathy, Somatic, and Affective subscores.Statistically significant positive associations are found with Global Mean T-score and 3 domain scores: Executive Function, SIP, and Working Memory.Positive and marginally-significant association with cholesterol was also found.
MT spot total area/object shows no associations with any of the outcomes, including Learning T-score, which yielded statistically non-significant association after an outlier was removed (r = −0.18,p = 0.274).
MT target average has marginally significant negative correlations of moderate effect size with BDI-II total score and its subscores.Correlation tests with T-scores did not yield significant results, with most interesting association being positive correlation with Executive Function T-score.It does not correlate with medical outcomes.
GFAP target average does not correlate with BDI-II scores, but has negative association with cognitive T-scores, with marginal significance for Verbal and Recall domains.Statistical significance for Learning domain did not hold after removal of the outlier (r = −0.261,p = 0.114).There were no associations with medical outcomes.

Associations between outcomes and covariates
Covariates with p-values below 0.20 level are considered for inclusion into multivariable models.The following summary focuses on results with p-values below 0.20.
Tests for non-linear effects of numeric covariates on outcomes did not yield significant results.
The effect of a large outlier (493 months) in ART duration was tested.Its removal did not lead to any significant associations between ART duration and outcomes.Only one p-value remains below 0.20 level: for association with triglycerides.
There is no association between covariates and BDI-II total score or its subscores.Affective domain score's parametric association with height is driven by the data minimum in height.Its removal resulted in p-value above 0.20 (r = 0.16, p = 0.345), making parametric and nonparametric tests more consistent with each other.
For cognitive scores, there is no covariates' effect on Global Mean T-score, as well as on domain scores in Verbal, Learning, and Recall.At 0.20 level, Executive Function and Working Memory T-scores are higher for those exposed to D'drugs; SIP T-score is, on average, higher for females; Motor T-score is, on average, lower for those with metabolic syndrome.
There is no statistical association between total cholesterol and any of the covariates at 0.20 level.Higher triglycerides levels were associated with non-white ethnicity, metabolic syndrome, longer duration of ART, and shorter height.

Associations between predictors and covariates
Covariates with p-values below 0.20 level are considered for inclusion into multivariable models.The following summary focuses on results with p-values below 0.20.
Tests for non-linear effects of numeric covariates on predictors did not yield significant results.
The effect of a large outlier (493 months) in ART duration was tested.Its removal did not change significance of associations between ART duration and predictors, except GFAP target average, where r = 0.221 (p = 0.177) without the outlier.Correlation with MT spot total area/object is r = 0.300 (p = 0.064) after the outlier removal.
Covariates are not statistically associated with MT spot average area.At 0.20 significance level, higher MT spot average intensity is associated with female sex, metabolic syndrome, and shorter height.At 0.20 significance level, higher levels of MT spot total area/object are associated with older age, longer duration on ART, and shorter height.Non-parametric tests also show association with ethnicity and metabolic syndrome.Although the multivariable model uses parametric tests, these 2 predictors will be considered for inclusion into multivariable models.At 0.20 significance level, higher levels of MT target average are associated with older age, metabolic syndrome, and shorter height.At 0.20 significance level, higher levels of GFAP target average are associated with older age, female sex, and longer ART duration.

Multivariable models
For multivariable models, two approaches are taken, because the sample size is small and a model with at most 4 predictors is recommended for n=40.
First approach investigates effect of a single covariate, regardless of its significance from univariable analyses, on each outcome-predictor association.This approach shows which, if any, of the covariates change either the significance, or the effect's magnitude, or the direction of the effect of the predictor on the outcome.
The second approach uses a backward model selection.For each outcome-predictor pair, the starting model regresses outcome on the predictor and those covariates that are significant at 0.20 level with either the outcome or the predictor.Stepwise backward model selection is implemented to reduce each model to those covariates that remain statistically significant at α = 0.20 level.This could result in a model with no covariates or a model already shown in the first approach.
Outliers that were found to be influential (e.g., longest ART duration) in univariable analyses were removed for multivariable analyses.This leads to sample sizes below n=40 for some models.
As shown in the results below, the effects of the predictors on triglycerides seems to be influenced by covariates as noted by large changes in the predictors' coefficients and R-squared values.However, none of the models for predicting triglycerides yield a significant p-value for any predictor.

Controling for one covariate.
The following tables show results of analyses, where each outcome was regressed on predictor with adjustment for one covariate.Results are grouped by predictor.For each predictor, 2 tables are shown.
The first table shows model coefficients (slopes) and p-values for the effect of the predictor on the outcomes.The first column of such table shows unadjusted results for comparison.
The second table shows effect size for each model, in a form of partial R-squared.The first column shows unadjusted R-squared for comparison.R-squared measures percent of variability in the outcome explained by the predictor.Partial R-squared estimates association between the outcome and predictor with the effect of covariate removed.In general, controlling for covariates does not change effect of MT spot average area on the considered outcomes.
The effect on Executive Functioning T-score strengthens and becomes statistically significant at 0.05 level when controlling for D'drug exposure.In general, controlling for covariates does not change effect of MT spot average intensity on the considered outcomes.Overall, there were no statistically significant results for models with MT spot total area/object as a predictor.The size effect, in a form of R-squared, remained fairly small (below 0.045 or 4.5%) for all models.
September 18, 2023  For MT target average, models adjusted for covariates show results consistent with results of univariable analyses.
P-values for predicting BDI-II Total and Somatic from this predictor become statistically significant with 0.05 level in presence of some covariates.Note that, when model selection was applied for these outcomes (see section 3.2), none of the covariates remained in the models.Overall, the multivariable models adjusting effect of GFAP target average for covariates yield results consistent with univariable analyses.
between outcomes & predictorsThe tests shown below are not corrected for multiple testing.The figures are clustered by predictors.

Table 1 :
Summary of the outcomes.

Table 2 :
Summary of the predictors.

Table 3 :
Summary of the covariates.

Table 4 :
Coefficients (p-values)for effect of MT spot average area on outcomes (rows) alone and with control for a covariate (columns).Covariates' effects are not shown.

Table 5 :
R-squared (first column) and partial R-squared (remaining columns) for effect of MT spot average area on outcomes (rows) alone and with control for a covariate (columns).Covariates' effects are not shown.

Table 6 :
Coefficients (p-values)for effect of MT spot average intensity on outcomes (rows) alone and with control for a covariate (columns).Covariates' effects are not shown.

Table 7 :
R-squared (first column) and partial R-squared (remaining columns) for effect of MT spot average intensity on outcomes (rows) alone and with control for a covariate (columns).Covariates' effects are not shown.

Table 8 :
Coefficients (p-values)for effect of MT spot total area/object on outcomes (rows) alone and with control for a covariate (columns).Covariates' effects are not shown.

Table 9 :
R-squared (first column) and partial R-squared (remaining columns) for effect of MT spot total area/object on outcomes (rows) alone and with control for a covariate (columns).Covariates' effects are not shown.

Table 10 :
Coefficients (p-values)for effect of MT target average on outcomes (rows) alone and with control for a covariate (columns).Covariates' effects are not shown.

Table 11 :
R-squared (first column) and partial R-squared (remaining columns) for effect of MT target average on outcomes (rows) alone and with control for a covariate (columns).Covariates' effects are not shown.

Table 12 :
Coefficients (p-values)for effect of GFAP target average on outcomes (rows) alone and with control for a covariate (columns).Covariates' effects are not shown.

Table 13 :
R-squared (first column) and partial R-squared (remaining columns) for effect of GFAP target average on outcomes (rows) alone and with control for a covariate (columns).Covariates' effects are not shown.