Predictors of ccf-mtDNA reactivity to acute psychological stress identified using machine learning classifiers: A proof-of-concept

Abstract

Objective: We have previously found that acute psychological stress may affect mitochondria and trigger an increase in serum mitochondrial DNA, known as circulating cell-free mtDNA (ccf-mtDNA). Similar to other stress reactivity measures, there are substantial unexplained inter-individual differences in the magnitude of ccf-mtDNA reactivity, as well as within-person differences across different occasions of testing. Here, we sought to identify psychological and physiological predictors of ccf-mtDNA reactivity using machine learning-based multivariate classifiers.

Method: We used data from serum ccf-mtDNA concentration measured pre- and post-stress in 46 healthy midlife adults tested on two separate occasions. To identify variables predicting the magnitude of ccf-mtDNA reactivity, two multivariate classification models, partial least-squares discriminant analysis (PLS-DA) and random forest (RF), were trained to discriminate between high and low ccf-mtDNA responders. Potential predictors used in the models included state variables such as physiological measures and affective states, and trait variables such as sex and personality measures. Variables identified across both models were considered to be predictors of ccf-mtDNA reactivity and selected for downstream analyses.

Results: Identified predictors were significantly enriched for state over trait measures (X2 = 7.03; p = 0.008) and for physiological over psychological measures (X2 = 4.36; p = 0.04). High responders were more likely to be male (X2 = 26.95; p < 0.001) and differed from low-responders on baseline cardiovascular and autonomic measures, and on stress-induced reduction in fatigue (Cohen's d = 0.38-0.73). These group-level findings also accurately accounted for within-person differences in 90% of cases.

Conclusion: These results suggest that acute cardiovascular and psychological indices, rather than stable individual traits, predict stress-induced ccf-mtDNA reactivity. This work provides a proof-of-concept that machine learning approaches can be used to explore determinants of inter-individual and within-person differences in stress psychophysiology.

Keywords: Machine learning; Mitochondria; Mitokine; Psychological stress; Stress reactivity; ccf-mtDNA.

Conflict of interest statement

Declarations of interest: BAK declares significant financial interest in GSK unrelated to the current study. All the other authors declare no conflict of interest.

Copyright © 2019 Elsevier Ltd. All rights reserved.

Figures

Fig. 1.. Circulating cell-free mitochondrial DNA (ccf-mtDNA) in response to induced psychological stress and operationalization of low and high responders.

(A) Socio-evaluative stress induces a 2-3 fold elevation in serum ccf-mtDNA 30 minutes after stress. (B) Distribution of the ccf-mtDNA reactivity for all participants across both sessions (n=74 total visits). The distribution of ccf-mtDNA response was divided into tertiles to define low, medium, and high responder groups. (C) A partial least squares discriminant analysis (PLS-DA) model using 56 physio-psychological trait and state variables available in the study produces partial separation of the low and high responders, as shown in the 3D plot of the first three PLS-DA components. Each datapoint is a participant at one of the two visits.

Fig. 2.. Identifying predictors of ccf-mtDNA response to acute psychological stress using multivariate classification algorithms.

(A) 56 total variables a priori classified as stable trait and variable state were used. Partial least squares discriminant analysis (PLS-DA) and random forest (RF) classification models were trained to distinguish between low and high ccf-mtDNA responders. (B) Following our pre-established analytic plan, we identified the top 15 variables predicting group affiliation ranked by variable importance in projection (VIP) score for the first component in PLS-DA (left), and mean decrease accuracy (MDA) score from the RF model (right). (C) Overlapping variables across both classifiers. Metrics for group differences are shown for comparative purposes only (not for statistical inference), using non parametric independent samples Mann-Whitney test for the continuous variables. Chi square was performed for the categorical variable (sex). Effect sizes were calculated as Cohen’s d. Significant (p<0.05) results are in shown bold, n = 46 visits across low and high reactivity tertiles. Abbreviations: BP: Blood Pressure, DBP: Diastolic blood pressure, SBP: systolic blood pressure, HF-HRV: High frequency heart rate variability, HR: Heart rate, HRV: Heart rate variability, LF-HRV: low frequency heart rate variability, RMSSD: root mean squared successive differences, RF: Random Forest.

Fig. 3.. Stress-induced changes in affect valence and activation stratified by low, medium high ccf-mtDNA responders.

Timecourse of affective response before (pre) and after (post) the stressor for the low, medium and high ccf-mtDNA responders arranged according to the circumplex model of emotion. The histograms show the reactivity (delta, post to pre) quantified for each tertile. Affect ratings were obtained using the profile of mood states (POMS) instrument, which assesses immediate mood states. Shown at the four poles are composite indices for low/high activation, and pleasant/unpleasant emotions. The central box also includes composite scores. Note the opposite direction of effects along the activation and valence axes. (n=74 visits)

Fig. 4.. Visualizing predictors of ccf-mtDNA reveal dose-response patterns.

(A) Proportion of women and men stratified by low, medium, and high ccf-mtDNA responder groups. (B) Scatterplot of mean ± standard error of the mean (SEM) by tertiles of ccf-mtDNA reactivity for baseline SBP and DBP, (C) baseline DBP and HR, (D) baseline SBP and HRV-RMSSD, (E) baseline and change HRV-RMSSD, and (F) change in LF-HRV and change in fatigue. Dotted arrows show progression from low to high responders. (n = 74 total visits) Abbreviations: DBP: diastolic blood pressure, SBP: systolic blood pressure, LF-HRV: low frequency heart rate variability, HR: Heart rate, HRV: Heart rate variability, RMSSD: root mean squared successive differences.

Fig. 5.. Individual profiles of divergent ccf-mtDNA responders and comparison with group-based predictions.

(A) Four participants (all males) exhibited low (green) and high (red) ccf-mtDNA responses on both occasions of testing and were thus identified as divergent responders. Shown are ccf-mtDNA levels before (Pre), immediately after (Post) and 30 min after (+30 min) the stressor for each participant. (B) Summary indicating the difference for each predictor between high and low ccf-mtDNA reactivity. Group-based findings for each predictor (black circle) establish the predicted direction of effects, shown as yellow boxes above or below the dotted line. For example, high responders are characterized by low baseline DBP and SBP, indicated by the black circles and boxes below 0. For each of the four divergent responders, the difference in each predictor between low and high reactivity sessions was plotted to assess the match in relation to group-based predictions. Individual data points going in the opposite direction than the group prediction (mismatch, 3 out of 28) are circled in red. Overall concordance between group-based predictions and individual divergent responses is 90%, compared to chance level (50%). Abbreviations: DBP: diastolic blood pressure, SBP: systolic blood pressure, LF-HRV: low frequency heart rate variability, HR: heart rate, HRV: heart rate variability, RMSSD: root mean squared successive differences.