Cortisol trajectory, melancholia, and response to electroconvulsive therapy

Abstract

While biomarkers have been used to define pathophysiological types and to optimize treatment in many areas of medicine, in psychiatry such biomarkers remain elusive. Based on previously described abnormalities of hypothalamic-pituitary-adrenal (HPA) axis function in severe forms of depression, we hypothesized that the temporal trajectory of basal cortisol levels would vary among individuals with depression due to heterogeneity in pathophysiology, and that cortisol trajectories that reflect elevated or increasing HPA activity would predict better response to electroconvulsive therapy (ECT). To test that hypothesis, we sampled scalp hair from 39 subjects with treatment-resistant depression just before ECT. Cortisol trajectory over the 12 weeks preceding ECT was reconstructed from cortisol concentrations in sequential hair segments. Cortisol trajectories varied widely between individuals, and exploratory analyses of clinical features revealed associations with melancholia and global severity. ECT non-responders showed a decreasing trajectory (mean change -25%, 95%-CI = [-1%,-43%]) during the 8 weeks preceding ECT (group-by-time interaction, p = 0.004). The association between cortisol trajectory and subsequent ECT response was independent of clinical features. A classification algorithm showed that cortisol trajectory predicted ECT response with 80% accuracy, suggesting that this biomarker might be developed into a clinically useful test for ECT-responsive depression. In conclusion, cortisol trajectory mapped onto symptoms of melancholia and independently predicted response to ECT in this severely depressed sample. These findings deserve to be replicated in a larger sample. Cortisol trajectory holds promise as a reliable, noninvasive, inexpensive biomarker for psychiatric disorders.

Keywords: Biomarker; Cortisol; Depression; Electroconvulsive therapy; Melancholia.

Figure 1.

Variation in cortisol trajectory in treatment-resistant depression. (A) Time-averaged hair cortisol concentration is plotted versus age for females (circles) and males (squares). Concentrations are plotted on a logarithmic scale. Median concentration, 7.2 pg/mg; mean concentration, 13.8 pg/mg; mean (SD) cortisol concentration in log₁₀ transformed units, 0.93 (0.41). (B) Hair cortisol concentration is shown for each hair segment. The distal, middle, and proximal hair segments represent time periods approximately 8–12, 4–8, and 0–4 weeks prior to ECT, respectively. (C) The data in (B) are re-plotted to show cortisol trajectories of individual subjects. (D) Change in hair cortisol concentration between segments is shown. Fold change is defined as the ratio of final over initial concentrations.

Figure 2.

Cortisol trajectory and melancholia. (A) Mean hair cortisol concentration is plotted versus hair segment for patients grouped by melancholia index: low (1st tertile, n=13, black), moderate (2nd tertile, n=15, blue), and high (3rd tertile, n=11, cyan). Error bars are 95%-confidence intervals. (B) Mean change in hair cortisol concentration between segments is plotted shown for patients grouped by melancholia index. Error bars are 95%-confidence intervals. (C) Change in hair cortisol concentration between distal and proximal segments is plotted versus melancholia index. Cortisol trajectories varied in a similar way as a function of global severity, which correlated with melancholia index (data not shown).

Figure 3.

Cortisol trajectory and response to electroconvulsive therapy (ECT). (A) Mean hair cortisol concentration is plotted versus hair segment for ECT responders (n=24, black) and non-responders (n=15, red). Error bars are 95%-confidence intervals. (B) Mean change in hair cortisol concentration between segments is shown for ECT responders and non-responders. Error bars are 95%-confidence intervals. (C) Change in hair cortisol concentration between middle and proximal segments is plotted versus cortisol concentration in the middle segment. The segregation apparent between non-responders (red) and responders (black) was quantified using the K-nearest-neighbors classifier (see text). For clarity, the outlier evident in Figures 1 and 2 is not shown; the results were essentially unchanged when the outlier was excluded from analyses.