Figure 4.

Connectivity biomarkers predict differential antidepressant response to rTMS. (a) Differing response rates to repetitive transcranial magnetic stimulation (rTMS) of the dorsomedial prefrontal cortex across patient biotypes (clusters) in n = 124 subjects. Response rate indicates percentage of subjects showing at least a partial clinical response to rTMS (χ² = 25.7, P = 1.1 × 10⁻⁵), defined conventionally as >25% reduction in symptom severity by HAMD. Full response rates (>50% reduction by HAMD, cross-hatched bars) also varied by biotype (χ² = 22.9, P = 4.3 × 10⁻⁵). (b) Boxplot of percent improvement in depression severity by biotype (P = 1.79 × 10⁻⁶, Kruskal–Wallis ANOVA), in which boxes denote the median and interquartile range and whiskers the minimum and maximum up to 1.5 × the IQR, beyond which outliers are plotted individually. Percent improvement = total HAMD score before treatment – total HAMD score after treatment/total HAMD score before treatment. **P = 0.00001–0.002 (Mann–Whitney), indicating significantly increased versus biotypes 2–4; *P = 0.007 (Mann–Whitney), indicating significantly increased versus biotype 4. (c) Functional connectivity differences in the DMPFC stimulation target in treatment responders versus nonresponders (Wilcoxon rank–sum tests, thresholded at P < 0.005). Warm colors represent increased and cool colors decreased functional connectivity in treatment responders as compared to nonresponders. The 12 ROIs depicted here were located within 3 cm of the putative DMPFC target site, estimated in a previously published report to be located at Talairach coordinates, x = 0, y = +30, z = +30 (ref.¹³). (d) The neuroanatomical distribution of the most discriminating connectivity features for the comparison of rTMS responders versus non-responders, summarized by illustrating the locations of the 25 (top 10%) most discriminating ROIs indexed by summing across all significantly discriminating connectivity features and colored by functional network as in Figure 1a. The red arrows denote the rTMS target site in the two (lower) medial panels. (e) Heat maps depicting differences in functional connectivity in patients who subsequently improved after receiving rTMS (n = 70), as compared to those who did not (n = 54). (f–i) Confusion matrices depicting the performance of classifiers trained to identify subsequent treatment responders on the basis of the most discriminating connectivity features (f), connectivity features plus biotype diagnosis (g), clinical symptoms alone (h) or connectivity features plus biotype diagnosis in an independent replication set (i, n = 30 patients with depression). NR, nonresponder; R, responder. (j) Summary of performance (overall accuracy) for classifiers in f–i. **significantly greater than clinical features alone (P < 0.001) and connectivity features alone (P = 0.003) by permutation testing; *P = 0.04 (significantly greater than clinical features alone by permutation testing). Cross-hatched bars depict classifier accuracy with more stringent data quality controls (Online Methods) and excluding equivocal classification outcomes (the 10% of subjects with the lowest absolute SVM classification scores). Error bars depict s.e.m. in a and 95% confidence intervals in j. All abbreviations as in Figures 1 and 2. See Supplementary Table 7 for MNI coordinates for ROIs in d.