Postpartum depression biomarkers predict exacerbation of OCD symptoms during pregnancy

Obsessive-compulsive disorder (OCD) symptoms in women often first occur or worsen during pregnancy and the postpartum, while the occurrence of obsessions and compulsions during the perinatal period is greater in women with comorbid depression (Fairbrother et al., 2016), suggesting a possible etiological overlap between perinatal OCD and depressive symptoms. We previously identified two biomarker loci (HP1BP3 and TTC9B) that, in conjunction with the proportion of monocytes in blood based on complete blood counts (CBCs), were capable of prospectively predicting antenatal and postpartum depression in humans with an area under the receiver operator characteristic curve (AUC) of 82% (Payne et al., 2020). The first objective of this study was to generate a proxy of monocyte proportion to enable model testing in cohorts without CBC information. The second objective was to assess the efficacy of PPD specific biomarker loci to identify exacerbations of OCD symptoms that occurred during pregnancy.

Human cohorts involved in the study for model building and proxy generation consisted of the Johns Hopkins University (JHU) Neuroimaging and Prospective cohorts described previously (Payne et al., 2020). Independent model evaluation for the second objective relied on a cohort of N = 48 participants (44 women in 22 matched pairs, age range 19–49 years old) selected from OCD-affected women that participated at the JHU site of the OCD Collaborative Genetics Study (2001–2006) (Samuels et al., 2006) or the OCD Collaborative Genetic Association Study (2007–2012) (Mattheisen et al., 2015).

We used the following strategy to perform the first objective. Using data from the GEO dataset GSE39981 that contains HM27 generated methylomes from FACs isolated cell types, we performed a probewise student’s t-test of N = 5 monocyte vs. N = 41 non-monocyte derived DNA methylation levels and found 6635 probes exhibited significant differences between groups after FDR based correction for multiple testing. To narrow this further, we took the top 5% of significant p values, which resulted in 41 probes. This set was narrowed further to 14 probes exhibiting significant association between DNA methylation and the ratio of monocytes to non-monocytes in N = 16 individuals with available differential CBC data from the JHU Prospective cohort. Of these probes, we selected probe cg10853416 located within 1500 bp of the transcription start site of the Membrane-Spanning 4-Domains, Subfamily A, Member 7 (MS4A7) for follow up analysis, as it is the only probe from the identified 14 expressed specifically in mature monocytes (Gingras et al., 2001). To validate this choice, we used pyrosequencing to measure DNA methylation of MS4A7 and identified a significant association with the differential CBC derived ratio of monocytes to non-monocytes in N = 21 samples from the independent JHU Neuroimaging cohort (β= −120.19 ± 41.6, F = 8.35, df=1/19, p = 0.0094), suggesting this probe may function as an adequate proxy of monocyte ratios for our modeling purposes.

With a proxy in hand, we next sought to generate a model capable of predicting exacerbation of OCD symptoms during pregnancy. Using a random forest machine learning approach, we trained a model on the presence (N = 4) or absence (N = 20) of obsessions occurring during pregnancy in the euthymic subset of women from the JHU Prospective Cohort. The random forest model was able to predict the training set with an AUC of 0.74 (95% CI: 0.47–1). We then applied the model to an independent test set and were able to distinguish a group of N = 21 women who experienced OCD symptom exacerbation from N = 20 women who did not with an AUC of 0.75 (95% CI: 0.59–0.91). To assess the importance of each model factor, we performed a Monte-Carlo card sorting task with 10,000 permutations, randomly shuffling each biomarker locus and recalculating the prediction AUC to create a null distribution of predictions for each locus. Permutation demonstrated that TTC9B and MS4A7 were most important to model performance (p = 0.0027, p = 0.034, respectively), while HP1BP3 exhibited a non-significant effect (p = 0.057).

Our work provides evidence for an etiological overlap between postpartum depression and perinatal OCD symptoms. Notably, the predictive efficacy of the model was tested for each biomarker, and TTC9B demonstrated the most significant evidence for contributing to OCD exacerbation. As discussed previously by Payne et al., (Payne et al., 2020), TTC9B epigenetic variation has been suggested by Shrestha et al., to mediate estrogen signaling, due in part to the demonstrated role of its close homologue, TTC9A, in mediating estrogen receptor alpha (ERα) signaling and by Lim et al., to alter rodent serotonin and behavior in the context of estrogen. Despite a lack of clear statistical significance, the permutation generated p value of 0.057 suggests that HP1BP3 is contributing at least some relevant biology. As previously discussed within Payne et al., Garfinkel et al., has suggested a role for HP1BP3 in mediating anxiety in female mice while the offspring of HP1BP3 KO mice had significantly lower survival rates due to a deficit in maternal care that could be reversed upon cross fostering. Cumulatively, the data suggest that HP1BP3 plays an important role in behaviorally mediated offspring survival and anxiety behaviors and warrants further study in the context of perinatal OCD.

This work suffers from several limitations, including a relatively small sample size of the training set, which was confined to the antenatal euthymic group of women. Furthermore, as our test set was comprised of approximately equal numbers of cases and controls, the predictive efficacy of any biomarker test based on these findings may not be representative of performance in the OCD population. Despite these caveats, our work provides suggestive evidence that a PPD biomarker model based on the association of epigenetic variation with gonadal hormone sensitivity is moderately predictive of OCD symptom exacerbation. These conclusions suggest that future etiologic studies of OCD should take into account hormonal variations and their associated downstream signaling pathways.