16:10-16:20 A combination of immune cell types identified through ensemble machine learning strategy detects altered profile in recurrent pregnancy loss

Abstract

Problem The underlying cause of recurrent pregnancy loss (RPL) remains largely unknown. As successful implantation and placentation depends on a tightly regulated immune response to facilitate adequate interaction with trophoblast cells, dysregulation of immunity might account for idiopathic RPL. So far, most studies focused on peripheral blood immune cells and performed single immunological parameter analysis which up till now did not allow for clear classification of RPL versus healthy pregnancies. Therefore, we here defined immune profiles in menstrual blood (MB) derived endometrial cells as well as peripheral blood (PB) of RPL patients and compared them to those of women who had healthy pregnancies. Method of Study PB and MB was obtained from women who suffered from at least three consecutive unexplained RPL (n=18) and aged-matched women without known disorders of reproduction and analyzed using flowcytometry. The proportions of 63 immune cell types defined by deep immune phenotyping were included in the analysis, next to age and CMV status. Results By harnessing the combined value of 8 machine learning classifiers in an ensemble strategy and recursive feature selection, we were able to determine a combination of immune parameters that separated RPL from controls. In peripheral blood, the combination of four cell types (non-switched memory B cells, CD8+CD4-T cells, CD56bright CD16- Natural Killer (NKbright) cells, CD4+ effector T cells) classified samples correctly to their respective cohort. The identified cell types differed from the results observed in MB, where a combination of 6 cell types (Ki67+CD8+ T cells, (HLA-DR+) regulatory T cells, CD27+ B cells, NKbright cells, Treg cells, CD24HiCD38Hi B cells) plus age. Based on the combination of these features, the average area under the curve of a receiver operating characteristics curve and the associated accuracy was >0.8 for both sample sources. Conclusion A combination of immune subsets for cohort classification allows for robust identification of immune parameters with possible diagnostic value and deserves further large-scale validation. In addition, the non-invasive source of menstrual blood holds many opportunities to assess and monitor reproductive health and further study the pathological mechanism of RPL.