Stratifying IVF population endometria using a prognosis gradient independent of endometrial timing

Background Independent of endometrial timing, there are molecular causes of implantation failure that disrupt the endometrial transcriptome in the mid-secretory phase. However, the molecular mechanisms disrupting the window of implantation (WOI) remain poorly understood. The molecular heterogeneity of this endometrial disruption must be characterized to develop personalized and more accurate diagnostic tools for preventive medicine, particularly for patients with a high risk of endometrial failure.

Objective(s) This study aimed to stratify and characterize the disrupted WOI patterns using endometrial timing-corrected whole gene expression and artificial intelligence (AI) models in in vitro fertilization (IVF) patients undergoing hormone replacement therapy (HRT).

Study design This multicenter prospective study was conducted between January 2019 and August 2022. Endometrial biopsies were collected during the mid-secretory phase for whole endometrial transcriptome analysis using RNA-Sequencing. To identify disruptions in the WOI, the transcriptomic variation due to cyclic endometrial tissue changes was removed. Out of 195 biopsies sequenced, 131 were derived from patients that met the clinical criteria to be classified as having a poor prognosis (≥3 implantation failures, n=32) or good prognosis (<3 implantation failures, n=99). The 131 patients were randomly allocated to training (n=105) and test (n=26) sets for biomarker signature discovery and assessment of predictive performance, respectively. The reproductive outcomes of the single embryo transfer immediately after biopsy collection were analyzed. Differential gene expression and functional analyses were performed to characterize molecular profiles. Finally, a quantitative polymerase chain reaction (qPCR) assay was used to corroborate the differential expression of six potential biomarkers.

Results With the dichotomous clinical classification of poor or good reproductive prognosis, there was no transcriptomic distinction between patients with a history of implantation failures during HRT endometrial preparation. Alternatively, using an AI model to stratify IVF patients based on the probability of endometrial disruption revealed molecular and clinical differences between profiles. Patients were stratified into four reproductive prognosis-related profiles, p1 (n=24), p2 (n=14), c2 (n=32) and c1 (n=61). The highest pregnancy rate (PR) was associated with c1 (91%) and the highest ongoing pregnancy rate (OPR) was associated with c2 (78%), linking these profiles to good reproductive prognoses. On the other hand, p1 had the highest biochemical miscarriage rate (BMR; 43%) while p2 had the highest clinical miscarriage rate (CMR; 43%). Notably, both p1 and p2 were related to lower PR and OPR, supporting that these profiles were associated with poor prognoses. Regarding the functional characterization in the poor prognosis profiles that were linked to miscarriages, p1 was associated with an excessive immune response against the embryo during early pregnancy stages, while p2 was initially immune-tolerant but rejected the fetus in later stages due to the lack of metabolic response.

Conclusion(s) This new AI-based prognostic stratification of IVF patients is promising for the clinical management of endometrial-factor infertility in precision medicine.