Non-invasive giant panda pregnancy and pseudopregnancy biomonitoring by integrated metabolomics and steroidomics

Abstract

Understanding the reproductive biology of giant pandas is crucial for their breeding success and conservation. Pregnancy monitoring, however, is challenging due to delayed implantation and obligatory pseudopregnancy, which limits the effectiveness of traditional immunoassays (IA). To remedy this, we combined polar metabolomics and steroidomics to enable a comprehensive view of the urinary molecular composition across six different reproductive phases spanning six pregnant and seven pseudopregnant cycles. Statistical comparisons revealed 696 discriminative features, including 174 features in the early luteal stages, well before the current pregnancy diagnostic window. Pregnant and pseudopregnant cycles showed differences in amino acid, energy, and steroid metabolism before and after CL reactivation, with androgen levels being significantly elevated in pregnant females specifically, suggesting a role in embryo implantation. Interestingly, we detected only one existing IA target metabolite, but identified other discriminative metabolites that may underlie IA signal detection. Finally, we demonstrated that classification models comprising biomarker panels may improve (early) pregnancy diagnosis with accuracies ranging from 0.763 to 1.000 across reproductive phases. These findings offer possibilities for assigning new biomarkers and optimizing IA target selection, thereby enhancing pregnancy monitoring sensitivity and reliability while improving our understanding of giant panda reproductive biology to support conservation efforts.

Keywords: Ailuropoda melanoleuca; Endocrine monitoring; Metabolomics; Pregnancy; Steroidomics; UHPLC-HRMS.

Fig. 1

Targeted polar metabolic feature abundances in pregnant and pseudopregnant giant pandas across the reproductive phases. Feature abundances are based on peak areas of the different samples per reproductive phase in pregnant (P, orange, n = 15 per phase) and pseudopregnant (PP, blue, n = 21 per phase) cycles. An anestrus, CLD corpus luteum dormancy phase, EAL early active luteal phase, LAL late active luteal phase.

Fig. 2

Untargeted polar metabolic feature abundances in pregnant and pseudopregnant giant pandas across the reproductive phases. Feature abundances are based on peak areas of the different samples per reproductive phase in pregnant (P, orange, n = 15 per phase) and pseudopregnant (PP, blue, n = 21 per phase) cycles. O-methoxyhippuric acid and n-aminodecanoic acid are putatively identified (Tier 3). An anestrus, CLD corpus luteum dormancy phase, EAL early active luteal phase, LAL late active luteal phase.

Fig. 3

Targeted steroid feature abundances in pregnant and pseudopregnant giant pandas across the reproductive phases. Feature abundances are based on peak areas of the different samples per reproductive phase in pregnant (P, orange, n = 18 per phase, except for the CLD1 phase (n = 17)) and pseudopregnant (PP, blue, n = 21 per phase, except for the anestrus and LAL phase (n = 20 and 18, respectively)) cycles. UA1 = unknown androgen 1, an analogue of 11β-hydroxyetiocholanolone. An anestrus, CLD corpus luteum dormancy phase, EAL early active luteal phase, LAL late active luteal phase.

Fig. 4

Untargeted steroid feature abundances in pregnant and pseudopregnant giant pandas across the reproductive phases. Feature abundances are based on peak areas of the different samples per reproductive phase in pregnant (P, orange, n = 18 per phase, except for the CLD1 phase (n = 17)) and pseudopregnant (PP, blue, n = 21 per phase, except for the anestrus and LAL phase (n = 20 and 18, respectively)) cycles. 21-deoxycortisol and 6,15-diketo-13,14-dihydro-PGF_1α are putatively identified (Tier 3). An anestrus, CLD corpus luteum dormancy phase, EAL early active luteal phase, LAL late active luteal phase.

Fig. 5

Accuracy dot plots of the GLM models with elastic net regularization per reproductive phase. (A) Models built with the metabolomics data. (B) Models built with the steroidomics data. (C) Models built with the combined metabolomics and steroidomics data. Acc = Accuracy, CI = 95% confidence interval. All models were built after QC filtering (CV > 30%) and exclusion of features significantly different in the anestrus phase (p < 0.05). Remaining features used to build the models encompassed 7486 for the metabolomics data, 7,397 for the steroidomics data, and 14,406 for the combined data. CLD corpus luteum dormancy phase, EAL early active luteal phase, LAL late active luteal phase.