Nurr1 deficiency orchestrates a coupled liver-gut pathological axis revealed by multi-omics and deep-learning histopathology.

The nuclear receptor Nurr1 (NR4A2) is a transcriptional regulator of inflammatory homeostasis, but its systemic effects on orchestrating inter-organ communications are largely unknown. Here we show that Nurr1 haplo-insufficiency results in a lethal coupled disorder across the liver-gut axis. Using a CRISPR-Cas9 generated murine model, we find that metabolically-activated heterozygous deficiency of Nurr1 results in profound hepatocellular necrosis and marked hepatic activation of inflammatory and pro-fibrotic genes coupled with dysregulation of the intestinal barrier, and severe small-intestinal dysbiosis. Multi-omics integration reveals a highly penetrant transcriptional signature of this herein termed liver-gut disorder, achieving up to 0.950 accuracy (SVM-RBF, 10-fold cross-validation) in classifying genotypes from integrated multi-omics features. Notably, we also demonstrate that these gene level perturbations in Nurr1 haplo-insufficiency can be thought of as learnable tissue 'morphologies' detectable by AI. Next, we created deep convolutional neural networks that accurately classify genotype from routine histopathology. Our algorithm achieves 99.50% accuracy in classifying hepatic fibrosis (Sirius Red), 99.20% in liver inflammation (H&E) and 92.31% in intestine (H&E). We provide the first multi-omics phenotype of Nurr1 deficiency, revealing its pivotal regulatory role in coordinating liver-gut homeostasis, and establishing a histopathological AI-driven framework. Grad-CAM saliency analysis confirms biological interpretability. Translational relevance is supported by human transcriptomic data (E-GEOD-61260) showing concordant upregulation of COL1A1 (log2FC= + 0.725, p < 0.01), TGFB1 (+ 0.429, p < 0.05), and MMP9 (+ 0.969, p < 0.01) alongside reduced NR4A2/NURR1 in human liver disease.