Prostate cancer frequently develops resistance to androgen deprivation therapy (ADT), with adenocarcinoma-neuroendocrine prostate cancer (NEPC) lineage plasticity representing a critical yet mechanistically unclear driver of therapeutic failure. Here, we employ single-cell transcriptomic analysis of 29,436 cells from transgenic adenocarcinoma of mouse prostate (TRAMP), recapitulating the full clinical progression of human prostate cancer-NEPC evolution, to dissect the molecular dynamics underlying lineage transition. We identify Hoxd11 as a key transcriptional driver of adenocarcinoma-NEPC plasticity, with its expression and regulatory activity markedly upregulated during early NEPC transformation. Genetic suppression of Hoxd11 reversed lineage plasticity and delayed ADT-induced neuroendocrine differentiation. Mechanistically, HOXD11 directly activates the NMDAR signaling pathway (via GRIN1/GRIN3A), which is hyperactivated in NEPC and correlates with poor prognosis. Pharmacological inhibition of NMDAR using the clinically available antagonist memantine significantly suppressed tumor growth in murine models. These findings position the HOXD11-NMDAR axis as a druggable pathway in NEPC, with memantine repurposing offering near-term therapeutic potential.

Sequencing Protocol:

scRNA-seq for TRAMP and human PCa tissue: Fresh tumor specimens were preserved in GEXSCOPE Tissue Preservation Solution under ice-cold transport conditions. Subsequent processing involved sequential steps: (1) washing in Hanks Balanced Salt Solution, (2) mechanical mincing coupled with enzymatic digestion using GEXSCOPE Tissue Dissociation Solution, (3) centrifugation and resuspension, followed by (4) erythrocyte depletion with GEXSCOPE Red Blood Cell Lysis Buffer. Single-cell suspensions derived from this workflow were converted into barcoded libraries employing the Chromium Single Cell Kit (10x Genomics). Final libraries were prepared with 10x Genomics Library Kits and subjected to paired-end 150 bp (PE150) sequencing on an Illumina NovaSeq 6000 platform, strictly adhering to manufacturer protocols.

RNA-seq for HEK-293T and LNCaP cells: Total RNA was isolated from HEK-293T cells using Trizol reagent according to the manufacturer's protocol, followed by purification and DNase I treatment to eliminate genomic DNA contamination. RNA concentration and purity were assessed spectrophotometrically using a NanoDrop ND-1000 instrument. Subsequently, 1 µg of high-quality RNA (A260/A280 ratio ≥ 1.8; A260/A230 ratio ≥ 2.0) underwent poly(A) selection using oligo-dT magnetic beads. Stranded RNA sequencing libraries were constructed from the enriched mRNA utilizing the KAPA Stranded mRNA-Seq Kit (Roche Sequencing Solutions, USA), incorporating unique dual indexing adapters. Library fragment size distribution was verified by capillary electrophoresis, and quantification was performed via qPCR (KAPA Library Quantification Kit). Equimolar pools of validated libraries were sequenced on an Illumina NovaSeq 6000 platform (Illumina, USA) employing a 150 bp paired-end (PE) read configuration to a target depth of approximately 30 million reads per sample.

Assay for Transposase-Accessible Chromatin with high throughput sequencing (ATAC-seq)：Library preparation for ATAC-seq was conducted according to the protocol of the Vazyme TD711 kit. Briefly, 5×10⁴ viable cells per reaction were washed twice with ice-cold PBS and then lysed in 50 μL of cold lysis buffer for 10 minutes on ice. A transposition reaction was performed in 50 μL of TTBL buffer containing 2.5 μL TTE Mix V50 for 30 minutes at 37 °C. The resulting DNA fragments were extracted with VAHTS DNA Clean Beads, involving two consecutive 80 % ethanol washes. Amplification of the library was achieved through 12 cycles of PCR. The final libraries were size-selected for fragments between 100 and 700 bp using VAHTS DNA Clean Beads, quantified with a Qubit 4.0 Fluorometer, and the fragment distribution was assessed on an Agilent 2100 Bioanalyzer prior to paired-end sequencing (150-300 bp) on the Illumina NovaSeq™ X Plus platform.

HOXD11-Flag CUT&Tag assay: The CUT&Tag assay was conducted with the Vazyme CUT&Tag Kit in accordance with the manufacturer's protocol. In brief, 1×10⁵ cells were harvested, washed twice with 1× PBS, and subsequently immobilized on activated ConA-coated magnetic beads. After permeabilization using Dig-wash buffer containing 0.05 % digitonin, the cells were incubated with a primary antibody against HOXD11-Flag overnight at 4 °C. This was followed by sequential incubations: first with a secondary antibody (1:50 dilution) for one hour at room temperature, and then with the pA/G-Tnp transposase complex for another hour. Tagmentation was then performed in the designated buffer at 37 °C for one hour. The resulting DNA fragments were purified using magnetic beads and amplified via a 12-cycle PCR with indexed primers. The final libraries were quantified on a Qubit 4.0 Fluorometer, and their quality was assessed with an Agilent 2100 Bioanalyzer to confirm insert sizes of 200-500 bp before paired-end sequencing on the Illumina NovaSeq™ X Plus platform.

Data processing to generate matrix:

RNA sequencing reads were processed via Cell Ranger (v3.0.3; 10x Genomics) for gene expression quantification, with alignment to the GRCm38 reference genome and Ensembl 92 gene annotation.

Raw RNA-seq reads were processed and aligned to the human reference genome (GRCh38) using HISAT2 (v2.2.1), and gene expression counts were generated with featureCounts (v2.0.6) based on GENCODE annotation.

Reads of ATAC-seq and CUT&Tag are aligned to human reference genome (GRCh38) using Bowtie2. The resulting BAM files are then processed to remove PCR duplicates. For ATAC-seq data, mitochondrial reads are filtered out. Both datasets are normalized and converted to BigWig format using deepTools' bamCoverage with RPKM/CPM normalization.