Longitudinal single-cell RNA sequencing of a neuroendocrine transdifferentiation model reveals transcriptional reprogramming in treatment-induced neuroendocrine prostate cancer

Neuroendocrine prostate cancer (NEPC), a lethal outcome of androgen deprivation, arises through neuroendocrine transdifferentiation (NEtD) of prostate adenocarcinoma (PRAD), yet its initiating drivers and their temporal order remain only partially defined. Clinical single-cell studies primarily capture established PRAD or NEPC at single timepoints, so they cannot directly resolve transitional states or the order and direction of transdifferentiation. We generated longitudinal single-cell RNA sequencing of the LTL331 patient-derived xenograft across eight timepoints spanning pre-castration PRAD to relapsed NEPC (51,726 cells) with matched whole-exome sequencing, integrating inferred single-cell copy-number profiles, trajectory inference, regulon and transcription-factor-activity analysis, and chromatin-accessibility data from a clinical metastatic castration-resistant prostate cancer atlas. Lineage states were tested for replication in two clinical single-cell cohorts by embedding-free, label-level comparison. NEtD proceeded by transcriptional reprogramming of a single clone, not clonal outgrowth: PRAD and NEPC shared copy-number architecture and a truncal TMPRSS2 :: ERG rearrangement, with the androgen receptor ( AR ) locus copy-number stable despite loss of AR transcription. Every PRAD-to-NEPC path funneled through one transient, obligate mesenchymal gateway (~0.25% of cells, peaking 12–16 weeks post-castration) that re-engaged the early neural-crest program in developmental order: the neural-plate-border regulator MSX1 peaked with the EMT-delamination program, and border specifier PAX7 persisted into the early neuroendocrine state. From this gateway the lineage bifurcated with strong bias toward a tuft-negative ASCL1− fate (≈0.94 versus ≈0.06; paired p = 4.8 × 10 −23 ; POU2F3 < 1%), diverging from the canonical ASCL1 / PHOX2B route and the ASCL1 -versus-tuft split of other NEPC models. MSX1 is a candidate priming regulator, supported by transcription-factor-activity analysis and AR−/NE+-specific footprints in patient NEPC chromatin. Adenocarcinoma and ASCL1 + neuroendocrine programs replicated in patient cohorts (AUROC 0.82–0.83), and neuroendocrine identity tracked the clinical neuroendocrine-differentiation (odds ratios 134.6 and 31.4). In the LTL331 patient-derived model, NEtD is a directional, stepwise transition rather than a stochastic loss of identity. To our knowledge, this study provides the first time-resolved evidence that prostate NEtD re-engages the early neural-crest program in developmental order before committing to a predominantly ASCL1− fate. The obligate gateway is a potential opportunity for intervention, and its regulators are candidate targets for intercepting NEtD. Not applicable.