Reading view

Therapeutic Co-targeting of Oxidative Phosphorylation and Pyrimidine Synthesis Restores Gemcitabine Response in Pancreatic Ductal Adenocarcinoma

By: Wenhui Ouyang · Jie Wang · Jie Chen · Zemin Fang · Ting Wei · Jiazi Qian · Shanying Gui

13 September 2026 at 18:00

Transl Res. 2026 Sep 13:S1931-5244(26)00195-7. doi: 10.1016/j.trsl.2026.09.009. Online ahead of print.

ABSTRACT

Gemcitabine resistance remains a major barrier to effective therapy in pancreatic ductal adenocarcinoma (PDAC), and current combination regimens show potential to overcome this resistance. Here, we identify the mitochondrial ribosomal proteins MRPS22 and MRPL3 as key metabolic gatekeepers that maintain mitochondrial OXPHOS and pyrimidine metabolism, thereby promoting pancreatic cancer cell proliferation and chemoresistance. Across independent cohorts, high MRPS22/MRPL3 expression associates with poorer survival. Depletion of either gene in PDAC curtailed cell proliferation and xenograft growth, which might be due to an impaired mitochondria function, including destabilized respiratory super-complex assembly, diminished ATP production, and increased oxidative stress. Multi-omics profiling revealed a broad reduction of central-carbon intermediates and a pronounced blockade of de novo pyrimidine synthesis at the dihydroorotate dehydrogenase (DHODH) node. MRPS22 depletion hampered nucleotide-pool generation, and exogenous deoxynucleotides partially rescued PDAC cell growth when MRPs were knocked down. Pharmacologic OXPHOS inhibition increased gemcitabine sensitivity, whereas gemcitabine-resistant derivatives exhibited heightened OXPHOS activity and upregulated mitochondrial ribosomal programs. Co-targeting OXPHOS (antimycin A) or DHODH (brequinar) with gemcitabine produced Loewe synergy in vitro and suppressed growth of gemcitabine-resistant xenografts without affecting body weight. Collectively, these findings established MRPS22/MRPL3 as translation-level drivers of PDAC metabolic fitness and nominate OXPHOS/DHODH blockade as a rational combination strategy to overcome gemcitabine resistance.

PMID:42732873 | DOI:10.1016/j.trsl.2026.09.009

On the Learning Dynamics of RLVR at the Edge of Competence

cs.AI, q-bio.NC updates on arXiv.org

By: Yu Huang · Zixin Wen · Yuejie Chi · Yuting Wei · Aarti Singh · Yingbin Liang · Yuxin Chen

17 February 2026 at 13:00

arXiv:2602.14872v1 Announce Type: cross Abstract: Reinforcement learning with verifiable rewards (RLVR) has been a main driver of recent breakthroughs in large reasoning models. Yet it remains a mystery how rewards based solely on final outcomes can help overcome the long-horizon barrier to extended reasoning. To understand this, we develop a theory of the training dynamics of RL for transformers on compositional reasoning tasks. Our theory characterizes how the effectiveness of RLVR is governed by the smoothness of the difficulty spectrum. When data contains abrupt discontinuities in difficulty, learning undergoes grokking-type phase transitions, producing prolonged plateaus before progress recurs. In contrast, a smooth difficulty spectrum leads to a relay effect: persistent gradient signals on easier problems elevate the model's capabilities to the point where harder ones become tractable, resulting in steady and continuous improvement. Our theory explains how RLVR can improve performance at the edge of competence, and suggests that appropriately designed data mixtures can yield scalable gains. As a technical contribution, our analysis develops and adapts tools from Fourier analysis on finite groups to our setting. We validate the predicted mechanisms empirically via synthetic experiments.