Stress-induced phosphorylation of eIF2α inhibits global protein synthesis to conserve energy for repair of stress-induced damage. Stress-induced translational arrest is observed in cells expressing a nonphosphorylatable eIF2α mutant (S51A), which indicates the existence of an alternative pathway of translational control. In this paper, we show that arsenite, heat shock, or ultraviolet irradiation promotes transfer RNA (tRNA) cleavage and accumulation of tRNA-derived, stress-induced small RNAs (tiRNAs). We show that angiogenin, a secreted ribonuclease, is required for stress-induced production of tiRNAs. Knockdown of angiogenin, but not related ribonucleases, inhibits arsenite-induced tiRNA production and translational arrest. In contrast, knockdown of the angiogenin inhibitor RNH1 enhances tiRNA production and promotes arsenite-induced translational arrest. Moreover, recombinant angiogenin, but not RNase 4 or RNase A, induces tiRNA production and inhibits protein synthesis in the absence of exogenous stress. Finally, transfection of angiogenin-induced tiRNAs promotes phospho-eIF2α–independent translational arrest. Our results introduce angiogenin and tiRNAs as components of a phospho-eIF2α–independent stress response program.

Liquid-liquid phase separation (LLPS) mediates formation of membraneless condensates such as those associated with RNA processing, but the rules that dictate their assembly, substructure, and coexistence with other liquid-like compartments remain elusive. Here, we address the biophysical mechanism of this multiphase organization using quantitative reconstitution of cytoplasmic stress granules (SGs) with attached P-bodies in human cells. Protein-interaction networks can be viewed as interconnected complexes (nodes) of RNA-binding domains (RBDs), whose integrated RNA-binding capacity determines whether LLPS occurs upon RNA influx. Surprisingly, both RBD-RNA specificity and disordered segments of key proteins are non-essential, but modulate multiphase condensation. Instead, stoichiometry-dependent competition between protein networks for connecting nodes determines SG and P-body composition and miscibility, while competitive binding of unconnected proteins disengages networks and prevents LLPS. Inspired by patchy colloid theory, we propose a general framework by which competing networks give rise to compositionally specific and tunable condensates, while relative linkage between nodes underlies multiphase organization.

Mammalian stress granules (SGs) contain stalled translation preinitiation complexes that are assembled into discrete granules by specific RNA-binding proteins such as G3BP. We now show that cells lacking both G3BP1 and G3BP2 cannot form SGs in response to eukaryotic initiation factor 2α phosphorylation or eIF4A inhibition, but are still SG-competent when challenged with severe heat or osmotic stress. Rescue experiments using G3BP1 mutants show that G3BP1-F33W, a mutant unable to bind G3BP partner proteins Caprin1 or USP10, rescues SG formation. Caprin1/USP10 binding to G3BP is mutually exclusive: Caprin binding promotes, but USP10 binding inhibits, SG formation. G3BP interacts with 40S ribosomal subunits through its RGG motif, which is also required for G3BP-mediated SG formation. We propose that G3BP mediates the condensation of SGs by shifting between two different states that are controlled by binding to Caprin1 or USP10.