The UPR is not a switch between protection and death. It is one pathway whose output changes with duration. In the acute phase, all three branches act to restore folding homeostasis: PERK attenuates translation and induces ATF4-dependent chaperones and antioxidant genes, ATF6 induces BiP and other folding components, and IRE1-XBP1s expands ER membrane and secretory machinery while ERAD clears the lumen. If the stress resolves within this window, the cell returns to baseline and the response shuts off.
If the stress does not resolve, the same branches begin to produce pro-apoptotic outputs. The decisive node is CHOP, a transcription factor induced downstream of ATF4 and also by ATF6. CHOP represses BCL-2, induces the pro-apoptotic BH3-only proteins BIM and PUMA, and drives expression of GADD34, a regulatory subunit of protein phosphatase 1 that dephosphorylates eIF2alpha. GADD34 is the negative feedback arm of the PERK branch: it restores translation, but if the folding deficit is still present, restored translation floods the ER with new clients and the deficit worsens. Sustained IRE1 signaling also contributes, through TRAF2-dependent activation of ASK1 and JNK, and through Bax-Bak-dependent calcium release from the ER that loads mitochondria and triggers intrinsic apoptosis.
The mechanistic point is that the pro-apoptotic arm is not a separate pathway. It is the same three transducers running past the point at which their outputs can restore homeostasis. The cell is not choosing to die; it is failing to restore folding capacity while the response continues to consume resources and generate signals that lower the apoptotic threshold. This is why chronic UPR activation damages rather than protects: the protective outputs saturate, the negative feedback arm restores translation into an unresolved deficit, and the pro-apoptotic outputs accumulate.