The inner mitochondrial membrane (IMM) is the site of bulk ATP generation in cells and has a broadly conserved lipid composition enriched in unsaturated phospholipids and cardiolipin (CL). While proteins that shape the IMM and its characteristic cristae membranes (CM) have been defined, specific mechanisms by which mitochondrial lipids dictate its structure and function have yet to be elucidated. Here we combine experimental lipidome dissection with multi-scale modeling to investigate how lipid interactions shape CM morphology and ATP generation. When modulating fatty acid unsaturation in engineered yeast strains, we observed that loss of di-unsaturated phospholipids (PLs) led to a breakpoint in IMM topology and respiratory capacity. We found that PL unsaturation modulates the organization of ATP synthases that shape cristae ridges. Based on molecular modeling of mitochondrial-specific membrane adaptations, we hypothesized that conical lipids like CL buffer against the effects of saturation on the IMM. In cells, we discovered that loss of CL collapses the IMM at intermediate levels of PL saturation, an effect that is independent of ATP synthase oligomerization. To explain this interaction, we employed a continuum modeling approach, finding that lipid and protein-mediated curvatures are predicted to act in concert to form curved membranes in the IMM. The model highlighted a snapthrough instability in cristae tubule formation, which could drive IMM collapse upon small changes in composition. The interaction between CL and di-unsaturated PLs suggests that growth conditions that alter the fatty acid pool, such as oxygen availability, could define CL function. While loss of CL only has a minimal phenotype under standard laboratory conditions, we show that its synthesis is essential under microaerobic conditions that better mimic natural yeast fermentation. Lipid and protein-mediated mechanisms of curvature generation can thus act together to support mitochondrial architecture under changing environments.

The goal when imaging bioprocesses with optical microscopy is to acquire the most spatiotemporal information with the least invasiveness. Deep neural networks have substantially improved optical microscopy, including image super-resolution and restoration, but still have substantial potential for artifacts. In this study, we developed rationalized deep learning (rDL) for structured illumination microscopy and lattice light sheet microscopy (LLSM) by incorporating prior knowledge of illumination patterns and, thereby, rationally guiding the network to denoise raw images. Here we demonstrate that rDL structured illumination microscopy eliminates spectral bias-induced resolution degradation and reduces model uncertainty by five-fold, improving the super-resolution information by more than ten-fold over other computational approaches. Moreover, rDL applied to LLSM enables self-supervised training by using the spatial or temporal continuity of noisy data itself, yielding results similar to those of supervised methods. We demonstrate the utility of rDL by imaging the rapid kinetics of motile cilia, nucleolar protein condensation during light-sensitive mitosis and long-term interactions between membranous and membrane-less organelles.