* The reported Toff for each variant is the lowest temperature at which fluorescence could be detected above noise. Tmax is the temperature at which fluorescence was maximal. Supplementary Table 2-Genetic constructs used in the study All plasmids were constructed using the pETDuet-1 backbone (EMD Biosciences) with the relevant thermal biosensor elements replacing multiple cloning sites 1 and 2.

Visualizing and modulating molecular and cellular processes occurring deep within living organisms is fundamental to our study of basic biology and disease. Currently, the most sophisticated tools available to dynamically monitor and control cellular events rely on light-responsive proteins, which are difficult to use outside of optically transparent model systems, cultured cells, or surgically accessed regions owing to strong scattering of light by biological tissue. In contrast, ultrasound is a widely used medical imaging and therapeutic modality that enables the observation and perturbation of internal anatomy and physiology but has historically had limited ability to monitor and control specific cellular processes. Recent advances are beginning to address this limitation through the development of biomolecular tools that allow ultrasound to connect directly to cellular functions such as gene expression. Driven by the discovery and engineering of new contrast agents, reporter genes, and bioswitches, the nascent field of biomolecular ultrasound carries a wave of exciting opportunities.

Rapid advances in synthetic biology are driving the development of genetically engineered microbes as therapeutic agents for a multitude of human diseases, including cancer. The immunosuppressive microenvironment of solid tumors, in particular, creates a favorable niche for systemically administered bacteria to engraft and release therapeutic payloads. However, such payloads can be harmful if released outside the tumor in healthy tissues where the bacteria also engraft in smaller numbers. To address this limitation, we engineer therapeutic bacteria to be controlled by focused ultrasound, a form of energy that can be applied noninvasively to specific anatomical sites such as solid tumors. This control is provided by a temperature-actuated genetic state switch that produces lasting therapeutic output in response to briefly applied focused ultrasound hyperthermia. Using a combination of rational design and high-throughput screening we optimize the switching circuits of engineered cells and connect their activity to the release of immune checkpoint inhibitors. In a clinically relevant cancer model, ultrasound-activated therapeutic microbes successfully turn on in situ and induce a marked suppression of tumor growth. This technology provides a critical tool for the spatiotemporal targeting of potent bacterial therapeutics in a variety of biological and clinical scenarios.

Ultrasound allows imaging at a much greater depth than optical methods, but existing genetically encoded acoustic reporters for in vivo cellular imaging have been limited by poor sensitivity, specificity and in vivo expression. Here we describe two acoustic reporter genes (ARGs)—one for use in bacteria and one for use in mammalian cells—identified through a phylogenetic screen of candidate gas vesicle gene clusters from diverse bacteria and archaea that provide stronger ultrasound contrast, produce non-linear signals distinguishable from background tissue and have stable long-term expression. Compared to their first-generation counterparts, these improved bacterial and mammalian ARGs produce 9-fold and 38-fold stronger non-linear contrast, respectively. Using these new ARGs, we non-invasively imaged in situ tumor colonization and gene expression in tumor-homing therapeutic bacteria, tracked the progression of tumor gene expression and growth in a mouse model of breast cancer, and performed gene-expression-guided needle biopsies of a genetically mosaic tumor, demonstrating non-invasive access to dynamic biological processes at centimeter depth.

Recent advances in molecular engineering and synthetic biology have made it possible for biomolecular and cell-based therapies to provide highly specific disease treatment. However, both the ability to spatially target the action of such therapies, and their range of effects on the target tissue remain limited. Here we show that biomolecules and cells can be engineered to deliver potent mechanical effects at specific locations inside the body under the direction of focused ultrasound. This capability is based on gas vesicles, a unique class of air-filled protein nanostructures derived from buoyant photosynthetic microbes. We show that lowfrequency ultrasound can convert these nanoscale biomolecules into micron-scale cavitating bubbles, as demonstrated with acoustic measurements and ultrafast optical microscopy. This allows gas vesicles targeted to cell-surface receptors to serve as remotely detonated cell-killing agents. In addition, it allows cells genetically engineered to express gas vesicles to be triggered with ultrasound to lyse and release therapeutic payloads. We demonstrate these capabilities in vitro, in cellulo, and in vivo. This technology equips biomolecular and cellular therapeutics with unique capabilities for spatiotemporal control and mechanical action.

No abstract

Peptide binding proteins play key roles in biology, and predicting their binding specificity is a long-standing challenge. While considerable protein structural information is available, the most successful current methods use sequence information alone, in part because it has been a challenge to model the subtle structural changes accompanying sequence substitutions. Protein structure prediction networks such as AlphaFold model sequence-structure relationships very accurately, and we reasoned that if it were possible to specifically train such networks on binding data, more generalizable models could be created. We show that placing a classifier on top of the AlphaFold network and fine-tuning the combined network parameters for both classification and structure prediction accuracy leads to a model with strong generalizable performance on a wide range of Class I and Class II peptide-MHC interactions that approaches the overall performance of the state-of-the-art NetMHCpan sequence-based method. The peptide-MHC optimized model shows excellent performance in distinguishing binding and non-binding peptides to SH3 and PDZ domains. This ability to generalize well beyond the training set far exceeds that of sequence only models, and should be particularly powerful for systems where less experimental data is available.

Peptide-binding proteins play key roles in biology, and predicting their binding specificity is a long-standing challenge. While considerable protein structural information is available, the most successful current methods use sequence information alone, in part because it has been a challenge to model the subtle structural changes accompanying sequence substitutions. Protein structure prediction networks such as AlphaFold model sequence-structure relationships very accurately, and we reasoned that if it were possible to specifically train such networks on binding data, more generalizable models could be created. We show that placing a classifier on top of the AlphaFold network and fine-tuning the combined network parameters for both classification and structure prediction accuracy leads to a model with strong generalizable performance on a wide range of Class I and Class II peptide-MHC interactions that approaches the overall performance of the state-of-the-art NetMHCpan sequence-based method. The peptide-MHC optimized model shows excellent performance in distinguishing binding and non-binding peptides to SH3 and PDZ domains. This ability to generalize well beyond the training set far exceeds that of sequence-only models and should be particularly powerful for systems where less experimental data are available.