AI-generated protein biosensors enable quantitative measurements of endogenous RAS activity in live cells

April 30, 2024, by Jason Z Zhang

Measuring the levels of active Ras in living cells poses a formidable challenge due to their low abundance in cells and the dynamic nature of Ras signaling. Ras undergoes rapid cycling between active (GTP-bound) and inactive (GDP-bound) states, demanding techniques that capture these swift transitions. Traditional biochemical assays often involve cell lysis, and lack the ability to perform “in cell biochemistry” where the spatial and temporal nuances of Ras are captured within intact cells. Additionally, the sheer diversity of Ras isoforms further complicates measurements, as each isoform exhibits distinct regulatory mechanisms. Advanced imaging techniques, such as fluorescence resonance energy transfer (FRET), or bioluminescence resonance energy transfer (BRET), offer glimpses into Ras dynamics within live cells^_1,2. Nevertheless, these methods are not without limitations, as they may interfere with normal cellular processes or lack the sensitivity required to detect subtle fluctuations as they typically require Ras overexpression to see meaningful readout changes. Thus, the difficulty in measuring Ras activity in living cells stems from the need to balance sensitivity, temporal precision, and spatial resolution, pushing researchers to continually innovate in pursuit of a comprehensive understanding of this pivotal signaling pathway.

In our recent work³, we show that the de novo designed LOCKR protein switch^4–6and AI-based computational protein design methodologies enable construction of a biosensor that reliably detects Ras activation within the physiologically relevant range of endogenous Ras-GTP levels, which has eluded previous biosensor design efforts. De novo proteins (proteins that have no sequence or structural similarity to native proteins) were used as novel sensor scaffolds, as we can mutate these protein domains without sacrificing specificity or functionality because they are orthogonal to the biological system of interest. Protein design methods and structure predictions (AlphaFold) of these de novo switches in silico predicted which mutations in the de novo proteins would alter the sensitivity of the sensor, which were then experimentally tested. Thus, these computational methods guided the design and optimization of the sensor. With this design methodology, we created two sensors (Figure 1): 

1. Fluorescence-based Ras-LOCKR-S that measures Ras activity (Ras-GTP levels) with the following properties:

a. Real-time measurement
b. Can be used in living cells
c. Single-cell resolution
d. Targetable to different subcellular regions for measuring Ras activity at particular subcellular compartments

2. Ras activity-dependent proximity labeler Ras-LOCKR-PL that profiles the environment around Ras-GTP with the following properties:

a. Profiles active Ras environments within 4 hours
b. Can be used in single-cell resolution if paired with proximity ligation assays⁷ 
c. Targetable to different subcellular regions for measuring Ras activity at particular subcellular compartments

Using these new sensors, we gained insight into the mechanisms of oncogenic signaling, and found that endogenous Ras is active not only at the PM but also at endomembranes (Golgi) and in membrane-less, cytosolic granules when a fusion receptor tyrosine kinase oncoprotein (EML4-Alk) was expressed. Targeting our Ras-LOCKR tools to EML4-Alk-containing granules₈ (seen in lung adenocarcinomas) enabled the discovery of several unanticipated factors inside these granules. We identified SAM68 as an upstream effector of aberrant cytosolic Ras activity inside EML4-Alk granules, thus providing a more complete picture of how Ras can be activated and is capable of driving oncogenic signaling in the absence of membranes. Our finding that co-inhibition of SAM68 and Alk leads to enhanced cancer cell death and inhibition of Ras-LOCKR-S-measured Ras activity suggests that co-treatment with Alk and SAM68 inhibitors could help overcome drug resistance in EML4-Alk-driven lung cancer. 

Overall, these results illustrate the power of our sensor development methodology to generate tools that map the activities, mechanisms, and functions of physiologically relevant molecules such as Ras. These Ras sensors are finally enabling researchers to interrogate Ras activity dynamics in its native cellular context. We envision that the tools described here can be used to answer many pressing questions in the Ras signaling and cancer fields.

Note: plasmids encoding Ras-LOCKR-S and Ras-LOCKR-PL are available on Addgene: https://www.addgene.org/browse/article/28244240.