Raman2RNA  – Advancing Single-Cell Analysis with Nondestructive Imaging

Traditional single-cell genomics and microscopy techniques, while providing detailed insights into gene programs and cell states, are inherently destructive. Techniques like scRNA-seq involve tissue fixation or cell lysis, precluding continuous tracking of molecular dynamics in live cells. Computational methods, such as pseudo-time algorithms and RNA velocity-based methods, attempt to infer dynamics from molecular snapshots but are limited by assumptions that are challenging to validate experimentally. Raman microscopy offers a nondestructive approach by reporting on the vibrational energy levels of molecules such as nucleic acids, proteins, and metabolites. However, until the development of R2R, Raman microscopy lacked the genetic and molecular interpretability provided by methods like scRNA-seq.

A new study published in the Journal Nature Biotechnology, led by Dr. Aviv Regev and colleagues at Genentech and the Broad Institute of MIT and Harvard, represents a significant breakthrough in the field of cellular biology and genomics. Their work introduces Raman2RNA (R2R), a novel method that infers single-cell RNA expression profiles in live cells through label-free hyperspectral Raman microscopy images. This development is particularly noteworthy because it circumvents the destructive nature of conventional single-cell RNA sequencing (scRNA-seq) and other profiling assays, thus preserving the viability and functionality of the cells under study.

R2R combines the strengths of Raman microscopy with scRNA-seq. It uses spatially resolved hyperspectral Raman images to predict single-cell RNA sequencing profiles in live cells. This method employs either anchor-based integration with single molecule fluorescence in situ hybridization (smFISH) or anchor-free generation with adversarial autoencoders. The success of R2R in inferring expression profiles nondestructively from Raman images far exceeds that of inference from brightfield images, as evidenced by its high cosine similarity scores.

A major advantage of R2R is its ability to track live cells while simultaneously analyzing their genomic profiles. This capability was demonstrated in the reprogramming of mouse fibroblasts into induced pluripotent stem cells and in tracking the differentiation trajectories of mouse embryonic stem cells. Such live-cell tracking with genomic analysis opens new avenues for studying cellular dynamics in real-time, which is crucial in understanding processes like differentiation, reprogramming, and response to stimuli.

R2R addresses the key limitations of current genomic profiling methods, namely the destructive nature of sample processing and the inability to monitor dynamic genomic changes in live cells. By providing a nondestructive means to assess cellular states and functions, R2R enhances the study of dynamic processes like cell growth, stress responses, and pathological conditions.

The development of a high-throughput multimodal Raman microscope as part of this study further enhances the applicability of R2R. This technology allows for automated acquisition of Raman spectra, brightfield, and fluorescent images, facilitating the simultaneous analysis of multiple cells and cellular states. The subcellular spatial resolution of this approach (<500 nm) is particularly significant, offering detailed insights into the molecular composition of cells. The predictive accuracy of R2R has been validated extensively. In the study, the method accurately distinguished cell types and predicted the binary expression of marker genes in a mixture of mouse fibroblasts and induced pluripotent stem cells. Furthermore, the ability of R2R to predict entire expression profiles at single-cell resolution was demonstrated, with significant correlations between predicted and measured expression profiles. The integration of Raman images with scRNA-seq profiles using Tangram and adversarial autoencoders represents a significant advancement in data integration techniques. These methods facilitate the translation of Raman profiles into single-cell expression profiles, thereby linking physical and genomic data at the cellular level.

R2R lays a foundation for the exploration of live genomic dynamics. Its ability to provide real-time insights into the molecular changes occurring within cells could revolutionize our understanding of various biological processes and disease mechanisms. Future developments could focus on increasing the throughput of Raman microscopy and enhancing the resolution further. Emerging vibrational spectroscopy techniques could play a role here, potentially increasing throughput to match that of massively parallel single-cell genomics. The use of more sophisticated machine learning models, such as contrastive learning or other domain translation architectures, could improve the stability and accuracy of the predictions made by R2R. Integrating R2R with other single-cell multi-omics approaches could enable a more comprehensive understanding of cellular states. This might involve projecting modalities like scATAC-seq onto Raman spectra, providing a more holistic view of the cell. The application of R2R in both clinical diagnostics and basic research could be transformative. In clinical diagnostics, R2R could be used for noninvasive monitoring of disease progression or response to treatment. In research, it could facilitate the study of complex biological processes in a variety of model systems.