Computational Chemical Imaging Reveals Dazzling Molecular Details of Living Cells and Organisms
Computing and machine learning are enabling imaging capabilities once considered “impossible”
By Kat J. McAlpine
For more than two decades, Ji-Xin Cheng has been determined to develop a new way for peering deeper into biology – one that doesn’t rely on dyes to illuminate the insides of cells and tissues.
To visualize what’s happening within cells on the molecular level, scientists have long relied on a technique called labeling. Typically, that process entails tailoring fluorescing dyes – proteins that glow when exposed to the light of a microscope – to attach to certain molecules, cells, or proteins of interest. A historically popular label has been a dye known as green fluorescent protein or GFP, which blazes green when under blue and ultraviolet light. It was first discovered in a bioluminescent jellyfish before scientists engineered it into a tool for “seeing” details within cells and tissues.
Cheng, a Boston University College of Engineering (ENG) professor and Photonics Center faculty member, says that although GFP and other labeling dyes have vastly contributed to biological research, there’s a limit to how much detail they can reveal about the inner workings of life. That’s because the size of these fluorescent labels – on a molecular scale – are usually much larger than the structures they’re designed to illuminate. They are so large that while they can adequately identify the presence of specific biological signatures, they obscure the exact location, shape, and other intricate details of their targets.
Instead, Cheng’s been interested in seeing the chemical bonds of biological samples without the need for fluorescent dyes.
“If we could use the intrinsic chemical bonds within every molecule to visualize biology, we could achieve better molecular information without relying on labeling,” Cheng says. “But this was a very difficult task because the signal we can capture from chemical bonds is many orders of magnitude weaker than the glow of fluorescent dyes. Years ago, people thought this would be impossible to do – but it’s been the great dream at the core of my career.”
Now, Cheng’s dream is becoming a reality.
Creating contrast between chemicals
In 2016, Cheng’s team invented a new method using mid-infrared light (which travels at wavelengths between microwaves and those that are visible to eye) to precisely heat up certain chemical bonds within cells and small organisms. As those bonds heat up under light, the change in temperature creates a visual contrast between different types of chemicals. Using photothermal imaging, Cheng and his team were able to see 3D molecular maps of cells and whole C. elegans worms without the use of traditional labels. Their approach, first published in Science Advances in 2016, could even capture super-fine details within a single virus, as the team recently reported in Nature Communications in December 2023.
To capture a mid-infrared photothermal image, however, could take up to 20 minutes using a laser to scan a sample point by point. “It isn’t fast enough to capture 3D images of living cells in real time,” Cheng says. “That’s where computation has come into the picture – we are pushing the limits of chemical imaging.”
He has teamed up with Lei Tian – an ENG assistant professor, Photonics Center faculty member, and director of the Computational Imaging Systems Lab (CISL) at BU – and together the duo and their research groups are developing a new imaging technique called bond-selective intensity diffraction tomography (BS-IDT). (See adjacent piece, “When Cutting-Edge Microscopes Meet Deep Learning Algorithms”, featuring Tian’s research).
Using one infrared laser and sixteen visible-spectrum lasers that shine light throughout a sample from different angles, their method stimulates chemicals bonds within the molecular structure of cells and organisms, changing their temperature and the way those bonds refract light. Once those signatures are detected, computational algorithms combine that information from all 16 angles to pinpoint the exact 3D location of chemical bonds.
The result? A highly detailed 3D digital reconstruction of the sample’s molecular makeup. In 2022, they published the first BS-IDT study in Nature Communications revealing the distribution of cancer-associated lipids – fatty compounds that are essential to all organic life – inside C. elegans worms.
“Understanding where lipids are and how they change over time can tell us new information about the aging and development of an organism,” Cheng says. And it could demystify the molecular interactions that spark and then spread cancer and other diseases.
Getting closer to real-time imaging
Together, Cheng and Tian are racing to make their BS-IDT method faster – bringing it up to real-time “video rate”. And their work has attracted significant support. With more than $1.3 million in funding from the Chan Zuckerberg Initiative, they are creating instrumentation and algorithms that will allow them to capture (and then computationally reconstruct) label-free molecular maps of a living sample at a rate of 30 captures per second. At that speed, once reconstructed digitally, it would be like watching a high-resolution movie of an organism’s molecular activity fluctuating right before your eyes.
Beyond cancer, Cheng is eager to use BS-IDT to study tau proteins responsible for Alzheimer’s disease. The accumulation and aggregation of tau proteins within brain tissue is closely correlated with onset and progression of Alzheimer’s. Like other proteins, as they aggregate, tau’s shape and function changes.
With conventional imaging methods, this change in molecular structure is not possible to see. Cheng and Tian’s BS-IDT method, utilizing chemical bonds which are 100 times smaller than traditional fluorescing dye molecules, can discern the precise structure of proteins like tau. They hope that BS-IDT could be used to detect and monitor how problematic proteins begin aggregating at the onset of Alzheimer’s and other diseases. Such insights could be game-changing for therapeutic research and development.
“This could transform drug screening and development,” Cheng says.
Lighting up even the scarcest molecules
Cheng’s lab is also developing another computational chemical imaging technique; one that is particularly powerful at visualizing molecules of interest even at extremely low concentrations.
“Many molecules found naturally in cells or introduced through drugs occur at very low concentrations throughout cells and tissues,” Cheng says. “With our new method, we can map out these low-concentration molecules – even seeing specific amino acids, for example, that reveal the metabolism and traffic of molecules inside cells.”
The imaging technique probes chemical bond signatures with Raman scattering, the idea that light scatters off different types of molecules in district ways. These distinct light-scattering signatures of each chemical bond can then be detected using microscopes. Since 2000, Cheng has led the field in using a technique called coherent Raman scattering microscopy to enable high-speed bond-selective imaging of living cells.
“Yet because Raman scattering is a very feeble effect, the sensitivity of a coherent Raman scattering microscope [has not been] sufficient to see low concentration molecules inside a cell,” Cheng says.
At high speeds, Raman scattering microscopy can be quite “noisy,” Cheng explains. “How do we know what signals are random and which are not? We’re using machine learning to help us differentiate true signal from noise, eventually achieving both high speeds and a high signal-to-noise ratio.”
This approach, which is supported with a R01 grant from the National Institutes of Health, relies on a machine learning algorithm that’s trained using hundreds of high-speed coherent Raman scattering microscopy images. Cheng’s team reported their technique in Nature Communications in 2021.
From seeing to probing
Cheng says the principles behind computational chemical imaging can also be harnessed for a different application – rather than just seeing the precise location of chemical bonds, laser excitation can be used to activate specific chemicals.
Recently, they’ve applied this to study protein phosphorylation, a chemical process where phosphate attaches to proteins inside cells. “This can lead to cancer and many other diseases,” Cheng says. “To see it in real time is very difficult – we’ve designed a probe to remove phosphates, triggering a chameleon-like shift in light diffraction from chemical bonds that we can then visualize microscopically.”
Using chemical bonds as probes, Cheng says, have key advantages over traditional molecular probes. First, they’re extremely small and precise. And, as the targeted chemical bonds within a cell shift in response to probing from extrinsic chemicals, their transition is very sharp, creating a strong signal that can be clearly detected and interpreted.
All these promising applications of computational chemical imaging are possible, Cheng says, thanks to collaboration and cross-disciplinary perspectives. His lab is situated on the same floor in the Photonics Center as Tian’s group. “We see each other every day and that helps us to always be problem solving and moving ahead,” he says.