to BioTechniques free email alert service to receive content updates.
Fluorescent Protein Biosensors Aren’t Bulletproof (Yet) | Microscopy Feature

Andrew S. Wiecek

There are a lot of fluorescent protein biosensors published, but most aren’t very useful. Andrew S. Wiecek investigates how protein engineers are trying to optimize these protein biosensors for the broader research community.

Bookmark and Share

At his Harvard Medical School lab, Gary Yellen is uncovering the metabolic mechanisms of a dietary therapy for epilepsy. The idea came from conversations with his wife, an epileptologist at Massachusetts General Hospital. One of the therapies she employs is the ketogenic diet. It can be as effective as most anticonvulsant drugs, but the therapeutic mechanisms remain unknown.

in an effort to better understand how to optimize the genetically encoded calcium indicator GCaMP2, Looger’s lab determined the structure of the sensor using X-ray crystallography. Source: Journal of Biological Chemistry

To understand how the diet works, Yellen wants to look at how it affects cellular conversion of glucose into pyruvate, a process that releases energy as adenosine triphosphate (ATP) and reduced nicotinamide adenine dinucleotide (NADH).

So, Yellen looked at published protein sensors for ATP, but most were based on luciferase, a luminescent enzyme that is incompatible with confocal microscopy. Also, it has a fairly dim signal and is easily affected by many environmental factors. So for Yellen, who wanted to study ATP in a cellular environment, a luciferase-based sensor was not very useful.

“There are a lot of published sensors, but only a fraction of published sensors are actually used,” says Yellen. “There are many challenges in biosensor design, and how well those challenges are met – matching their properties to the actual measurements that need to be made in cells and making sure they are well-behaved – determines how usable they are down the road.”

While such sensors promise to provide new information at the cellular level in living tissue, designing protein sensors that can be used by the broad research community has been rather challenging. So now, the problem is not how to develop these sensors but rather how to optimize them to provide robust signals for all researchers.

No Plug-and-Play

Because metabolic pathways were defined using biochemistry methods that average multiple responses from homogeneous cell populations in culture, these charts cannot accurately predict how an intact biochemical pathway in a single cell will actually respond to different types of regulation.

“When you put a dozen reactions together, it won’t predict the overall behavior. It’s a crapshoot unless you actually measure the ups and downs of single cells in intact systems,” says Yellen.

And that’s the promise of protein biosensors, to allow the imaging of an individual cells’ behavior in a natural setting, surrounded by other types of cells in vivo. This is particularly important in tissues consisting of multiple cell types, like the brain. Specifically for Yellen, an appropriate protein biosensor could help fill in the mechanistic gaps in their hypothesis about the ketogenic diet by mapping ATP and NADH with single-cell resolution.

At the Howard Hughes Medical Institute’s Janelia Farm Research Campus, group leader Loren Looger is developing protein biosensors to detect neurotransmitters for in vivo imaging. His lab has previously worked on the genetically encoded calcium indicator GCaMP2.

“Where I think sensors are going to change things is in vivo imaging of calcium voltage and neurotransmitters, where signals take place very fast,” says Looger. For example, the neurotransmitter glutamate is only in high concentration in the synapse for 100 microseconds. At these time scales, imaging is the only approach quick enough to measure it in real-time.

But despite all of this promise, the hundreds of published protein sensors have not provided much biological insight to date because not many can get them to work properly. “If you read a review that summarizes all the papers, you could come up with about 250 sensors, but the number that are actually useful, is probably more like five,” says Looger.

It’s fairly straightforward to create a fluorescent protein sensor. First, find a bacterial transcription factor or other protein that responds to the target analyte. Then, fuse that protein to a genetically encoded fluorescent protein in such a way that its fluorescence changes when the transcription factor binds to the analyte.

But because these protein sensors are dynamic, there’s plenty that can go wrong. In the first place, the bacterial proteins could bind to other molecules besides the target analyte or not have the right affinity for the analyte in the appropriate range for measurement. Then, the signal change might not be drastic enough for detection. And then there are the environmental sensitivities that can affect the signal as well.

By and large, the usability problem lies not with the creation but with the optimization of these protein sensors. Most protein sensor developers are less interested in making robust tools for the research community rather than answering their own experimental questions.

“Optimization is a lot of work. It’s not surprising that the vast majority of people who develop sensors make them so that they are good enough for the experiments that they themselves want to do,” says protein sensor developer Amy Palmer, assistant professor at the University of Colorado.

As a result, these protein biosensors are not plug-and-play tools. If you want to use these them, you first have to become familiar with them. That means doing the less-than-glorious work of testing variables and setting up controls.

“What biosensor developers need to do is make the sensors a little more bulletproof, so people who don’t spend their days thinking deeply about them,” says Yellen. “We have to provide appropriate warnings or clean them up enough so that people can use them without being fooled.”

Screw the Peptides

At Janelia Farms, Looger is using a systematic approach to raising the bar for protein sensors, hoping to use the resources at the HHMI to make useful tools to enable others to reveal new biological insights. Source: HHMI

At Janelia Farms, Looger is using a systematic approach to raising the bar for protein sensors, hoping to use the resources at the HHMI to make useful tools to enable others to reveal new biological insights.

In 2009, in an effort to better understand how to optimize the genetically encoded calcium indicator GCaMP2, Looger’s lab determined the structure of the sensor using X-ray crystallography. His team found many subtle surprises, particularly with regards to local rearrangements around the fusion area. Using these insights, Looger’s team improved the signal-to-noise ratio of the GCaMP2 through protein engineering and published their results in the Journal of Biological Chemistry (1).

Since then, Looger’s team has done the same with several other protein sensors, and a pattern has emerged. Now, they can predict similar rearrangements and functionality in other protein sensors that they targeted for optimization. “We think we understand the sort of structural rearrangements and know what to screen for,” says Looger.

In addition, Looger has also focused on reducing the environmental sensitivities of the sensors. Many protein sensor developers only test their sensors in vitro or in human embryonic kidney (HEK) cells, systems where it’s relatively easy to make a sensor look good. So, when another group attempts to use the sensor in their system of interest, the results are unpredictable. Often, the sensor is too dim or photobleaches quickly.

Because Looger’s lab itself does not have the expertise to do in vivo imaging, he works with other labs that can validate his protein sensors in a variety of models, including worm, fly, fish, plants, retina, and other relevant systems. By publishing the results of these experiments, end-users can be more assured that the sensor will work in their system of choice, avoiding a waste of time and money.

Additionally, Looger’s lab is involved with the HHMI Genetically Encoded Calcium Indicator (GECI) Project. The goal is to engineer fluorescent sensors that can image neuron activity. But instead of just testing these calcium sensors in vitro and then in HEK cells, the project joins together four lab heads at Janelia Farms to test these sensors in actual neurons.

“There is just no proxy in knowing how it’s going to do in a neuron,” says Looger. “Screw the HEK cells and screw some stupid little in vitro model. We’re just going to have to pound through rat brains and just test 10,000 sensors in cultured neurons.”

All things considered, it’s a huge undertaking. The job involves producing rat litters, dissociating the neurons from their brains, getting the sensors into a lentivirus delivery system, growing the cells, and using a powerful microscope to image all of them. In the end, it produces extraordinarily high-quality of data for thousands of sensor variants in the exact cell type of interest.

“It makes you wonder how we limped along before with just testing things in Escherichia coli and then thinking we had a reasonably good understanding of our system,” says Looger.

Optimization Bottleneck

At the University of Colorado, Palmer is also part of a new class of researchers hoping to democratize protein biosensors. Source: University of Colorado/Casey Cass

At the University of Colorado, Palmer is also part of this new class of researchers hoping to democratize protein biosensors. In order for these sensors to be more broadly applicable, she believes they need to have responses that are robust enough to be used in non-optimized systems.

“People that are experts in sensor development can get really beautiful data and can make really interesting biological discoveries with sensors with small responses,” says Palmer. “They are very trained at doing that and know what variables to control.”

When a group engineers a protein biosensor, they tailor it to their system, which usually includes a microscope system with very sensitive camera and specific optics. Outside of that specialized set-up, the magnitude of signal change and the signal-to-noise ratio is diminished. Instead, Palmer believes that these sensors need to be developed with the broader community in mind, and that means making them work with common microscopes and experimental set-ups.

“The biggest obstacle is how do we optimize these sensors. It’s very much an ad hoc, trial and error process,” says Palmer. “There’s a real bottleneck in optimizing the sensors with very few systemic ways to screen libraries in a high-throughput way.”

To fill this technical gap, Palmer’s lab is designing microfluidic systems that screen the responses and kinetics of large numbers of protein sensors. In a paper published in the Journal of the American Chemical Society last month (2), Palmer and colleagues described a microfluidic flow cytometer that characterizes the response of metal-ion sensors in mammalian cells at a rate of 15 cells per second. Using the instrument, the team has already examined the response amplitude and kinetics of zinc and calcium sensors in cells. This data will be useful in improving the design and development of future generations of sensors.

Getting Lucky

Back at Harvard, Yellen and his Harvard colleagues decided to develop their own ATP sensor from scratch, combining a variant of green fluorescent protein with the bacterial regulatory protein GlnK1 from Methanococcus jannaschii. In a 2009 paper published in Nature Methods (3), the team published a description of their biosensor – called Perceval – which does a good job of sensing ATP:ADP ratio, but also has a sensitivity to pH, a common problem for protein sensors.

So, in a paper published in the Journal of the American Chemical Society last year (4), Yellen’s lab described a secondary sensor based on red fluorescent protein that would simultaneously measure pH. This secondary sensor would allow Yellen’s team to correct for pH dependency with their ATP sensor.

Since then, Yellen’s lab has learned some new tricks for overcoming pH sensitivity. In a paper published in Cell Metabolism (5), Yellen’s lab described a sensor for NADH, another piece to their bigger puzzle of the ketogenic diet, but this time they used protein engineering to overcome pH sensitivity.

They found that pH sensitivity occurred with the fluorescent protein, as well as with the binding affinity. So, Yellen used a fluorescent protein with a different pKa, so that pH wouldn’t affect the signal within the range of detection. To solve the binding affinity issue, Yellen’s team got lucky. They found that one amino acid accounted for 99% of that pH sensitivity. They changed it, and, as a result, their NADH sensor became unusually resistant to pH fluctuation.

And now, Yellen’s lab can measure changes in ATP and NADH levels in neurons that occur with changes in fuel source. And they’ve already seen that fuel sources do indeed affect the ATP levels, and, in turn, the electrical activity of these cells, supporting their ideas of how the ketogenic diet works.


  1. Akerboom, J., J. V. D. Rivera, M. R. M. Guilbe, E. A. C. Malavé, H. H. Hernandez, L. Tian, S. A. Hires, J. S. Marvin, L. L. Looger, and E. R. Schreiter. 2009. Crystal structures of the GCaMP calcium sensor reveal the mechanism of fluorescence signal change and aid rational design. The Journal of Biological Chemistry 284(10):6455-6464.
  2. Ma, H., E. A. Gibson, P. J. Dittmer, R. Jimenez, and A. E. Palmer. 2012. High-Throughput examination of fluorescence resonance energy Transfer-Detected Metal-Ion response in mammalian cells. Journal of the American Chemical Society (January).
  3. Berg, J., Y. P. Hung, and G. Yellen. 2009. A genetically encoded fluorescent reporter of ATP:ADP ratio. Nature Methods 6(2):161-166.
  4. Tantama, M., Y. P. Hung, and G. Yellen. 2011. Imaging intracellular pH in live cells with a genetically encoded red fluorescent protein sensor. Journal of the American Chemical Society 133(26):10034-10037.
  5. Hung, Y. P., J. G. Albeck, M. Tantama, and G. Yellen. 2011. Imaging cytosolic NADH-NAD(+) redox state with a genetically encoded fluorescent biosensor. Cell Metabolism 14(4):545-554.