Background : Perceval, the ATP:ADP sensor

12 03 2009

Recently, Brain Windows mentioned the report A genetically encoded fluorescent reporter of ATP:ADP ratio. We invited Dr. Jim Berg, the lead author of the study to provide additional background to our readers. Below, Jim provides a fascinating look at rationale behind sensor development.  I really like that they came at this problem with a biological question in mind, something I would recommend before anyone start the development of a genetically encoded indicator.


A pixel-by-pixel ratio of the 490 nm excitation image by the 430 nm excitation image from two cultured HEK293 cells expressing Perceval during control conditions (left) and after 40 min of metabolic inhibition with 5 mM 2-deoxyglucose (right)

A pixel-by-pixel ratio of the 490 nm excitation image by the 430 nm excitation image from two cultured HEK293 cells expressing Perceval during control conditions (left) and after 40 min of metabolic inhibition with 5 mM 2-deoxyglucose (right)


Here’s a little insight into why we decided to develop a fluorescent sensor for cellular energy, and how Perceval evolved. One of the primary research interests of the Yellen lab is the interaction between diet and epilepsy. The ketogenic diet, a high fat, low carbohydrate regimen, is remarkably effective at reducing seizure number. We are investigating how the transition in brain metabolism from glucose to a mixture of glucose and ketone bodies (the metabolically active byproduct of fat metabolism) could lead to a change in neuronal excitability. Previously, we described how acute application of ketone bodies reduces the excitability of substantia nigra neurons, an effect that relies on the opening of ATP-sensitive potassium (KATP) channels. Our hypothesis is that the inhibition of glycolysis by ketone body metabolism leads to a reduction in sub-membrane ATP, resulting in an opening of KATP channels and a decrease in neuronal excitability. This relies on the controversial idea that sub-membrane ATP is provided by glycolysis (possibly by glycolytic enzymes tethered to the membrane), and that the diffusion of ATP is restricted between the submembrane space and bulk cytoplasm, and concept known as “compartmentation”. To fully test this hypothesis, we required an optical sensor for ATP levels.

When planning these experiments, our first thought was to use Luciferase to detect different subcellular ATP levels. For a number of reasons, primarily Luciferase’s weak signal, we decided that a fluorescent sensor for ATP would be much more useful for our application. Our initial approach was a FRET-based design, with CFP and YFP tethered to a bacterial periplasmic binding protein that dimerized upon ATP application. Although these sensors gave some encouraging results, we never got the change in signal that would be required for cellular assays. We then adopted the ‘circularly permuted fluorescent protein (cpFP) approach that had previously produced sensors for calcium (pericam) and hydrogen peroxide (HyPer). We inserted the yellow fluorescent protein cpmVenus into the loop of the bacterial ATP binding protein, GlnK1 (involved in the regulation of ammonia transport) and found that application of small amounts of ATP to the purified sensor led to a substantial change in the excitation spectrum of the sensor. The affinity of the sensor for ATP was extremely high, orders of magnitude more sensitive than would be appropriate for cellular assays. We also found that our sensor responded to ADP application, only with a much smaller fluorescence change. It was then that we determined that these two perceived negatives (too high affinity and ADP binding) would lead to a sensor that reports the ratio of ATP to ADP. In a bit of good fortune, our design for an ATP sensor had in fact given us a sensor for the more valuable ATP:ADP ratio. After tinkering with our sensor by semirandom mutagenesis of the GlnK1 portion of the protein, we expressed the improved sensor, which we named Perceval (for permuted reporter of cellular energy value) into cultured cells and monitored a change in fluorescence with metabolic inhibition.

Right now, we are excited to use Perceval to investigate neuronal/glial metabolism in mammals. We may target subcellular ATP by either tethering Perceval to a membrane protein, or by using TIRF microscopy. In addition, we are continuing to design improved versions of Perceval, as well as sensors for other metabolic intermediates. We also hope that these sensors will be useful in applications beyond neuronal metabolism, from studies of cancer cells to bacterial metabolism.

ATP and FRET – A cautionary note, confirmed?

6 03 2007

In the February issue of Nature Biotechnology, a correspondence piece noted some very surprising findings regarding the sensitivity of genetically-encoded fluorescence resonance energy transfer (FRET) indicators to adenosine triphosphate.

During the development of CFP-YFP based FRET reporters for adenosine nucleotides, Willemse et. al. discovered that all of their FRET constructs, including putatively non-responsive controls had a significant response to millimolar levels of ATP. Increasing levels of ATP appeared to quench the acceptor chromophore. The effect appeared specific to ATP, as 10mM ADP had no effect and 10mM GTP had a very small one. They also tested whether there was a direct effect on either the CFP or YFP and found none. Only constructs that underwent FRET showed ATP responsiveness. The authors suggested that the effect was due to “a direct quenching of the energy-transfer step coupled to energy-induced charge displacement in the phosphate groups.”

One must be very cautious in interpreting anomalous results in FRET imaging. Genetically-encoded FRET reporters have been used for over 10 years, and their specificity has been validated by dozens of high-quality imaging labs. Furthermore, a large variety of potential confounds, including pH shifts, osmotic effects, non-linear photo-bleaching, protein precipitation and secondary effectors, must be carefully accounted for when doing FRET experiments. However, fluorescent imaging in cell biology is a rapidly developing field, and the discovery of surprising photophysics and photochemical effects is not uncommon. Therefore, I took a rather skeptical, but open-minded approach to the published results and interpretation. I needed to see for myself.

Effect of ATP on GluSnFR emission

I performed the following experiment. 5ul of 100uM pure, soluble GluSnFR (a custom genetically-encoded FRET reporter for glutamate) was diluted into 3mL of HBSS containing 0, 3.3 or 10mM ATP. pH was balanced to exactly 7.35 in each solution. The emission spectra of the solutions were then measured on a SPEX spectrophotometer with excitation set to 420nm, selective to CFP excitation. To my surprise, there was a strong effect of increasing ATP concentration on the emission spectra of the FRET construct. YFP/CFP changed from 2.12 @ 0mM ATP to 1.87 @ 3.3mM to 1.51 @10mM. There was a suprisingly small increase in the donor emission @ 474nm, perhaps reduced due to slight variations in GluSnFR concentration. Although I did not perform several possible control experiments, this data does corroborate the findings of the authors.

The authors interpretation of a quenching of the energy-transfer step seems to be a novel mechanism not supported by my understanding of FRET theory.  If the mechanism were siphoning off FRET to a quenched acceptor (ATP), then there should be a direct effect on CFP upon ATP addition.  In the supplemental figures, the authors show there is not. My expectation for mechanism would be a direct effect of ATP on the conformation of the sensor that does not depend on the linker. The presence of CFP and YFP is common between all the sensors tested.  Perhaps ATP reduces the tendency of CFP and YFP to dimerize, hence the time-averaged FRET ratios are reduced. Does this effect persist after subsitution of other variants of CFP and YFP? Certainly, more technical work needs to be done to nail down the mechanism of the effect.

Why is this finding important to the field of neuroscience? Average ATP concentrations in a neuron fluctuate between approximately 0.8 and 1.4mM, although some neurons may reach as high as 5mM. Slow changes of these levels are not likely to have a significant impact on the readout of intracellular FRET reporters in neurons. However, the authors make the point that subcellular compartments of neurons may have ATP levels as high as 20mM which undergo more dynamic fluctuations. If this is the case, and if the generality of the quenching effect holds up to future scrutiny, then FRET reporters targeted to these compartments may have their readout significantly confounded by changes in ATP levels. Development of ATP-specific sensors genetically targeted to these compartments of potentially high ATP flux will provide essential insight to the impact these findings will have on the general field of cellular FRET imaging.