Biosensors are at the forefront of exciting developments in bioengineering and biomedical research due to their applicability in the diagnosis and treatment of debilitating diseases such as diabetes, cancer, and ALS. They allow us to take advantage of pre-existing mechanisms in nature to detect chemicals in the body that can serve as diagnostic markers for disease. Using high performance computing we modelled the behavior of a glucose sensing biosensor at a high resolution at the atomic level. The biosensor we modelled has the ability to fluoresce in the presence of glucose, and therefore serves as an effective monitor of blood sugar levels - a critical biomarker used for diabetes treatment. This has monumental applications in the treatment of diabetes. Such a biosensor could be potentially coupled to an insulin producing circuit to automatically deliver needed medicine to diabetics without the use of invasive needles and injections. Our modelling approach can be applied to simulate related biosensors, testing many iterations of possible biosensor designs without the need to perform an wet-lab experiment that would produce hazardous waste. Our team has produced an in silico optimization and debugging biosensor template which allows for a majority of testing to be performed prior to entering a wet-lab facility. By reducing the amount of time spent in the wet-lab, our modelling approach provides a safer, more eco-friendly testing environment. It’s as simple as saving on pipette tips - we don’t have to throw away hundreds of plastic pipet tips for one experiment. Biosensors are a rapidly developing treatment and diagnosis tool in biomedical research, and our team has been able to utilize high-performance computer modelling to efficiently test these revolutionary devices.

Our project modeled the interaction between a glucose molecule and a previously engineered and tested glucose biosensor (Fig. 1).

Figure 1 - Our biosensor in its environment. The protein, colored yellow and in the drawing method New Cartoon, is in an aqueous environment, which is shown by the blue molecules. Glucose is bound to the biosensor, and is colored red.

We modeled the biosensor in the presence and absence of its target, glucose (Fig. 2, 3). The biosensor we researched contains a periplasmic binding proteins and protein fluorophores. These proteins and other proteins are made up of long chains containing a specific sequence of amino acids. There are 21 core amino acids within the human body, each containing a carboxyl group, amino group, and an R-group all bonded to a central carbon atom. The R-groups are what determine the identity of the amino acid, and it is made up of different atoms.

Atoms are the smallest unit of matter from which most of the world, including our biosensor, is built. Three properties largely contribute to the behavior of atoms: charge, sterics, and free energy. In a neutral atom, the number of positively-charged protons and negatively-charged electrons are equal, resulting in a zero net-charge. However, the removal or addition of electrons throws this balance off, creating a positively or negatively charged atom. Similar to magnets, positive atoms and negative atoms will attract each other, while atoms of the same charge repel. Sterics involves the spatial arrangement of atoms. When atoms get too close to each other, their electron clouds repel each other, creating an unstable molecule. Finally, Gibbs free energy measures the ability to do work. Atoms try to move towards a lower energy potential, and lower their ability to do work. All three of these factors affect the behavior of atoms, and therefore the properties of proteins.

Figure 2 - Our biosensor without a glucose molecule. The biosensor is colored yellow (representation: NewCartoon).

Figure 3 - Our biosensor bound to a glucose molecule. The biosensor is colored yellow (representation: NewCartoon) and the glucose molecule is colored blue and red (representation: VDW).

Our biosensor requires the presence of glucose to induce a change in the periplasmic binding protein (Fig. 4, 5). This change in shape allows the protein fluorophore to illuminate and provide a visual indication of the presence of glucose. The proteins are illuminated using FRET, or fluorescence resonance energy transfer. FRET allows one protein to transfer energy so another protein, which then fluoresces. This procedure is dependent on distance, therefore the change in shape is necessary to bring proteins close enough together to allow for illumination.

Figure 4 - This is the catalytic domain of our original biosensor without glucose bound to it.

Figure 5 - This is the catalytic domain of our original biosensor with glucose.

Our team used the program Amber to produce the input files for our simulations, specifically the program LEaP, a branch within Amber. The pdb file containing our biosensor was first loaded into the program. We then edited the molecule’s environment by adding 113 ions to the aqueous environment. LEaP was also used to bond a glucose molecule to our biosensor, thus creating two varying environments, one with glucose present, and the other without, and allowed us to run two simulations and compare the results. The edits made to the biosensor and its environment were saved as prmtop and inpcrd files. The prmtop file contained the topology of the model, while the inpcrd file contained the coordinates. These files were then loaded into NAMD for the simulation.

NAMD was an essential part of our process, as it allowed us to load the environments we had created and the parameters we had set in AMBER, and test our biosensor’s functionality in each of the 12 environments we modeled.

We used VMD to provide us with a visual representation of our final project. Most VMD default parameters were used, however when representing building the biosensor we used a different drawing method called CPK. This allowed us to see every bonded atom making up the biosensor. Different representations allow for us to see different aspects of the biosensor. For example, using the representation New Cartoon allows one to see the molecule’s alpha helices and beta sheets, not each individual atom.

VMD was a useful tool in our project because not only does it represent standard molecules read from PDB (Protein Data Bank) files, but with its integration with NAMD, it could represent more complex systems; a large protein or molecule in a specified environment. VMD was not used for applying any parameters on either the system or the molecule itself; it was only used to created a 3D model of the system.

Using these three programs -- AMBER, NAMD, and VMD -- we were able to not only build our system, but model the biosensor in it, adjust the environment to a realistic one modeling the blood stream, and visualize our biosensor.

Figure 6 - Total energy of the model systems (kcal/mol) plotted against time.

Even so, our models with glucose molecules appeared to exhibit little visible conformational change as a result of their interaction with glucose (Fig. 2, 3).

Additionally, for each repetition we found the root mean squared deviation (RMSD) and standard deviation (SD) of atomic movement in the molecule. RMSD measures the distance that the atoms within the biosensor have moved during the simulation from their original placement in the protein. It is measured in angstrom, which is 10-10 m. SD, on the other hand, is a measure of the variability in this amount of movement from atom to atom over time. In each case, we found the average RMSD to be around 1.25 to 1.50 angstrom (with a standard deviation of about 0.45 to 0.80), corroborating our visual findings that the insertion of a glucose molecule into the system had minimal impact conformational impact on the biosensor molecule. Thus, so far we have not observed the expected degree of conformational change in the biosensor in any of the three systems; the small changes in atomic position that we observed would likely not be sufficient to induce fluorescence through Forster resonance energy transfer (FRET).

However, despite this, it is quite possible that slight modifications in the design of the simulations could produce a working biosensor model in the future. For instance, we could run the simulations with more iterations or try other starting configurations of the molecule. Additionally, it may be quite useful to more fully model and run analysis on the system without glucose so that we can detect differences between the glucose-bound and non glucose-bound molecule that may otherwise be difficult to determine. For example, if there is a significantly higher variability in the position of the system without glucose (as measured by the RMSD and SD), the significantly smaller variability once the biosensor is bound to glucose would indicate that there is in fact a significant structural change that occurs. We could also test different (including different sizes of) linker molecules between the various polypeptide chains in the molecule to determine whether they influence the amount of conformational change that the biosensor protein undergoes in response to the presence of glucose.

Table 1 - Root mean square deviation (RMSD) and standard deviation (SD) were used to evaluate the magnitude of change the systems underwent during minimization, heating and equilibration in NAMD. Multiple systems (2-4) were generated describing unique random starting conformations of the molecules. Several repetitions of each system were examined to ensure consistency in the results.