A cartoon summary of my research at Merck in 2022, which involved using Surface Plasmon Resonance (SPR) and Differential Scanning Fluorimetry (DSF), two commonly used biophysical assays.
The research group which I was part of during my 2022 internship. Photo taken in occasion of National Intern Day 2022.
My first wet-lab research experience was during my sophomore-year internship (Summer 2022) with the Computational and Structural Chemistry group at Merck, at their Kenilworth, NJ site. My project was to use the biophysical assays known as surface plasmon resonance (SPR) and differential scanning fluorimetry (DSF) to profile around 120 different small molecule inhibitors of a particular enzyme, and to differentiate between covalent reversible binders (those that formed a covalent bond with the enzyme at its active site, and the bond could be broken) and covalent irreversible binders (those that formed a permanent covalent bond at said location). The reversible binders are desired because they exhibit the sigmoidal dose-dependent response (in this case, inhibition) that is expected of an effective drug, and also because an irreversible binder tends to be cytotoxic (i.e., toxic to the human cell). But before I delve deeper into the project, I'd like to share why the entire internship experience has been so pivotal to my career.
The core project itself, to me, felt like an application of everything I had learned in school so far, from the simple "C1 * V1 = C2 * V2" equation in General Chemistry, to advanced concepts like protein binding and Kd from Biochemistry, and even non-technical skills such as time management and networking. All of this served as a foundation for more advanced skills such as assay design, data analysis, and so on. Speaking of networking, we interns got to network with many people from different areas of the company. I allocated time blocks to schedule meetings with scientists in departments related to the one I was in. I realized you can network even with people outside the research realm, simply by sitting down with them at lunch! The casualness with which people mentioned technical things like peptides and antibodies in-between their life discussions, and the ease with which people understood them, was awe-inspiring. We took part in a weekly seminar about the different stages of drug discovery, and had numerous social events dispersed all throughout the summer. There is even a mentoring program where I was paired with a mentor from a completely different department! We had discussions around resume building, soft skills, future career pathways, and much more.
My heartfelt gratitude goes out to all the people who made this internship such a lovely experience. In fact, I loved the company culture and the environment here so much, I chose to intern at Merck again! (More on that in my upcoming post for Merck 2023, coming soon! 😄)
And now, without further ado, let's get into the essence of the project!
Surface plasmon resonance (SPR) is used to study the kinetics (rates) of binding and dissociation between two molecules that bind. One of these two molecules, the "ligand" (in this case, our protein) is fixed onto a biosensor, which records the change in mass during binding (the "response"). The other molecule, the "analyte" (in this case, the small molecule) is introduced to the biosensor environment in varying concentrations. In the diagram on the top right, as more analyte binds to the ligand, the response curve rises and levels off as the system reaches equilibrium (i.e., the rates of binding and dissociation become equal). The analyte solution flow is then replaced with a "blank" buffer, causing the analyte to dissociate, and the curve to go down. The slopes of the association and dissociation curves, and the concentrations used, can be used to calculate the rates of association and dissociation, and the binding affinity (Kd, see the Biochemistry page).
In my SPR runs (see first column in the cartoon schematic), some of the 120-something compounds tested showed dissociation curves that decreased as expected. Those, I understood, were clearly reversible binders. Others, however, showed curves that didn't seem to decrease once they levelled off. I wasn't sure whether they were actually irreversible, or whether they were reversible but came off very, very slowly. That was where DSF came in.
Figure credit: Gao et al. 2020.
Figure credit: Cytiva.
Differential scanning fluorimetry (DSF) is a "thermal shift" assay where the protein, small molecule and a fluorescent dye are heated together over a specified temperature range. As the temperature increases, the protein starts to denature (lose its shape) and unfold. The dye, which is hydrophobic (fat-soluble, "greasy") starts binding to the hydrophobic inner regions of the protein that become exposed, and causes an increase in signal. The dye, in essence, tracks how much of the protein has unfolded. The Tm, or "melting temperature" is defined as the temperature at which 50% of the proteins have been denatured. In the raw response-vs-temperature curve, the Tm lies in the inflection point (point of greatest slope), but as the figure on the bottom left shows, when the first-derivative of the curve is taken, the inflection point becomes a peak, or valley, which makes getting the Tm much more intuitive.
It turns out that when ligands (such as the small molecules in our example) bind to the protein, the Tm changes. Reversible compounds change the Tm in a dose-dependent manner, meaning higher compound concentrations led to greater Tm shifts. Irreversible compounds exhibited a dose-independent Tm shift, which means the new Tm stayed the same no matter how much or how little compound was added (see second column of cartoon above). Here, now, was a phenomenon I could finally exploit to place the compounds into the buckets I wanted! The only catch...was that the DSF instrument software only let you assign the Tm's manually by eye. I knew there had to be some way of automating this process. Thus began the second half, the dry-lab half, of the project.
I wrote a script on Python that took in the first-derivative curves from the instrument files and assigned the Tm's to each concentration of each compound based on the location of the valley. I took those Tm shifts and plotted them against concentration so that each compound now had only one characteristic curve instead of five (see third column in cartoon). The final step was to group the curves together by their shape to place them in their appropriate buckets. The result of all this was a reduction in the retest rate per set of compounds. Now, in the event that the user is unsure of the mechanism of action, they now only need to retest only 4 or 5 of the compounds instead of all 120.