Here’s another one for the Brute Force File, always noting that brute repetitious force is what machines are here for. A joint MIT/Broad Institute effort reports on a platform for combinatorial drug screening in nanodroplets, in this case looking for known compounds that potentiate the effect of antibiotics on gram-negative bacteria. Testing drug combinations is a conceptually simple problem, simple enough that you can immediately see what a tremendous pain it turns into on scale. For instance, as the paper notes, there are about 2,000 drugs approved by the FDA. If you wanted to test each two-drug combination, that comes to around two million separate experiments, with each compound at a fixed dose. It would be even better to take each member of each pair through a dose/response sequence, but at two million runs per data point that gets out of hand real fast, if it wasn’t at the very start.
In this case, they’re doing 4,1600 drugs and known active substances combined with 10 different antibiotics, with a dose-response, which takes you out, with control experiments, to about four million individual wells. That’s enough to keep a person occupied – more to the point, it’s also enough to keep a conventional robotic screening system occupied for a bit, too, and the authors have come up with a different approach. Instead of setting things up to run each experiment through a defined sequence of compounds and concentrations, they have a more let-four-million-flowers-bloom approach.
First, they create stock solutions of the compounds, with a special twist: each of these also contains three fluorescent marker compounds (three different wavelengths) in a unique ratio. This fluorescent barcode will come in handy. That’s because the next step is to emulsify a sample from each well into a haze of roughly 1-nanoliter aqueous droplets in a fluorocarbon oil/fluorosurfactant mix (a technique that’s been worked out before). These emulsions get pooled and re-pipetted so that each new well has two random droplets in it, and then the array is read out by automated fluorescence microscopy, with those dye ratios determined for the two droplets in each well. Now you know which compounds are in which well, although you didn’t dispense them out in an orderly array. (There will, naturally, be duplicate wells for each combination across the microarray). The droplets are then fused by an AC electric field, and cells (in this case, bacteria) are added to each well for the assay.
Now, there are several ways that this can go wrong, and to their credit, the team tried to control for them. One tricky bit with nanodroplets is the diffusion of compounds out of them (and into the fluorocarbon) and their possible re-entry into the droplets next door, which will mix things up pretty badly. They looked for this problem with test compounds and optimized their emulsion handling, washing, etc., to try to avoid it, and in the end, there seems to have been a low false-positive rate in the screen, which is a good sign. They also tested three common bacterial pathogens (P. aeruginosa, S. aureus, and E. coli) under the culture conditions with known antibiotics that were dispensed via the nanodroplet technique, looking for dose-response and overall reproducibility. And they also tested a pair of compounds that will absolutely synergize: ampicillin and sulbactam. That’s a a beta-lactam antibiotic and a beta-lactamase inhibitor paired up, and the dispensing/fusion technique showed that they do indeed show up as a positive combination. That combination, along with an erythromycin/tetracycline one, was used as a positive control in each microarray, along with blank negatives.
With everything tuned up, they ran the four million experiment microarray screen, which took about ten days total. About 85% of the runs actually worked, and 49 compounds dropped out due to trouble with droplet formation, fluorescence barcoding, pooling difficulties, etc. Overall, there were 28 compound/antibiotic pairs that came out positive (representing 20 different compounds), and further screening with more stringent cutoffs narrow that down to six. Five of these have never had any reported antibiotic effects, and there were some interesting patterns. For example, many of the hits would synergize with erythromycin and/or novobiocin, but the combinations with vancomycin were all over the place, ranging from potentiation to completely wiping out the antibiotic’s effect (!)
It’s going to be interesting to figure out how these potential antibiotic potentiators work; their canonical mechanisms aren’t too enlightening. And it shouldn’t be too hard to test these combinations in vivo, since small animal models for this sort of thing are well established, and many of the compounds tested have been that far in development earlier on. Further on, there’s no reason to confine this technique to antibiotics and bacteria. Many other cell types can be handled this way, and there is no shortage of potential combinations to try. It’s actually quite rare to systematically look for such effects, and that means that there’s a lot of unexplored territory out there, some of which could be quite useful indeed.