Sanofi has signed a deal with a Stanford-born startup called DiCE Molecules, looking for small-molecule inhibitors of protein-protein interactions. So who are these folks and what route do they have into this perpetually promising-and-challenging area?
DiCE grew out of work at Pehr Harbury’s group at Stanford. The company’s web site makes some bold claims, but that’s what startup web sites are for:
DiCE Molecules’ technology selects and optimizes drug-like ligands to any given target, beginning with libraries containing billions of individual molecules. Unique among currently available options, DiCE Molecules’ technology restores the libraries to their original ligand concentration after each round of screening, revealing the full landscape of binding molecules and allowing them to easily be selected for enhanced potency, selectivity and drug-like properties through testing with proprietary assays. This novel approach may address long-standing chemistry issues and would enable monoclonal antibodies to be replaced by orally-administered medicines.
Unlike the older technology of DNA tagged libraries in which the nucleic acids serve merely to identify the compounds, with DiCE Molecules’ approach DNA is utilized in a manner to that analogous to biological directed evolution: to encode the synthesis of the compound using a physically linked replicable polymer (DNA), enabling one to amplify weak signals out of a vast background of irrelevant noise. It is impossible to overstate the importance of this distinction.
I think one can be forgiven for not quite grasping the point of the first paragraph, but trying to unravel the scientific details from an “About Us” web page is almost always futile. The second paragraph is more parseable, at least to me, but also raises a thirst for more details. Some of those can doubtless be found in this 2004 paper, which is probably the first iteration of their technology (see also this 2007 review). It starts off with an excellent question which I’m sure has occurred to everyone who’s given some thought to chemical biology and evolution:
. . .Multiple generations of selective pressure and reproduction transform a diverse population into one consisting only of molecules fit to survive. Life on this planet thus emerged from a limited chemical palette, comprising proteins, nucleic acids, sugars, lipids, and metabolites. Over the last two decades, technologies that recapitulate this process in the test tube have been developed, and have produced an amazing collection of biopolymers with unprecedented recognition and catalytic properties (reviewed in Roberts and Ja 1999). At present, however, these in vitro selection techniques cannot be applied to compounds of nonbiological origin and have therefore not affected most areas of molecular discovery. The question arises: what would become possible if in vitro selection were applied to chemical populations of arbitrary composition?
Exactly. I’ve been kicking that thought around in my head for years as well. The evolutionary motor driving molecular biology is so relentlessly powerful that you can’t help but want to harness it for diverse small molecules as well, but that is no easy task. Much of the time these ideas lead you into coming up with conjugates of DNA and small molecules or proteins and small molecules, thus to hijack the molecular machinery as it does its thing. And there have been plenty of useful idea that have come out of these approaches: DNA-encoded libraries, yeast hybrid screening, antibody conjugates, phage and RNA display technologies, and others. But if you push this concept too far, you keep bumping into the fact that evolution has built this amazing heap of machinery to shuttle information from nucleic acid polymers into proteins, not to give you a platform for small-molecule diversity. All the tools are tuned up to work on DNA, RNA, amino acids, and proteins, not your library of drug-like small molecules. So most of the techniques in this area are indirect ones, bounce shots that try to harness the natural system somehow.
This paper details a DNA-encoded library, but with a twist: it’s not a static library of screening compounds, each with its own DNA bar code. In this case, the DNA tag starts off longer than usual, with several regions for hybridization, and it stays on through iterative rounds of screening as the small molecule on the end is elaborated. In this case, it’s a peptide, but there are many other chemistries that are available, as the conventional DNA-encoded library folks have shown.
In this case, they demonstrated making a ligand for a known antibody (3-E7) which has been used as a test bed before in these sorts of experiments. They start off by making pentapeptides on the end of their single-stranded DNA template, with ten amino acids to vary at each position. The N-terminus is either left as NH2 or is capped with one of nine small organic acids. The ssDNA is converted into double-stranded form, and the library is screened against the antibody, and the selected/amplified DNA coming out of that process is used as the input for the next round. What they found was that the process, after two rounds, converged on Leu-enkephalin, the known binding partner for the 3-E7 antibody. You can see it picking up amino acids along the way until it arrives at the actual pentapeptide. A second experiment started with every codon in the initial DNA sequence coding for a different amino acid than in the first run, but that one converged on Leu-enkephalin as well after round 2, so this wasn’t a lucky hit because of DNA coding bias.
The hope is that this process will zero in on promising binders without having to make them all up front:
Diversification between rounds of selection by recombination makes possible in vitro evolution of libraries with complexities exceeding the physical library size. Thus, a “best” molecule can be pinpointed without exhaustive testing of all potential species. Starting with a working population of compounds that sparsely sample a chemical space, molecules containing parts of an optimal molecular solution often have a selective advantage relative to siblings, and become enriched. Subsequent recombination processes splice together fragments from the numerous partially optimal molecules to form a globally optimal molecule. Thus, the best structure is found, even if the odds were negligible that it existed in the initial working population.
And that, I would have to guess, is the fundamental technology behind DiCE, although it’s no doubt gone through several refinements in recent years. I like it, but I can think of one possible pitfall (and there are surely others). This is, in a way, a sort of DNA-directed fragment optimization. But there are times when a good molecule cannot seem to be dissected into fragments – there’s no apparent path, because the binding doesn’t seem to add up piece by piece, but rather ends up being more synergistic. This is not a killer objection to DiCE, though, because it’s just a source of false negatives, and presumably there’s a reasonable universe of true positives to pick from. You’re looking for a technique that will give you some real hits, not all the real hits that could ever be.
It’s a long way from these ideas to replacing monoclonal antibodies with synthetic small molecules (although I’ll bet that the molecules will turn out to be only small-ish by the time they get optimized). But Sanofi has seen enough to be interested, and that should give DiCE some good funding (and good incentives) to see what they can do with it.