DFG-out and Type II kinase inhibitors: where selectivity comes from

Roughly two-thirds of marketed kinase inhibitors bind the active, ATP-ready conformation. The remaining third bind a conformation the kinase only adopts when it is switched off, and that conformational switch is where a lot of the selectivity story in this class lives. The distinction goes by an inelegant three-letter shorthand: DFG-in versus DFG-out. Once you know what the DFG motif does, the difference between a Type I and a Type II inhibitor stops being arbitrary and starts being a structural prediction you can act on.

The DFG motif and the activation loop

Every protein kinase has an activation loop, and at the N-terminal end of that loop sits a conserved Asp-Phe-Gly tripeptide. The aspartate coordinates a magnesium ion that positions the gamma phosphate of ATP for transfer. The phenylalanine is the structural switch. In the catalytically competent state, that phenylalanine packs into a hydrophobic pocket adjacent to the C-helix, the aspartate points into the ATP site, and the kinase is ready to phosphorylate substrate. This is the DFG-in conformation.

In the inactive DFG-out conformation, the phenylalanine flips by roughly 180 degrees and rotates out of the back pocket. The aspartate rotates with it, and a previously occluded hydrophobic cavity behind the gatekeeper residue opens up. Drug designers call this newly exposed cavity the allosteric or back pocket, and it is the pocket that Type II inhibitors exploit.

Type I, Type II, and the cousins

• Type I inhibitors bind DFG-in. They sit in the ATP cleft, make the standard hinge hydrogen bond, and compete with ATP. Most kinase inhibitors are Type I — gefitinib, erlotinib, dasatinib, crizotinib, sunitinib in one of its binding modes, and the entire FGFR and JAK families. Selectivity is harder because every kinase looks roughly similar in its ATP-bound state.
• Type II inhibitors bind DFG-out. They straddle the gatekeeper, reach into the back pocket, and pick up interactions that simply do not exist in the DFG-in state. Imatinib, nilotinib, ponatinib, sorafenib, regorafenib, and BIRB-796 are the canonical examples. The back pocket has significantly more sequence variation across kinases than the ATP site does, which is why Type II inhibitors can be more selective even though they hit a larger surface.
• Type III inhibitors are pure allosteric — they bind outside the ATP cleft entirely. Trametinib (MEK), asciminib (BCR-ABL myristate pocket), and the SHP2 allosteric inhibitors fit here. The reachable chemistry is small but the selectivity can be exquisite.
• Type IV and covalent inhibitors add the warhead dimension. Osimertinib, ibrutinib, and sotorasib are all covalent Type I-derivatives. Covalent Type II is rare but exists.

Why this matters for docking

If you take a DFG-in crystal structure and dock a Type II inhibitor against it, the dock will look terrible. The back pocket is closed, the phenylalanine occupies the volume the drug wants to fill, and the scoring function will reward poses that contort into the accessible ATP cleft rather than the pocket the drug actually binds. You will see scores 3 to 5 kcal/mol worse than they should be and conclude the molecule is inactive.

The fix is to dock against the right conformation. For ABL kinase, the canonical DFG-out structure is 1IEP (imatinib bound). For BRAF, it is 1UWH (BAY-43-9006 bound). For p38-alpha, it is 1KV2 (BIRB-796 bound). Several public databases (KLIFS, KinCoRe) annotate every deposited kinase structure with its DFG state, and an ensemble docking approach that samples both states will recover Type II binders that single-structure docking misses. The Liganx ensemble feature does exactly that.

The harder case is when the protein has no deposited DFG-out structure for your target. AlphaFold predictions are biased toward the active state, so an AlphaFold model of a kinase is almost certainly DFG-in. You either need to switch to a homology model from a DFG-out cousin, run an explicit MD-driven conformational sampling step, or accept that your docking will only catch Type I binders for that target.

Try the docking yourself

The canonical ABL DFG-out structure is 1IEP with imatinib bound, and the DFG-in counterpart for ABL is 2GQG with dasatinib bound. Open Studio and pick ABL from the target catalog to dock candidates against both states side-by-side. Liganx is molecular docking online, free in the browser, and the ensemble option runs both DFG conformations so you can see the score differential directly. For a Type II compound like imatinib or ponatinib the back-pocket structure will score 2-4 kcal/mol better than the ATP-only structure; for a Type I compound like dasatinib the ranking flips. Molecular docking with the right conformation is the difference between a usable score and a misleading one.

Primary sources

• Schindler T, et al. Structural mechanism for STI-571 inhibition of abelson tyrosine kinase. Science 289, 1938-1942 (2000). doi:10.1126/science.289.5486.1938
• Liu Y, Gray NS. Rational design of inhibitors that bind to inactive kinase conformations. Nat Chem Biol 2, 358-364 (2006). doi:10.1038/nchembio799
• Zuccotto F, et al. Through the “gatekeeper door”: exploiting the active kinase conformation. J Med Chem 53, 2681-2694 (2010). doi:10.1021/jm901443h
• Kanev GK, et al. KLIFS: an overhaul after the first 5 years of supporting kinase research. Nucleic Acids Res 49, D562-D569 (2021). doi:10.1093/nar/gkaa895