How does roginolisib selectively inhibit PI3Kδ by stabilizing helix kα12?

A shared hinge bond but a different kα12 contact

Both roginolisib and idelalisib are ATP-competitive PI3Kδ inhibitors, and both hydrogen-bond to the hinge backbone of Val828 in the ATP pocket. The 2.75 Å co-crystal structure of roginolisib with PI3Kδ shows where they part ways: roginolisib does not occupy the so-called PI3Kδ specificity pocket formed by Met752, Pro758, Trp760 and Ile777, but instead engages Met752 and Trp760 through van der Waals and π-π interactions and reorients their side chains relative to the idelalisib complex.

The structure also shows roginolisib forming a hydrogen bond from its sulphone to Lys779 (3.1 Å) and a fluorine hydrogen bond to Ser754 (2.9 Å), neither of which idelalisib addresses. Most relevant to the paper's argument, roginolisib shifts the local arrangement around Asp911 and the DRH motif: Asp897 hydrogen-bonds His895 at 2.6 Å, whereas in the idelalisib structure His895 hydrogen-bonds Asp893 at 3.3 Å. That repositioning is what allows the C-terminal helix kα12 to be drawn into the inhibitor's interaction network.

Basic kα12 residues tether the inactive state

Helix kα12 is the last helix of the catalytic domain and part of the regulatory arch (kα10–kα11–kα12) that must open for PI3Kδ to become fully active at the membrane. In molecular dynamics simulations, the basic kα12 residues Lys1040 and Arg1043 interact with the γ-phosphate of ATP in the active state, with Asp911 of the DFG motif coordinating the magnesium ion. When roginolisib is bound, Lys1040 and Arg1043 instead interact with Asp897 and Asp911, and the inhibitor's positioning orients Asp911 toward the C-terminal helix.

The simulations quantify this as a probability distribution of the distance between the Lys1040 side-chain nitrogen and the Asp911 carboxyl carbon across replicates, excluding the first 500 ns as equilibration. The authors interpret the resulting restraint on kα12 flexibility as the mechanism that keeps the arch closed and the enzyme inactive. This is a model-based mechanistic claim supported by the crystal structure and by the hydrogen-deuterium exchange data below, not a direct observation of a single molecule.

HDX-MS shows kα12 protection only in the inactive enzyme

Hydrogen-deuterium exchange mass spectrometry (HDX-MS) reports how quickly backbone amides exchange with solvent deuterium; slower exchange, or protection, indicates a more buried or more rigid region. Comparing apo p110δ/p85α with inhibitor-bound complexes, roginolisib and idelalisib both change exchange in the kα1–kα2 loop, but they differ at the C-terminus. Protection of the kα12 helix (residues 1022–1044) appears with roginolisib in the wild-type heterodimer.

To test whether roginolisib prefers the inactive state, the authors used the p85α-Y685A mutation, which removes the cSH2 inhibitory contact with the elbow between kα11 and kα12 and produces a partially activated heterodimer. HDX-MS confirmed that this mutation exposes the C-terminal kα11–kα12 helices, the end of substrate binding loop 927–940 and the nearby 855–868 loop. In that activated background, idelalisib still affected the same regions as in the wild type, but roginolisib no longer protected kα12. The authors read this as roginolisib making additional contacts with the C-terminal region of the inactive kinase that are lost upon activation.

Two δ/β-specific residues explain the selectivity profile

The selectivity argument rests on sequence, not on the conserved ATP pocket. Within kα12, Lys1040 and Arg1043 are unique to the δ and β isoforms, so the interactions that stabilize the roginolisib-induced kα12 conformation cannot form equivalently in α or γ. The paper also notes that the p85α cSH2 domain does not inhibit p110α, and that the p101 regulatory subunit has no interactions with the C-terminus of the γ isoform, removing the autoinhibitory architecture that roginolisib exploits.

The authors therefore predict selectivity for p110δ and, to a lesser extent, p110β. They report this is consistent with in vitro IC50 measurements showing over 500-fold selectivity against α and over 1,000-fold against γ, with comparatively low selectivity against β. Those IC50 figures are cited from prior work rather than measured in this study, so they corroborate the structural rationale without independently establishing it.

kα12 inserts into membranes during activation

Because kα12 is both the roginolisib contact point and a membrane-facing element, the paper examines what kα12 does in the active enzyme. In MD simulations of ATP-bound p110δ started 2 nm from a PI(4,5)P2-containing membrane approximating the leukocytic plasma membrane, the protein associated spontaneously. The first contacts came from the C2 domain β5/β6 loop, whose nine basic residues engaged negatively charged lipid headgroups within about 50 ns.

kα12 followed: Lys1028 interacted with PIP2 and the helix inserted into the membrane, remaining embedded between lipid acyl chains for the rest of a 3 μs simulation, with its C-terminal end becoming more ordered. Across five replicates totaling 15 μs, PIP2 lipids approached the p110δ active site. The authors connect this to earlier findings that truncating kα12 drastically reduces lipid kinase activity. The functional implication is that locking kα12 with roginolisib interferes with a helix that the active enzyme needs to deploy at the membrane.

Inhibitor binding barely changes membrane affinity

A separate question is whether the conformational changes roginolisib induces alter how tightly PI3Kδ binds membranes. Using protein-lipid FRET assays with vesicles mimicking plasma membrane composition and an excess of PDGFR phosphopeptide to mimic receptor tyrosine kinase engagement, neither roginolisib nor idelalisib changed membrane interaction. The authors conclude that the altered conformations of the membrane-interacting kα1–kα2 and kα12 helices have only a limited effect on lipid binding, at least for the pY-stimulated heterodimer.

This result narrows the mechanism: roginolisib's advantage is not explained by keeping the enzyme off membranes. It is consistent with the proposed action being conformational, holding kα12 and the regulatory arch in the inactive arrangement rather than blocking recruitment.

More sustained PIP3 suppression in CLL patient cells

To connect binding mode to cellular output, the authors quantified PIP3 in lysates from chronic lymphocytic leukaemia (CLL) patient samples treated ex vivo with roginolisib or idelalisib at two fixed doses and time points, using LC-MS with TMS-derivatization and internal-standard normalization. Samples came from treatment-naïve patients and from patients refractory to BTK and BCL-2 inhibitors. Roginolisib was more effective than idelalisib at decreasing PIP3 at the 2 h time point at the high dose.

Responses were heterogeneous. Idelalisib produced variable responses, and one naïve and two double-refractory patient samples showed more than a 2-fold increase in peak PIP3 production versus vehicle, possibly indicating treatment resistance. All tested refractory samples had decreased PIP3 after 2 h of roginolisib, significantly lower than after 24 h, which the authors suggest reflects a differential feedback response to the two inhibitors. The sample sizes are small (n = 5, 4 and 3 across the panels), so these are ex vivo observations in patient-derived cells, not clinical outcomes.

Why the inactive-state mechanism matters most

The paper's distinctive claim is not simply that roginolisib is a potent PI3Kδ inhibitor but that it achieves selectivity by stabilizing an inactive conformation through kα12, a mode the authors state has not been described before for PI3Ks. That reframes where selectivity can come from: not the conserved ATP pocket, but the regulatory arch and its isoform-specific basic residues. The comparison that carries the argument is internal to the paper — roginolisib protects kα12 in the wild-type heterodimer but not in the activated p85α-Y685A background, while idelalisib behaves the same in both.

The practical priority that follows is to read roginolisib's profile as state-dependent: its kα12 contacts depend on the enzyme being autoinhibited, and its selectivity rationale depends on Lys1040 and Arg1043 being present in δ and β but absent in α and γ. The CLL data add a functional signal, more consistent PIP3 suppression at 2 h in refractory samples, but with small patient numbers and ex vivo treatment. The mechanistic case is strongest where crystallography, HDX-MS and simulations converge on kα12; the clinical advantage remains a hypothesis the paper motivates rather than demonstrates.

Sources used in this answer