PI3K inhibitors are finally coming of age

Abstract

Overactive phosphoinositide 3-kinase (PI3K) in cancer and immune dysregulation has spurred extensive efforts to develop therapeutic PI3K inhibitors. Although progress has been hampered by issues such as poor drug tolerance and drug resistance, several PI3K inhibitors have now received regulatory approval - the PI3Kα isoform-selective inhibitor alpelisib for the treatment of breast cancer and inhibitors mainly aimed at the leukocyte-enriched PI3Kδ in B cell malignancies. In addition to targeting cancer cell-intrinsic PI3K activity, emerging evidence highlights the potential of PI3K inhibitors in cancer immunotherapy. This Review summarizes key discoveries that aid the clinical translation of PI3Kα and PI3Kδ inhibitors, highlighting lessons learnt and future opportunities.

Figure 1 –. General overview of signalling by class I PI3K isoforms.

The class IA PI3K catalytic subunits (p110α, β and δ) bind the p85 regulatory subunits which keep the p85/p110 complex in an inactive, cytosolic form. The p85 subunits have two SH2 domains that allow the p85/p110 heterodimers to bind to phosphorylated tyrosine residues in membrane-associated proteins, such as receptors and adaptor proteins, thereby recruiting the PI3K heterodimer to its lipid substrates while simultaneously disinhibiting its enzymatic activity. Mammals have three genes for p85 regulatory subunits, namely PIK3R1 (encoding p85α, p55α and p50α), PIK3R2 (encoding p85β) and PIK3R3 (encoding p55γ). p110γ, the sole member of the class IB PI3Ks, binds p101/p84 regulatory subunits which do not have homology to p85 or other proteins, and which permit p110γ to engage with Gβγ subunits downstream of GPCRs. Class I PI3Ks can also engage with small GTPases such as members of the Ras (p110α, p110δ, p110γ) or Cdc42, Rac or Rab5 families (p110β). Unlike PI3Kα and PI3Kδ, PI3Kβ is also activated by Gβγ subunits downstream of GPCRs and appears to require more inputs to become fully activated compared to PI3Kα. (Insert): overall domain structure of the p110 catalytic subunits. Class I PI3Ks phosphorylate the 3-position of the inositol ring of a specific phosphatidylinositol (PtdIns) lipid, namely phosphatidylinositol-(4,5)-bisphosphate (PtdIns(4,5)P₂), converting it to phosphatidylinositol-(3,4,5)-trisphosphate (PtdIns(3,4,5)P₃, or PIP₃). PIP₃ can be converted to PtdIns(3,4)P₂ following dephosphorylation of the 5’-position by the 5-phosphatases SHIP1 and SHIP2. Together, PIP₃ and PtdIns(3,4)P₂ function as second messengers downstream of class I PI3Ks by interacting with 3-phosphoinositide-binding pleckstrin homology (PH) domains found in diverse proteins, including protein kinases (AKT, BTK), adaptor proteins and regulators of small GTPases. The tumour suppressor phosphatase and tensin homolog (PTEN) 3-phosphoinositide phosphatase dampens class I PI3K signalling, by dephosphorylating PIP₃ and PtdIns(3,4)P₂. PTEN is frequently somatically inactivated in cancer, through a wide range of mechanisms, including loss-of-expression and/or mutation. PTEN inactivation is also the cause of a developmental syndrome known as PTEN Hamartoma Tumour Syndrome (PHTS) in which one gene copy of PTEN has been partially or fully inactivated. Individuals with PHTS are predisposed to benign overgrowths, neurodevelopmental abnormalities as well as specific cancers in adulthood.

Figure 2 –. Key features of the interaction between PI3Ks and pan- and PI3Kα-selective inhibitors.

The native shape of PI3K enzymes is taken to be that observed by crystallography for ATP-bound p110γ (2a; PDB:1E8X) or the very similar apo forms observed for p110γ (PDB:1E8Y), p110δ (PDB:2WXR) and PI3Kα (in complex with a partial p85α fragment, PBD:2RD0). Peptides are shown as ribbons with key residues shown in stick representation. Ligands are shown in stick representation. Colour coding of atoms in stick representations - carbon: cyan, oxygen: red; nitrogen: blue; fluorine: green; phosphorus: purple; colour coding of ligands - ATP: green, copanlisib: dark blue, alpelisib: pink, idelalisib: red; hydrogen bonds are shown in blue dashed lines, metal interactions in orange dashed lines. a) (Top panel): ATP (carbon atoms bright green) bound in p110γ (1E8X); p110γ shown as brown ribbon with sidechains shown in cyan for residues mentioned in text. The adenine makes an acceptor-donor pair of hydrogen bonds with the NH of hinge Val882 and carbonyl of Glu880 whilst the triphosphate is bound by two metal ions, the terminal ammonium groups of Lys807 and Lys833 and a hydrogen bond from Ser806.(Lower panel): 2D representation of the interactions of ATP with the binding pocket of p110γ. b) (Top panel) copanlisib (dark blue) bound in p110γ (yellow ribbon, 5G2N). The pendant aminopyrimidine group of copanlisib fits into the affinity pocket and forms H-bonds with Asp836 and Asp841 via the amino group and receives a H-bond from Lys833 to one of the ring nitrogen atoms. The morpholinopropyl moiety extends towards solvent and does not make any significant interactions; its role in the molecule is mainly as a solubilising group. (Lower panel): 2D representation of copanlisib indicating the H-bonds made with PI3Kγ. c) (Top panel): alpelisib (pink) bound in PI3Kα (green ribbon, 4JPS). Note the multiple H-bonds: to hinge Val851; involving the primary carboxamide of alpelisib with Gln859 in p110α (Asp862, Lys890, Asn836 in p110β, γ and δ, respectively) and the backbone carbonyl of Ser854; the water-mediated H-bond to the pyridine N from Asp810 and Asp933. The charged terminal amine of Lys802 is close to the CF₃ group. (Lower panel): 2D representation of the major interactions of alpelisib in PI3Kα. d) 2D representation of the PI3Kα/δ inhibitor taselisib with H-bonding interactions observed in the crystal with PI3Kα and, in italic, with PI3Kδ. The ether oxygen of taselisib makes the key hinge interaction with both PI3Kα and PI3Kδ. Taselisib has a primary amide that can make the same interactions with p110α as alpelisib, but in p110δ a rotation of the side chain places this amide differently, where it can still interact with the backbone carbonyl of Ser831 and places the terminal carbonyl of taselisib towards solvent (PDB:5T8F). In the affinity pocket, taselisib appears to be capapble of accepting H-bonds from Lys779 (PI3Kδ numbering) to N2 and from a putative water molecule located between Asp787 and Tyr813 (PI3Kδ numbering) to N4. e) 2D representation of the PI3Kα-selective inhibitor inavolisib with H-bonding interactions observed in the crystal with PI3Kα. A carbonyl group in inavolisib accepts an H-bond from Tyr836 in p110α and a difluoromethyl group interacts with the hydroxyl of Ser774 in p110α. Although both of these residues are conserved in all class I PI3K isoforms, the combination of these structural features with a primary amide interacting with the non-conserved Gln859 of p110α results in very high PI3Kα isoform selectivity. f) Structure of the PI3Kα-selective inhibitor serabelisib. Although a crystal structure has not been disclosed for this molecule it is probable that the binding mode mimics that of copanlisib (Figure 2b) with the nitrogen of the imidazopyridine accepting a H-bond from the hinge Val851 and the aminobenzoxazole making interactions with the residues in the affinity pocket (hashed arrows). g) Structure of PI3Kα inhibitor MEN1611 showing the observed hydrogen bonds in PI3Kγ.

Figure 3 –. Interactions of flat PI3Kδ-selective inhibitors with PI3Kδ

a) (Upper panel): nemiralisib (brown) bound in p110δ (purple ribbon, 5AE8) showing H bonds with the hinge Val828 and adjacent Glu826 plus Asp787. Note that the isopropyl group, though not making any specific interactions, occupies the space above Trp760 in p110δ that is occluded in the other isoforms where the residues corresponding to Thr750 (coloured in green) are larger (Arg770, Lys777, Lys802 in p110α, β and γ, respectively) (Lower panel): 2D representation of nemiralisib, with H-bonding interactions and the isopropyl group occupying the tryptophan shelf over Trp760 as observed in the crystal with p110δ. b) 2D representation of the PI3Kδ-selective inhibitor leniolisib, whose quinazoline 1-N accepts an H-bond from Val828 of the hinge. The substituted pyridine occupies the affintiy pocket while the propanoyl pyrrolidine occludes Trp760 giving isoform selectivity in a similar manner to nemiralisib.

Figure 4 –. Interactions of selected propeller-shaped PI3Kδ-selective inhibitors with PI3Kδ

a) Inhibitor-induced specificity pocket in PI3K, illustrated by idelalisib binding to p110δ. Left panel: structure of idelalisib from 4XEO drawn to emphasise the propeller shape, thus the three ring systems of the hinge-binding purine, the quinazolinone amd the phenyl are approximately mutually orthogonal in an orientation organised by a combination of the chiral ethyl group and the phenyl ring. Middle panel: apo structure of p110δ (2WXR) with Met752 packing against Trp760. The blue arrow indicates the relative motion of Met752 in the flexing of the enzyme in solution that can open up the selectivity pocket. Right panel: crystal structure of idelalisib bound in p110δ (4XEO) with the purine making the hinge interaction with the NH of Val828 and the carbonyl of Glu826. The electron deficient quinazolinone ring system fits into the induced selectivity pocket between Met752 and Trp760 and makes a face to edge interaction with the electron rich indole of Trp760. b) 2D representation of idelalisib showing the major interactions with p110δ. c) 2D representation of the PI3Kγ/δ inhibitor duvelisib, with H-bonding interactions observed in the crystal with p110δ. Note the similarity to idelalisib. d) 2D representation of the PI3Kδ-selective inhibitor seletalisib. This is another propeller-shaped PI3Kδ inhibitor, in this case it is probable that the 1 N atom accepts an H-bond from the hinge Val828, with a non-classical H-bond being formed from the CH of the adjacent pyridine ring. e) Structure of PI3Kδ/CK 1ε inhibitor umbralisib. A crystal structure of this has not been published; however, based on the similarity with other propeller inhibitors the structural features can be identified with confidence. The 3-fluoro-4-isopropoxyphenyl ring is similar to substituents in SW13 and SW14 for which crystal structures are known; this occupies the affinity pocket and may be responsible for the high isoform selectivity observed. f) Structure of the PI3Kδ-selective inhibitor parsaclisib with proposed H-bonding interactions based on molecular docking. Note the additional interactions made by the pendant lactam that accepts two H-bonds from both the hydroxyl of Thr750 (p110δ, Arg770, Lys777, Lys802 in p110α, β and γ, respectively) and the terminal ammonium of Lys708 (p110δ, Gln728, Arg735, Ser760 in p110α, β and γ), respectively; other propeller inhibitors do not have an equivalent group. Despite the multiple structural differences with other PI3Kδ inhibitors, parsaclisib still forms a propeller shape. g) Structure of PI3Kδ-selective inhibitor AMG319 showing the hinge interactions with PI3Kδ based on a crystal structure in PI3Kγ.

Figure 5 –. Multi-pronged anti-cancer activity of PI3Kα inhibition in solid tumours

a. Pleiotropic effect of PIK3CA mutation in solid tumours, inducing both cancer-cell intrinsic and paracrine effects. b. Proposed mechanisms for the combinatorial anti-tumour effect of anti-PI3Kα and anti-oestrogen therapy in HR⁺/HER2⁻ breast cancer. Anti-proliferation induced by PI3K inhibitionleads to a compensatory expression of the estrogen receptor (ER) and increased dependency on estrogen. The increase in ER transcription can occur via enhanced FOXO3A activity (which is no longer inhibited by active PI3K/Akt) and an epigenetic mechanism through the histone methyltransferase KMT2D which is inhibited upon phosphorylation by AKT^,. Blockade of AKT by PI3Kα inhibition enhances KMT2D activity, leading to a more open chromatin state that facilitates ER-dependent transcription. This epigenetic mechanism can also be transcriptional as it is proposed that KMT2D affects the occupancy of the transcription factor FOXA1, a key regulator of ER binding in breast cancer.

Figure 6 –. Multi-pronged anti-cancer activity of PI3Kδ inhibition in cancer

a. Proposed triple mode-of-action of PI3Kδ inhibition in CLL: (1) a cancer-cell intrinsic impact, with PI3Kδ dampening signalling by the BCR and a range of cytokines, chemokines, co-stimulatory molecules and adhesion receptors; (2) inhibition of stromal cells that support the leukaemic cells, such as myeloid-derived nurse-like cells, mesenchymal fibroblast-like cells and leukaemia-associated T-cells, and (3) a host anti-leukaemia adaptive immune response, as a consequence of dampening of Treg function upon PI3Kδ inhibition. b. Documented effects of PI3Kδ inhibition in FL: (1) a cancer-cell intrinsic impact, with PI3Kδ dampening signalling by the BCR, the CD40/CD40L pathway as well as restoration of FL cell dependence on the BCL2 anti-apoptotic protein, resulting in a predisposition to FL cell death and sensitivity to BLC2 inhibitors; (2) dampening of recruitment of T-follicular helper cells and Treg cells through downmodulation of the CCL22 chemokine; and downregulation of proteins involved in B–T-cell synapses, leading to an inefficient crosstalk between FL cells and T-follicular helper cells; (3) dampening of follicular dendritic cell-FL interactions related to angiogenesis, cell adhesion and transendothelial migration in FL patients that show a clinical response to PI3Kδ inhibition. (4) a host anti-leukaemia adaptive immune response, as a consequence of dampening of Treg function upon PI3Kδ inhibition. Such a PI3Kδ-inhibition induced immune response has to be formally documented in FL. c. Effect of PI3Kδ inhibition in solid tumours: (1) a cancer cell-intrinsic impact: some solid tumours (such as breast and melanoma) express high levels of PI3Kδ which may provide sensitivity to PI3Kδ inhibition. (2) dampening of the immuno-suppressive effects of MDSCs and macrophages, and dampening of cancer-stimulating fibroblasts and macrophages, and (3) preferential inhibition of Treg cells, allowing a CD8⁺ T-cell immune response to develop. The question marks in the figure indicate that the role of PI3Kδ in the indicated responses requires further validation, with inhibition of PI3Kγ likely to have a stronger suppressive impact on macrophages than inhibition of PI3Kδ, and blockade of PI3Kα and/or PI3Kβ having a stronger impact than PI3Kδ inhibition on fibroblasts.