In Retrospect: Root-Cause Analysis of Structure–Activity Relationships in IRAK4 Inhibitor Zimlovisertib (PF-06650833)

Abstract

In this paper, we disclose insights on the root causes of three structure–activity relationship (SAR) observations encountered in the discovery of the IRAK4 inhibitor Zimlovisertib (PF-06650833). The first is a nonlinear potency SAR encountered with the isoquinoline ether substituent, the second is a potency enhancement introduced by fluorine substitution on the lactam, and the third is a slight potency preference for all-syn (2S,3S,4S) stereochemistry in the fluorine-substituted lactam. We present new data that help to inform us of the origins of these unexpected SAR trends.

The development of structure–activity relationships (SARs) is the foundation of medicinal chemistry research directed to the development of small-molecule drug candidates. While SAR development and discussion are a feature of nearly every research paper in medicinal chemistry, the root causes of those relationships are frequently not disclosed and are rarely subjected to further study. Recently we described the discovery of Zimlovisertib (PF-06650833),¹ an Interleukin 1 Receptor Associated Kinase 4 (IRAK4) inhibitor clinical candidate (Figure 1, 1) that has been shown to reduce levels of inflammation markers in clinical trials.^2,3 IRAK4 is a central point in the signaling processes of innate immunity, responding directly to signaling from the Interleukin-1 (IL-1) family of receptors and Toll-like receptors.⁴ IRAK4 has also been reported to mediate signaling between innate immunity and adaptive immunity.⁵⁻⁸ Medicinal chemistry activities by various groups on the identification of IRAK4 inhibitors have been summarized in review articles.^9,10

Figure 1.

Structure of Zimlovisertib, 1.

The evolution of 1 from the fragment screening hit 2 may be briefly summarized as shown in Scheme 1 and Table 1 and was more fully described previously.¹ Most of the structure–activity relationships that were developed were understood in terms of well-established principles of both medicinal chemistry and physical organic chemistry. These principles allowed us to improve the overall profile of this series with respect to both target engagement (potency) and the absorption, distribution, metabolism, and elimination profiles of the compounds as we advanced from the fragment screening hit 2 to the isoquinoline lactam 3 as previously described.

Scheme 1. Selected Key Compounds in the Discovery of 1.

Table 1. Enzyme and Cellular Assay Potencies of the Compounds Shown in Scheme 1.

Compound	IRAK4 IC50 (nM)a	PBMC IC50 (nM)b
2	55 μM	Not tested
3	4.6	133
4	7.6	347
5	0.5	9.0
1	0.2	2.4

^aIRAK4 enzyme potency was measured in a DELFIA assay using activated full-length IRAK4 protein in the presence of 600 μM ATP (the ATP K_m for IRAK4) and assessing phosphorylation of a peptide substrate.

^bIRAK4 cell potency was assessed by measuring R848-stimulated TNFα production in peripheral blood mononuclear cells (PBMCs) isolated from human blood. All experiments to determine IC₅₀ values were performed in at least duplicate at each compound concentration dilution unless otherwise noted, and the geometric mean of all the IC₅₀ values is provided when IC₅₀s were determined from two or more independent experiments. See ref (1) for further details.

However, there were three structure–activity relationship findings related to compound potency that emerged after the discovery of compound 3 that we could not conclusively explain. These findings were:

• 1.The unexpectedly poor potency of the ethyl ether 6 relative to the starting isopropyl ether 3 and methyl ether 4 (Table 2). This was not predicted using computational tools and could not be rationalized by examination of the previously reported protein–ligand crystal structure of 3 in IRAK4.
• 2.The potency enhancement conferred by the fluorine substituent on the lactam, which was not observed with other substituents at the lactam 3-position. Again, this was not predicted using computational tools, and the protein–ligand crystal structure of neither 5 nor 1 in IRAK4 offered conclusive information.
• 3.The slight improvement in potency observed with the all-syn (2S,3S,4S) lactam stereochemistry of 1 relative to the (2S,3S,4R) isomer 7 (Table 4) in which the fluorine substituent is anti to the other two substituents. This phenomenon was also observed with the (3S) methyl lactams 8 and 9 (see Table 3) and was not predicted based on previous SAR or computation.

Table 2. Potency Dependence on the Isoquinoline Ether.

Compound	R	IRAK4 IC50 (nM)a	LipEb	PBMC IC50 (nM)a
3	i-Pr	4.6	6.6	133
6	Et	23.9	6.2	672
4	Me	7.6	7.9	347

^aSee notes for Table 1.

^bLipophilic efficiency.

Table 4. Effect of Fluorine Substitution on Other Lactam Analogues.

Compound	Rsyn	Ranti	IRAK4 IC50 (nM)a	PBMC IC50 (nM)a
5	H	H	0.5	9.0
1	F	H	0.2	2.4
7	H	F	0.1	6.5
13	H		2.4	52.1
14	F		0.6	16.8

^aSee notes for Table 1.

Table 3. Effect of syn and anti Lactam 4-Substituents on the Potency of (2S,3S-Methyl) Derivatives.

Compound	Rsyn	Ranti	IRAK4 IC50 (nM)a	PBMC IC50 (nM)a
8	F	H	0.3	12.7
9	H	F	1.9	29.5
10	H	H	2.7	52.3
11	Me	H	29.2b	2836b
12	H	Me	3.7	87.1

^aSee notes for Table 1.

^bn = 1.

While we had identified and reported on our clinical candidate, we remained unsatisfied with our lack of understanding of these three points and set out to determine why these phenomena were observed. Our findings are reported below.

1. As the ether substituent on the isoquinoline was made smaller i-Pr → Et → Me (Table 2), we were surprised to find that the potency in both enzyme and cell assays did not follow a linear progression. Instead, ethyl ether 6 stood out as less potent than either the isopropyl or the methyl ether. This was also observed in other matched triads of Me, Et, and i-Pr ethers (Supporting Information, Tables S2 and S3).

To understand this, we obtained protein–ligand crystal structures of 4 and 6 as previously described to compare to that of 3.¹ The crystal structure of IRAK4 in complex with compound 4 at 2.0 Å resolution shows that the compound makes an intramolecular hydrogen bond between the ether oxygen and the amide and forms a two-point hinge binding interaction with Met265 and Val263 backbones. The methyl ether on the isoquinoline has a torsion angle of ca. 172 degrees and is close to the gatekeeper Tyr262 making a van der Waals interaction (Figure 2).

Figure 2.

Structure of methyl ether 4 in complex with human IRAK4.

The crystal structure of compound 6 at 1.8 Å resolution shows an unambiguous electron density corresponding to the ethyl ether (Supporting Information). Interestingly, the ethyl ether has an unusual torsion angle of ca. 44 degrees, which is not favorable (Figure 3).

Figure 3.

Structure of ethyl ether 6 in complex with human IRAK4.

Modeling¹¹ the ethyl ether 6 into IRAK4 with a torsion angle of 180 deg, which is the most frequent and lowest energy conformation, causes a clash with the gatekeeper residue (Figure 4). The observed 44° torsion angle (red line, Figure 5) is almost not populated in the CCDC. This conformation is between the eclipsed (0° torsion) and gauche (60° torsion) conformations and is higher in energy than the anti (180° torsion) conformation, explaining the loss of potency suffered by 6.

Figure 4.

Model of 6 in IRAK4 with 180-degree torsion.

Figure 5.

Frequency of observed ethyl ether torsion angles in CCDC. Vertical scale is occurrences in thousands.

By contrast, the isopropyl group of compound 3 (PDB: 5UIT)¹ adopts favorable van der Waals contacts with the gatekeeper (Figure 6), adopting a preferred conformation (Figure 7).

Figure 6.

Structure of isopropyl ether 3 in complex with human IRAK4.

Figure 7.

Frequency of observed isopropyl ether torsion angles in CCDC. Vertical scale is occurrences in hundreds.

2. The introduction of the fluorine substituent yielded a potency enhancement over other substituents that were examined.¹² We had studied the SAR of substitution at this position using the 3-methyl substituted lactam 10 as our starting point in order to keep the lipophilicity, and therefore potential oxidative metabolism, of our target analogues to a minimum. We weighted the cell potency as measured in the PBMC assay in our decision making because the lower limit of the enzyme DELFIA assay was reached with IC₅₀ values below 1 nM. Analysis of the protein crystal structure of 3 suggested that there may be sufficient space to accommodate a very small substituent on the lactam adjacent to the carbonyl. Introduction of a methyl group syn to the lactam 3-methyl (11, Table 3) caused a significant loss in potency, while the anti methyl analogue 12 was not detrimental. However, addition of a fluorine substituent adjacent to the lactam carbonyl, as in compounds 8 and 9, afforded a marked increase in potency in both the enzyme and cell assays (see Supporting Information Table S4 and Note S1 for additional fluorinated lactam derivatives).

This approximately threefold further increase in potency upon syn-fluorine substitution was observed in the 4-ethyl-substituted lactams 1 and 7 and in the cyclopropane-substituted lactam matched pair 13 and 14 (Table 4).

In our original paper, we hypothesized that the increased potency conferred by fluorine substitution might be due to increased hydrogen bond donor capability of the lactam.¹³ We thought to study the ¹³C NMR spectra of the compounds in question, having recently described how ¹³C NMR was used in conjunction with cyclic voltammetry to study the reduction potentials of α,β-unsaturated lactams.¹⁴ Comparison of the proton-decoupled ¹³C NMR chemical shifts for the lactam carbonyl carbons of the compounds in Tables 3 and 4 revealed that the ¹³C NMR chemical shifts of the carbonyl carbon atoms in the fluorinated lactams were upfield of the carbonyl resonances of the nonfluorinated lactams (Table 5).

Table 5. ¹³C NMR Chemical Shifts of Lactam Carbonyl Carbon Atoms.

Compound	R3	Rsyn	Ranti	13C CO, δ ppm	IRAK4 IC50 (nM)a	PBMC IC50 (nM)a
10	Me	H	H	176.82	2.7	52.3
11	Me	Me	H	176.95	29.2	2836
12	Me	H	Me	180.66	3.7	87.1
8	Me	F	H	171.62b	0.3	12.7
9	Me	H	F	171.06b	1.9	29.5
5	Et	H	H	176.71	0.5	9.0
1	Et	F	H	171.02b	0.2	2.4
7	Et	H	F	171.45b	0.1	6.5
13		H		176.78	2.4	52.1
14		F		171.44b	0.6	16.8

^aSee notes for Table 1.

^bDoublet due to ¹³C–¹⁹F coupling.

In each case, the resonance of the lactam carbonyl carbon in the fluorinated compound occurred slightly more than 5 ppm upfield of the nonfluorinated analogues. By contrast, the primary amide carbonyl resonance remained essentially unchanged in all compounds. This phenomenon is not unique to these lactams. For example, a comparison of reported ¹³C NMR spectra of acetanilide 15 (Figure 8, CO δ = 168.7)¹⁵ and 2′-fluoroacetanilide 16 (CO δ = 165.4)¹⁶ reveals an upfield shift of 3.3 ppm in the fluorinated compound. Further fluorine substitutions result in further upfield shifts of the carbonyl resonances (Supporting Information, Table S5). The effect may be somewhat magnified in the lactam analogues due to improved orbital overlap in the lactams as a result of the conformational restraint.

Figure 8.

Structures of acetanilide 15 and 2′-fluoroacetanilide 16.

At first, one might expect the electron-withdrawing fluorine substituent to deshield the carbonyl carbon, rather than shielding it. However, amide groups can accommodate the electron-withdrawing demand of the fluorine substituent by withdrawing electron density from the nitrogen atom. Thus, the net effect of the fluorine substituent is to increase the charge separation between the amide N and C atoms, leading to an increased positive charge on the N atom and a decreased positive charge on the C atom. Accordingly, the lactam nitrogen atoms of 5 and 1 should have different ¹⁵N chemical shifts. In fact, the lactam N of 5 had a resonance at 123.8 ppm, while that of 1 had a resonance of 127.0 ppm (see Supporting Information for relevant spectra). Thus, the lactam N of 1 is indeed more deshielded than that of 5, which is consistent with the hypothesis that the 1 lactam has more positively charged character on nitrogen than does 5, making the NH of 1 a better H-bond donor to the backbone carbonyls of Ala315 and Asn316.

These conclusions are further supported by calculation of the electrostatic potentials of the atoms in the lactam ring (Table 6, Supporting Information Note S2). These results show that an increased positive charge on the N atom and a decreased positive charge on the C atom are expected. What may be a bit surprising is that the negative charge on the amide O atom is not significantly increased, suggesting that the fluorine substituent does not simply amplify the contribution of the charge-separated amide tautomer and improve the hydrogen bond acceptor character of the lactam carbonyl, which accepts a hydrogen bond directly from Ser328 and indirectly from Lys213 through a water molecule.

Table 6. Calculated Electrostatic Potentials of Lactam Atoms.

Atom →	1	2	3	4	5	6	7	8	9
X = H, 5	–0.727	0.153	0.178	–0.661	0.747	–0.554		0.106	–0.682
X = F, 1	–0.697	0.092	0.035	–0.074	0.628	–0.511	–0.175	0.107	–0.684

Taken together, these results support the conclusion that the fluorinated lactam in 1 is a better hydrogen bond donor than the lactam in 5. This increase in the hydrogen bond donor character of the fluorinated species is again mirrored in the comparison of acetanilide 15 to 2′-fluoroacetanilide 16. 2′-Fluoroacetanilide (pK_a 13.1)¹⁷ is 9 pK_a units more acidic than is acetanilide (pK_a 22.3),¹⁸ a result of the increased positive charge on the nitrogen and therefore further polarization of the N–H bond (see Supporting Information Table S5 for further data).

3. We initially imagined that the slight but significant potency preference for the syn fluorine isomer was due to an improvement in the hydrogen bond acceptor capability of the lactam carbonyl, with the syn fluorine atom enhancing the hydrogen bond acceptor capability somewhat more than the anti fluorine isomer due to differences in orbital overlap or ring conformation between the two isomers.^19,20 However, the calculations above revealed that one could expect little difference between the effects of the two fluorine isomers on the lactam carbonyl. The potency preference for the syn fluoro isomers could not be attributed entirely to a steric effect, inasmuch as the anti methyl substituent in 12 was tolerated. Nor could the potency preference be attributed to the syn fluorine atom itself accepting a hydrogen bond,^21,22 since the protein–ligand crystal structure of 1 revealed no protein residues capable of doing so within the necessary distance and geometry. This potency preference for the syn fluorine isomer was observed in other matched pairs (Supporting Information, Table S4) and was not unique solely to compound pairs 1/7 and 8/9.

Closer inspection of the protein–ligand crystal structure of 1 revealed that the void volume into which the corresponding anti fluorine isomer would project is occupied by solvent (Figure 9). However, at the boundary of this solvent space are found two aspartate residue side chains, Asp 311 and Asp329, with Asp329 in particularly close proximity (3 Å) to the location of an anti fluorine atom. Together, the aspartate residues create a localized environment of negative charge which is repulsive to the fluorine atom,²³ and this is the root cause of the slightly better potency observed with the syn fluoro isomers. That the repulsive effect of the anti fluoro isomer is not larger may be attributed to the improvement in the hydrogen bond donor capability of the lactam nitrogen, which by calculation is expected to be present in a similar magnitude regardless of the fluorine stereochemistry.

Figure 9.

Crystal structure of 1 with IRAK4 (PDB: 5UIU).

Three previously unexplained features of the SAR in the progression of 3 to 1 are now resolved. First, it was found by protein–ligand crystallography that ethyl ether 6 made an unexpected negative torsion upon binding with IRAK4, which was compensated for in isopropyl ether 3. Second, it is clear from ¹³C and ¹⁵N NMR, calculation, and examination of the literature that the fluorine substituent does indeed render the lactam in 1 a better H-bond donor than the unsubstituted lactam in 5. Third, the anti fluoro isomers were found to project into a pocket with an overall partially negatively charged environment, resulting in a partially repulsive interaction relative to the syn fluoro isomers.

In conclusion, the structure–activity relationships in the development of the IRAK4 inhibitor 1 (PF-06650833) are now fully understood. We believe that nonlinear, nonadditive, or unpredicted structure–activity relationships, while frequently encountered in medicinal chemistry, should not be simply accepted as phenomena but instead represent rewarding learning opportunities that can contribute to the successful development of a new chemical lead.

Glossary

Abbreviations

CCDC

Cambridge Crystallographic Data Centre

IRAK-4

Interleukin 1 Receptor Associated Kinase 4

LipE

lipophilic efficiency

PBMC

peripheral blood mononuclear cells

SAR

structure–activity relationships

Data Availability Statement

The data underlying this study are available in ref (1), this published article, and its Supporting Information.

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsmedchemlett.4c00036.

• General NMR information. Line lists, ¹H, ¹³C and HN-HMBC spectra of compounds 5 and 1. Comparison of carbon and nitrogen chemical shifts around the lactam ring of compounds 5 and 1. Data collection, refinement statistics and methods for IRAK4 crystallization, and electron density maps of 4 and 6. Comparison of ether substituent matched triads, comparison of matched fluoro substituted lactams, comparison of C=O ¹³C NMR chemical shifts and pK_a of acetanilide derivatives, notes (PDF)

Accession Codes

Crystal structure of IRAK4 in complex with compounds 1 PDB 5UIU, 3 PDB 5UIT, 4 PDB 8W3W, 6 PDB 8W3X.

Author Contributions

S.W.W. ideated the research presented, performed the calculations, and prepared the manuscript. K.L.L. developed the SAR trends presented. S.H. and J.D.K. conducted the protein–ligand crystallography and analysis. K.A.F. performed the ¹H, ¹³C, and ¹⁵N NMR work and assigned the spectra.

The authors declare no competing financial interest.