Aurora A–Selective Inhibitor LY3295668 Leads to Dominant Mitotic Arrest, Apoptosis in Cancer Cells, and Shows Potent Preclinical Antitumor Efficacy

This article is featured in Highlights of This Issue, p. 2183

The Aurora kinases are composed of three highly conserved serine/threonine kinases: Aurora A, B, and C that regulate mitotic and meiotic processes (1–3). The proteins contain a highly conserved central kinase domain with a variable N-terminal domain critical for unique protein–protein interactions and a C-terminal domain carrying a protein stability regulation motif (4).

Aurora A is critical to centrosome maturation/separation, mitotic spindle formation, and mitotic spindle checkpoint activation (5–7). Aurora A mRNA/protein expression and kinase activity are cell-cycle regulated, with basal level at interphase and expression/activity dramatically increased from late G₂ to telophase in mitosis. The kinase activity requires Aurora A autophosphorylation at Thr288 in the activation loop. Aurora A associates with centrosomes at interphase cells as well as with mitotic spindle in mitosis (8). Aurora A regulates mitotic progression by interacting with and phosphorylating multiple mitotic kinases, including targeting protein for Xklp2 (TPX2), p53, large tumor suppressor kinase 2 (LATS2), Rho-associated, coiled-coil–containing protein kinase (p160ROCK), and others (8–11).

Aurora B controls chromosome alignment in mitosis and cytokinesis as a catalytic unit in the chromosome passenger complex, which also contains Survivin, Borealin, and INCENP (12–14). Similar to Aurora A, Aurora B expression and localization are cell-cycle regulated, with expression and kinase activity peaking at mitosis and localization at kinetochores and centromeres from prophase to metaphase and then relocating to the cleavage furrow and midbody during cytokinesis. Meanwhile Aurora B kinase activity requires phosphorylation at Thr232 in the activation loop so that it can phosphorylate histone H3 at Ser10, and other substrates for chromosome condensation and spindle checkpoint (13, 15). Aurora C expression is restricted to the testis and functionally mimics Aurora B (16).

Although the three Aurora kinases share high sequence similarity, especially at the kinase domain, they function distinctly in cell-cycle progression (17). Aurora A is a mitotic kinase whose depletion primarily leads to mitotic spindle formation defects and consequently prometaphase arrest in human cancer cell lines, evidenced by the increase of histone H3 Ser10 phosphorylation as a mitotic marker. Aurora B/C function as chromosome passenger proteins such that kinase inactivation primarily induces polyploid cells. This ploidy is caused by cytokinesis failure, as evidenced by the failure of daughter cell separation and the Aurora B–mediated inhibition of histone Ser10 phosphorylation (18–20). Aurora B/C inactivation phenotypes are also epistatic to these of Aurora A. Concomitant inactivation of Aurora A and B, or all Aurora isoforms, demonstrates the Aurora B/C-dominant cytokinesis failure and polyploidy phenotypes. Polyploidy and subsequent chromosome segregation error-induced genomic instability are potential mechanisms of tumorigenesis and drug resistance. Thus, Aurora B/C inhibitory activity may counter the beneficial effects of Aurora A inhibition and should be avoided with a selective Aurora A inhibitor.

Although more than a dozen inhibitors have been developed, for example MLN8237 (alisertib), AZD1152 (barasertib), and others (21–25), most are either pan-Aurora or Aurora B/C inhibitors (26). As we show here, the most advanced Aurora inhibitor alisertib (MLN8237) has narrow Aurora A versus Aurora B selectivity, and at reported clinically efficacious concentrations (27), the Aurora B phenotype would dominate (27). Here we report development of a highly Aurora A–selective kinase inhibitor, LY3295668, which potently inhibits proliferation in a broad panel of cancer cell lines, induces apoptosis, and is highly efficacious in vivo, for example, in SCLC and other xenograft and patient-derived tumor (PDX) preclinical tumor models.

NCI-H446, NCI-H69 (both from small-cell lung carcinoma), HeLa (cervical adenocarcinoma), Calu6, NCI-H358 (both from lung carcinoma), and Detroit 562 (pharyngeal carcinoma) were purchased from ATCC first in 2012 and later as needed. The panel of cell lines tested in Fig. 4 is listed in the Supplementary Information. The cell lines were pathogen tested and authenticated by short tandem repeat (STR)-based DNA profiling. All cell lines were cultured in conditions recommended by manufacturers at 37°C with or without 5% CO₂ in a humidified atmosphere and used within 2 months from thaw. Growth curves were determined for all cell lines. Cell densities were optimized to achieve robust cell growth during assays.

The following antibodies against corresponding proteins were purchased and used according to the manufacturer's instructions: α-tubulin, Centrin 1 were from Abcam; actin, β-tubulin, Aurora A P-Thr288, Aurora B, P-Aurora A/B/C (P-Thr288/Thr232/Thr198, respectively), E-cadherin, vimentin, Sox2, cyclin B1, cleaved PARP, and Histone H3 were from Cell Signaling Technology; histone H3 P-Ser10 was from EMD Millipore, Aurora A was from R&D Systems; β-tubulin, actin, and GAPDH were from Sigma and Aldrich; horseradish peroxidase (HRP)-conjugated anti-rabbit and anti-mouse immunoglobulin G (IgG) were from GE Healthcare; and goat anti-mouse IgM Alexa Fluor-488, anti-mouse Alexa Fluor-647, anti-rabbit Alexa Fluor-488, and Hoechst 33342 from Thermo Fisher Scientific. The immunofluorescence microscopy was done with Leica DMI6000B or Zeiss LSM880 microscopes. For quantification of P-Aurora A and P-H3 in mitotic cells, images were acquired with the 40× water objective of an Opera Phenix and analyzed using Columbus software (Perkin Elmer).

LY3285668 was synthesized at Lilly Research Laboratories chemistry labs (example 1 in ref. 28). MLN8237 (LI-R034) was purchased from Shanghai ChemPartner Co. Ltd. AZD1152 (HM5015062) from Hutchison MediPharma Ltd., flavopiridol from AmFineCom, Inc. TGFβ was from R&D Systems and SB431542 from Sigma-Aldrich. All compounds were dissolved in DMSO as 10 mmol/L stock solution for in vitro assays. For in vivo animal studies, LY3295668 was formulated as a solution in 20% 2-hydroxypropyl-β -cyclodextrin (HPBCD) in 25 mmol/L phosphate buffer pH 2, adjusted with hydrochloric acid.

For details, see Supplementary Data.

For details, see Supplementary Data.

For details, please see Supplementary Data.

See Supplementary Data and ref. 40 for details.

NCI-H446 xenograft tumor lysates were used for in vivo target inhibition (IVTI). For details, see Supplementary Data.

For details, see Supplementary Data.

For details, see Supplementary Data and ref. 28.

All in vivo studies were performed according to the Institutional Animal Care and Use Protocols of the party or provider conducting the experiments. NCI-H446, NCI-H69, and Detroit 562 xenograft and the small-cell lung cancer (SCLC) PDX models (in Oncotest GmbH; Charles River Laboratories) were used. For details, see Supplementary Data.

For details, see Supplementary Data.

For details, see Supplementary Data.

LY3295668 [(2R,4R)-1-[(3-chloro-2-fluoro-phenyl)methyl]-4-[[3-fluoro-6-[(5-methyl-IH-pyrazol-3-yl)amino]-2-pyridyl]methyl]- 2-methylpiperidine-4-carboxylic acid] was identified during a medicinal chemistry program seeking to discover highly selective inhibitors of Aurora A kinase (Fig. 1A; ref. 29). It is a reversible ATP-competitive inhibitor with high aqueous solubility at 1.89 mg/mL. It also shows excellent physiochemical properties with LogD (pH 7.4) of 2.38 and measured fraction unbound of 1.0%, 1.1%, 1.4%, and 4.5% in human, dog, rat, and mouse sera, respectively.

LY3295668 potently inhibited Aurora A and showed excellent kinase selectivity in binding of a panel of 489 protein kinases in vitro, and carries significant binding activity for only 3 other protein kinases with IC₅₀s less than 0.1 μmol/L (Fig. 1B). Enzymatically, LY3295668 showed remarkable selectivity against Aurora B (1,339-fold; Fig. 1C, left). In NCI-H446 cells, LY3295668 was a potent (IC₅₀ = 0.00059 μmol/L) inhibitor on Aurora A kinase autophosphorylation at Thr288 (Fig. 1C, middle). Aurora B inhibition potency, as determined by histone H3 Ser10 phosphorylation, was very poor (IC₅₀ = 1.42 μmol/L; Fig. 1C, middle), indicating a >2,000-fold selectivity of Aurora A over Aurora B in this cell-based assay. Consistent with LY3295668 Aurora A inhibitory activity, 24-hour treatment led to an increase in histone H3 Ser10 phosphorylation (mitotic arrest) as well as in G₂–M DNA content (Fig. 1C, right) in HeLa cells. Accordingly, NCI-H446 and HeLa cell proliferations were inhibited (Fig. 1C, right; Supplementary Table S1). The correlation of Aurora A biochemical IC₅₀ with P-H3 elevation, 4N DNA content increase, and cell proliferation inhibition indicates that LY3295668 Aurora A inhibition leads to mitotic arrest as well as cell proliferation inhibition.

Using a chemoproteomics platform (KiNativ; ref. 30) to detect the compound kinase-binding profile in NCI-H446 cancer cells, LY3295668 treatment at 1 μmol/L for 24 hours demonstrated an excellent Aurora A potency with inhibition efficiency at 98%, as well as kinase selectivity with only one other kinase (EphA2) showing significant inhibition (92%) among 203 kinases tested (Supplementary Table S2). Under this condition, PLK1, Aurora B, and EGFR kinase binding increased 256%, 265%, and 851% (n = 2), respectively. Both PLK1 and Aurora B expression are cell-cycle regulated, and it is expected that they would increase dramatically in mitosis (Supplementary Table S2), indicating LY3295668 treatment induces G₂–M arrest.

We further used the KiNativ chemoproteomics platform to examine the LY3295668 kinase inhibition profile in vivo with the NCI-H446 xenograft model. When the animals were treated at 50 mg/kg every day for 3 hours, only Aurora A and Aurora B showed evidence of significant inhibition out of 195 kinases detectable (Supplementary Fig. S1A, left; Supplementary Table S3). When mice were treated with LY3295668 at 50 mg/kg twice a day (at 0 and 12 hours) and then tumors collected at 24 hours, the compound only inhibited Aurora A significantly, whereas PLK1, AURKB, and three other kinase expression increased significantly (Supplementary Fig. S1A, right), again indicating LY3295668 treatment in vivo led to Aurora A inhibition-induced mitotic arrest. In the mouse spleen, LY3295668 treatment at 50 mg/kg every day for 3 hours only significantly inhibited Aurora A (Supplementary Fig. S1B; Supplementary Table S4). Two treatments of LY3295668 (50 mg/kg twice a day, at 0 and 12 hours) again demonstrated that Aurora A is the only kinase significantly inhibited and a Serine/Threonine-Protein Kinase A-Raf is the only kinase with expression significantly increased (Supplementary Table S4). Taken together, LY3295668 consistently demonstrates selective Aurora A inhibition in vitro and in vivo.

Several different Aurora A and Aurora B kinase substrates in cells have been identified and used to characterize Aurora kinase inhibitors (26). Here, we are the first to use an Aurora A P-Thr288 inhibition assay to screen for an Aurora A–selective inhibitor. Pairwise analysis from 2,940 compounds from our structure–activity relationship (SAR) library indicates that Aurora A cell-based P-T288 inhibition is most highly correlated with Aurora A enzymatic assay not Aurora B (Spearman ρ = 0.7264; Supplementary Table S5). The more revealing correlation comes from multiple linear regression analysis. For Aurora A cell assay (Fig. 1D, left), improving Aurora A biochemical potency 10-fold (1 log unit) would have an expected improvement 10-fold (log Aurora A has estimate of 10^1.083) in Aurora A cell potency, while Aurora B biochemical 10-fold improvement can only contribute 1.6-fold (log Aurora B has estimate of 10^0.202). For Aurora B cell-based activity (Fig. 1D, middle), Aurora A enzymatic activity potency increase does not contribute at all (10^−0.043), whereas 10-fold Aurora B enzymatic potency increase contributes to 5-fold increase in the Aurora B cell potency (10^0.639). So both pairwise and multiple regression correlation analyses support our flow scheme to identify Aurora A–selective inhibitors (Fig. 1D; Supplementary Table S5).

Aurora A and B inhibition are not equal in inducing mitotic arrest (Fig. 1D, right): improving Aurora A biochemical potency 10-fold contributes to mitotic arrest 2.9-fold (10^0.457) while Aurora B only 1.9-fold (10^0.268). These data indicate that Aurora A inhibition was much more effective in mitotic arrest induction, whereas Aurora B inhibition–induced endoreduplication and polyploidy phenotype may interfere with the mitotic arrest phenotype.

To determine the Aurora A inhibition potency required inducing significant cell proliferation inhibition and thus significant antitumor efficacy in vitro and in vivo, a statistical analysis was performed which indicates that 90% Aurora A inhibition (Aurora A P-T288 IC₉₀) can effectively induce cell proliferation inhibition as well as mitotic arrest (Supplementary Fig. S2).

The potency and selectivity of LY3295668 against Aurora A translated to mitotic arrest measured by histone H3 P-Ser10 marker and an increase in 4N DNA content in HeLa, NCI-H446, Calu6, and other tumor cell lines (Figs. 1B and 2A, left; Supplementary Fig. S3). Several Aurora inhibitors are being tested clinically, notably alisertib (MLN8237) and barasertib (AZD1152; refs. 21–23, 25). Each of these Aurora inhibitors was compared with LY3295668 in cell-based assays (Fig. 2). Alisertib is an Aurora A/B dual inhibitor with approximately 16-fold Aurora A selectivity versus Aurora B in cell-based assay (Fig. 2A, middle). Although mitotic arrest from LY3295668 treatment was still evident up to 20 μmol/L (the highest tested concentration), alisertib only showed Aurora A–selective inhibition between 0.05 and 0.25 μmol/L, but exhibited Aurora B–related P-H3 inhibition at concentrations above 0.25 μmol/L (Fig. 2A, middle). Barasertib also demonstrated an Aurora B–selective inhibitor dominated phenotype (Fig. 2A, right).

All three Aurora inhibitors potently inhibited HeLa cell proliferation (Fig. 2B, left). LY3295668 also demonstrated potent apoptosis induction with significant caspase 3/7 increase reaching maximal at 0.087 μmol/L and higher (Fig. 2B, right), correlating to its Aurora A inhibition, mitotic arrest, and cell proliferation inhibition potency (Fig. 2A, left). In contrast, alisertib showed most dramatic apoptosis induction only from 0.087 to 0.250 μmol/L, matching its Aurora A–selective cell activity range (Fig. 2A, middle). At concentrations of alisertib shown to be dominated by effects of Aurora B inhibition (>0.25 μmol/L), the apoptosis induction actually slightly decreased (Fig. 2B, right). Barasertib did not induce considerable apoptosis, reflected by less caspase 3/7 induction (Fig. 2B, right).

The cell growth inhibition and apoptosis induction difference among LY3295668 and other Aurora inhibitors (2.2 μmol/L) can also be observed from cell morphology images (Supplementary Fig. S4) taken from the same cell culture plates used for IncuCyte quantitation in Fig. 2B. The majority (96.4%) of HeLa cells were killed by LY3295668 showing caspase 3/7 positivity (green dots). However, there were still significant numbers of surviving cells showing enlarged/multinuclei and cytoplasm in alisertib or barasertib treatment, indicating they had gone through endoreduplication and cytokinesis failure mediated by Aurora B inhibition (Supplementary Fig. S4).

Immunofluorescent staining of HeLa cells treated with 1 μmol/L of LY3295668 clearly indicates mitotic arrest at prometaphase with abnormally shortened microtubule spindle network, disorganized arrays of microtubules (β-tubulin staining and Aurora A staining, which localizes to mitotic spindle and centrosomes), as well as misaligned condensed chromosomes (DAPI staining; Fig. 3A, top and middle). LY3295668 treatment also inhibited Aurora A P-Thr288 (P-AurA) staining at 94% (±0.468), with only a minor effect on histone H3 P-Ser10 (P-H3, 16% ± 0.2093), consistent with its Aurora A–selective inhibition (Fig. 3A, top and middle and B). Alisertib treatment also arrested cells in mitosis with disarrayed microtubule spindle and misaligned chromosomes, but abolished both Aurora A P-Thr288 (95% ± 0.433) and histone H3 P-Ser10 (93% ± 0.313) staining. This is indicative of an Aurora A/B dual inhibition phenotype (Fig. 3A, top and middle and B). Although the gross network of mitotic spindle was near normal, barasertib treatment led to misaligned chromosomes as well as corrupted mitotic spindle (Fig. 3A, top and middle). Barasertib inhibited histone H3 P-Ser10 (90% ± 4.054), but only marginally on Aurora A P-Thr288 (21% ± 2.976), representative of an Aurora B–dominant phenotype (Fig. 3A, top and middle, Fig. 3B).

PLK4 inhibition was also observed with LY3295668 (Fig. 1C) and can produce mitotic arrest. To distinguish between the Aurora A versus PLK4 inhibition–induced mitotic phenotype, cells were stained with Centrin 1 because centriole duplication is PLK4 dependent (31). Neither LY3295668, nor alisertib or barasertib perturbed centriole duplication and maturation with duplicated Centrin staining localized at the center of centrosomes in all the mitotic cells (Fig. 3A, bottom). These data indicate that although LY3295668 may inhibit PLK4 enzyme activity in vitro, this activity did not translate into significant inhibition of PLK4 in cells (Fig. 3A, bottom).

Western blot analysis further revealed the Aurora A versus Aurora B inhibition difference on mitotic arrest and apoptosis (Fig. 3C). At 0.25, 1, and 5 μmol/L, LY3295668-treated cells for 24 hours showed Aurora A, Aurora B, histone H3 P-Ser10, and cyclin B1 increases in a dose-dependent manner, indicating it effectively induced mitotic arrest (Fig. 3C, left). Although there was a transient increase in the expression of these endpoints by alisertib at 0.25 μmol/L (Aurora A inhibition), the increases were quickly lost due to transition to an Aurora B inhibition–dominant phenotype at concentrations >0.25 μmol/L. Cyclin B1 levels decreased from those at 0.25 μmol/L to those at 1 and 5 μmol/L correlated with its cytokinesis defects with Aurora B inhibition, because cyclin B1 degradation is cell-cycle regulated toward telophase and cytokinesis. Barasertib as an Aurora B–selective inhibitor did not show these target gene expression increase, and quite opposite, it induced histone H3 P-Ser10 decrease from its potent Aurora B kinase inhibition (Fig. 3C, left). LY3295668 exhibited dose-dependent PARP cleavage (apoptosis induction) across the concentration range tested (0.25–5 μmol/L), whereas alisertib induced apoptosis dose-dependently up to 0.25 μmol/L (most Aurora A–selective concentration), which then decreased at 1 and 5 μmol/L (Aurora B inhibition), and barasertib only induced moderately (Fig. 3C). Longer treatment for 48 hours showed similar trend (Fig. 3C, right).

Several recent studies show Aurora A functions in EMT and tumor stemness (32–34). We tested whether LY3295668 could also prevent EMT and reduce cancer stemness in a validated model cell line, NCI-H358 (28). As demonstrated previously, TGFβ inhibitor SB431542 reversed the TGFβ-induced epithelial marker E-cadherin decrease (Fig. 3D). LY3295668, but not alisertib or barasertib, was able to partially prevent the TGFβ-induced E-cadherin decrease at 0.1 μmol/L (Fig. 3D). Mesenchymal marker vimentin also showed a similar trend as seen with E-cadherin, albeit to a lesser degree (Supplementary Fig. S5). LY3295668 and alisertib inhibited Sox2 expression at the same concentration (Fig. 3D), indicating Aurora A inhibitors can reduce cancer cell stemness. However, Aurora B inhibitor barasertib did not cause a significant Sox2 decrease at the same concentration (Fig. 3D).

The Aurora A versus Aurora B selectivity of these compounds also impacted genomic stability as determined by flow cytometry analysis (Fig. 3E). LY3295668 induced strong G₂–M arrest after 24 hours in HeLa cells (4N peak). The mitotic arrest persisted up to 48 hours without significant progression to 8N DNA content (tetraploid from the second round of genome replication). Although both alisertib and barasertib initially caused mitotic arrest by 24 hours treatment, they accumulated tetraploid (8N) cells after 48-hour treatment (2 doublings for HeLa cells) through endoreduplication and cytokinesis failure; a profile attributed to Aurora B inhibition. These polyploid cells can survive multiple rounds of endoreduplication without fully engaging apoptosis or other cell death pathway activation, which explains the surviving cells in Supplementary Fig. S4.

LY3295668 showed significant inhibition of cancer cell viability across a panel of 80 tumor lines (Fig. 4), mostly from breast and lung cancers. The potency of LY3295668 in cell lines ranged from IC₅₀ single-digit nmol/L to greater than 20 μmol/L (Fig. 4A). Fifty-five of 80 (68%) cell lines displayed significant sensitivity (IC₅₀ < 1 μmol/L) to LY3295668 with an average IC₅₀ of 0.048 μmol/L (Fig. 4A). Twenty-eight of 34 (82%) of breast, as well as 18/27 (67%) of lung cancer cell lines, are sensitive to LY3295668 treatment and have IC₅₀s below 1 μmol/L. The average IC₅₀s were 0.039 and 0.034 μmol/L, respectively, for the sensitive breast and lung panels (Fig. 4B and C). In breast cancer cells, triple-negative breast cancer was the most sensitive subtype with average IC₅₀ 0.164 μmol/L, and 13/15 (87%) lines had IC₅₀ below 1 μmol/L, while luminal was the most insensitive with an average IC₅₀ 3.653 μmol/L and 13/19 (68%) had IC₅₀ less than 1 μmol/L (Fig. 4B). In lung cancer cells, SCLC was the most sensitive tumor type with an average IC₅₀ of 0.016 μmol/L and all 6 (100%) SCLC cell lines were sensitive to LY3295668 (Fig. 4C). Twelve of 21 (57%) of non–small cell lung cancer were sensitive to LY3295668 with an average IC₅₀ of 4.694 μmol/L (Fig. 4C).

A series of 10 compounds with Aurora A and/or Aurora B activity were tested in a bone marrow colony formation assay to better understand the heme toxicity relative Aurora kinase selectivity. LY3295668 showed an IC₉₀ of 1.4 μmol/mL, while alisertib and barasertib gave IC₉₀s at 0.2 and 0.1 μmol/mL, respectively. Multiple regression analysis indicates that human bone marrow cell inhibition was correlation with Aurora B cell potency, but not Aurora A (Supplementary Fig. S6). This would suggest that the heme toxicity observed in rodents and clinically (alisertib, barasertib) may be driven by the Aurora B activity. As such, an Aurora A–selective inhibitor might be better tolerated (also see ref. 35).

LY3295668 demonstrated potent Aurora A inhibition in vivo in the NCI-H446 xenograft model using nude mouse and rat (Fig. 5A) at 3 hours after oral dosing. At 30 mg/kg oral dosing in mice, LY3295668 also showed persistent Aurora A target inhibition (>90% P-Aurora A inhibition) for more than 8 hours and induced significant mitotic arrest as indicated by increases in mitotic maker histone H3 Ser10 phosphorylation (Fig. 5B, left). At 50 mg/kg, LY3295668 effectively inhibited Aurora A > 90% for 12 hours (96.0%, 95.1%, 94.5%, 90.7%, and 77.5% at 1, 4, 8, 12, and 24 hours, respectively). Correspondingly, mitotic arrest persisted up to 16 hours postdosing with histone H3 P-Ser10 increasing by 237%, 323%, 330%, 245%, and 328% at 1, 4, 8, 12, and 24 hours, respectively. LY3295668 has a calculated threshold effective dose (TED₅₀: dose that produces an effect equal to 50% P-AurA inhibition) at 2.13 mg/kg and threshold effective concentration (TEC₅₀: concentration that produces an effect equal to 50% P-AurA inhibition) of 0.07 μmol/L. We estimated that with twice-a-day dosing schedule, a dose of 30 mg/kg might be expected to achieve >90% Aurora A target inhibition for more than 16 hours daily, whereas a dose of 50 mg/kg would achieve this level of target engagement for the entire dosing period. At exposures comparable with those in the mouse, a 5 mg/kg oral dose in rats, LY3295668 showed persistent Aurora A target inhibition and a corresponding histone H3 P-Ser10 increase (Fig. 5B, right).

LY3295668 at these doses did not inhibit histone H3 P-Ser10 at the 1-hour time point (Fig. 5B), indicating it did not show Aurora B inhibition in xenograft tumors. Taken together with the previous kinase profiling data (Fig. 1; Supplementary Fig. S1), LY3295668 consistently demonstrated that it is a potent and selective Aurora A inhibitor in vivo.

Inhibition of cell proliferation inhibition by LY3295668 was time-dependent, that is potency increased with longer incubation time (4–48 hours) and maximal inhibition reached after 24 to 48 hours (Supplementary Table S1; Supplementary Fig. S7). After 48 hours of LY3295668 treatment, proliferation IC₅₀s were 0.090 ± 0.021 (n = 2), 0.108 ± 0.034 (n = 2), and 0.076 ± 0.033 (n = 2) μmol/L, respectively, in NCI-H446, HeLa, and Calu6 cells (Supplementary Table S1). The time-dependent cell proliferation inhibition in combination with the notion that 90% phospho-Aurora A inhibition was needed (Supplementary Fig. S2) enabled us to design an optimized in vivo efficacy schedule. We subsequently conducted in vivo efficacy studies by achieving >90% phospho-Aurora A inhibition for at least 16 hours/day (30 mg/kg) or 24 hours/day (50 mg/kg) in mice (Fig. 6A and B).

LY3295668 demonstrated significant tumor growth inhibition in SCLC NCI-H446 xenograft mode in a dose-dependent manner (Fig. 6A, left). Tumors treated with 30 mg/kg LY3295668 regrew once dosing had stopped, whereas the tumors receiving 50 mg/kg LY3295668 did not show significant regrowth until 14 days postdosing (day-55 total duration, Fig. 6A, left). LY3295668 was well tolerated, with no significant body weight loss in the study.

We varied dosing schedule to understand the duration of LY3295668 dosing needed to reach durable efficacy in SCLC NCI-H69 xenograft model. LY3295668 at 50 mg/kg, dosed either as (BID×7, rest 14) × 2, (BID×14, rest 7) × 2, or (BID×21) × 2 schedules, all produced significant tumor growth inhibition (Fig. 6A, middle). Although the effect of (BID×7, rest 14) schedule did not persist after dosing stopped, both (BID14, rest 7) and BID×21 dosing schedules similarly produced sustained tumor growth inhibition; lasting more than 40 days (109-day total duration) postdosing (Fig. 6A, middle). We also tested comparable daily dose scenarios by dosing LY3295668 at 90 mg/kg BID×7, 50 mg/kg BID×14, or 30 mg/kg BID×21 (Fig. 6A, right). The results confirmed that constant, complete target inhibition (50 mg/kg BID×14) produced the most robust antitumor efficacy (Fig. 6A right) that persisted for at least 40 days postdosing (day-80 total duration). Should a dosing holiday be required, LY3295668 administered at concentrations achieving >90% Aurora A inhibition on a 14-day-on/7-day-off schedule would still be expected to reach maximal efficacy in SCLC (Fig. 6A).

We further demonstrated LY3295668 antitumor efficacy using SCLC patient-derived xenografts (PDX) models. In LXFS538, LY3295668 treatment induced 61.1% of tumor regression at day 56 of treatment whereas the standard-of care (etoposide + cisplatin) was only moderate at 43.1% tumor growth inhibition (Fig. 6B, left). When there was a significant tumor regrowth from 1 of the 5 tumors at day 70, LY3295668 retreatment at 50 mg/kg showed tumor regression again (Fig. 6B, left). In LXFS2156, LY3295668 treatment at 50 mg/kg BID×28 showed 97.2% of tumor growth inhibition whereas the standard-of-care did not show any significant tumor inhibition (Fig. 6B, right). However, the tumor inhibition in LY3295668-treated group did not persist beyond dosing period (Fig. 6B, right).

Efficacy of LY3295668 was also explored in combination with various standards of care. In LXFS2156, LY3295668 at 40 mg/kg showed significant efficacy, whereas standard of care (Etoposide + Cisplatin) did not significantly affect tumor growth (Fig. 6C, left). LY3295668 in combination with standard of care produced better-than-additive effect than either monotherapy (Fig. 6C, left). In LXFS573, LY3295668 also showed significant combination benefit with standard-of-care agents (Fig. 6C, middle). In these studies, 10% to 15% of body weight losses were observed in combination groups.

Multiple lines of data indicate potential synergism between Aurora A and EGFR inhibitors (36). Consequently, LY3295668 was investigated for potential synergy in combination with cetuximab and cisplatin in a head and neck squamous cell carcinoma (HNSCC), the Detroit 562 xenograft model (Fig. 6C, right). At the end of dosing period (day 44 total duration; 4-day postdosing regimen completed), LY3295668 at 30 mg/kg alone showed significant tumor growth inhibition, so did cetuximab or cetuximab plus cisplatin treatment. More importantly, LY3295668 in combination with cetuximab or with both cetuximab plus cisplatin exhibited a synergistic effect in comparison with either monotherapy, produced much prolonged tumor growth inhibition beyond the dosing period, especially LY3295668 with cetuximab + cisplatin group (Fig. 6C, right).

Although Aurora kinases A, B, and C are highly homologous, they are not functional orthologs. Aurora A–selective inhibition leads to mitotic arrest in many cancer cells. The mitotic arrest often further progresses to cancer cell apoptosis/death in sensitive cancer cell lines. We contend that an Aurora A–selective inhibitor such as LY3295668 may be more effective than Aurora B or Aurora A/B dual inhibitors (alisertib, barasertib), as it would avoid DNA endoreduplication and polyploidy, which could persist for multiple cell divisions without completely killing the cancer cells (37). A couple of Aurora A inhibitors have been identified over the years; however, they suffer either the lack of Aurora A inhibition potency or lack of enough selectivity against Aurora B (Fig. 2; refs. 24, 26). LY3295668 is a potential best-in-class Aurora A inhibitor with >1,000-fold Aurora A versus Aurora B selectivity. Furthermore, we contend that at the pharmacologically relevant concentrations, LY3295668 only inhibits Aurora A in vitro and in vivo, not Aurora B.

Using LY3295668, we discovered that Aurora A versus Aurora B/C inhibitors show profound differences in mitotic arrest phenotypes and especially apoptosis induction. Although LY3295668, alisertib, and barasertib each can potently inhibit HeLa cell growth, only LY3295668 consistently promoted apoptosis across the range of tested concentrations, as well as inhibits EMT and cancer cell stemness markers (Figs. 2 and 3). Interestingly, alisertib inhibits Aurora A selectively in a range of 0.080–0.250 μmol/L. This corresponds to the same range of concentrations needed to induce maximal apoptosis in HeLa cells. At concentrations >0.25 μmol/L, alisertib reduced its apoptosis induction capacity as it switched to an Aurora B–dominant phenotype (Figs. 2 and 3). So at a clinically relevant human plasma exposure of 1 to 2 μmol/L (38–41), alisertib behaves phenotypically as an Aurora B inhibitor in vivo (42). This phenotype has been observed previously in sarcoma cells, although alisertib showed an Aurora A inhibition phenotype (increased phospho-H3) in their model (42). Although alisertib is a dual Aurora A/B inhibitor in vitro, the appearance of the Aurora B phenotype is exposure-dependent, that is, at lower plasma concentrations, only Aurora A may be inhibited.

Aurora B inhibition–induced endoreduplication and subsequent polyploidy are retinoblastoma protein-dependent and can lead to chromosome segregation errors and thus genomic instability through aberrant mitosis (Figs. 2 and 3; refs. 17, 18, 23, 26, 35, 43). As such, Aurora B inhibitor–induced tetraploid cells, which show better survival after compound treatment (Figs. 2 and 3), could significantly increase the refractory/resistant cell population, especially when the inhibitor in vivo exposure is at subthreshold levels during dosing intervals. Aneuploidy/genomic instability is one of the most prevalent signatures of cancer (44, 45). Polyploidy, including tetraploidy, is thought to be a major pathway to induce aneuploidy and genomic instability. For example, RAS oncogene induces tetraploidy as an intermediate step in its tumorigenesis (46). As for cancer therapy, Aurora A–selective inhibitor may be the best strategy and Aurora B activity should be avoided. LY3295668 has now advanced to phase I clinical testing to test the translation of this hypothesis in patients (NCT03092934).

J. Du is a research advisor at Eli Lilly & Company. L. Yan is a research scientist at Eli Lilly & Company. R. Torres is a senior research scientist Eli Lilly & Company. X. Gong is a principal research scientist at Eli Lilly & Company. C. Marugán is a research associate at Eli Lilly & Company. C. Baquero is a research associate at Eli Lilly & Company. S.C. Chapman is a senior research scientist and has ownership interest (including patents) in Eli Lilly & Company. Y. Zeng is a consultant at Eli Lilly & Company. G.P. Donoho is a director, oncology research, In Vivo Pharmacology and has ownership interest (including patents) in Eli Lilly & Company. R.D. Van Horn is a consultant biologist and has ownership interest (including patents) in Eli Lilly & Company. D.S. Barnard is a consultant biologist and has ownership interest (including patents) in Eli Lilly & Company. R.P. Beckmann is a research scientist and has ownership interest (including patents) in Eli Lilly & Company. L.-C. Chio is a biology consultant at Eli Lilly & Company. A. Aggarwal is a Head, Oncology Genomics and Bioinformatics and has ownership interest (including patents) in Eli Lilly & Company. L.F. Stancato is a research fellow at Eli Lilly & Company. P.W. Iversen is a senior research advisor at Eli Lilly & Company. J.R. Manro is a senior research scientist and has ownership interest (including patents) in Eli Lilly & Company. J.L. Walgren is an advisor of regulatory affairs at Eli Lilly & Company. B.W. Halstead is a consultant oncology BDS&E at Eli Lilly & Company. M.Z. Dieter is a principal research scientist at Eli Lilly & Company. S.V. Bhagwat is a senior research advisor at Eli Lilly & Company. E.L. Kreklau is a senior director at Eli Lilly & Company. B.K.R. Patel is a research fellow and has ownership interest (including patents) in Eli Lilly & Company. C. Reinhard is a senior director at Eli Lilly & Company. G.D. Plowman is a vice president, oncology research at Eli Lilly & Company. D.A. Barda is a research scientist and has ownership interest (including patents) in Eli Lilly & Company. J.R. Henry is a senior research fellow and has ownership interest (including patents) in Eli Lilly & Company. S.G. Buchanan has ownership interest (including patents) in Eli Lilly & Company. R.M. Campbell is a research fellow and has ownership interest (including patents) in Eli Lilly & Company. No potential conflicts of interest were disclosed by the other authors.

Conception and design: J. Du, L. Yan, X. Gong, S.C. Chapman, R.P. Beckmann, M.S. Marshall, P.W. Iversen, E.L. Kreklau, M.J. Lallena, X.S. Ye, C. Reinhard, G.D. Plowman, D.A. Barda, J.R. Henry, S.G. Buchanan, R.M. Campbell

Development of methodology: J. Du, L. Yan, R. Torres, X. Gong, H. Bian, K. Boehnke, R.T. Foreman, R.P. Beckmann, M.S. Marshall, L.-C. Chio, S. Chu, S. Bhattachar, L.F. Stancato, J.L. Walgren, S.V. Bhagwat, E.L. Kreklau, M.J. Lallena, X.S. Ye

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J. Du, R. Torres, X. Gong, H. Bian, C. Marugán, K. Boehnke, C. Baquero, Y.-H. Hui, Y. Yang, Y. Zeng, S.M. Bogner, A. Capen, G.P. Donoho, R.D. Van Horn, D.S. Barnard, M.S. Marshall, Y. Qian, A. Aggarwal, M.S. Dowless, J.L. Walgren, M.Z. Dieter, M.J. Lallena

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J. Du, L. Yan, R. Torres, X. Gong, C. Marugán, K. Boehnke, C. Baquero, Y.-H. Hui, S.C. Chapman, R.D. Van Horn, Y.W. Webster, A. Aggarwal, L.F. Stancato, P.W. Iversen, J.R. Manro, J.L. Walgren, B.W. Halstead, R. Martinez, M.J. Lallena, B.K.R. Patel, D.A. Barda, J.R. Henry, S.G. Buchanan, R.M. Campbell

Writing, review, and/or revision of the manuscript: J. Du, R. Torres, Y.-H. Hui, S.C. Chapman, G.P. Donoho, R.P. Beckmann, A. Aggarwal, L.F. Stancato, P.W. Iversen, J.R. Manro, B.W. Halstead, M.Z. Dieter, R. Martinez, E.L. Kreklau, M.J. Lallena, X.S. Ye, B.K.R. Patel, G.D. Plowman, D.A. Barda, J.R. Henry, S.G. Buchanan, R.M. Campbell

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J. Du, R. Torres, Y.-H. Hui, J.A. Dempsey, J.R. Manro, D.A. Barda

Study supervision: J. Du, A. Capen, G.P. Donoho, X.S. Ye, C. Reinhard, G.D. Plowman, S.G. Buchanan, R.M. Campbell

We would like to thank Drs. James Starling, Alfonso de Dios, Michal Vieth, Daphne L. Farrington, Harold B. Brooks, John Toth, Farhana Merzoug, Karen L. Cox, Susan A. Konicek, and Teresa Burke for their guidance and support. This work was supported by Eli Lilly & Company.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.