MDM2 antagonist clinical response association with a gene expression signature in acute myeloid leukaemia

Acute myeloid leukaemia (AML) is uniquely sensitive to p53 activation ^{1, 2} as ≈90% of patients carry wild-type TP53 and frequent MDM2 overexpression.³ MDM2 blocks p53 transactivation and targets p53 for ubiquitin-dependent degradation.^{4, 5} Nutlins have been characterized as potent and selective small-molecule MDM2 antagonists.^{1, 6–8} RG7112 was the first such MDM2 antagonist to undergo clinical assessment in solid tumors and leukaemia trials.^{1, 2, 9} As not all patients with functional p53 will respond to MDM2 antagonists, diagnostic tools may identify patients likely to respond.

To establish an in vitro MDM2 antagonist therapy-predictive mRNA signature, we assessed genome-wide associations between growth inhibitory effects of RG7112 (IC₅₀s) among 287 human cancer cell lines and pretreatment RNAseq-derived transcript levels (Table S1, Figure S1). Thirty-five candidate genes were identified with significance at false discovery rate 0.05 through two approaches: comparing mRNA expressions between sensitive (IC₅₀<1µM) and resistant cell lines (IC₅₀>10µM); or Spearman correlation between mRNA expressions and IC₅₀s (Table S2). Functional annotation indicated that many significant genes were known regulators of the MDM2-p53 interaction or downstream p53 pathways, including cell-cycle arrest and apoptosis. Among these, MDM2 demonstrated association between overexpression and in vitro sensitivity to RG7112 (Spearman correlation coefficient=−0.39; P<0.001). A multiple logistic regression classifier was identified comprising high expression of MDM2, XPC, BBC3 (PUMA), and low expression of tumor suppressor gene CDKN2A (Table S3). This signature score (G_MDM2 + G_XPC + G_BBC3 – G_CDKN2A at baseline), associated with cell-line response to MDM2 antagonist (P<0.001) and discriminated sensitive from resistant cell lines (area under the curve [AUC] = 0.92; 95% CI, 0.87–0.97; Table 2, S3 and Figure S2). In addition to MDM2, the other 3 signature components are regulators of the MDM2–p53 interaction or downstream p53 pathways. XPC is key in repairing damaged DNA. BBC3 (PUMA) is induced by exposure to DNA-damaging agents and by activated p53, and mediates apoptosis. CDKN2A gene, comprising p16 and p14ARF, is linked to tumor suppressor pathways, inhibiting MDM2 function by nucleolus sequestering.¹⁰ Cell lines with low signature scores trended with p53 mutation, whereas cell lines with high signature score trended with p53 wild type (P<0.001; Figure S2). Multivariate logistic models indicated signature scores remained significant (P<0.001) when adjusted for TP53 mutation status.

Table 2.

Predictions from various predictive biomarkers.

	Oncology Cell Lines Collectionsa	NO21279b (derived from blood samples)	NP28679b (derived from blood samples)
Score
AUC (95% CI)	0.92 (0.87–0.97)	0.86 (0.71–1.00)	0.90 (0.76–1.00)
Specificityc	0.9	0.65	0.71
Sensitivityd	0.87	1	1
TP53
AUC (95% CI)	0.87 (0.81–0.93)	0.61 (0.52–0.70)	0.56 (0.36–0.74)
Specificitye	0.95	0.22	0.25
Sensitivityf	0.8	1	0.86
MDM2
AUC (95% CI)	0.84 (0.77–0.90)	0.60 (0.36–0.83)	0.77 (0.52–1.00)
Specificityc	0.75	0.35	0.86
Sensitivityd	0.8	1	0.71

AUC, area under the curve; IC₅₀, half maximal inhibitory concentration; MDM2, murine double minute 2.

^aResponders defined as IC₅₀ <1 and nonresponders defined as IC₅₀ >10 in cell lines.

^bResponders defined as patients having bone marrow blasts < 5% after treatment and nonresponders defined as patients having bone marrow blasts >= 5% after treatment in NO21279 and NP28679.

^cSpecificity: proportion of nonresponders who had scores or MDM2 expression lower than the corresponding Youden index.

^dSensitivity: proportion of responders who had scores or MDM2 expression higher than the corresponding Youden index.

^eSpecificity: proportion of nonresponders who had TP53 mutations.

^fSensitivity: proportion of responders who had wild-type TP53.

RG7112 was assessed in phase 1 dose escalation trial NO21279 (patients with relapsed/refractory leukaemia; Figure S1;Table 1). Enrollment criteria are detailed in Supplement. Clinical response in NO21279 was as follows: responders were patients whose bone marrow blasts were <= 5% after treatment; >5% blasts were non-responders. mRNA expressions in blood leukaemia samples and bone marrow aspirate samples were profiled from 28 evaluable patients treated at the maximum tolerated dose (1500 mg/m² twice daily × 10 days) at pretreatment, after a single dose (cycle 1 day 2 [C1D2], blood only), and on last day of dosing (cycle 1 day 10 [C1D10]). Signature scores from pretreatment blood samples associated with clinical response (P=0.005; Table 1; Figure S2) and with pharmacodynamic biomarker response, defined as change in MDM2 mRNA expression in blood (Spearman correlation coefficient 0.41; P=0.02; Figure S3). Signature scores distinguished response with AUC=0.86 (95% CI, 0.71–1.00); higher than AUCs of TP53 mutation status or MDM2 mRNA expression in blood as individual biomarkers (Table 2). Using a signature score cutpoint selected by Youden index, patients were classified by response prior to MDM2-antagonist therapy with 100% sensitivity and 65% specificity. TP53-mutant patients showed a trend of lower signature scores than TP53–wild-type patients, although not significant (P=0.068; Table S4). Furthermore, signature scores of TP53-wild-type responders are significantly higher than TP53-wild-type non-responders (P=0.006; Table S4), demonstrating additional discriminative power of the proposed signature in TP53-wild-type patients. Correlation (P=0.02) was observed between clinical response and signature score in multiple logistic regression with both TP53 mutation status and signature. Taken together, these data indicate the signature score can potentially serve as an indicator of MDM2–p53 pathway function, with added predictive value beyond TP53 status for AML patients.

Table 1.

Cohort Characteristics of AML Patients in the two clinical trials

NO21279 (N=28)
	Non-Responder	Responder	P value
Sample size	23	5
Median (IQR) age (years)	60.0 (34.5, 67.0)	58.0 (48.0, 65.0)	0.83a
Female, n (%)	7 (44)	3 (60)	0.32b
TP53 mutations, n (%)	5 (22)	0	0.51b
Median (IQR) mRNA signature score at baseline derived from microarray measurements in blood samples	15.2 (14.8, 15.8)	16.4 (16.0, 16.5)	0.005a
NP28679 (N=21)
	Non-Responder	Responder	P value
Sample size	14	7
Median (IQR) age (years)	64.0 (52.0, 73.5)	70.0 (61.5, 71.5)	0.55a
Female, n (%)	7 (50)	4 (57)	1.00b
TP53 mutations, n (%)	3 (25)	1 (14)	1.00b
Median (IQR) mRNA signature score at baseline derived from RT-PCR meansurements in blood samples	4.0 (3.5, 4.6)	5.2 (5.0, 5.5)	0.001a

C1D10, cycle 1 day 10; IQR, interquartile range; MDM2, murine double minute 2; RT-PCR, real-time polymerase chain reaction.

^aP values are derived by Wilcoxon rank-sum test.

^bP values are derived by Fisher exact test.

We also sought to determine if the 4-gene signature may provide pharmacodynamic metrics for assessing clinical activity consistent with the intended mechanism of action Relative median expression of MDM2, XPC, BBC3, and CDKN2A mRNA in blood samples from C1D10 were 2.51-fold higher (fold change [FC]; interquartile range [IQR], 1.69–5.05), 1.75 FC (IQR, 1.25–2.07), 1.62 FC (IQR, 1.10–2.01), and 0.73 FC (IQR, 0.62–0.92) over baseline, respectively, consistent with the intended MDM2-antagonist mechanism of action for . The mRNA signature scores significantly differ based on response when measured on C1D2 (P=0.013) and on C1D10 (P=0.01; Figure S4). The mRNA signature score showed consistence in blood samples and bone marrow aspirate for the same patient at baseline (R=0.50; P=0.016; Figure S5). Furthermore, strong concordance between MDM2 expressions in 28 patients measured under 2 platforms, microarray and quantitative real-time polymerase chain reaction, was observed (R=0.5; P=0.019; Figure S6).

We further evaluated the signature with a pharmacologically optimized next-generation MDM2 antagonist RG7388 using pretreatment specimens from a phase 1 study NP28679 (AML patients with relapsed/refractory disease following induction chemotherapy or unsuitable for standard induction therapy;Table 1 and S4). Twenty-one patients receiving RG7388 in combination with cytarabine (Figure S1) were evaluable. Clinical endpoints were defined with the same criterion to NO21279. Consistent with previous findings, the signature scores, derived from qRT-PCR of MDM2, XPC, BBC3, CDKN2A in blood leukaemia samples at baseline, were associated with clinical response (P=0.001;Table 1; Figure S2). The signature scores distinguished responders from non-responders with AUC=0.90(95% CI, 0.76–1.00;Table 2); higher than AUCs of TP53 mutation status or MDM2 mRNA expression (Table 2). Using a signature score cutoff selected by the Youden index, patients may be discriminated by predicted response prior to the therapy with 100% sensitivity and 71% specificity. Correlation was observed again between signature score and response (P=0.02) in multiple regression with TP53 mutation status and signature.

In summary, we demonstrate a biological classifier discriminates response broadly to MDM2-antagonist therapy. The level of evidence attained by cell line efficacy modeling and response assessments in trials NO21279 and NP28679 (with MDM2 antagonists RG7112 and RG7388, respectively) adds substantial weight to the validity of this panel.