Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2023 Feb 1.
Published in final edited form as: Leukemia. 2022 Jun 10;36(8):2022–2031. doi: 10.1038/s41375-022-01622-0

A Ten-Gene DNA-Damage Response Pathway Gene Expression Signature Predicts Gemtuzumab Ozogamicin Response in Pediatric AML Patients Treated on COGAAML0531 and AAML03P1 Trials

Mohammed O Gbadamosi 1, Vivek Shastri 1, Abdelrahman H Elsayed 1, Rhonda Ries 2, Oluwaseyi Olabige 1, Nam Nguyen 1, Angelica De Jesus 1, Yi-Chen Wang 3, Alice Dang 3, Betsy A Hirsch 4, Todd A Alonzo 3,5, Alan Gamis 6, Soheil Meshinchi 2, Jatinder K Lamba 1,7
PMCID: PMC9357169  NIHMSID: NIHMS1813919  PMID: 35688939

Abstract

Gemtuzumab ozogamicin (GO) is an anti-CD33 monoclonal antibody linked to calicheamicin, a DNA damaging agent, and is a well-established therapeutic for treating acute myeloid leukemia (AML). In this study, we used LASSO regression modeling to develop a 10-gene DNA damage response gene expression score (CalDDR-GEx10) predictive of clinical outcome in pediatric AML patients treated with treatment regimen containing GO from the AAML03P1 and AAML0531 trials (ADE+GO arm, N = 301). When treated with ADE+GO, patients with a high CalDDR-GEx10 score had lower complete remission rates (62.8% vs. 85.5%, P = 1.77*10−5) and worse event-free survival (28.7% vs. 56.5% P = 4.08*10−8) compared to those with a low CalDDR-GEx10 score. However, the CalDDR-GEx10 score was not associated with clinical outcome in patients treated with standard chemotherapy alone (ADE, N = 242), implying the specificity of the CalDDR-GEx10 score to calicheamicin-induced DNA damage response. In multivariable models adjusted for risk group, FLT3-status, white blood cell count, and age, the CalDDR-GEx10 score remained a significant predictor of outcome in patients treated with ADE+GO. Our findings present a potential tool that can specifically assess response to calicheamicin-induced DNA damage preemptively via assessing diagnostic leukemic cell gene expression and guide clinical decisions related to treatment using GO.

Keywords: acute myeloid leukemia, DNA-damage, gemtuzumab ozogamicin, calicheamicin, gene expression, pediatric

Introduction

Acute myeloid leukemia (AML) is a heterogeneous malignancy characterized by impairment in the proliferation, differentiation, and self-renewal capabilities of myeloid stem and progenitor cells. For over four decades, the standard induction regimen for AML has been a combination of cytarabine and anthracyclines, however, this regimen fails to induce remission in roughly 10–20% of children, and among those who achieve remission, roughly 40% relapse1. Over the past few years, new agents have been approved for the treatment of AML with one of the most well-established agents being gemtuzumab ozogamicin (GO) which was recently indicated for use in pediatric AML patients. GO is a CD33-directed antibody-drug conjugate (ADC) consisting of a humanized anti-CD33 monoclonal antibody, hP67.6, covalently linked to N-acetyl-γ-calicheamicin (hereafter referred to as calicheamicin), a potent antitumor cytotoxin2,3. In 2017, GO received FDA reapproval for the treatment of newly diagnosed and relapsed/refractory AML4,5. While the future of GO as a therapeutic in AML is bright, studies have shown that clinical response to GO is subject to interpatient variability6. Hitherto, efforts to understand the interpatient variation in response to GO have focused on the association between CD33 cell surface expression levels and outcome, and more recently the association between genetic variations in CD33 and ABCB1 and response to GO have been reported7–11. In particular, our previous findings identified significant associations between the rs12459419 CD33 splicing polymorphism and the rs1045642 ABCB1 polymorphism and response to GO9,11. Nevertheless, efforts to understand the interpatient variability regarding the DNA-damaging effects of calicheamicin and response to those effects have been limited.

One of the critical components of GO is calicheamicin, an enediyne antitumor antibiotic that binds to the minor groove of DNA at preferred sites of AGGA, TCCT, and TCCA. Once bound, calicheamicin induces DNA damage, which is mostly double-stranded breaks (DSBs), leading to cellular apoptosis12–14. In response to DSB DNA damage, three major pathways can be activated, the homologous recombination pathway, and the canonical and alternative non-homologous end joining (NHEJ) pathways. Given that calicheamicin typically induces DSBs containing just three base pairs of homology, the latter two pathways are more often than not activated in response to calicheamicin-induced DNA damage15,16. The NHEJ pathways rely upon docking proteins for the recruitment of nucleases, ligases, and other downstream checkpoint and apoptosis signaling proteins to mediate DNA damage repair and the subsequent cellular responses17. Among these factors are DNA-damage sensors, such as ATM and ATR, p53 downstream targets including various members of the Bcl-2 family, and various proteins for processing DNA damage such as DNA-PK and XRCC418. In the context of GO, expression, and activity of some of these genes, such as AKT1, CHEK1 and CHEK2, PRKDC, and various anti-apoptotic Bcl-2 family members, have been shown to impact the efficacy of calicheamicin in mechanistic in vitro studies; however, there are no studies to date evaluating the expression levels of these genes and response to GO19–22.

In this study, we evaluated the impact of diagnostic leukemic cell gene expression levels of 18 genes in DNA-damage response pathways on the clinical outcome of pediatric AML patients treated who received GO in addition to standard chemotherapy (ADE+GO, N = 301).

Results

CalDDR-GEx10 score development in patients treated with ADE+GO:

The overall study design is shown in Fig 1. As mentioned above, 10 genes passed the 95% threshold and were represented in at least 950 of the 1000 cross-validated iterations (Supplemental Fig. S1) of the LASSO Cox regression model using gene expression levels and event free survival (EFS) data from AML patients treated with ADE+GO (N = 301). The CalDDRGEx10 score was thus computed for each patient by multiplying the expression levels of each of the respective gene by the final coefficient values obtained from the average of the coefficient estimates from the 1000 cross-validated iterations as defined in the equation below:

Figure 1.

Figure 1.

Open in a new tab

Overall study design.

CalDDR-GEx10score=AKT1×−0.192+ATR×0.345+BAD×0.306+BAK1×0.152+BCL2L1×−0.056+CASP9×−0.190+H2AFX×−0.116+PARP1×−0.112+PRKDC×−0.078+XRCC4×0.247.

Using recursive partitioning, patients in ADE+GO group were further dichotomized into high (N = 94; 31%) and low (N = 207; 69%) CalDDR-GEx10 score groups (score cutoff = 0.207). Likewise, patients treated in the ADE arm of AAML0531 were also dichotomized into high (N = 61; 25%) and low (N = 181; 75%) score groups using the same CalDDR-GEx10 score cutoff value. Table 1 shows the patient characteristics by high and low score groups for patients treated with ADE+GO and patients treated with ADE alone. Other than initial risk group stratification within ADE+GO treated patients, no significant difference between low and high score groups was observed. Of note, there were no significant differences in the distribution of the score between patients treated with ADE or ADE+GO (Supplemental Fig. S2).

Table 1.

Baseline characteristics of patients in the high and low CalDDR-GEx10 score groups dichotomized by treatment regimen.

ADE + GO (N = 301) ADE (N = 242)
CalDDR-GEx10 Low (N = 207) CalDDR-GEx10 High (N = 94) P CalDDR-GEx10 Low (N = 181) CalDDR-GEx10 High (N = 61) P
Sex (N)
 Female 100/207 (48%) 47/94 (50%) 0.8 88/181 (49%) 31/61 (51%) 0.5
Age (N)
 <10 97/207 (47%) 49/94 (52%) 0.4 92/181 (51%) 31/61 (51%) 0.9
 >10 110/207 (53%) 45/94 (48%) 89/181 (49%) 30/61 (49%)
Race (N)
 American Indian or Alaska Native 1/197 (0.5%) 0/91 (0%) 0.8 1/173 (0.6%) 0/56 (0%) 0.9
 Asian 7/197 (3.6%) 4/91 (4.4%) 11/173 (6.4%) 2/56 (3.6%)
 Black or African American 23/197 (12%) 13/91 (14%) 19/173 (11%) 7/56 (12%)
 Other 3/197 (1.5%) 0/91 (0%) 13/173 (7.5%) 3/56 (5.4%)
 Native Hawaiian/Pacific Islander 6/197 (3.0%) 4/91 (4.4%) 0/173 (0%) 0/56 (0%)
 White 157/197 (80%) 70/91 (77%) 129/173 (75%) 44/56 (79%)
 Unknown 10 3 8 5
Risk Group (N)
 Low 93/202 (46%) 25/89 (28%) 0.002 74/177 (42%) 23/59 (39%) 0.5
 Standard 87/202 (43%) 42/89 (47%) 70/177 (40%) 28/59 (47%)
 High 22/202 (11%) 22/89 (25%) 33/177 (19%) 8/59 (14%)
 Unknown 5 5 4 2
Clinical Trial (N)
 AAML03P1 48/207 (23%) 14/94 (15%) 0.1 0 0 NA
 AAML0531 159/207 (77%) 80/94 (85%) 181 61
Clinical Features ({n/N} {%})
WBC 0.8 0.8
 <30% 93/207 (45%) 41/94 (44%) 86/181 (48%) 28/61 (46%)
 >30% 114/207 (55%) 53/94 (56%) 95/181 (52%) 33/61 (54%)
FLT3 Status 0 >0.9
 Wildtype 156/207 (75%) 63/94 (67%) 139/181 (77%) 49/61 (80%)
 Point Mutation 24/207 (12%) 4/94 (4.3%) 8/181 (4.4%) 2/61 (3.3%)
 ITD 27/207 (13%) 27/94 (29%) 34/181 (19%) 10/61 (16%)
Cytogenetic Features (N)
 Normal 49/201 (24%) 26/87 (30%) 0.09 45/175 (26%) 12/59 (20%) 0.9
 inv(16) 37/201 (18%) 6/87 (6.9%) 24/175 (14%) 9/59 (15%)
 t(8;21) 31/201 (15%) 10/87 (11%) 24/175 (14%) 9/59 (15%)
 KMT2A 33/201 (16%) 18/87 (21%) 29/175 (17%) 11/59 (19%)
 Other 51/201 (25%) 27/87 (31%) 53/175 (30%) 18/59 (31%)
 Unknown 6 7 6 2
CD33 rs12459419 ({n/N} {%})
 CC 99/197 (50%) 49/88 (56%) 0.4 90/179 (50%) 22/60 (37%) 0.067
 CT+TT 98/197 (50%) 39/88 (44%) 89/179 (50%) 38/60 (63%)
 Unknown 10 6 2 1
ABCB1 rs1045642 ({n/N} {%})
 CC 52/198 (26%) 23/88 (26%) >0.9 54/178 (30%) 15/60 (25%) 0.4
 CT+TT 146/198 (74%) 65/88 (74%) 124/178 (70%) 45/60 (75%)
 Unknown 9 6 3 1
Open in a new tab

Abbreviations: ADE, standard AML induction therapy using cytarabine, daunorubicin, and etoposide; GO, gemtuzumab ozogamicin; CalDDR-GEx10, Calicheamicin DNA Damage Response Gene Expression Score; WBC, white blood cell count; FLT3, fms-like tyrosine kinase 3; ITD, internal tandem duplication; KMT2A, Lysine Methyltransferase 2A;

The CalDDR-GEx10 score predicts outcome in patients treated with ADE+GO:

Patients within the high CalDDR-GEx10 score group had a significantly lower CR rate as compared to patients within the low CalDDR-GEx10 score group when treated with ADE+GO (62.8% CR vs. 85.5% CR, P = 1.77*10−5; Fig. 2A). The extent of this association was reduced in patients treated with ADE alone, where no significant difference in CR rate was observed between CalDDR-GEx10 score groups (high vs. low score group; 65.6% CR vs. 75.1% CR, P = 0.20; Fig. 2B). When evaluated as a continuous variable, patients treated with ADE+GO who did not achieve CR had a significantly higher CalDDRGEx10 score as compared to those that did achieve CR (Not in CR vs. CR: CalDDRGEx10 score mean ± SD: 0.167 ± 0.329 vs. 0.006 ± 0.335, P = 7.83*10−4; Supplemental Fig. 3A). Such an association was not observed for patients treated with ADE alone (Not in CR vs. CR: CalDDR-GEx10 score mean ± SD: 0.018 ± 0.373 vs 0.015 ± 0.316, P = 0.67; Supplemental Fig. 3B).

Figure 2.

Figure 2.

Open in a new tab

CalDDR-GEx10 score group is associated with complete remission (CR), event-free survival (EFS), and overall survival (OS) in patients treated with ADE +GO but not in patients treated with ADE alone. Stacked bar plots depict CR status by CalDDRGEx10 score group for patients treated with ADE+GO (A) and for patients treated with ADE alone (B). Kaplan-Meier curves depict EFS (C), OS (D), and (relapse risk (RR) E) by CalDDR-GEx10 score group for patients treated with ADE+GO and EFS (F), OS (G), and RR (H) by CalDDR-GEx10 score group for patients treated with ADE alone.

Consistent with these results, patients in the high CalDDR-GEx10 score group had inferior EFS (28.7% vs. 56.5%; HR = 2.43, 95% CI = 1.77–3.34, P = 4.08×10−8; Fig. 2C), overall survival (OS, 51.1% vs 67.6%; HR = 1.85, 95% CI = 1.27–2.71, P = 1.42*10−3; Fig. 2D), and relapse risk (RR, HR = 2.12, 95% CI = 1.39–3.25, P = 5.30*10−4; Fig. 2E) as compared to the patients in low CalDDR-GEx10 score group when treated with ADE+GO. However, no significant difference between CalDDR-GEx10 score groups was observed for EFS (45.9% vs 49.2%; HR = 0.89, 95% CI = 0.59–1.33, P = 0.56; Fig. 2F), OS (64.6% vs 60.7%; HR = 1.03, 95% CI = 0.65–1.65, P = 0.89; Fig. 2G), or RR (HR = 0.97, 95% CI = 0.58–1.64, P = 0.91; Fig. 2H) in patients treated with ADE alone. Consistent results were observed when the CalDDR-GEx10 score was used as a continuous variable. In an unadjusted univariate analysis, each unit increase in the score was significantly associated with inferior EFS (HR = 3.91, 95% CI = 2.40–6.36, P = 3.99*10−8), OS (HR = 2.92, 95% CI = 1.81–5.12, P = 1.81*10−4), and RR (HR = 3.25, 95% CI = 1.94–6.33, P =3.30*10−5) in patients treated with ADE+GO. Such an association between CalDDR-GEx10 score and EFS, OS, or RR was not observed in patients treated with ADE alone (EFS: HR = 1.06, 95% CI = 0.64–1.74, P = 0.83; OS: HR = 1.15, 95% CI = 0.63–2.12, P = 0.65; RR: HR = 1.07, 95% CI = 0.55–2.05, P = 0.85). Similar results were observed in risk-adjusted models as well (Table 2). In multivariable logistic regression models that included well-established prognostic factors, such as molecular and cytogenetic risk group, FLT3 mutational status, age, and WBC at diagnosis, the CalDDR-GEx10 score group remained a significant and independent predictor of the likelihood to fail to achieve complete remission for patients treated with ADE+GO (OR = 2.90, 95% CI = 1.57–5.37, P < 0.001; Supplemental Fig S4A). Similar results were observed when the CalDDR-GEx10 score was tested as a continuous variable (OR = 3.42, 95% CI = 1.35–8.98, P = 0.01; Supplemental Fig S4B). In multivariable hazard models that included the same prognostic factors the CalDDR-GEx10 score group remained a significant and independent predictor of inferior EFS for patients treated with ADE+GO (HR = 1.97, 95% CI =1.40–2.75, P < 0.001; Fig. 3A) and RR (HR = 1.96, 95% CI 1.21–3.17, P = 0.006; Table 3), however, for OS we did not see a significant difference (HR = 1.40, 95% CI = 0.94–2.09, P = 0.10; Fig. 3B). When tested as a continuous variable in multivariable proportional hazard models, the CalDDR-GEx10 score remained a significant and independent predictor of EFS, OS, and RR in response to ADE+GO treatment (EFS: HR = 2.87, 95% CI = 1.73–4.75, P < 0.001; Supplemental Fig. S5A OS: HR = 2.13, 95% CI = 1.18–3.85, P = 0.012; Supplemental Fig. S5B; RR: HR = 3.31, 95% CI 1.71–6.40, P < 0.001; Table 3).

Table 2.

Univariate Cox-proportional hazard regression models of CalDDR-GEx10 score and other variables

Event Free Survival (EFS) Overall Survival (OS) Relapse Risk (RR)
ADE+GO (N=301) ADE (N=242) ADE+GO (N=301) ADE (N=242) ADE+GO (N=234) ADE (N=175)
HR (95% CI) P-value HR (95% CI) P-value HR (95% CI) P-value HR (95% CI) P-value HR (95% CI) P-value HR (95% CI) P-value
Numerical CalDDR-GEx10 score
Unadjusted analysis 3.91 (2.40–6.36) 3.99*10−8 1.06 (0.64–1.74) 0.83 2.92 (1.67–5.12) 1.81*10−4 1.15 (0.63–2.12) 0.65 3.25 (1.94–6.33) 3.30*10−5 1.07 (0.55–2.05) 0.85
Risk group adjusted analysis 2.90 (1.75–4.80) 3.74*10−5 1.10 (0.68–1.80) 0.69 2.09 (1.15–3.77) 0.01 1.17 (0.65–2.12) 0.61 3.17 (1.68–5.99) 3.90*10−4 1.05 (0.51–2.17) 0.90
Categorical CalDDR-GEx10 score
Low CalDDR-GEx10 score - - - - - - - - - - - -
High-CalDDR-GEx10 score 2.43 (1.77–3.34) 4.08*10−8 0.89 (0.59–1.33) 0.56 1.85 (1.27 – 2.71) 1.42*10−3 1.03 (0.65–1.65) 0.89 2.12 (1.39–3.25) 5.30*10−4 0.97 (0.58–1.64) 0.91
Risk Group
Low Risk - - - - - - - - - - - -
Standard Risk 1.81 (1.24–2.63) 2.10*10−3 2.66 (1.74–4.05) 5.47*10−6 2.09 (1.31–3.34) 2.03*10−3 3.49 (2.00–6.07) 9.87*10−6 2.10 (1.32–3.33) 1.60*10−3 2.73 (1.68–4.44) 5.40*10−5
High Risk 3.34 (2.13–5.25) 1.60*10−7 3.64 (2.23–5.95) 2.65*10−7 3.78 (2.20–6.49) 1.43*10−6 4.08 (2.18–7.64) 1.13*10−5 2.31 (1.16–4.61) 0.02 2.35 (1.23–4.48) 9.30*10−3
Cytogenetic Group
Cytogenetically Normal - - - - - - - - - - - -
Inv(16) 0.56 (0.32–0.97) 0.04 0.74 (0.39–1.41) 0.37 0.47 (0.23–0.96) 0.04 0.38 (0.14–1.01) 0.05 0.57 (0.30–1.10) 0.10 1.58 (0.74–3.37) 0.24
t(8;21) 0.49 (0.27–0.89) 0.02 0.53 (0.27–1.07) 0.08 0.60 (0.30–1.19) 0.14 0.47 (0.19–1.18) 0.11 0.22 (0.08–0.61) 3.50*10−3 0.94 (0.43–2.05) 0.88
KMT2A 0.81 (0.50–1.32) 0.40 2.09 (1.25–3.47) 0.005 0.77 (0.43–1.37) 0.37 1.78 (0.96–3.31) 0.07 0.76 (0.41–1.41) 0.38 4.12 (2.08–8.17) 5.00*10−5
Other aberrations 1.05 (0.70–1.58) 0.81 1.28 (0.79–2.06) 0.32 1.03 (0.64–1.67) 0.9 1.61 (0.92–2.82) 0.09 1.06 (0.63–1.76) 0.83 1.53 (0.76–3.10) 0.23
FLT3 Status
FLT3-WT - - - - - - - - - - - -
Point Mutation 0.72 (0.38–1.33) 0.29 1.56 (0.72–3.37) 0.26 0.77 (0.37–1.60) 0.48 0.53 (0.13–2.16) 0.37 0.65 (0.31–1.34) 0.24 1.49 (0.65–3.45) 0.35
ITD 1.94 (1.34–2.80) 4.06 *10−4 1.53 (1.00–2.34) 0.05 1.99 (1.30–3.04) 1.54*10−3 1.28 (0.77–2.13) 0.34 1.42 (0.83–2.42) 0.20 1.33 (0.73–2.41) 0.35
Clinical Features
WBC > 30 (G/L) 1.48 (1.07–2.04) 0.02 1.35 (0.96–1.92) 0.09 1.14 (0.78–1.65) 0.5 1.13 (0.74–1.72) 0.57 1.19 (0.80–1.79) 0.40 1.08 (0.70–1.68) 0.71
Age > 10 Years 1.32 (0.96–1.81) 0.08 1.12 (0.79–1.58) 0.53 1.18 (0.81–1.70) 0.39 1.58 (1.03–2.41) 0.03 1.10 (0.73–1.66) 0.64 0.90 (0.58–1.39) 0.63
Open in a new tab

Abbreviations: ADE, standard AML induction therapy using cytarabine, daunorubicin, and etoposide; GO, gemtuzumab ozogamicin; CalDDR-GEx10, Calicheamicin DNA Damage Response Gene Expression Score; WBC, white blood cell count; FLT3, fms-like tyrosine kinase 3; ITD, internal tandem duplication; KMT2A, Lysine Methyltransferase 2A;

Figure 3.

Figure 3.

Open in a new tab

Forest plots of multivariable cox proportional hazard models that includes CalDDR-GEx10 score groups, risk-group assignment, age, FLT3 mutational status, and white blood cell count (WBC) at diagnosis, for association with event-free survival (EFS) (A) and overall survival (OS) (B) in patients treated with ADE+GO. Hazard ratios are shown within the plots with their respective 95% CIs and P values listed adjacent.

Table 3.

Adjusted Gray’s test models of CalDDR-GEx10 score and other variables for association with Relapse Risk

Relapse Risk (RR)
ADE+GO (N=227) ADE (N=168)
HR (95% CI) P-value HR (95% CI) P-value
Numerical CalDDR-GEx10 score
Risk Group Adjusted Analysis 3.31 (1.71 – 6.40) 3.90*10−4 1.10 (0.53 – 2.29) 0.80
Categorical CalDDR-GEx10 score
Low CalDDR-GEx10 score - - - -
High-CalDDR-GEx10 score 1.96 (1.21 – 3.17) 6.00*10−3 1.05 (0.61 – 1.80) 0.90
Open in a new tab

Our group has previously shown that a splicing SNP in CD33 (rs12459419) and a well-studied synonymous SNP (rs1045642) in drug transporter, ABCB1 (implicated in efflux of calicheamicin) are associated with GO response9,11. Thus, we performed additional multivariable hazard models that included factors mentioned above and the genotype groups for these 2 SNPs. Similar results were observed in these models for CalDDR-GEx10 score groups (EFS: HR = 1.98, 95% CI = 1.39–2.80, P < 0.001; Supplemental Figure S6A; OS: HR = 1.29, 95% CI = 0.85–1.96, P = 0.24; Supplemental Figure S6B; RR: HR = 2.06, 95% CI = 1.28–3.32, P = 0.003; Supplemental Table S3;) and for the CalDDR-GEx10 score as a continuous variable (EFS: 2.89, 95% CI = 1.70–4.91, P < 0.001; Supplemental Figure S6C; OS: HR = 1.88, 95% CI = 1.02–3.46, P = 0.044; Supplemental Figure S6D; RR: HR = 3.39, 95% CI = 1.71–6.75, P <0.001; Supplemental Table S3).

The CalDDR-GEx10 score predicts outcome in different risk groups for patients treated with ADE+GO:

Given that we observed differences in the distribution of risk group categories within CalDDR-GEx10 score groups (Table 1), we further evaluated the prognostic capabilities of the score within each risk group for patients treated with ADE+GO and ADE alone (Fig. 4). The high CalDDR-GEx10 score group had significantly worse prognosis as compared to the low score group within standard-risk group patients treated with ADE+GO (EFS: HR = 2.65, 95% CI = 1.67–4.22, P = 3.60*10−5; Fig. 4B). Within low-risk and high-risk group patients treated with ADE+GO, the high CalDDR-GEx10 score group had an inferior EFS compared to patients in the low CalDDR-GEx10 score group although it was not statistically significant (Low-risk HR = 1.61, 95% CI = 0.82–3.13, P = 0.16; Fig. 4A; high-risk HR = 1.46, 95% CI = 0.74–2.88, P = 0.28; Fig. 4C). In patients treated with ADE alone, the CalDDR-GEx10 score was not associated with EFS (Low-risk, HR = 0.99, 95% CI = 0.45–2.19, P = 0.97) Fig 4F; Standard-risk, HR = 1.00, 95% CI = 0.58–1.73, P = 0.99 Fig. 4E; High-risk, HR = 0.64, 95% CI = 0.25–1.68, P = 0.37, Fig. 4D). Similar results were obtained for OS within standard-risk group patients where patients in the high CalDDR-GEx10 score group had worse outcomes when treated with ADE+GO as compared to patients in the low CalDDR-GEx10 score group (HR = 1.79, 95% CI = 1.04– 3.08, P = 0.04; Supplemental Fig. S7); however, for low-risk and high-risk group patients within treated with ADE+GO no significant association of CalDDR-GEx10 score with OS was observed. No significant association was observed between the CalDDR-GEx10 score and OS with respect to patients treated with ADE alone.

Figure 4.

Figure 4.

Open in a new tab

EFS by CalDDR-GEx10 score groups within low, standard and high-risk group patients treated with ADE+GO (A-C) or ADE alone (D-F).

Patients within the low CalDDR-GEx10 score group benefit from ADE+GO therapy:

To assess if our score has the potential to guide the selection of patients to be treated using GO, we compared the outcome by treatment regimen (ADE+GO vs. ADE) within the high and low CalDDR-GEx10 score groups. A significantly higher proportion of patients treated with ADE+GO within the low score group achieved CR compared to patients treated with ADE alone (85.5% vs 75.1%, P = 0.01). Similar results were also observed in multivariable logistic regression models adjusting for risk group, age, FLT3 mutational status, and WBC. Patients within the low CalDDR-GEx10 score group were significantly more likely to achieve complete remission when treated with ADE+GO as compared to ADE alone (OR = 0.55, 95% CI = 0.55–0.93, P = 0.027; Supplemental Fig. S8A). In contrast, no difference in CR between ADE+GO and ADE treatment regimens was observed for the high CalDDR-GEx10 score group in univariate analysis (62.8% vs 65.6%, P = 0.85; Fig. 2A–B) or multivariable logistic regression models (OR = 1.11, 95% CI = 0.54–2.28, P = 0.77; Supplemental Figure S8B). Consistently, the low CalDDR-GEx10 score group had significantly better EFS and RR outcomes when treated with ADE+GO compared to treatment with ADE alone (EFS: HR = 0.69, 95% CI = 0.52–0.92, P = 0.01; Fig. 5A; RR: 0.63, 95% CI 0.44–0.90, P = 0.01; Fig. 5B). In contrast, patients in the high CalDDR-GEx10 score group had significantly inferior EFS when treated with ADE+GO compared to treatment with ADE alone (HR = 1.85, 95% CI = 1.20–2.83, P = 4.90*10−3; Fig. 5C) but RR was not significantly different by treatment arms (Fig. 5D). In multivariable Cox-proportional hazard models including risk group, age, FLT3 mutational status, and WBC at diagnosis, the low CalDDR-GEx10 score group had significantly better EFS outcome when treated with ADE+GO compared to treatment with ADE alone (HR = 0.66, 95% CI = 0.49–0.89, P = 0.007; Fig. 5E) while patients in the high CalDDR-GEx10 score group trended towards inferior EFS when treated with ADE+GO compared to treatment with ADE alone (HR = 1.54, 95% CI = 0.99–2.39, P = 0.06; Fig. 5F). Likewise, patients in the low CalDDR-GEx10 score group had a significantly lower RR when treated with ADE+GO as compared to treatment with ADE alone (HR = 0.57, 95% CI = 0.39–0.84, P = 0.004; Supplemental Table S4). In light of the interesting observation that patients with a high CalDDR-GEx10 score tended to have significantly worse EFS outcomes when treated with GO, we examined CalDDR-GEx10 score for association with treatment-related mortality (TRM) by treatment arms. Overall, there were insignificant differences in TRM based on treatment regimen for patients with a high CalDDR-GEx10 score (ADE alone vs ADE+GO: 2.6% vs. 8.5%; TRM, HR=3.21, 95%CI=0.37–27.5, P= 0.29; Supplemental Figure S9A) or patients with a low CalDDR-GEx10 score (ADE alone vs ADE+GO: 3.7 vs. 4.6%; TRM, HR=1.16, 95%CI=0.38–2.58, P=0.79; Supplemental Figure S9B). Similar results were observed in risk-adjusted and multivariable models (Supplemental Table S5).

Figure 5.

Figure 5.

Open in a new tab

Patients in the low CalDDR-GEx10 score perform better when treated with ADE+GO vs ADE alone. Kaplan-Meier curve depict event-free survival (EFS) (A) and relapse risk (RR) (B) by treatment regimen for patients in the low and EFS (C) and RR (D) by treatment regimen for patients in the high CalDDR-GEx10 score groups. Forest plot for corresponding multivariable Cox-proportional hazard models for association of treatment regimen with EFS for patients within the low (E) and high (F) CalDDR-GEx10 score groups. Hazard ratios are shown within the plots with their respective 95% CIs and P values listed adjacent.

Discussion

In September 2017, the FDA announced reapproval of GO, a CD33-directed ADC, for the treatment of newly diagnosed and relapsed/refractory AML setting the pace for a new era of personalized therapeutic options for AML. The efficacy of GO is rooted in its unique mechanism of action which exploits the myeloid restricted expression profile of CD33 and its internalization capabilities to specifically deliver calicheamicin payloads to CD33+ AML cells. Once released inside the cell, calicheamicin binds to the minor groove in DNA causing DSBs thus triggering cell death. Previous studies in our group have identified variations in CD33 structure and expression that are associated with outcomes in patients treated with ADE+GO9,10,23. Likewise, we also identified associations between expression of the ABCB1 drug efflux transporter and clinical outcome in patients treated with ADE+GO, highlighting cellular response to calicheamicin as an important factor that modulates response to GO overall11. Herein we sought to further establish the importance of cellular responses to calicheamicin by investigating the association between the leukemic cell expression of genes involved in DNA damage response and clinical outcome. We selected 18 genes with previous evidence in calicheamicin response or those involved in DNA damage response (Supplemental Table S2). Using a LASSO regression modeling approach, we developed the CalDDR-GEx10 score, consisting of 10 genes: AKT1, ATR, BAD, BAK1, BCL2L1, CASP9, H2AFX, PARP1, PRKDC, XRCC4. Of these ten genes, a few have previously been associated with GO response in literature while some are novel associations yet to be reported (Supplemental Table S2).

The CalDDR-GEx10 score was associated with CR after induction I, specifically in patients treated with ADE+GO, but no association was observed for patients treated with ADE alone. Likewise, the CalDDR-GEx10 score was also an independent predictor of EFS, OS, and RR outcomes among other well-established prognostic factors for AML for patients treated with ADE+GO but not for patients treated with ADE alone. Similarly, the CalDDR-GEx10 score remained an independent predictor of EFS, OS, and RR the rs12459419 CD33 splicing polymorphism, and the rs1045642 ABCB1 polymorphism, which we previously identified as predictors of response to GO. In our analysis, we also identified that patients with a low CalDDR-GEx10 score have better clinical outcomes when treated with ADE+GO. Interestingly, there was a difference in the abundance of low-risk patients (46% vs 28%) and patients with a FL3-ITD mutation (13% vs 29%) within the low and high CalDDR-GEx10 score groups for patients treated with ADE+GO, whereas there was no such difference for patients treated with ADE. Previous work has established that low-risk cytogenetics and the presence of the FLT3-ITD mutation are associated with increased benefit from GO and thus the imbalance in the abundance of these characteristics may confound the predictive abilities of CalDDR-GEx10 score24. However, our analysis of the association between EFS and the CalDDR-GEx10 score within different cytogenetic and molecular risk groups showed the potential of the CalDDR-GEx10 score to predict response to GO within each risk group (Fig. 4). Furthermore, risk-adjusted and multivariable models established that the CalDDR-GEx10 score was a predictor of response to GO (Supplemental Fig. S4), outcome of patients treated with GO (Fig. 3 and Supplemental Fig. S5), and clinical benefit of treatment using GO (Supplemental Fig. S8 and Fig. 5E–5F). Within high CalDDR-GEx10 score group, EFS was significantly poor, OS trended towards poor and RR and TRM (of note this was limited by sample size and the overall low proportional contribution of TRM to the total number of events) did not differ in patients treated ADE+GO as compared to ADE, suggesting that the difference in EFS is potentially due to a combination of differences in TRM, RR, and OS between the two treatment arms although none of these independently were significant. Taken together, these results highlight the importance of interindividual variations in cellular responses to calicheamicin in patient response to GO and the potential utility of the CalDDR-GEx10 score as a tool to define and categorize these differences in a clinically relevant manner allowing for the determination of which patients are better candidates for treatment regimens which include GO.

Unfortunately, due to the withdrawal of GO in 2010, data from AML patients treated with GO are limited and thus we did not have a suitable cohort for validation of the associations observed herein. Nonetheless, our results from the analysis of the association between the CalDDR-GEx10 score and clinical outcome in AML patients treated with ADE+GO or ADE alone and demonstrate the prognostic significance of the CalDDR-GEx10 score as a predictor of response to GO. Furthermore, the results presented herein provide rationale for a subsequent prospective study utilizing the CalDDR-GEx10 score in conjunction with previously identified prognostic markers for response to GO to identify patients best suited for treatment using GO and stratify them into various treatment arms accordingly (Supplemental Fig. S10). In fact, given the use of calicheamicin as the cytotoxin in inotuzumab ozogamicin, a CD22-directed ADC used in the context of B-cell acute lymphoblastic leukemia, we hypothesize the CalDDR-GEx10 score may hold broader utility outside the context of AML and GO therapy alone. Likewise, given the moonlighting functionalities of these genes in the context of DNA damage response, it is possible that the CalDDR-GEx10 score may hold utility in regard other therapeutics with similar pharmacodynamic and cellular responses to calicheamicin.

In conclusion, we developed a 10 gene transcriptomic score based on genes involved in response to calicheamicin-induced DNA damage response. The CalDDR-GEx10 score specifically predicted response to GO as measured by CR status after induction 1 and survival outcome in patients treated with ADE+GO. The CalDDR-GEx10 score also holds promise as a tool for determining which patients are better candidates for treatment using GO and may complement other biomarkers for GO response such as SNPs in CD33 and ABCB1 which hold promise in similar capacities. Our ongoing studies are focused on building a robust classification system by integrating pharmacogenomics and transcriptomics scores to accelerate clinical utility of these markers.

Methods

Patient cohorts:

Patients from Children’s Oncology Group trials AAML0531 (NCT01407757, ADE+GO arm, N = 239; ADE arm, N = 242) and AAML03P1 (NCT00070174, all treated with ADE+GO N = 62) with both gene expression levels from diagnostic bone marrow specimen and clinical outcome data were included in this study. Briefly, the AAML03P1 trial was a pilot safety study in which pediatric AML patients received standard chemotherapy consisting of cytarabine, daunorubicin, and etoposide (ADE) with the addition of two 3 mg/m2 doses of GO administered once in induction I and second time in intensification II25. The AAML0531 trial randomized patients to standard therapy with (ADE+GO arm) or without (ADE arm) the addition of GO. Details of the study design, treatment regimens, risk stratification criteria, and clinical outcomes from these studies have been published previously25,26. Briefly, risk group stratification was based on presence or absence of the molecular cytogenetic features with low-risk group patients characterized by t(8:21), inv(16) or t(16:16); high-risk group patients characterized by the presence of monosomy 7, monosomy 5/5q deletion, FLT3-ITD, or persistent disease at the end of induction I; while patients lacking these features were categorized as intermediate or standard risk. The characteristics of patients included in this study are summarized in Supplemental Table S1. Overall, there was no difference in the distribution by age, gender, race, risk group, cytogenetic features, or receipt of stem cell transplantation. For both studies, institutional review boards of all the participating sites approved the clinical trial and all patients provided informed consent. With respect to the outcome endpoints used in this study, response after induction I was based on morphologic examination of bone marrow blasts with complete remission (CR) being defined as having < 5% bone marrow blasts after induction therapy. Event-free survival (EFS) was defined as the time from study entry until death, induction failure, or relapse of any type, and overall survival (OS) was defined as the time from study entry until death. Relapse risk (RR) was defined as the time from the end of induction II for patients in CR to relapse, where death without relapse was considered a competing event. Treatment-related mortality (TRM) was defined as the time from either study entry, or from end of induction II for patients in CR, to deaths without a relapse where relapse was considered as a competing event.

RNA-seq data adjustment and normalization:

We preselected 18 genes of pharmacodynamic relevance to calicheamicin-induced DNA damage and subsequent response pathways using reports from the literature and the PharmGKB database (https://www.pharmgkb.org/pathway/PA166115250) (Supplemental Table S2 provides a list of these genes along with supporting literature references). Of the 543 patients with both RNA-seq gene expression and clinical data available, 301 were treated with ADE+GO (AAML0531, N = 239 and AAML03P1, N = 62), and 242 were treated with ADE alone. The expression levels of the 18 genes were extracted from RNA-seq that was performed on samples obtained at diagnosis on the Illumina HiSeq2000 platform and raw reads were normalized to obtain transcripts per kilobase million (TPM). To combine the RNA-seq data from two datasets with different read depths and adjust for the related batch effects, we used ComBat implemented within the sva package in R. After combining the datasets using ComBat, all RNA-seq values were log2 normalized (log2(TPM+1)) and for all subsequent analysis.

Statistical Analysis:

The LASSO regression algorithm, as implemented in the glmnet package for R (4.1.0), was applied to the gene expression levels of 18 preselected genes and the EFS data from the 301 patients treated with ADE+GO. LASSO regression uses both shrinkage (the process of shrinking estimators or coefficients to a more centralized mean value) and regularization (penalizes the data fitting criteria to avoid overfitting) to generate simpler sparse models and is advantageous when the variables used within the model have a high correlation. The variability and reproducibility of the LASSO Cox model estimates were evaluated using previously defined methods27. In brief, the LASSO Cox regression fitting process was repeated in 1000 cross-validated iterations. Ten genes that appeared in 95% of the iterations with a non-zero coefficient were included in the final model. The final coefficient value for the included genes was the average of coefficient estimates obtained from the 1000 cross-validated iterations. Subsequently, the CalDDR-GEx10 score was obtained for each patient using an equation defined as the summation of the expression value for each gene multiplied by its respective coefficient. Patients treated with ADE+GO were subsequently dichotomized into low or high CalDDR-GEx10 score groups using recursive partitioning as implemented in the rPart package following methods from our previous work27–29. EFS and OS probabilities were estimated using the Kaplan-Meier method. Estimates of RR and TRM were calculated by cumulative incidence, considering competing events. The associations between the CalDDR-GEx10 score and survival outcomes were evaluated using Cox-proportional hazard models. Multivariable Cox-proportional hazard models were used to assess the independent effect of the score groups on EFS and OS while adjusting for prognostic factors. Proportional subdistribution hazards regression models (Fine-Gray models) were used to assess both univariate and multivariable associations between groups and the RR or TRM endpoints respectively. The Chi-square test was used to examine the differences between categories while the Kruskal-Wallis test was used to assess the differences in average values where appropriate.

Supplementary Material

1813919_Sup_Material

Acknowledgements:

This work was supported by the NIH (R21CA155524), The Leukemia Lymphoma Society (6610-20), The St Baldrick’s Foundation, University of Florida Health Cancer Center, and College of Pharmacy, University of Florida. NIH awards U10CA180899, U10CA180886, U10CA98413, and U10CA098543 supported the clinical trial.

Data Availability

The transcriptomic and clinical data used in this study were obtained from the TARGET database (https://target-data.nci.nih.gov/Public/AML/mRNA-seq), however, data is only publicly available for N = 128 treated with ADE + GO and N = 175 patients treated with ADE alone. Other controlled data used in this study was obtained and used with permission from the Children’s Oncology Group.

References

  • 1.Döhner H, Weisdorf DJ, Bloomfield CD. Acute Myeloid Leukemia. N Engl J Med 2015; 373: 1136–1152. [DOI] [PubMed] [Google Scholar]
  • 2.Sievers EL, Appelbaum FR, Spielberger RT, Forman SJ, Flowers D, Smith FO et al. Selective ablation of acute myeloid leukemia using antibody-targeted chemotherapy: a phase I study of an anti-CD33 calicheamicin immunoconjugate. Blood 1999; 93: 3678–84. [PubMed] [Google Scholar]
  • 3.Hamann PR, Hinman LM, Hollander I, Beyer CF, Lindh D, Holcomb R et al. Gemtuzumab ozogamicin, a potent and selective anti-CD33 antibody - Calicheamicin conjugate for treatment of acute myeloid leukemia. Bioconjug Chem 2002; 13: 47–58. [DOI] [PubMed] [Google Scholar]
  • 4.Jen EY, Ko C-W, Lee JE, Del Valle PL, Aydanian A, Jewell C et al. FDA Approval: Gemtuzumab ozogamicin for the treatment of adults with newly-diagnosed CD33-positive acute myeloid leukemia. Clin Cancer Res 2018. doi: 10.1158/1078-0432.CCR-17-3179. [DOI] [PubMed] [Google Scholar]
  • 5.Norsworthy KJ, Ko C, Lee JE, Liu J, John CS, Przepiorka D et al. FDA Approval Summary: Mylotarg for Treatment of Patients with Relapsed or Refractory CD33-Positive Acute Myeloid Leukemia. Oncologist 2018. doi: 10.1634/theoncologist.2017-0604. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 6.Gbadamosi M, Meshinchi S, Lamba JK. Gemtuzumab ozogamicin for treatment of newly diagnosed CD33-positive acute myeloid leukemia. Futur Oncol 2018; 14. doi: 10.2217/fon-2018-0325. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7.Pollard JA, Alonzo TA, Loken M, Gerbing RB, Ho PA, Bernstein ID et al. Correlation of CD33 expression level with disease characteristics and response to gemtuzumab ozogamicin containing chemotherapy in childhood AML. Blood 2012; 119: 3705–3711. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Pollard JA, Loken M, Gerbing RB, Raimondi SC, Hirsch BA, Aplenc R et al. CD33 expression and its association with gemtuzumab ozogamicin response: Results from the randomized phase III children’s oncology group trial AAML0531. J Clin Oncol 2016; 34: 747–755. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Lamba JK, Chauhan L, Shin M, Loken MR, Pollard JA, Wang YC et al. CD33 splicing polymorphism determines gemtuzumab ozogamicin response in de novo acute myeloid leukemia: Report from randomized phase III children’s oncology group trial AAML0531. J Clin Oncol 2017; 35: 674–2682. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10.Chauhan L, Shin M, Wang Y-C, Loken M, Pollard J, Aplenc R et al. CD33_PGx6_Score Predicts Gemtuzumab Ozogamicin Response in Childhood Acute Myeloid Leukemia: A Report From the Children’s Oncology Group. JCO Precis Oncol 2019; : 1–15. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11.Rafiee R, Chauhan L, Alonzo TA, Wang YC, Elmasry A, Loken MR et al. ABCB1 SNP predicts outcome in patients with acute myeloid leukemia treated with Gemtuzumab ozogamicin: a report from Children’s Oncology Group AAML0531 Trial. Blood Cancer J 2019; 9. doi: 10.1038/s41408-019-0211-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12.Zein N, Sinha A, McGahren W, Ellestad G. Calicheamicin yI : An Antitumor Antibiotic That Cleaves Double-Stranded DNA Site Specifically. Science (80- ) 1988; 240: 198–201. [DOI] [PubMed] [Google Scholar]
  • 13.Dedon PC, Salzberg AA, Xu J. Exclusive production of bistranded DNA damage by calicheamicin. Biochemistry 1993; 32: 3617–3622. [DOI] [PubMed] [Google Scholar]
  • 14.Elmroth K, Nygren J, Mårtensson S, Ismail IH, Hammarsten O. Cleavage of cellular DNA by calicheamicin gamma1. DNA Repair (Amst) 2003; 2: 363–374. [DOI] [PubMed] [Google Scholar]
  • 15.Dedon PC, Goldberg IH. Free-radical mechanisms involved in the formation of sequence-dependent bistranded DNA lesions by the antitumor antibiotics bleomycin, neocarzinostatin, and calicheamicin. Chem Res Toxicol 1992; 5: 311–332. [DOI] [PubMed] [Google Scholar]
  • 16.Pannunzio NR, Watanabe G, Lieber MR. Nonhomologous DNA end-joining for repair of DNA double-strand breaks. J Biol Chem 2018; 293: 10512–10523. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17.Lieber MR. The mechanism of double-strand DNA break repair by the nonhomologous DNA end-joining pathway. Annu Rev Biochem 2010; 79: 181–211. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.Shibata A, Jeggo PA. DNA Double-strand Break Repair in a Cellular Context. Clin Oncol 2014; 26: 243–249. [DOI] [PubMed] [Google Scholar]
  • 19.Amico D, Barbui AM, Erba E, Rambaldi A, Introna M. Differential response of human acute myeloid leukemia cells to gemtuzumab ozogamicin in vitro: role of Chk1 and Chk2 phosphorylation and caspase 3. Blood 2003; 101: 4589–4597. [DOI] [PubMed] [Google Scholar]
  • 20.Rosen DB, Harrington KH, Cordeiro JA, Leung LY, Putta S, Lacayo N et al. AKT Signaling as a Novel Factor Associated with In Vitro Resistance of Human AML to Gemtuzumab Ozogamicin. PLoS One 2013; 8: e53518. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21.Carr MI, Zimmermann A, Chiu L-Y, Zenke FT, Blaukat A, Vassilev LT. DNA-PK Inhibitor, M3814, as a New Combination Partner of Mylotarg in the Treatment of Acute Myeloid Leukemia. Front. Oncol. 2020; 10: 127. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22.Godwin CD, Bates OM, Jean SR, Laszlo GS, Garling EE, Beddoe ME et al. Anti-apoptotic BCL-2 family proteins confer resistance to calicheamicin-based antibody-drug conjugate therapy of acute leukemia. Leuk Lymphoma 2020; : 1–5. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Papageorgiou I, Loken MR, Brodersen LE, Gbadamosi M, Uy GL, Meshinchi S et al. CCGG deletion (rs201074739) in CD33 results in premature termination codon and complete loss of CD33 expression: another key variant with potential impact on response to CD33-directed agents. Leuk Lymphoma 2019; 60. doi: 10.1080/10428194.2019.1569232. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24.Xu Q, He S, Yu L. Clinical Benefits and Safety of Gemtuzumab Ozogamicin in Treating Acute Myeloid Leukemia in Various Subgroups: An Updated Systematic Review, Meta-Analysis, and Network Meta-Analysis. Front. Immunol. 2021; 12. 10.3389/fimmu.2021.683595. [DOI] [PMC free article] [PubMed]
  • 25.Cooper TM, Franklin J, Gerbing RB, Alonzo TA, Hurwitz C, Raimondi SC et al. AAML03P1, a pilot study of the safety of gemtuzumab ozogamicin in combination with chemotherapy for newly diagnosed childhood acute myeloid leukemia: A report from the Children’s Oncology Group. Cancer 2012; 118: 761–769. [DOI] [PubMed] [Google Scholar]
  • 26.Gamis AS, Alonzo TA, Meshinchi S, Sung L, Gerbing RB, Raimondi SC et al. Gemtuzumab ozogamicin in children and adolescents with de novo acute myeloid leukemia improves event-free survival by reducing relapse risk: Results from the randomized phase III children’s oncology group Trial AAML0531. J Clin Oncol 2014; 32: 3021–3032. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27.Elsayed AH, Rafiee R, Cao X, Raimondi S, Downing JR, Ribeiro R et al. A six-gene leukemic stem cell score identifies high risk pediatric acute myeloid leukemia. Leukemia 2020; 34: 735–745. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28.Therneau TM, Atkinson EJ, Foundation M. An Introduction to Recursive Partitioning Using the RPART Routines (Long). 2022; : 1–52. [Google Scholar]
  • 29.Breiman L, Friedman JH, Olshen RA, Stone CJ. Classification and regression trees. Routledge, 2017. [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

1813919_Sup_Material
NIHMS1813919-supplement-1813919_Sup_Material.pdf (2.2MB, pdf)

Data Availability Statement

The transcriptomic and clinical data used in this study were obtained from the TARGET database (https://target-data.nci.nih.gov/Public/AML/mRNA-seq), however, data is only publicly available for N = 128 treated with ADE + GO and N = 175 patients treated with ADE alone. Other controlled data used in this study was obtained and used with permission from the Children’s Oncology Group.

RESOURCES