Bortezomib is significantly beneficial for de novo pediatric AML patients with low phosphorylation of the NF-κB subunit RelA

Abstract

The addition of the proteasome inhibitor (PI) bortezomib to standard chemotherapy (ADE: cytarabine [Ara-C], daunorubicin and etoposide,) did not improve overall outcome in the Children’s Oncology Group AAML1031 phase 3 randomized clinical trial (AAML1031). Bortezomib prevents protein degradation, including RelA via the intracellular NF-kB pathway. In this study, we hypothesized that subgroups of pediatric AML patients benefitting from standard therapy plus bortezomib (ADEB) could be identified based on pre-treatment RelA expression and phosphorylation status. RelA-total and phosphorylation at serine 536 (RelA-pSer⁵³⁶) levels were measured in 483 patient samples using reverse phase protein array technology. In ADEB-treated patients, low-RelA-pSer⁵³⁶ was favorably prognostic when compared to high-RelA-pSer⁵³⁶ (3-yr overall survival (OS): 81% vs. 68%, p=0.032; relapse risk (RR): 30% vs. 49%, p=0.004). RR in low-RelA-pSer⁵³⁶ patients significantly decreased in ADEB compared to ADE (RR: 30% vs. 44%, p=0.035). Correlation between RelA-pSer⁵³⁶ and 295 other assayed proteins identified a strong correlation with HSF1-pSer³²⁶, another protein previously identified as modifying ADEB response. The combination of low-RelA-pSer⁵³⁶ and low-HSF1-pSer³²⁶ was a significant predictor of ADEB response (3-yr OS: 86% vs. 67%, p=0.013). Thus, bortezomib may improve clinical outcome in a subgroup of AML patients identified by low-RelA-pSer⁵³⁶ and low-HSF1-pSer³²⁶.

Introduction

Nuclear factor kappa-B (NF-κB) is a protein complex formed by combinations of the five proteins in the Rel family: RelA (p65), RelB, c-Rel, p50 (NFκ-B1) and p52 (NF-κB2). The subunits RelA, p50, and c-Rel form heterodimers and are held inactive in the cytosol when sequestered by the “inhibitor of kappa B” (IκB) protein. Activation of NF-κB via the canonical pathway is stimulated by multiple pro-inflammatory stimuli or cytokines (e.g., tumor necrosis factor α) that leads to phosphorylation of IκB at serine 32 or 36 by IκB kinase (IKK). Phosphorylation of IκB results in its ubiquitination and proteasomal degradation, releasing the NF-κB complexes from IκB. Additional activation of NF-κB is achieved through phosphorylation of the core component RelA at serine 536 (RelA-pSer⁵³⁶), as well as through other post-translational modifications via IKK, [1–3] allowing the untethered complex (RelA/p50) to translocate to the nucleus. [4,5] Here, NF-κB acts as a transcription factor that binds to the κB enhancer motif on the DNA. In several cancers, NF-κB is constitutively active as a result of chromosomal translocations and mutations encoding NF-κB and IκB or IKK proteins. [4] Overexpression of NF-κB target genes (e.g., CREB, c-JUN, GSK3, AP-1) [2,6] is a key survival factor in cancer and plays a key role in cell proliferation, apoptosis [5] and programmed cell death. [7] In AML, NF-κB protein binding is increased in nuclear extracts of leukemic CD34+, but not in those of normal CD34+ cells, suggesting increased NF-κB activation. [8]

Survival of pediatric AML remains guarded with 5-year overall survival (OS) rates of around 70% in high-income countries. Although survival exceeds that of solid tumors in pediatric, there are significant long-term sequelae including heart dysfunction, decreased fertility and second malignancies. [9] Better therapies are needed to increase survival, reduce relapse and to reduce long-term side effects. [10] Recently, the Children’s Oncology Group (COG) evaluated the efficacy of adding proteasomal inhibition (PI) to standard chemotherapy (ADE: cytarabine (Ara-C), daunorubicin, etoposide). by adding bortezomib (ADEB) in a randomized phase 3 clinical trial for newly diagnosed pediatric AML patients (AAML1031). [11] While the study showed no improvement in OS or event-free survival (EFS) across the entire cohort, [12] we however, previously showed that the ADEB was beneficial in subgroups of patients with decreased phosphorylation of heat shock factor 1 at serine 326 (HSF1-pSer³²⁶), [11] as well as in patients demonstrating upregulation of histone modifying enzymes (HME) with more transposase-accessible chromatin (submitted).

One of the mechanisms thought responsible for the efficacy of PIs is their antiproliferative and proapoptotic property via manipulation of NF-κB. [13] Initially, it was thought that PIs cause NF-κB inhibition by preventing proteasomal degradation of IκB. [5] However, some studies showed that bortezomib also induced NF-κB activity via phosphorylation of IKK, causing a release of RelA/p50 via increased IκB degradation. [14,15] This indicates that the relationship between NF-κB (in)activity and bortezomib is complex. Given this interaction between NF-κB and bortezomib, we questioned if protein expression of total RelA (RelA-total) or activated RelA (RelA-pSer⁵³⁶) was prognostic of clinical response to ADEB, and if we could identify an additional subgroup of pediatric AML patients that would benefit from ADEB.

Methods

Pediatric AML patient samples

Peripheral blood samples were obtained from 483 de novo pediatric AML patients that participated in the COG AAML1031 (#NCT01371981) phase 3 clinical trial, and 30 CD34+ bone marrow samples obtained from healthy donors; of these 20 were pediatric and ten were adults CD34+ samples. Samples from adults were included to allow future comparison to adult AML samples which were published previously. [16] Samples were acquired during routine diagnostic assessments prior to, 10 hours (hr) and 24hr after the initiation of systemic chemotherapy, and were collected between July 2011 and February 2017. Written informed consent was obtained in accordance with local IRB review boards and the Declaration of Helsinki.

Outcome data was available for 410 of the 483 patients enrolled on the AAML1031 study. One hundred and sixty-four patients received standard ADE induction therapy, 210 patients received ADEB, and 36 patients with known FLT3-ITD mutations received ADE plus sorafenib (ADES). Because we were particularly interested in the association between RelA-total and RelA-pSer⁵³⁶ protein expression in relation to treatment with ADE plus bortezomib (ADEB), and since outcome after ADES and ADE did not differ [12], patients treated with ADE and ADES were combined in our proteomic analysis (n=200). Three hundred forty-eight (85%) patients achieved complete remission by the end of the second course, 31 (8%) patients were refractory or died (failed therapy), 156 (45%) patients relapsed after remission, and 286 were still alive at the end of their follow-up (70%). Survival analysis was performed as published previously. [11] Patients had mutation data available for CEBPA, FLT3-ITD, c-KIT and NPM1.

RPPA methodology

The methodology and validation of the RPPA technique have been described elsewhere. [17–21] Briefly, fresh samples were obtained from patients at local sites and were shipped to the central processing lab at Baylor College of Medicine by overnight courier. Mononuclear cells were isolated from peripheral blood by centrifugation using lymphocyte separation solution (Sigma) and enriched for leukemic cells by CD3/CD19 depletion (Miltenyi Biotech, Cologne, Germany). Protein preparations were normalized to a concentration of 1×10⁷ cells/mL and printed in five serial dilutions onto slides along with normalization and expression controls. Slides were probed with 301 antibodies listed in Supplementary Table 1, including a primary validated antibody against total RelA (Cell Signaling, Danvers, MA, Cat. #3034) and RelA-pSer⁵³⁶ (Cell Signaling, Cat. #3033). Five antibodies were excluded for different reasons yielding a final of 296 antibodies used for analysis. [11] Stained slides were analyzed using Microvigene® Software (version 3.0, Vigene Tech, Carlisle, MA). SuperCurve algorithms were used to generate a single concentration value from the five serial dilutions. [22] Loading control [23] and topographical normalization [24] procedures were performed to account for protein concentration and background antibody staining variations on each array.

Transcriptome sequencing data

Additional mutation profiles were obtained from transcriptome sequencing data for 390 of the 483 patients (81%) and generated elsewhere. [25] Mutations that were present in ≥ 10 patients were KRAS (n=29), NRAS (n=99), GATA2 (n=12), PTPN11 (n=28), MYH11 (n=17) and IDH1 or IDH2 (n=14)).

Pathway analysis

STRING software (String version 11.0; http://string-db.org) was used to determine protein associations. [26]

Statistical analysis

Data were frozen as of June 30, 2019 for outcome analyses. Estimates of OS and EFS were calculated using the Kaplan–Meier method. OS and EFS were defined as time from study entry until death or until relapse, secondary malignancy, or death, respectively. RR was calculated using methods of competing events and was defined as the time from the end of two courses of induction (for patients in complete remission) to relapse, where deaths without a relapse were considered competing events. The significance of predictor variables was tested with the log-rank statistic for OS, EFS and with Gray’s statistic for RR. Cox proportional hazard models were used to estimate hazard ratios for univariable and multivariable analyses of OS and EFS. Competing risk regression models were used for analyses of RR. Outcome by treatment arm was based on intention-to treat analysis. Correlation between the patient cohorts and categorical clinical variables were compared using Pearson’s Chi-square test, and for continuous variables using the Kruskal-Wallis test. Pearson’s correlation analyses were performed to correlate RelA-pSer⁵³⁶ protein expression levels with expression of the other 295 proteins. P-values were adjusted using false discovery rate correction. All statistical analyses were performed in R Version 1.3.959 (2009–2020, RStudio, Inc., Boston, MA) or SAS version 9.4 (SAS Institute, Inc., Cary, NC).

Results

RelA-total and RelA-pSer⁵³⁶ protein expression across newly diagnosed pediatric AML patients

Relative RelA-total and RelA-pSer⁵³⁶ protein expression levels were measured in bulk leukemia blasts from 483 pediatric AML patients, as well as in 30 CD34+ samples from healthy controls. All samples used for outcome analysis were collected at time of diagnosis, prior to exposure to systemic chemotherapy. RelA-total was relatively homogeneously expressed, with individual normalized RelA-total expression ranging from −1.84 to +0.94 log2. Overall, RelA-total was not different in pediatric AML compared to normal CD34+ cells (Supplementary Figure 1A, p=0.63), and only 2% of patients had RelA-total levels significantly higher than normal CD34+ (95% CI normal CD34+ [−0.50; +0.66 log2]). RelA-pSer⁵³⁶, in contrast, had significantly lower expression relative to normal CD34+ cells (p<0.001, Supplementary Figure 1B), and had a larger expression range (−2.6 to +2.0 log2). Eighty-one percent of the samples had RelA-pSer⁵³⁶ expression significantly lower and 6% had expression significantly higher than normal CD34+ (95% CI normal CD34+ [−0.49; +0.49 log2]). RelA-total roughly correlated with RelA-pSer⁵³⁶, with the most variability in phosphorylation occurring in samples with the highest RelA-total (Supplementary Figure 2A, r=0.16, p<0.001).

RelA-total and RelA-pSer⁵³⁶ were also measured in different myeloid cell populations with varying degrees of stem cell characteristics. Compared to leukemic bulk and CD3/CD19 depleted mononuclear cells, RelA-total was lower in CD34+, CD34+CD38+ and stem cell enriched CD34+CD38− populations (p<0.001, p=0.006 and p=0.019 respectively, Supplementary Figure 1C). RelA-pSer⁵³⁶ expression was lowest in CD34+ (p<0.001) and highest in CD34−CD38− (p<0.001) and CD34−CD38+ (p<0.001) compared with the CD3-CD19− leukemic bulk cells (Supplementary Figure 1D). RelA-pSer⁵³⁶ median expression trended slightly higher in cells with markers of leukemia stem cells (LSC) (LSC-like; CD34+CD38−) compared to bulk cells (n=33, p=0.053). Yet, a significant difference between the variances of RelA-pSer⁵³⁶ was observed in LSC vs. bulk cells (Bartlett test of homogeneity of variances, p=0.002). This suggests that RelA-pSer⁵³⁶ levels tend to be higher in LSC-enriched AML blasts compared to bulk and CD3-/CD19− cells, although median expressions are not significantly different.

In addition, RelA mRNA transcriptome levels were available for 390 of the 483 AML patients. There was no significant correlation between RelA mRNA and RPPA protein levels for RelA-total (Pearson’s coefficient r=0.057, p=0.27) or between RelA mRNA and RelA-pSer⁵³⁶ (r=0.027, p=0.6) (Supplementary Figure 2B–C).

Correlation between RelA, patient characteristics and disease features

To associate RelA-total and RelA-pSer⁵³⁶ expression levels with patient characteristics (Table 1) and molecular AML features (Table 2), patients were divided into two groups based on median RelA expression across the 483 patients of both RelA antibody-targets. While the frequency of most variables was not different between the two cohorts, low-RelA-total-patients were more frequently associated with t(8;21) translocation and c-Kit mutation, and less frequently with CEBPA and GATA2 mutation. Among all the clinical features that were compared, patients with low-RelA-pSer⁵³⁶ only differed from high-RelA-pSer⁵³⁶ patients in frequency of IDH1/2 (6% vs. 1%, p=0.02).

Table 1.

AML patient characteristics (n=483)

		All cases	Low-RelA-total	High-RelA-total	p	Low-RelA-pSer536	High-RelA-pSer536	p
Number		100%	50%	50%		50%	50%
Gender (n=481)	Female	50%	50%	50%	0.49	51%	49%	0.62
Age (years old)	≤ 1	12%	10%	14%	0.25	12%	12%	0.13
	2–10	33%	35%	31%		37%	29%
	≥ 11	55%	56%	55%		51%	60%
White blood cell count (at study entry)	>100000	24%	21%	27%	0.19	27%	20%	0.09
CNS (n=480)	Positive	39%	36%	43%	0.14	40%	38%	0.64
Ethnicity (n=480)	Hispanic	19%	21%	18%	0.55	18%	21%	0.46
Race	Black	12%	11%	13%	0.71	13%	10%	0.53
AAML1031 risk group† (n=469)	High risk	29%	29%	28%	0.99	30%	27%	0.63
Complete remission at end of induction II	Yes	85%	87%	83%	0.30	81%	88%	0.07

Table 2.

Pediatric AML molecular features (n=483)

		All cases	Low-RelA-total	High-RelA-total	p	Low-RelA-pSer536	High-RelA-pSer536	p
Number		100%	50%	50%		50%	50%
Cytogenetics (n=476)	t(8;21)	16%	22%	10%	0.001	16%	16%	0.93
	inv16	14%	13%	14%	0.89	16%	11%	0.16
	Normal karyotype	28%	25%	31%	0.22	28%	27%	0.88
	t(9;11)(p22;q23)/11q23	18%	14%	21%	0.06	14%	22%	0.05
	−5, −7, or +8	9%	8%	10%	0.75	10%	8%	0.44
	Other	15%	17%	14%	0.45	14%	16%	0.87
NPM1 mutation	Mutant	10%	12%	9%	0.30	7%	13%	0.05
CEBPA mutation	Mutant	9%	5%	13%	0.001	12%	6%	0.06
FLT3-ITD	Mutant	22%	21%	23%	0.64	22%	21%	0.84
High-allelic FLT3-ITD ratio	Yes (≥ 0.4)	74%	79%	77%	0.58	77%	71%	0.57
c-Kit mutation (Exon 8) (n=399)	Mutant	4%	6%	2%	0.19	5%	3%	0.47
c-Kit mutation (Exon 17) (n=399)	Mutant	8%	11%	5%	0.09	7%	9%	0.80
c-Kit mutation (Exon 8 or 17) (n=399)	Mutant	12%	16%	8%	0.026	12%	12%	1.00
KRAS (n=390)	Mutant	7%	6%	9%	0.44	8%	7%	1.00
NRAS (n=390)	Mutant	25%	26%	25%	0.82	26%	25%	0.91
KRAS or NRAS mutated (n=390)	Mutant	31%	31%	30%	1.00	30%	31%	1.00
PTPN11 (n=390)	Mutant	7%	9%	6%	0.33	6%	9%	0.30
MYH11 (n=390)	Mutant	4%	5%	4%	0.62	6%	3%	0.34
GATA2 (n=390)	Mutant	3%	1%	6%	0.008	4%	3%	0.80
IDH1 & 2 (n=390)	Mutant	4%	3%	5%	0.41	6%	1%	0.02

^†AAML1031 protocol risk group definition:

· Low risk: inv(16)/t(16;16) or t(8;21), or NPM or CEBPα mutation;

· High risk: FLT3/ITD+ with high allelic ratio ≥ 0.4, or monosomy 5/del5q or 7, without low-risk features;

· Risk status unknown for 10/410

^*NA/ unknown values not considered in p-value calculations and are excluded from the results

Low-RelA-pSer⁵³⁶ correlates with favorable outcome in patients treated with ADEB

To investigate the effect of RelA-total and RelA-pSer⁵³⁶ on outcome, survival analysis was performed for 410 of the 483 patients with available outcome data. OS, EFS and RR were calculated between low- and high-RelA-total, and low- and high-RelA-pSer⁵³⁶. No prognostic effects were observed for RelA-total across the entire cohort, or when stratified by ADE vs. ADEB therapy (Supplementary Figure 3). However, when we performed the same analysis in RelA-pSer⁵³⁶, low-RelA-pSer⁵³⁶ was significantly favorable in ADEB treated patients (Figure 1A, solid lines). In ADEB-treated patients, 3-yr-OS was 81% in the low-RelA-pSer⁵³⁶ patients compared with 68% in high-RelA-pSer⁵³⁶ (p=0.032), and 3-yr post-induction RR was 30% in low-RelA-pSer⁵³⁶ patients vs. 49% in high-RelA-pSer⁵³⁶ patients (p=0.004, Figure 1C). A similar trend was observed when RelA-pSer⁵³⁶ was split into thirds. Low-RelA-pSer⁵³⁶ patients treated with ADEB also trended toward better 3-yrs EFS, with 60% 3-yrs EFS vs. 47% in high-RelA-pSer⁵³⁶ (p=0.058, Figure 1B). Differences in outcome between low- and high-RelA-pSer⁵³⁶ were not observed in patients treated with ADE (p=0.985, Figure 1B). Comparison of outcome between the treatment arms in low- and high-RelA-pSer⁵³⁶ patients separately, showed a significant decreased in RR after ADEB in low-RelA-pSer⁵³⁶ compared with ADE (30% vs. 44%, p=0.035). OS and EFS also trended to be better after ADEB, but differences were not significant (70% vs 81%, p=0.159; 50% vs 60%, p=0.257) (Figure 1C, gray lines). In high-RelA-pSer⁵³⁶ patients, RR tended to show the reverse, with better outcome after ADE vs. ADEB (RR, p=0.051, Figure 1C, black lines). When restricted to low-risk AAML1031 patients only [12], survival analysis again identified low-RelA-pSer⁵³⁶ as favorable prognostic indicator in ADEB-treated patients (OS, p=0.045, EFS p=0.057, RR p=0.010) (Supplementary Figure 4). Multivariate analysis revealed high-RelA-pSer⁵³⁶ as an independently unfavorable prognostic variable for RR with ADEB treatment (Supplementary Table 2).

Figure 1. Kaplan-Meier survival analysis for low and high-RelA-pSer⁵³⁶.

A) OS, B) EFS and C) RR for low-RelA-pSer⁵³⁶ patients treated with ADE or ADES (grey dashed line) or ADEB (grey solid line), vs. high-RelA-Ser⁵³⁶ patients treated with ADE or ADES (black dashed line) or ADEB (black solid line).

RelA-pSer⁵³⁶ phosphorylation increases following systemic chemotherapy

To gain insight into the effect of bortezomib on RelA expression and activation state in AML, we analyzed baseline RelA-total and RelA-pSer⁵³⁶ expression, and compared this to expression in samples collected 10hr and 24hr after the initiation of systemic treatment. Although, RelA-total did not show any change after ADE or ADEB treatment over 24hr (Supplementary Figure 5), we observed that RelA-pSer⁵³⁶ expression was increased 24hr after exposure to both ADE and ADEB compared to pre-treatment (Supplementary Figure 6A). Because we had seen that outcome of ADEB-treated patients significantly improved in low-RelA-pSer⁵³⁶ patients but not in high-RelA-pSer536, we evaluated changes in expression after chemotherapy in low-RelA-pSer⁵³⁶ and high-RelA-pSer⁵³⁶ patients. This showed an increase in RelA-pSer⁵³⁶ after 24hr in low-RelA-pSer⁵³⁶ patients independent of their received treatment (p=5.2e-06, p=1.3e-07, respectively, Supplementary Figure 6B), but not in high-RelA-pSer⁵³⁶ patients (p=0.69, ADEB; p=0.42, Supplementary Figure 6C). Expression in patients with low-pretreatment RelA-pSer⁵³⁶ did not reach the pre-treatment levels of the high- RelA-pSer⁵³⁶ patients.

RelA-pSer⁵³⁶ positively correlates with HSF1-pSer³²⁶

RelA-pSer³⁵⁶ expression was correlated with 295 proteins on the same array. Correlations were performed by calculating Pearson’s correlation coefficient. The strongest correlation to RelA-pSer536 was MAPK14-pThr180_Tyr182 (r=0.65, p<0.001), a kinase known to facilitate activation of transcription factors, including RelA in the NF-κB complex.^27,28 Twenty-four proteins were identified as being positively correlated with RelA-pSer⁵³⁶, and four proteins (PRKCB, GRP78, MYH11, ELK1.pSer³⁸³) were negatively correlated (r>0.25, Bonferroni adjusted p<0.05) (Figure 2A). None of the proteins was associated with cell death. All but two positively correlated proteins were phosphorylated.

Figure 2. Waterfall plot and network analysis of correlated proteins with RelA-pSer⁵³⁶.

A) Waterfall plot showing the 28 significantly correlated proteins with RelA-pSer536 (r ≥ 0.25, or r ≤ −0.25). * Denotes antibodies directed against PTM-sites. B) STRING networks analysis for the 24 proteins that correlated with both RelA-pSer⁵³⁶ and HSF1-pSer³²⁶. Only interactions with either RELA or HSF1 are shown.

Interestingly, low expression of the protein HSF1-pSer³²⁶, previously found to be associated with the beneficial effect of ADEB in pediatric AML, was among the strongest correlated proteins with low RelA-pSer⁵³⁶ (r=0.59, p<0.001). [11] Furthermore, twenty-four of the 28 proteins that correlated with RelA-pSer⁵³⁶ also significantly correlated with HSF1-pSer³²⁶ (Supplementary Figure 7). Although a search in the STRING database revealed no previously known relationship between these proteins, network analysis connected RelA and HSF1 via MAPK1, both directly and through several other proteins (Figure 2B). To see if this same relationship exists in other hematological malignancies, we performed the identical analysis in 361 T-cell acute lymphoblastic leukemia patient samples (n=268 pediatric, n=93 adult). Again, HSF1-pSer³²⁶ was found to be the most significantly correlated protein with RelA-pSer⁵³⁶ (r=0.56, p<0.001). Also, out of the 22 positively correlated proteins, 14 proteins contained post-translationally modified protein sites (Supplementary Figure 8).

The combination of low-RelA-pSer⁵³⁶ and low-HSF1-pSer³²⁶ augments the beneficial effect of bortezomib

Because RelA-pSer⁵³⁶ and HSF1-pSer³²⁶ were correlated, and because both proteins were identified as individually prognostic in pediatric AML patients treated with ADEB, we hypothesized that the combination of RelA-pSer⁵³⁶ and HSF1-pSer³²⁶ might have increased survival or lower relapse rates. Therefore, patients were clustered into four groups based on the combination of the expression of these two proteins; low-RelApSer⁵³⁶-low-HSF1-pSer³²⁶ (n=183/483, 38%), low-RelA-pSer⁵³⁶-high-HSF1-pSer³²⁶ (n=59, 12%), high-RelA-pSer⁵³⁶-low-HSF1-pSer³²⁶ (n=59, 12%), high-RelA-pSer⁵³⁶-high-HSF1-pSer³²⁶ (n=182, 38%). The unequal distribution of patients among these four groups suggests a strong linkage between RelA-pSer⁵³⁶ and HSF1-pSer³²⁶ expression (Chi-Square test, p<0.01). Patient characteristics based on these four protein groups are shown in Supplementary Table 3.

Survival analysis, restricted to patients treated with ADEB, showed that the combination of low-RelApSer⁵³⁶-low-HSF1-pSer³²⁶ had a superior 3-yr OS and EFS compared to the other three groups (Figure 3, OS; 86% vs. 67%, p=0.001, EFS: 66% vs. 45%, p=0.002). RR was also significantly lower in patients with the combination of low-RelApSer⁵³⁶-low-HSF1-pSer³²⁶ compared to the other patient groups (26% vs. 48%, p=0.002). Again, this effect was not seen in ADE-treated patients (Supplementary Figure 9)

Figure 3. Kaplan-Meier survival analysis for RelA-pSer⁵³⁶ and HSF1-pSer³²⁶ combined in ADEB-treated patients.

A) OS, B) EFS, and C) RR for patients stratified based on their RelA-pSer⁵³⁶ and HSF1-pSer³²⁶ expression levels. Solid grey: low-RelA-pSer⁵³⁶, low-HSF1-pSer³²⁶; dashed grey: low-RelA-pSer⁵³⁶, high-HSF1-pSer³²⁶; solid black: high-RelA-pSer⁵³⁶, low-HSF1-pSer³²⁶; dashed black: high-RelA-pSer⁵³⁶, high-HSF1-pSer³²⁶.

The combination of low-RelA-pSer⁵³⁶ and low-HSF1-pSer³²⁶ is an independent prognostic marker in ADEB

To investigate whether the prognostic effect of the combination of RelA-pSer536 and HSF1-pSer326 in ADEB-treated patients was independent of other variables, multivariate analysis was performed. Only variables found to be significantly prognostic in univariate analysis were considered. This analysis showed a significant contribution on outcome for RelA-pSer⁵³⁶ with HSF1-pSer³²⁶, age at diagnosis and AAML1031 risk groups. Hazard ratio of low-RelA-pSer⁵³⁶ with low HSF1-pSer³²⁶ (low-low) was set as reference (HR = 1), with associated increase HR for the remaining three subsets (low-high, high-low, high-high, Table 3).

Table 3.

Multivariate analysis in ADEB-treated patients, including the combination of RelA-pSer⁵³⁶ and HSF1-pSer³²⁶.

		OS from study entry				EFS from study entry			RR from end of course 2
		N	HR	95% CI	p	HR	95% CI	p	N	HR	95% CI	p
RELA-pSer536 & HSF1-pSer326	Low-low	71	1			1			62	1
	Low-high	25	3.74	1.50 – 9.36	0.005	2.46	1.29 – 4.71	0.007	17	1.85	0.76 – 4.50	0.177
	High-low	25	2.26	0.88 – 5.79	0.090	1.51	0.78 – 2.90	0.220	23	1.88	0.91 – 3.89	0.089
	High-high	83	2.96	1.38 – 6.32	0.005	1.97	1.19 – 3.26	0.008	76	2.62	1.44 – 4.76	0.002
Age at Dx	0–1 yr.	27	2.22	1.05 – 4.71	0.038	2.26	1.23 – 4.13	0.008	22	2.02	0.97 – 4.23	0.062
	2–10 yr.	67	1			1			57	1
	11+ yr.	110	0.72	0.39 – 1.31	0.283	0.77	0.50 – 1.21	0.257	99	0.71	0.42 – 1.22	0.214
Risk group (AAML1031 definition)	Low	155	1			1			146	1
	High	49	2.80	1.58 – 4.97	<0.001	2.31	1.47 – 3.64	<0.001	32	1.53	0.82 – 2.85	0.186

Discussion

In this study we showed for the first time that low-RelA-pSer⁵³⁶ was associated with better outcome compared to high-RelA-pSer⁵³⁶ in pediatric AML patients treated with bortezomib added to ADE, and that patients with low-RelA-pSer⁵³⁶ did better after treatment with ADEB vs. ADE alone, whereas high-RelA-pSer⁵³⁶ did worse with ADEB. We also showed that RelA-pSer⁵³⁶, but not RelA-total, increased following chemotherapy in the low-RelA-pSer⁵³⁶ patients but not in high-RelA-pSer⁵³⁶ patients. Previously we demonstrated that subgroups of pediatric AML patients benefitted from ADEB, including those with low-HSF1-pSer³²⁶ as well as those with high expression of HME (manuscript submitted).

In addition to low-RelA-pSer⁵³⁶, we also showed that the combination of low-RelA-pSer⁵³⁶ and low-HSF1-pSer³²⁶ improved prognosis compared with either protein alone. We suggest that the combination of these two proteins could be used to a priori identify a group of AML patients (38% of patients in this study) that benefit from ADEB. Of note, while the combination with low/high expression of HME did not affect patient prognosis with both low-RelA-pSer⁵³⁶ and low-HSF1-pSer³²⁶ expression, HME stratification could potentially identify another small group of low-RelA-pSer⁵³⁶ and high-HSF1-pSer³²⁶ expression with a favorable prognosis after ADEB, while the low-RelA-pSer⁵³⁶ and high-HSF1-pSer³²⁶ following ADE did not (OS; p=0.025, EFS; p=0.007, RR; p=0.085, Supplementary Figure 10).

Studies have shown that proteasome inhibition has two separate effects on the NF-κB. Bortezomib and other PIs can either block proteasomal degradation of IkB, resulting in decreased NF-kB activation, or phosphorylate the NF-κB regulator IKK, which leads to increased NF-κB activation. [5,27] To examine this, we evaluated RelA expression and activation (i.e., phosphorylation) across pediatric AML subtypes and measured RelA expression across a variety of AML cell types. Our data showed that only 2% of bulk leukemic cells had RelA-total significantly higher than in normal CD34+ and only 9% of bulk leukemia had higher RelA-pSer⁵³⁶ than in normal expression of RelA-pSer⁵³⁶. Previous work has shown that RelA is often up-regulated in AML, but those studies were conducted in tumor cell lines and may not accurately reflect the biology in primary cells, or measured RelA-activity in the leukemic stem cell population, which are known to have higher NF-κB expression levels compared to bulk cells. Despite small numbers (n=33), our data showed a similar trend toward higher expression of RelA-pSer⁵³⁶ in LSC-enriched AML cells, which supports the finding that NF-κB expression/activation is likely higher in leukemia initiating stem cells. [8]

Secondly, when we correlated RelA expression with outcome using multivariate analysis, we identified RelA-pSer⁵³⁶ as a prognostic indicator in patients treated with ADEB. Patients with low-RelA-pSer⁵³⁶ expression had a significantly better survival compared to those with the high RelA-pSer⁵³⁶ levels (+13% OS, −19% RR at 3-yrs). We think that as bortezomib primarily inhibits NF-kB activation and in turn induces cell death, low-RelA-pSer⁵³⁶ can enhance the effect of low NF-κB activity.

If we are able to identify this subset of low-RelA-pSer⁵³⁶ patients prior to treatment, those could potentially be treated with ADEB, resulting in higher survival rates. However, this requires quick measurement of RelA-pSer⁵³⁶ expression at the time of diagnosis to accurately distinguish high from low-RelA-pSer⁵³⁶ patients (e.g., by using an enzyme-linked immune sorbent assay or immunohistochemistry). In addition, the use of a small molecule inhibitor to directly inhibit RelA or an upstream activating kinase of the NF-κB/RelA pathway, would be another approach to sensitize high-RelA-pSer⁵³⁶ patients to ADEB, or to further increase sensitivity of those who are already sensitive. For instance, in chronic inflammatory diseases, where a disproportional activation of RelA is often part of the underlying inflammatory pathology, there has already been tremendous focus on manipulating RelA. [28,29] A challenge in creating RelA as a therapeutic target remains proper intracellular delivery and selective targeting without significant off-target effects. A promising drug that recently entered a phase I clinical trial for hematological malignancies, including AML, myelodysplastic syndrome and non-Hodgkin lymphoma, is the interleukin-1 receptor associated kinase (IRAK) inhibitor CA-4948. IRAK inhibitors manipulates NF-κB mediated transcription via inhibition of IRAK, which is an upstream activating kinase of the IKK complex. [30]

As in various hematological malignancies, treatment with PIs was shown to induce activation of NF-kB in a time- and dose-dependent manner, [31–33] we compared RelA-total and RelA-pSer⁵³⁶ expression pre-treatment to expression 10 and 24hr post-treatment exposure. Whereas RelA-total levels did not change overtime, RelA-pSer⁵³⁶ significantly increased 24hr following treatment regardless of treatment regimen. We hypothesize that chemotherapy, regardless of the addition of bortezomib, induces RelA phosphorylation as an attempt to respond to stress caused by the chemotherapy. In addition, phosphorylation levels of low-RelA-pSer⁵³⁶ patients significantly increased after both ADE and ADEB treatment, but this effect did not occur in the high-RelA-pSer⁵³⁶ patients. Low-RelA-pSer⁵³⁶ patients might be more dependent on this stress-response than high-RelA-pSer⁵³⁶ for survival, which is speculatively blocked by the PI addition, leading to more cell stress and cell death resulting in better clinical outcome. Although increased, the change after 24hr in low-RelA-pSer⁵³⁶ cells was not strong enough to reach the baseline levels seen in high-RelA-pSer⁵³⁶ patients. Schematic summary of our hypothesis is shows in Figure 4.

Figure 4. Schematic summary of RelA-pSer⁵³⁶ dependence in bortezomib treatment.

A) Patients were split based on pre-treatment RelA-pSer⁵³⁶ expression levels into low (blue) and high (red). B) Expression increases in low- RelA-pSer⁵³⁶ patients after treatment with chemotherapy (purple), but does not reach the same levels as found in the high- RelA-pSer⁵³⁶. In high- RelA-pSer⁵³⁶ patients, no increase in RelA-pSer⁵³⁶ was seen. C) Solid line represents baseline state of the leukemia cells with a tendency toward cell proliferation. After chemotherapy, this balances over to more apoptosis. We hypothesize that RelA-pSer536 plays a role in stress response caused by chemotherapy which is speculatively blocked by PI therapy, and that low- RelA-pSer⁵³⁶ patients are more dependent on this response. Size of the black arrow represents the increase in apoptosis. D) Increased cell stress result in cell death and cell death eventually leads to patient survival. Survival of low-RelA-pS⁵³⁶ increases after treatment with ADEB (solid line) vs. ADE (dashed line).

Finally, we found a strong correlation between RelA-pSer⁵³⁶ and HSF1-pSer³²⁶, which was strengthened by a similarly strong association in a cohort of 358 T-cell acute lymphoblastic leukemia patients, increasing the probability that there is a real, but previously unrecognized, relationship between these proteins. Since low levels of both protein modifications were favorably prognostic in ADEB-treated patients, the combination of low-RelA-pSer⁵³⁶ and low-HSF1-pSer³²⁶ appears to increase our ability to predict which patients are likely to respond to PI-containing chemotherapy. The favorable prognostic effect of low expression of RelA-pSer536 and HSF1-pSer326 was abolished when the expression of either protein was high. We hypothesize that as both proteins become active in response to cell stress, the combination of low-RelA/HSF1 represents the intrinsically most “stressless” state. This suggests that these low-low AML cells may be more sensitive to increased stress caused by PI inhibition. The buildup of misfolded proteins in AML cells with low RelA/HSF1 may prevent adaption to homeostasis disruption, resulting in cell death. This study suggests a functional HSF1-RelA axis in AML in response to proteasome inhibition. We were able to link RelA to HSF1 via MAPK1, directly or via several other proteins (Figure 2). The observation that a better prognosis in pediatric AML after ADEB in patients with both HSF1 and RelA provides further evidence that a RelA-NF-kB pathway may be relevant to chemotherapy containing PI. Rao et al. found that knockdown of HSF1 results in inhibition of the NF-κB pathway, [34] but more research is needed to verify the existence of a RelA-HSF1 axis and to confirm its role in AML PI sensitivity.

In conclusion, in this study we have identified low-RelA-pSer⁵³⁶ as favorable prognostic factor in ADEB treated pediatric AML patients. This finding was even stronger in combination with HSF1-pSer³²⁶. As about one third of the patients expressed low-RelA-pSer⁵³⁶ and low-HSF1-pSer³², we hypothesize that a priori identification and treatment with ADEB of these patients may result significant improvement of OS in pediatric AML.

Supplementary Material

fS1

Figure S1. Relative RelA-pSer536 protein expression. Relative A) RelA-total and B) RelA-pSer⁵³⁶ protein expression in 483 AML patient samples, and 30 CD34+ non-malignant bone marrow samples. C) RelA-total and D) RelA-pSer⁵³⁶ per cell type in pediatric AML. Bulk cells and CD3/CD19 depleted cells are showed in dark blue (reference set). Cell populations that had significantly lower RelA compared to bulk and CD3-CD19− cells are shown in light blue. Cell populations with RelA-pSer⁵³⁶ expression levels higher than bulk and CD3-CD19− cells are indicated in red. Cell populations shown in gray were not significantly different from bulk/CD3-CD19− cells.

fS3

Figure S3. Outcome stratification by high and low RelA-total levels in all (n=410, A, D and G), ADE/ADES (n=200, B, E and H) and ADEB treated (n=210, C, F and I). Upper panel is demonstrated by overall survival (OS), middle panel by event-free survival (EFS) and the bottom panel by relapse risk (RR) after second round of induction. Blue curves represent low-RelA-total patients (n=200, n=100 and n=110 for all, ADE/ADES and ADEB respectively in OS/EFS and n=174, n=84 and n=90 for RR in all, ADE/ADES and ADEB respectively. The orange lines represent high-RelA-total patients (n=210, n=100 and n=100 for OS/EFS curves and n=174, n=87 and n=88 for RR curves in all, ADE/ADES and ADEB treated patients respectively.

fS2

Figure S2. A) Correlation between RelA-pSer⁵³⁶ and RelA-Total. B) RelA-total and C) RelA-pSer⁵³⁶ protein correlation with RelA mRNA expression (transcripts per kilobase million; TPM).

fS5

Figure S5. Log2 RelA-total protein levels in ADE (left panel) and ADEB treated (right panel) at baseline (0 hr) and 10, and 24 hours after chemotherapy exposure (ADE; n=234, n=229 and n=228, ADEB; n=213, n=203 and n=196).

fS6

Figure S6. Log2 change in RelA-pSer⁵³⁶ expression following chemotherapy exposure. A) RelA-pSer⁵³⁶ expression at baseline (0hr), 10hr and 24hr after receiving ADE (n=234, n=229 and n=228) and ADEB (n=213, n=203 and n=196). RelA-pSer⁵³⁶ expression levels at baseline (0hr) and 24hr after receiving ADE or ADEB in B) low-RelA-pSer⁵³⁶ (ADE; both timepoints n=107 and ADEB; both timepoints n=98) and B) high-RelA-pSer⁵³⁶ patients (ADE; both timepoints n=104 and ADEB; both timepoints n=92).

fS4

Figure S4. Outcome stratification by high and low RelA-pSer⁵³⁶ in low risk (upper panel; A-C) and high-risk pediatric AML patients (bottom panel; D-F). A and D represent overall survival, B and E event-free-survival and C and F show relapse risk after second round of induction.

fS7

Figure S7. Waterfall plot for HSF1-pS326 in pediatric AML

fS8

Figure S8. Waterfall plot for RelA-pSer536 in T-ALL

fS9

Figure S9. Kaplan-Meier survival analysis for RelA-pSer⁵³⁶ and HSF1-pSer³²⁶ combined in ADE-treated patients. A) OS, B) EFS, and C) RR for patients stratified based on their RelA-pSer⁵³⁶ and HSF1-pSer³²⁶ expression levels. Solid grey: low-RelA-pSer⁵³⁶, low-HSF1-pSer³²⁶; dashed grey: low-RelA-pSer⁵³⁶, high-HSF1-pSer³²⁶; solid black: high-RelA-pSer⁵³⁶, low-HSF1-pSer³²⁶; dashed black: high-RelA-pSer⁵³⁶, high-HSF1-pSer³²⁶.

fS10

Figure S10. Kaplan-Meier survival analysis for RelA-pSer⁵³⁶, HSF1-pSer³²⁶ and histone methylation enzyme (HME) groups combined. OS (A), EFS (B) and RR (C) for patients stratified by RelA.pSer⁵³⁶, HSF1.pSer³²⁶ and HME-group.

tS1

Table S1. List of antibodies used on the array.

tS2

Table S2. Multivariate variate analysis in ADEB-treated patients including variables that were significantly associated with outcome in univariate analysis.

tS3

Table S3. Patient characteristics of patients stratified by RelA-pSer⁵³⁶ and HSF1-pSer³²⁶.

Data availability statement

RPPA data used for the analysis was deposited at https://www.leukemiaatlas.org/).