Comprehensive study on ERG gene expression in normal karyotype acute myeloid leukemia: ERG expression is of limited prognostic value, whereas the accumulation of adverse prognostic markers stepwise worsens the prognosis

The clinical course of normal karyotype acute myeloid leukemia (CN-AML) is very heterogeneous and partly re ﬂ ected by speci ﬁ c molecular abnormalities. 1 The most useful markers implicated in prognostication are FLT3 internal tandem duplication ( FLT3 -ITD), NPM1 mutations (mut), biallelic CEBPA mut and RUNX1 mut, with the latter three being now integrated in the updated WHO classi ﬁ cation. 2 – 4 Beside these, considerably more molecular alterations have been identi ﬁ ed in CN-AML, the prognostic relevance of which is not as clear. Deregulated expression of ERG (ets-related gene) represents one of these alterations, since high ERG expression has been allocated to lower complete remission (CR) rates and shorter disease-free survival, event-free survival (EFS) and overall survival (OS) in some studies, 5 – 7 whereas another study of Marcucci et al. 8 only reported an adverse effect of high ERG expression on the achievement of CR and on EFS. Besides the prognostic value of single alterations, it becomes increasingly important to consider individual markers in their genetic context, as the prognostic impact of the aforementioned parameters may vary depending on the presence (or absence) of other molecular markers. The best validated example is repre-sented by NPM1 mut and FLT3 -ITD, as only NPM1 mut patients without FLT3 -ITD (low-risk) have, in contrast to their FLT3 -ITD positive counterparts, a comparatively better outcome and would therefore no longer bene ﬁ t from allogeneic stem cell transplantation. 2,9 To re ﬁ ne risk-adapted models, the analysis of recently described molecular alterations in the light of other relevant

The clinical course of normal karyotype acute myeloid leukemia (CN-AML) is very heterogeneous and partly reflected by specific molecular abnormalities. 1 The most useful markers implicated in prognostication are FLT3 internal tandem duplication (FLT3-ITD), NPM1 mutations (mut), biallelic CEBPAmut and RUNX1mut, with the latter three being now integrated in the updated WHO classification. [2][3][4] Beside these, considerably more molecular alterations have been identified in CN-AML, the prognostic relevance of which is not as clear. Deregulated expression of ERG (ets-related gene) represents one of these alterations, since high ERG expression has been allocated to lower complete remission (CR) rates and shorter disease-free survival, event-free survival (EFS) and overall survival (OS) in some studies, [5][6][7] whereas another study of Marcucci et al. 8 only reported an adverse effect of high ERG expression on the achievement of CR and on EFS. Besides the prognostic value of single alterations, it becomes increasingly important to consider individual markers in their genetic context, as the prognostic impact of the aforementioned parameters may vary depending on the presence (or absence) of other molecular markers. The best validated example is represented by NPM1mut and FLT3-ITD, as only NPM1mut patients without FLT3-ITD (low-risk) have, in contrast to their FLT3-ITD positive counterparts, a comparatively better outcome and would therefore no longer benefit from allogeneic stem cell transplantation. 2,9 To refine risk-adapted models, the analysis of recently described molecular alterations in the light of other relevant molecular prognosticators is needed. The aim of the present study therefore was to reveal putative associations of altered ERG gene expression to other molecular alterations and to assess the impact of deregulated ERG expression on outcome, either alone and moreover in the context of the previously defined molecular alterations.
A total of 325 younger ( o 65 years) de novo CN-AML patients (169 female, 156 male; median age 53 years, range 18-65 years) were investigated. Of these, 295 patients received intensive treatment according to German standard AML protocols 10 and were subject to prognostic analysis. The diagnosis was made according to World Health Organization criteria. 11 Chromosome banding analysis was performed for all patients according to standard procedures. ERG expression was measured in 64 peripheral blood and 261 bone marrow samples for consistency with our previous analysis, in which the same patients had been characterized for BAALC expression. This previous study aimed at evaluation of the prognostic value of BAALC expression and did not include data on ERG expression. 12 Alterations in ASXL1, CEBPA, DNMT3A, FLT3 (ITD and mutations in the tyrosine kinase domain (TKD)), IDH1, IDH2, MLL, NPM1, NRAS, RUNX1, TET2 and WT1 were analyzed by either polymerase chain reaction, Sanger sequencing or an amplicon deep-sequencing approach. Further details on patient characteristics and the study methodology are provided in the Supplementary Material.
In diagnostic CN-AML samples, the expression of ERG varied within a wide range from 0.1 to 1008% ERG/ABL1 with a median of 189%. First, we evaluated associations of ERG expression levels, as continuous variable, with patient characteristics and molecular markers. In terms of patients characteristics, only a slightly negative correlation of ERG expression levels to age was revealed (r = − 0.235, P o 0.001; Supplementary Table S1). Regarding molecular alterations, ERG expression levels were found to overlap between the different genetic subgroups. Nevertheless, substantial differences in mean ERG expression levels were revealed. Higher ERG expression levels were significantly associated with high BAALC expression, high FLT3-ITD to FLT3wt ratios (⩾ 0.5; further termed FLT3-ITD ⩾ 0.5) and WT1mut as well as with the absence of NPM1mut and IDH1R132mut (Figure 1a). These results are consistent with published data in terms of BAALC and FLT3-ITD, though ERG has been analyzed as a categorical parameter in these previous studies. [6][7][8] Regarding the molecular intermediate-risk group of NPM1wt or FLT3-ITD ⩾ 0.5, mean ERG expression levels were significantly higher as compared with the low-risk group ( Figure 1b). Thus, overall an association of unfavorable prognostic parameters with high ERG expression levels was observed.
Given the strong correlation of high ERG expression levels to high BAALC expression as well as to different molecular genetic alterations, we analyzed correlations of expression of both genes, ERG and BAALC, to molecular mutations grouped into functional biological categories. Again, expression levels of both genes were found to overlap between the different functional subgroups. Slightly higher ERG expression levels were found in patients harboring mutations in one of the myeloid transcription factors, CEBPA and RUNX1, as compared with the patients without these mutations (Figure 1b). Also for BAALC, higher expression levels were significantly related to a mutated status in the myeloid transcription factor group. Further, substantially lower BAALC expression levels were observed in patients harboring mutations in genes involved in DNA methylation, including DNMT3A, TET2, IDH1 and IDH2 (Supplementary Figure S1). Interestingly, aside from the strong correlation to FLT3-ITD neither ERG nor BAALC revealed significant correlation to the activated signaling/proliferation group (Supplementary Figure S1). Therefore, in case of FLT3-ITD, the specific single gene association seems more important than a correlation to activated signaling/proliferation in general.
The impact of different parameters on OS and EFS was assessed by Cox regression analyses. The prognostic value of BAALC expression as a categorical variable (defining high and low expressers at certain cutoff levels 7,12,13 ) has been shown before and could be corroborated, when analyzing BAALC expression as a continues variable, using log transformed    Abbreviations: CI, confidence interval; HR, hazard ratio; ITD, internal tandem duplication; mut, mutation; n.s., not significant; PTD, partial tandem duplication; Pts, patients; TKD, tyrosine kinase domain. a Per 10 years of increase.
Letter to the Editor expression levels (Table 1). ERG expression levels as a continuous log transformed parameter did neither affect OS nor EFS. This is in line with the study of Diffner et al., 14 where ERG expression analyzed as a continuous parameter did not impact on survival, but opposes the aforementioned studies, [5][6][7][8] where ERG expression has been associated with outcome, when dichotomized at certain cutoff levels (median or 75th percentile). Therefore, we performed Kaplan-Meier analysis dichotomizing ERG expression at distinct cutoff levels. A significant correlation to shorter EFS and a trend toward inferior OS was observed for ERG expression levels above the median (Figure 1c). As ERG expression strongly correlates with NPM1wt and FLT3-ITD, we assessed the prognostic value in the respective low-and intermediate-risk groups. As anticipated, no differences in EFS and OS were observed, which contrasts previous studies. 5,6 On the other hand, we found BAALC expression to strongly impact on EFS and OS in the intermediate-risk group of NPM1wt or FLT3-ITD, when dichotomized at the median (further termed low or high BAALC, respectively; Supplementary Figure S2). This result provides important prognostic information as the patients with NPM1 wildtype or FLT3-ITD and high BAALC expression rather reflect OS of the ELN intermediate II-risk group, whereas the respective low BAALC expressers resemble outcome of the favorable-risk group. 15 To clarify whether the sole accumulation of prognostic markers -in contrast to the above-tested specific genetic context of NPM1 and FLT3-worsens prognosis, we determined the number of independent adverse prognostic parameters for each patient and performed survival analyses (Table 1). We defined four subgroups according to the number of adverse prognostic factors, namely high BAALC, FLT3-ITD ⩾ 0.5, MLL-PTD and WT1mut for EFS as well as ASXL1mut, high BAALC, FLT3-ITD ⩾ 0.5, MLL-PTD and WT1mut for OS; with group A: no adverse marker, group B: 1 adverse marker, group C: 2 adverse markers, group D: 3 or 4 adverse markers as none of the patients harbored concomitant alterations in all five adverse prognostic factors. The distribution of the adverse markers within these subgroups is given in the Supplementary Figure S3. Kaplan-Meier analysis revealed a 3-year EFS of 52% in group A, 36% in group B, 26% in group C and 7% in group D and a 3-year OS of 72% in group A, 61% in group B, 35% in group C and 13% in group D (Figure 1d). For EFS, group B and group C did not differ significantly, whereas substantial differences were shown for all other comparisons. Regarding OS, significant differences were shown for all comparisons except for group A versus group B (Figure 1d). In particular, Cox regression analyses revealed that EFS and OS were remarkably related to the number of adverse prognostic parameters (for both P o0.001; HR: 1.54 and HR: 1.70 per unfavorable marker positive, respectively). Thus, EFS and OS differed according to the number of adverse prognostic markers, suggesting that a comprehensive screening of molecular genetic alterations provides additional information for risk assessment in CN-AML. Furthermore, we performed multivariate analysis with the numbers of unfavorable markers (ASXL1mut (only for OS), high BAALC, FLT3-ITD ⩾ 0.5, MLL-PTD and WT1mut: 0 to 4 adverse prognostic markers) and age. Both parameters were independently associated with shorter EFS (for both P o 0.001; HR: 1.70 per unfavorable marker positive, HR: 1.35 per decade) and OS (for both Po 0.001; HR: 1.97 per unfavorable marker positive, HR: 1.51 per decade).
In conclusion, we found ERG expression levels to correlate with specific molecular alteration and moreover to impact on EFS and OS though this impact was dependent on other molecular alterations. Besides the assessment of ERG expression, we were able to demonstrate that both the pattern of molecular alterations as well as the number of independent adverse markers, namely ASXL1mut, high BAALC, FLT3-ITD ⩾ 0.5, MLL-PTD, WT1mut, are relevant for risk stratification in CN-AML.