The Use of DNA Methylation Profiling to Assess the Significance of Different Types of Mutations in CEBPA-Mutated AML

In AML, the favorable prognosis associated with mutations in the CEBPA gene is restricted to those cases with double CEBPA mutations (CEBPA^DM), consistent with the loss of normal CEBP/alpha activity from both alleles. Current recommendations are that CEBPA^DM-mutated patients should not receive a stem cell transplant in first remission. In general, these cases have 2 ‘classical’ mutations, an N-terminal out-of-frame insertion/deletion that leads to loss of the full-length p42 protein and increased levels of the p30 isoform translated from an internal start site, coupled with a C-terminal in-frame insertion/deletion in the DNA binding domain (DBD) or leucine zipper domain (LZD) that interferes with DNA binding or dimerization. However, in our study of 1427 younger adult patients, 26% of mutations did not fit this classical description due to either the location or type of mutation. Furthermore, of the CEBPA^DM cases, 20% had a classical plus a ‘non-classical’ mutation or a homozygous non-classical mutation. It will be important to understand the functional consequences of these atypical mutations if CEBPA genotype is to be used to determine patient management. As methylation profiling has shown that CEBPA^DM cases form a distinct hypermethylated cluster, we investigated whether this can provide information about non-classical cases.

A test set of 40 diagnostic samples were analyzed on the Illumina Infinium 27K Human Methylation Array, all normal karyotype with wild type (WT) NPM1, FLT3^ITD and FLT3^TKD; 10 were CEBPA^DM, 30 CEBPA^WT. Unsupervised cluster analysis showed that the 10 CEBPA^DM cases clustered within a group of 16 hypermethylated cases that separated from 24 hypomethylated cases. A methylation signature was created from the 25 most-differentially methylated CpG sites between the CEBPA^DM and CEBPA^WT cases and used to examine a validation set of 95 samples analyzed on the Illumina Infinium 450K Human Methylation Array (31 CEBPA^DM, 38 single-mutated CEBPA [CEBPA^SM], 26 CEBPA^WT). This included 38 cases with non-classical mutations, 14 of them CEBPA^DM. On unsupervised cluster analysis, most CEBPA^DM cases (81%) fell in a hypermethylated group that was distinct from CEBPA^SM and CEBPA^WT cases, with no segregation between the latter. We derived a genotype predictor by comparing the % methylation in a sample at each of the 25 CpG sites with that in the CEBPA^DM and CEBPA^WT signatures to determine which signature the sample data most approximated. This correctly predicted 25/31 (81%) of the CEBPA^DM cases, including 2 with missense DBD/LZD mutations (A295P, N321S) coupled with a classic N or C mutation, 2 with homozygous classic C mutations, indicating that presence of the p30 isoform is not required for the methylation profile, and 5 with a classic N mutation coupled with a truncating mutation in the middle of the gene, consistent with the presence of the p30 isoform alone. This data was supported by functional evaluation of mutant constructs in a luciferase reporter assay to assess DNA binding and transactivation activity (TA). Classic CEBPA^DM constructs all had significantly lower TA than CEBPA^WT (mean 12%, 27%, 15% of CEBPA^WT for homozygous N, homozygous C and N+C constructs). Combination of a classic N mutation with missense DBD mutations (A295P, R297P, R300P), a LZD truncation (K313fs) or a middle region truncation (Q209fs, A238fs), all led to ≤15% activity consistent with almost complete loss of CEBP/alpha activity.

Of the 6 CEBPA^DM cases that did not cluster as expected, 1 with classic N+C mutations had a lower mutant level (mean 28% for the 2 mutations compared to 45% for 9 other pairs with available data) and 1 had a homozygous missense LZD mutation that did not show reduced TA that could explain the discrepancy. The other 4 all had high mutant level (mean level ≥39%) and biallelic mutations as assessed by cloning, and relevant constructs showed low TA (≤19%). The reason for their misclassification is therefore not apparent, although we cannot exclude the possibility of other coincident mutations influencing methylation.

These data indicate that the hypermethylated profile associated with CEBPA^DM cases holds true for most of the CEBPA mutations identified in patients and can be used to support predicted functional consequence of the mutations. This may be particularly useful in determining management in CEBPA^DM cases with non-classical mutations.