Letter to the Editor
Acute myeloid leukemia (AML) is a clonal hematopoietic disease caused by both inherited
and acquired genetic alterations. Conventional cytogenetic analysis reveals chromosomal
aberrations in about 50% of AML cases. These chromosomal rearrangements promote leukemia
by affecting cell proliferation, differentiation, and/or survival. In addition to
these genetic alterations, cooperating mutations are required for leukemogenesis (1–3).
Among subtypes of AML, pediatric acute megakaryoblastic leukemia (AMKL) is primarily
associated with three cytogenetic subtypes: one with trisomy 21 plus mutation of GATA1,
one with the t(1;22) translocation (4, 5), and a third recently reported recurrent
cryptic inversion of chromosome 16 (inv(16)(p13.3q24.3)) (6, 7). AMKL with t(1;22)
is typically an aggressive disease, presenting in early infancy with a median survival
of just 8 months. Translocation between chromosomes 1p13 and 22q13 results in the
fusion gene, RBM15-MKL1. The mechanism(s) by which the fusion protein promotes leukemia
is unknown. RBM15-MKL1 knock-in mice develop AMKL, but with delayed onset (16–18 months)
and incomplete penetrance (8). Induced RBM15-MKL1 overexpression in cell lines causes
cell death (9). Together, these findings suggest that RBM15-MKL1 requires cooperating
mutations to induce AMKL. To identify potential cooperating mutations in t(1;22) AMKL,
we performed whole exome capture followed by next-generation sequencing in diagnostic
and remission samples from a t(1;22)-AMKL patient. An 11 week-old boy presented with
loss of appetite and fussiness. Samples of bone marrow and peripheral blood were obtained
at presentation. Morphological and immunophenotypic analyses were performed at the
UC Davis Pathology Department according to standard protocols. FISH and cytogenetic
analyses were performed by the Mayo Clinic Laboratory (Children’s Oncology Group designated
laboratory). The remaining diagnostic sample was frozen. The blood had a white blood
cell count (WBC) of 38,000/mm3, hemoglobin of 6.1g/dL, and platelet count of 99,000/mm3.
The peripheral blood smear showed immature blast cells (18% of WBC). Bone marrow examination
revealed severe fibrosis (Figure 1A, B). Peripheral blood immunophenotyping showed
15% leukemic blasts based on CD45 expression and SSC profile. These cells expressed
CD33 (75%), CD34 (44%), CD41 (63%) and CD61 (79%) and cytogenetic analyses showed
46,XY,t(1;22)(p13;q13) (Figure 1C). Based on the fibrotic marrow, megakaryocytic markers,
and translocation, the diagnosis of AMKL was assigned. He responded very well to the
treatment with cytarabine, etoposide, daunorubicin, mitoxantrone, L-asparaginase,
and Gemtuzumab, the Children’s Oncology Group AAML0531 protocol. Because he did not
have an HLA matched family donor, he did not undergo allogeneic hematopoietic stem
cell transplantation. He is currently disease free 4 years after completing therapy.
Peripheral blood was collected and cryopreserved 6 and 9 months after remission was
achieved.
The quality of extracted genomic DNA from the fixed cells of the diagnostic sample
using the Gentra Puregen kit (Qiagen) was highly comparable to that of freshly thawed
peripheral blood of the remission sample (A260/280 = 1.89 vs. 1.86, leukemia and remission,
respectively) and no difference in quality was seen in the further analyses, including
exome capture, PCR amplification, and sequencing. To delineate the breakpoints on
chromosomes 1 and 22, PCR was performed as described (5). Multiplex PCR was performed
with combinations of primers (as shown in Figure 1D) to narrow down the breakpoint
region. When single primer pairs were then used for individual PCR reactions, no amplification
occurred with OTT4 and In4.1-3, but 0.5 kb and 2.2 kb products were amplified with
OTT4 and In4.4 and In4.5, respectively (Figure 1E) confirming that the translocation
occurred between chromosome 1 and 22 with the breakpoint in intron 1 of OTT and intron
4 of MKL1.
To identify potential cooperating mutations in this patient sample, exome capture
followed by deep sequencing was performed on the leukemia and remission blood samples
from the patient. A total of 88,807,512 and 79,716,978 reads were obtained from the
leukemic and remission samples, respectively (Supplementary Table S2). Picard software
identified and removed PCR duplicates (14.0% of the leukemia and 9.4% of the blood
sample). About 96% of the remaining unique reads from both samples mapped to only
one location in the genome with ~ 94% of targeted bases with >10× coverage (Table
S2). Varscan identified 97,926 single nucleotide variations (SNVs) with a minimum
variant frequency in the sample of >0.02 and minimum average quality score >20. Of
these variants, 22,326 were called somatic variations, which were filtered to require
<0.2 p-value that the SNV was different in leukemia vs. remission, leaving 4,580.
Of the 4,580 SNV, 2,123 were in genic regions and 95.3% were classified as coding
variants, with 75.2% being non-synonymous. SIFT predicted 36.0% and 62.9% of the coding
variants as damaging and tolerated variations, respectively (Supplementary Table 3).
Of 1,956 coding variants, 1,876 were novel SNPs based on dbSNP. Of the 1,956, 377
SNVs remained after further filtering to require <0.05 p-value and damaging/tolerated
SIFT predictions, and to exclude known dbSNPs. To overcome the low blast purity (15%
of cells), we used a less stringent filter to maximize the chance of detecting variations
within the leukemic cells, which resulted in false positive variants from the initial
analysis. Therefore, it was necessary to carefully inspect the reads manually using
IGV. After the manual inspection of the 377 SNVs for base quality displayed in the
Integrated Genomics Viewer (IGV v2.1,(10)), we found 12 variants to be likely candidates
and performed further validation for these 12 variants (Table 1). Neither known AML-associated
mutations nor mutations in tumor suppressor genes were found. However, small indels
would not be detectable in our data due to the low blast percentage.
To validate the candidate mutations, Sanger sequencing was performed, but proved to
be technically challenging due to the low percentage of blasts. By Sanger sequencing,
the expected peak for the variant alleles was not distinguishable except for one in
HERC2, which proved to exist in both the leukemic and remission samples at approximately
50%, suggesting a unique SNP in the patient (data not shown). To improve the sensitivity,
10 of the remaining 11 variants were further validated with ddPCR (1 failed in assay
design due to repetitive sequences and high GC content). Despite multiple repetitions
and assessment using modified primer pairs, of the 10 variants assessed using ddPCR,
only MMP8 was validated (Supplementary Figure S1), while the other 9 variants were
not detectable in the leukemic sample (Table 1). Given that, for example, a candidate
mutation with 7% frequency was found in CDK9 using exome sequencing, we had hypothesized
that this was a heterozygous mutation in the leukemic blasts. However, this was not
validated with ddPCR even after analysis of over 14,000 molecules of this region from
the diagnostic DNA sample (Supplementary Figure S1). While CDK9 is known to play a
role in megakaryocyte maturation and to regulate both the cell cycle and transcription,
further inspection of the exome data showed that the sequence quality was low at the
mutant bases. Other unvalidated variant calls had low number of mutant reads, low
mutant base quality, or possible mapping problems. The ddPCR validation demonstrated
that more stringent p-value, mutant base quality and read number filtration is necessary
when analyzing challenging low blast % samples.
The MMP8 mutation in our patient sample, a substitution from C to T on chr11:102592188,
resulted in amino acid change MMP8G189D. This mutation is predicted to be damaging
and deleterious by SIFT and PROVEAN (11, 12), respectively. This region is universally
conserved in all vertebrates except elephant and frog (http://genome.uscs.edu/, (13)).
MMP8 encodes matrix metalloproteinase-8, which breaks down type I, II and III collagens
in the extracellular matrix, and requires both calcium and zinc ions for activity
(14). The variation is located at a calcium ion binding site in the catalytic domain
(Figure S2). To determine the effect of this mutation on enzymatic activity, we generated
constructs containing either wild-type or mutant MMP8 and expressed them in HEK 293T
cells. No difference was observed in expression level between the constructs (Figure
1F). Type 1 collagen substrate gel zymography revealed significantly reduced proteolytic
activity of the mutant compared to wildtype MMP8 (Figure 1G and H, p-value=0.017).
To further assess whether MMP8 may play a role in AMKL leukemogenesis, we evaluated
MMP8 expression in primary AMKL samples. Although RNA was not available from this
case, gene expression data from 41 pediatric AML cases including 16 cases of AMKL
demonstrated 8 of the 16 AMKL cases to have detectable MMP8 expression (Figure S3)
(7). All AMKL samples in this study had >90% purity, an important consideration given
that MMP8 is known to be highly expressed in mature myeloid cells. Transcriptome sequencing
was available for 14 of these AMKL cases and the sample with the highest signal by
exon array also had significant levels of the MMP8 transcript by sequencing. In an
independent cohort of 60 AMKL cases also analyzed by gene expression profiling, a
subset of patients had evidence of MMP8 expression (15). While the purity of these
samples was extremely variable (12–94%), several samples with low purity (M7122, 12%;
M7026, 28%) had low levels of MMP8 expression demonstrating that the variability in
MMP8 expression cannot solely be due to myeloid cell contamination. Furthermore, there
are cases with elevated expression that are relatively pure (M7040, 70%; M7072, 90%).
Although the same mutation was not found in any of the samples, MMP8 is aberrantly
expressed in a subset of AMKL cases. In addition, MMP8 mRNA is expressed in murine
megakaryocyte-erythroid progenitors based on the ENCODE database (GSE40522). Given
that MMP8 was the only candidate gene with 14% variation frequency in the exome sequencing,
the patient had few cooperating mutations in coding regions, which might explain the
excellent response to treatment and survival. In conclusion, we provide the first
data identifying a novel somatic mutation that may provide a key insight to understanding
the pathogenesis of RBM15-MKL1 in AMKL.
Supplementary Material
1
2
3