This prospective study, the first to apply genome-wide SNP array analysis to paired bone marrow and normal DNA specimens from a large cohort of patients with MDS, identified cryptic chromosomal changes with both clinical and molecular genetic implications. Our findings differ in important aspects from those in retrospective studies based on DNA extracted from archived bone marrow cell samples (11
). Importantly, the overall rate of genomic abnormalities in MDS patients with normal bone marrow karyotypes was much lower in our analysis than in earlier studies. We attribute this discrepancy to the lack of paired normal DNA samples in most previous analyses resulting in the spurious identification of inherited CNVs or regions of apparent homozygosity as somatically acquired genomic alterations within the MDS clone. Indeed, we demonstrate that the majority of apparent abnormalities detected by analysis of bone marrow cell DNA alone reflect inherited genomic diversity rather than clonal somatic abnormalities. Thus, a straightforward comparison with matched normal DNA is required to identify unequivocally the more subtle aberrations arising from acquired clonal genomic changes. Hence, an attempt to correlate UPD or copy number changes with clinical outcome (14
) in patients who have been incorrectly classified due to uncontrolled assessment of genomic variation will not yield valid results.
Aside from its role in assessing bone marrow cell clonality in newly diagnosed cases of MDS, SNP array analysis promises to help unravel the molecular pathogenesis of these complex stem cell diseases. Indeed, identification of discrete clonal molecular changes in MDS by high-density SNP array analysis will likely provide the best starting point for the discovery of new genetic mutations and signal transduction pathways involved in MDS, leading to the development of more effective targeted therapies. We would emphasize that the smallest regions of copy number alteration or LOH are often the most informative in implicating individual genes for detailed sequence analysis, but these small regions are also the most difficult to distinguish from nonpathologic human germline variations. Hence, consistent use of paired bone marrow and normal DNA samples would also be expected to accelerate the pace of disease allele discovery in MDS.
We also addressed the question of whether CD34+ cell selection is required to assess the clonal aberrations in MDS. All of the aberrations found in the CD34+ fraction were also identified in unselected mononuclear cells, indicating that the clonality of the MDS bone marrow is remarkably high. Our results suggest that the cells lacking MDS-associated clonal changes in the bone marrow-derived samples are mainly circulating blood T-lymphocytes (see Supplementary Fig. 2
), indicating that depletion of mature T cells might be a useful strategy for removing residual normal cells from the bone marrow aspirate. In any event, SNP array analysis in MDS can be performed with the mononuclear fraction of whole bone marrow cells, and is therefore clinically feasible even when, as is often the case, the MDS bone marrow sample contains low numbers of cells.
We report the identification of UPDs in four of 33 patients with a normal karyotype. Notably, all segmental UPDs discovered in this study were terminal (see Supplementary Fig.5
for summary). This suggests that break-induced replication (BIR (26
)) might be a dominant mechanism by which a cell duplicates a somatically acquired event such as a mutation, a microdeletion or an epigenetically suppressed region and consequently become homozygous for this segmental region. BIR appears to be a common repair mechanism at stalled or broken replication forks, however, only the reduplication of a region that contains a genetic or epigenetic alteration conferring a growth advantage to the cell would allow for its clonal dominance and selection, leading to its detection by LOH analysis. In this regard, our findings of nonrandom segmental UPDs help to clarify models purporting to explain the pathogenetic basis of large deletions that are typically observed in MDS cases. Certain large deletions, such as the del(5q), are thought to arise from haploinsufficiency for one or more genes within the targeted regions (21
). Indeed, haploinsufficiency for the RPS14
gene has recently been linked to the pathogenesis of MDS associated with the 5q- syndrome(7
); homozygous inactivation is not observed in such cases because the RPS14 protein is essential for cellular protein synthesis. Consistent with this model of molecular pathogenesis, we did not observe any UPD affecting 5q in MDS cases. Apparently, deletion of one allele of 5q is sufficient for pathogenesis, and loss of the wild-type alleles with reduplication of mutated or microdeleted target genes on 5q, as would occur with a UPD, is strongly selected against, because it is lethal to the MDS stem cell. By contrast, we did observe UPD affecting the same deleted region that is affected in cases with cytogenetically apparent loss of the long arm of chromosome 7. This implies that at least one mutated or epigenetically suppressed gene in this region is likely reduplicated together with loss of the normal allele in cases with 7q UPD, fulfilling Knudsen's hypothesis for the homozygous inactivation of classical tumor suppressor genes (33
We hypothesize that genes within the region of UPD on chromosome 7 are likely to harbor inactivating point mutations that will eventually be identified by judicious sequencing of the involved genes in specific cases with UPD. By contrast, other regions of UPD that do not correspond to regions of cytogenetic deletion may harbor activating mutations that are duplicated by UPD and thus provide a growth advantage (34
mutations in acute myeloid leukemia (AML) fit the latter category, in that they are frequently identified on both alleles in AML cases with a normal karyotype. Thus, SNP array analysis can provide critical information needed to pinpoint and identify mutated genes and altered signal transduction pathways in MDS.
A major goal of our study was to detect clonal genomic abnormalities in MDS patients that could be used to improve the clinical management of these disorders. Although longer follow-up times are needed to determine the association of our SNP array findings with treatment outcome, several results appear to have immediate relevance to patient management. The ability to detect clonal genomic aberrations in cases with a normal bone marrow cell karyotype using SNP array analysis can distinguish MDS from other causes of bone marrow cell dysplasia and pancytopenia due to the effects of drugs, environmental toxins or aberrant immune responses, thus aiding in the initial diagnosis. Moreover, our identification of UPDs affecting chromosome 7 in two patients with low-risk IPSS scores who later showed rapidly deteriorating clinical courses is intriguing. UPD selects for homozygosity of a specific genomic region, and the same region of chromosome 7q affected by UPD is also very often deleted in MDS cases. When identified by cytogenetic analysis, abnormalities of chromosome 7q by themselves signify an increased IPSS score. Thus, it is possible that the 7q UPDs we identified are equivalent to a deletion in this region and may constitute a high-risk feature in MDS (17
). Whether or not UPDs affecting this and other genomic regions convey important prognostic information is an important question to address in additional prospective studies using SNP array technology to evaluate patients with MDS entering clinical therapeutic trials.