Leukemogenesis in infants and young children with trisomy 21

Children with Down syndrome (DS) due to constitutional trisomy 21 (T21) have a more than 50-fold increased risk of developing acute leukemia before their fifth birthday compared to children without DS.^1,2  Few genetic disorders show such a strong link to leukemia and have offered so many clues to clinicians and scientists about why this should be so. While it is clear that the hematologic problems in DS must be underpinned by the presence of a supernumerary copy of chromosome 21 (Hsa21), the more we learn about potentially relevant mechanisms, the more questions remain and the more enigmatic the answers become. Nevertheless, clinical and mechanistic studies have together significantly improved our understanding of DS leukemias and the way we manage these disorders in patients. Understanding DS leukemogenesis is important given that worldwide approximately 200 000 children every year are born with DS.³  Here, I describe how observations made almost 40 years ago by Alvin Zipursky have triggered a wealth of further studies that provide insight into the prenatal origin of leukemias in DS as well as the fundamental processes governing the biological properties and gene expression of hematopoietic cells.⁴ 

Population-based studies show an increased frequency of both acute myeloid leukemia (AML) and acute lymphoblastic leukemia (ALL) (Table 1).^1,2  Given that the incidence of nonhematologic cancers is half that reported in individuals without DS,^1,2,5  this points to the specific susceptibility of hematopoietic cells to the leukemogenic effects of T21 and also hints at effects on multiple lineages. A second distinctive feature of DS leukemias is the age at presentation, which indicates, particularly for the specific subtype of AML known as ML-DS, that there is a defined time window during which T21-driven leukemic transformation occurs. ML-DS is now known to originate in utero and usually presents with a self-limiting, neonatal leukemic syndrome known as transient abnormal myelopoiesis (TAM), or transient myeloproliferative disorder, that is virtually unique to DS; while most cases of TAM resolve within 3 to 4 months, subsequent transformation to full-blown leukemia is limited to the first 4 years of life (Figure 1).^2,4-6  By contrast, ALL is rare in infants with DS but otherwise has an age distribution similar to ALL in children without DS.^5,7  DS leukemias also exhibit distinct immunophenotypic characteristics. Most notably, blast cells in ML-DS coexpress erythroid and megakaryocytic markers, an uncommon finding in children without DS,⁸  while DS-ALL, unlike non-DS ALL, is almost exclusively B-lineage.⁹  Finally, DS leukemias are molecularly distinct, as described in detail later.

In the 2022 World Health Organization classification, TAM and ML-DS are collectively known as “myeloid proliferations associated with DS typically associated with exon 2 or 3 GATA1 mutations.”¹⁰  The key discovery of the link between somatic N-terminal mutations in the transcription factor gene GATA1 and TAM was made by John Crispino's lab, which had also first demonstrated the same type of GATA1 mutation in children with DS acute megakaryoblastic leukemia (now termed ML-DS).^11,12  Further, his lab showed that such mutations result in the exclusive production of a short GATA1 protein (GATA1s) that lacks the N-terminal activation domain. As the GATA1 gene is on the X chromosome, cells bearing GATA1s-causing mutations also lose expression of the full-length GATA1 protein, thus disrupting its normal activity in regulating hematopoietic cell development, particularly of erythroid and megakaryocyte (MK) lineages, by altering its ability to bind to and regulate downstream targets.^13-16  Although almost all cases of TAM occur in neonates with DS, an identical clinical presentation can also affect neonates with mosaic DS.^17,18  These babies lack the characteristic phenotypic features of DS, but peripheral blood karyotyping shows that some or all of the hematopoietic cells harbor T21, suggesting that both T21 and GATA1s-producing mutations are essential for the development of TAM/ML-DS. Indeed, in the single reported case of TAM with a GATA1 mutation and no evidence of T21, the GATA1 mutation was a large deletion rather than a producer of GATA1s.¹⁹  In addition, inherited GATA1s-producing mutations are not leukemogenic in disomic individuals,²⁰  although rare families with germline GATA1s mutations are reported where the development of ML-DS–like acute leukemia was accompanied by acquired trisomy or tetrasomy 21,^21,22  further supporting the importance of GATA1s/T21 as a potent oncogenic driver.

Studies on prenatal and cord blood samples from babies with DS have shown that somatic GATA1 mutations are acquired before birth, most likely during the second trimester.^23-25  Furthermore, multiple GATA1 mutations occur in approximately 25% of cases, suggesting a strong selective advantage for GATA1s in fetal cells.^6,24  We previously showed that T21 itself causes a striking increase in fetal liver megakaryocytic-erythroid stem and progenitor cells in DS in the absence of GATA1s mutations, suggesting that it is the T21-mediated perturbation of fetal hematopoiesis that underlies the leukemogenic effects of GATA1s in neonates with DS.²⁶  Paired ML-DS and TAM samples show the same GATA1 mutations, confirming that ML-DS and TAM are clonally linked conditions and suggesting that other factors, in addition to GATA1s, are needed to effect a full leukemic transformation from a transient neonatal syndrome to ML-DS, a condition that is fatal in the absence of chemotherapy.^24,27,28  Whole-genome and targeted sequencing in more than 300 patient samples not only confirmed the presence of GATA1s mutations in ML-DS but also identified loss-of-function mutations in cohesin genes as the principal secondary genetic events in ML-DS; mutations in genes encoding epigenetic regulators and components of tyrosine kinase signaling pathways were also common.^27,28  While these secondary events can occur in other leukemias, their coacquisition with GATA1s is virtually unique to ML-DS. Thus, together these findings support a multistep model of leukemogenesis in ML-DS in the vast majority of cases (Figure 1) and provide a framework both for understanding the natural history of ML-DS and for investigating the precise mechanisms responsible.

TAM is a fetal and neonatal condition. Most cases present at or just after birth and always by 3 months of age.²⁵  Although TAM is usually a short-lived disease that resolves without specific therapy, in fact it encompasses a wide spectrum of severity, from clinically silent disease to rapidly fatal multiorgan failure due to the leukemic infiltration of multiple tissues, particularly the liver and lungs.^17,18,29  While clinical studies report that approximately 10% of neonates with DS have TAM,^4,6,17,18  prospective studies, including our own, documenting the frequency of GATA1s mutations in DS neonates using sensitive next generation sequencing indicate an even higher frequency of TAM, of approximately 30%, although many of these cases are clinically and hematologically silent.⁶  As the estimated incidence of DS worldwide is 1 per 500 to 1 per 2000 live births,³⁰  TAM is by far the most common leukemia in neonates. The characteristic hematologic features of TAM are key to the diagnosis and to our understanding of the pathogenesis of the disease. While neonates with silent TAM due to small GATA1s clones have no reliable hematologic markers of disease, most cases of clinical TAM have leukocytosis and increased peripheral blood blasts (>10%) (Table 2). While our group uses a blast percentage of greater than 10%, others recommend defining TAM on the basis of blasts greater than or equal to 5%.³¹  Importantly, however, the blast percentage can only serve as a guide for the presence of GATA1s mutations because more than one-third of DS neonates without GATA1s mutations have blasts greater than or equal to 5%, automated counts are unreliable, and manual counts require considerable experience. Typical blast cells in TAM resemble immature or partially differentiated megakaryoblasts (Figure 2), but they are often pleomorphic and may display features of erythroid, basophil, or eosinophil differentiation,^32-34  reflecting the roles of GATA1 in multiple lineages.³⁵  Consistent with this, in addition to CD117 (c-kit) TAM blasts coexpress variable proportions of CD34, CD7, CD36, CD41/42b, and CD235a (glycophorin A).⁸  Morphologic evidence of the perturbation of megakaryopoiesis by GATA1s is almost always seen on blood smears in TAM (Figure 2), although thrombocytopenia is not a reliable diagnostic indicator of TAM because the platelet count may be normal, reduced, or even increased.²⁹  While anemia is uncommon, the median hematocrit (Hct) is lower in neonates with TAM compared with DS neonates without TAM (Table 2). This is consistent with some impairment of erythropoiesis by GATA1s, although less than seen in disomic models of GATA1s function, perhaps due to compensation by the underlying T21-associated increase in erythropoiesis in neonates with DS.^6,13,14,29 

For neonates with severe TAM (5%-20%), factors for early death include preterm delivery, ascites, a leukocyte count greater than 100 × 10⁹/l, hepatomegaly, and coagulopathy.^17,18,29  For such cases the mainstay of treatment is supportive care and the judicious use of cytarabine as recommended in recent guidelines.^29,36  While cytarabine reduces the risk of early death in severe TAM, there is no evidence yet that treatment prevents ML-DS.^37,38  In most neonates, TAM resolves within 3 to 4 months, even when the disease is severe. As 10% to 20% of cases of clinical TAM subsequently develop ML-DS, even after apparent complete remission, it seems sensible for children with TAM to be offered regular review until the age of 4 years, the time window during which these children are at risk of ML-DS.^2,5,17,18,29  At present, no clinical, hematologic, or molecular features at birth reliably predict which children develop ML-DS, although molecular or flow cytometric evidence of residual disease at 3 months confers an increased risk of subsequent ML-DS.^37,38  In our experience, the persistence of residual GATA1s clones beyond 4 months makes subsequent ML-DS almost inevitable, although such clones are usually too small to produce detectable molecular, hematologic, or clinical evidence of their presence. Although the median age at diagnosis of ML-DS is 12 to 18 months, the evolution of TAM to ML-DS usually manifests as slow, progressive pancytopenia over several months with occasional circulating blasts. However, in some cases leukemic transformation is much more acute, with rapidly increasing blasts, so any change in the blood count during follow-up should prompt a careful review of the blood smear. Adaptations to ML-DS chemotherapy regimens to minimize the increased toxicity of many drugs for patients with DS have improved disease-free survival to 90%, although the outcome for children who relapse remains poor.³⁹ 

In contrast to ML-DS, children with DS-ALL have an inferior outcome to those without DS, both because of treatment-related toxicity and intrinsic differences in ALL biology.^9,40  A recent large international study matched for cytogenetic subgroup showed that compared to children without DS, those with DS had worse 5-year event-free survival (75% vs 88%), overall survival (77% vs 94%), and post-induction treatment-related mortality (12.2% vs 2.7%).⁴⁰  Survival after relapse is also inferior, even after transplantation, although chimeric antigen receptor T-cell therapy has shown promise.⁴¹  While the clinical features are similar to non-DS ALL, the pattern of cytogenetic abnormalities in DS-ALL is distinct, consistent with the leukemogenic role of T21 in this disease. The frequency of “good prognosis” cytogenetic abnormalities, such as ETV6-RUNX1 and high hyperdiploidy, is lower in DS-ALL, and intriguingly, BCR-ABL translocations and KMT2A gene rearrangements are rare or absent in most series.^9,40  Instead, the most common genomic alteration in DS-ALL is the rearrangement of CRLF2 (CRLF2r), which is found in approximately 60% of cases, half of which also have activating mutations in JAK2.^42-44  A higher proportion of RAS/MAPK pathway mutations is also reported in DS-ALL, usually in JAK2 wild-type cases or subclones.^45,46  In addition, increasing evidence suggests the potential importance of heritable risk loci (such as PAX5 or IKZF1 deletion) both in the risk of developing ALL in children with DS and also in treatment outcome, particularly for IKZF1 deletion in which a potential mechanism involving decreased enhancer activity, differential protein binding, and effects on the cell growth of IKZF1 variants has been reported.^40,47  Finally, it is notable that in children without DS, T21 is one of the most frequent chromosomal abnormalities in B-ALL, particularly in high hyperdiploid ALL, where up to 100% of cases have 1 to 3 additional copies of Hsa21,⁴⁸  supporting a key role for increased Hsa21 dosage in lymphoid leukemogenesis.

Although immunophenotypic and transcriptome analysis suggests that DS-ALL blast transcriptomes are mainly shaped by their molecular cytogenetic subgroup,^45,49  clinical and mechanistic studies also support an important role for T21 at several stages in leukemogenesis. Indeed, the acquisition of a supernumerary Hsa21 is always a nonrandom event restricted to certain cytogenetic subtypes of DS-ALL (eg, ETV-RUNX1 and P2RY8-CRLF2); abnormalities of Hsa21 have also been estimated to affect up to 65% of all childhood B-ALL cases.^50,51  The existence of a prenatal stage of DS-ALL, although postulated, has never been demonstrated and might seem unlikely given the rarity of DS-ALL in children under 12 months of age.⁹  On the other hand, there is increasing evidence that T21 severely impairs fetal B-progenitor development in humans, due both to intrinsic hematopoietic stem/progenitor defects and extrinsic regulation through an altered hematopoietic microenvironment,^26,52  and that quantitative and qualitative defects in B cells are common in children with DS. This suggests that prenatal perturbation of B lymphopoiesis by T21 could shape the cellular context for leukemia initiation.

While clinical studies provide irrefutable evidence that T21 itself is an important factor in DS leukemogenesis, model systems have been used to more precisely decipher how the supernumerary copy of Hsa21 in hematopoietic cells provides such a strong leukemogenic stimulus. Hsa21 has approximately 230 protein- coding genes, many of which have key roles in normal and malignant hematopoiesis (such as DYRK1A, ERG, ETS2, and RUNX1), as well as almost twice as many non–protein-coding genes, including 5 microRNAs (miRNAs).³  Most studies of DS leukemogenesis have focused on the dosage of genes on Hsa21, either using mouse models based on segmental trisomies or induced pluripotent stem cells (iPSC) (Table 3).^14,53,54  While some Hsa21 genes and tissues exhibit the expected 1.5-fold higher expression levels than the corresponding disomic tissues, differences in gene expression are highly tissue-specific and cell population-specific. This may explain why, even though the increased dosage of several Hsa21 genes has been shown to be leukemogenic in model systems, direct links between altered gene expression and hematopoietic phenotypes in DS have been so difficult to establish and why none of these phenotypes has so far been explained by a single Hsa21 gene acting alone. Furthermore, many studies now show that the perturbation of gene expression by T21 is genome-wide.^3,52,55  This suggests that epigenetic mechanisms of cellular adaptation to aneuploidy are likely to be involved, particularly during fetal life when the exquisite control of gene expression is crucial for normal development. In line with this, recent work has found genome-wide effects on the DNA methylation of hematopoietic cells in cord blood from neonates with DS as well as significant changes in the methylation of regions close to genes key to normal hematopoietic development, such as RUNX1 and FLI1.⁵⁶ 

Clinical studies indicate that leukemia initiation in DS, at least for ML-DS, critically depends upon a fetal cell context as well as the presence of T21. Euploid mouse models highlight developmental-specific differences in GATA1s function, but these do not develop leukemia.¹⁵  Similarly, several mouse models based on segmental trisomies, iPSC, or the overexpression of a single Hsa21 ortholog have not so far fully recapitulated human TAM/ML-DS despite identifying several potent oncogenic candidates (eg, DYRK1A, ERG, ETS2) that can trigger myeloproliferative or leukemia disorders,^57,58  perhaps because they lack key features of human fetal cells. In line with this, models based on insertional mutagenesis in murine fetal liver cells more closely mimic the immunophenotypic and molecular characteristics of TAM/ML-DS.^27,59,60  Furthermore, Wagenblast and colleagues used clustered regularly interspaced short palindromic repeats-Cas9 editing targeting GATA1 and STAG2 in human DS fetal liver cells to investigate the mechanism of cooperation between T21, GATA1s, and cohesin loss in DS leukemogenesis and successfully created a model that recapitulated many aspects of TAM and ML-DS.⁶¹  Using a combination of genome-wide analysis of GATA1-binding sites, miRNA profiling, and overexpression and knockout (KO) experiments, they went on to show that a subset of Hsa21 miRNAs (miR-99a, miR-125b-2, and miR-155) was responsible, at least in part, for a GATA1s-induced TAM-like leukemia in fetal cells,⁶¹  supporting earlier data showing a role for Hsa21 miRNAs.^59,60  Intriguingly, in their model, T21 appeared unnecessary for the development of ML-DS as both DS and normal fetal liver cells could be transformed to a similar degree when dual edited to produce GATA1s in the setting of STAG2 KO. If T21 is mostly important for leukemia initiation in fetal cells in DS myeloid leukemias, as these data suggest, this may explain why TAM and the acquisition of GATA1s mutations are confined to fetal and neonatal life and why ML-DS is restricted to a time window during which fetal cells can still survive postnatally.

The role of T21 in leukemia initiation and progression in DS-ALL has mainly been studied using mouse models in tandem with the molecular analysis of primary patient samples.^7,62-65  Several genes on Hsa21 have been shown to be important for normal B-cell growth and differentiation, including DYRK1A, ERG, ETS2, HMGN1, and RUNX1; overexpression of DYRK1A and also of HMGN1 have been directly linked to the pathogenesis of B-ALL.^62-65  Increased expression of DYRK1A has been reported in a number of leukemias, including B-ALL, and has recently been shown to be necessary for the growth of B-ALL cells.⁶³  Although not specifically investigated in DS-ALL, this work is particularly interesting because it showed that small-molecule inhibitors targeting DYRK1A were effective in animal models of B-ALL.⁶³  Also taking differential gene expression in primary patient cells as a starting point, Lane et al. performed experiments in the Ts65Dn mouse, one of the most well-characterized segmental T21 mouse models, and not only narrowed down the putative region of Hsa21 (21q22) to a region of 31 genes sufficient to alter B-cell self-renewal and differentiation but also attributed it to overexpression of just 1 of these, HMGN1, which encodes a nucleosome-binding protein.⁶⁴  They went on to demonstrate that overexpression of HMGN1 caused both a global increase in histone H3K27 acetylation and downstream B-lineage gene expression and also promoted the development of B-ALL, albeit in cooperation with BCR-ABL or a combination of 4 DS-ALL cytogenetic events (CRLF2 overexpression, an activating JAK2 mutation (JAK2^R683G), Pax5 haploinsufficiency, and expression of a dominant negative Ikzf isoform).^64,65 

Clinical and experimental studies, only some of which are summarized here, have together yielded fascinating insights into DS leukemogenesis and underscore the complexity of the role of T21 in hematopoietic cells. The effects of T21 are developmental stage–specific and are most evident before birth, when striking abnormalities of erythro-megakaryopoiesis and B lymphopoiesis are present in the liver and bone marrow (Figure 3). The presence of T21 and somatic GATA1s mutations in fetal hematopoietic cells is essential for leukemia initiation in TAM/ML-DS. Whether rewiring of gene expression programs in fetal early lymphoid or B progenitors persists into early childhood and cooperates with aberrant CRLF2/JAK2 and/or RAS function to initiate DS-ALL is not known. Many questions remain to be answered, not least the reasons for the high frequency of somatic GATA1 mutations in DS fetal blood cells and of CRLF2/JAK2 aberrations in DS-ALL. Evidence for a mutagenic phenotype or a generalized defect in DNA repair in DS tissues is sparse⁶⁶  and precisely how hematopoietic cells to adapt to the potentially devastating effects of aneuploidy, and the extent to which this varies in different cell types, remains to be determined.

Irene Roberts: no competing financial interests to declare.

Irene Roberts: nothing to disclose.