Introduction

Chromosomal rearrangements involving the human MLL (mixed lineage leukemia) gene are recurrently associated with the disease phenotype of acute leukemias.1, 2 The presence of distinct MLL rearrangements is an independent dismal prognostic factor, while very few MLL rearrangements display either a good or intermediate outcome.3, 4 It became also clear from recent studies that the follow-up of patients during therapy by minimal residual disease (MRD) monitoring has a very strong impact on outcome.5, 6, 7 For this purpose, we established a diagnostic network that allowed different study groups and clinical centers to obtain genomic MLL breakpoint sequences that can be directly used for quantifying MRD levels in patients. The current work flow to identify MLL rearrangements includes a prescreening step (cytogenetic analyses,8, 9 split-signal fluorescence in situ hybridization10, 11, 12 or reverse transcription-polymerase chain reaction (PCR) in combination with long-distance inverse-PCR that was performed on small amounts (∼1 μg) of isolated genomic DNA.13 This allowed us to identify readily reciprocal translocations, complex chromosomal rearrangements, gene-internal duplications, deletions or inversions on chromosome 11q, and MLL gene insertions into other chromosomes, or vice versa, the insertion of chromatin material into the MLL gene.

To gain insight into the frequency of distinct MLL rearrangements, all prescreened samples of infant, pediatric and adult leukemia patients was sent for analysis to the Frankfurt Diagnostic Center of Acute Leukemia (DCAL). Prescreening tests were performed at different European centers (Aarhus, Berlin, Bordeaux, Bratislava, Brest, Bristol, Catania, Copenhagen, Frankfurt, Giessen, Granada, Graz, Grenoble, Haifa, Hamburg, Hanover, Heidelberg, Jena, Jerusalem, Kiel, Lille, Lisbon, Madrid, Minsk, Montpellier, Monza, Munster, Munich, Nancy, Nantes, Newcastle upon Tyne, Olomouc, Padua, Paris, Porto, Prague, Reims, Rotterdam, Tampere, Tel Hashomer, Toulouse, Turku, Tubingen, Vienna, Yekaterinburg, Zabrze and Zurich) and centers located outside of Europe (Boston, Buenos Aires, Hong Kong, Houston, Rio de Janeiro, Seoul, Sydney and Tohoku), where acute leukemia patients are enrolled in different study groups. All prescreened MLL rearrangements were successfully analyzed at the Frankfurt DCAL and patient-specific MLL fusion sequences for MRD monitoring were obtained.

On the basis of the results obtained in this and previous studies,13, 14, 15 a total of 79 direct translocation partner genes (TPGs) and their specific breakpoint regions have now been identified. Seven additional loci have been cloned where the 5′-portion of MLL was not fused to another gene. In 19 other cases, we were not able to identify a der(11) fusion gene. This could be either attributed to a technical problem (such as a too long genomic fragment) or to the fact that no der(11) exists in these few patients. However, in all of these 19 cases, we successfully identified a reciprocal MLL fusion allele. The latter subgroup was allocated to the group of ‘complex MLL rearrangements’ (n=182) because of the extending class of ‘reciprocal MLL fusion genes’ (63 loci, 119 fusion genes). Finally, there were still 35 chromosomal translocations of the human MLL gene that were characterized in the past by cytogenetic methods, but that were never analyzed at the molecular level. Thus, the MLL recombinome presently comprises 121 different ‘direct TPGs’ (decoding the MLL N terminus), whereas the 182 ‘reciprocal TPGs’ (decoding the MLL C terminus) derive from complex rearrangements that involved already known ‘direct TPGs’. It is worth noting that in nearly all of the investigated cases the 3′-MLL gene portion was not lost, except the very few cases (n=4 out of 1622) that were interstitial deletions at 11q23 causing a direct fusion of the 5′-MLL gene portion with a gene portion localized telomeric to MLL, or where we were able to demonstrate that only an MLL spliced fusion exists (n=3 out of 1622). Besides the number of direct and reciprocal MLL fusions, we tried to analyze all available patient data for interesting association between age, sex, disease type, secondary leukemia and breakpoint localization. All these data and their analyses is here presented and discussed.

Patients and methods

Patient material

Genomic DNA was isolated from bone marrow and/or peripheral blood samples of leukemia patients and sent to the DCAL (Frankfurt/Main, Germany). Patient samples were obtained from study groups (the AMLCG-study group, Munich; the GMALL study group, Berlin; Polish Pediatric Leukemia and Lymphoma Study Group; Zabrze; I-BFM network) or other diagnostic centers (Aarhus, Berlin, Bordeaux, Boston, Bratislava, Brest, Bristol, Buenos Aires, Catania, Copenhagen, Frankfurt, Giessen, Granada, Graz, Grenoble, Haifa, Hamburg, Hanover, Heidelberg, Hong Kong, Houston, Jena, Jerusalem, Kiel, Lille, Lisbon, Madrid, Minsk, Montpellier, Monza, Munster, Munich, Nancy, Nantes, Newcastle upon Tyne, Olomouc, Padua, Paris, Porto, Prague, Reims, Rio de Janeiro, Rotterdam, Seoul, Sydney, Tampere, Tel Hashomer, Tohoku, Toulouse, Turku, Tubingen, Vienna, Yekaterinburg, Zabrze and Zurich). Informed consent was obtained from all patients or patients’ parents/legal guardians and control individuals.

Long distance inverse-PCR experiments

All DNA samples were treated and analyzed as described.13, 14, 15 Briefly, 1 μg genomic patient DNA was digested with restriction enzymes and religated to form DNA circles before long-distance inverse-PCR analyses. Restriction polymorphic PCR amplimers were isolated from the gel and subjected to DNA sequence analyses to obtain the patient-specific fusion sequences. This genomic DNA fusion sequence is idiosyncratic for each leukemia patient and was made available to the sender of the DNA sample. The average processing time was around five working days.

Data evaluation and statistical analyses

All clinical and experimental patient data were implemented into a database program (FileMaker Pro) for further analysis. Information about all individual patients was used to compare all defined subgroups and to perform statistical analyses to retrieve important information or significant correlations. χ2 Tests were performed to identify significant deviations from mean values.

Results

The study cohort

To analyze the recombinome of the human MLL gene, 1622 prescreened acute leukemia samples were obtained from the above-mentioned centers over a period of one decade (2003–2013). Successful analysis of the direct MLL fusion could be performed for all patient samples except 19 cases, where only a reciprocal MLL fusion allele could be characterized. In these cases we identified only the reciprocal MLL fusion allele to guarantee MRD experiments. Of those 1622 cases, 1590 entered this study because we obtained all the critical information that was necessary for data processing (gender, age at diagnosis, disease type and subtype or information about de novo or secondary leukemia). A total of 32 cases was excluded from our study because relevant information about these patients were missing; they had the following MLL rearrangements: 9 × MLL-MLLT3/AF9; 5 × MLL-AFF1/AF4; 4 × MLL-MLLT1/ENL; 4 × MLLMLLT10/AF10, 3 × MLL-MLLT4/AF6, 2 × MLL-MLLT6/AF17, 1 × MLL-GAS7, 1 × MLL-EPS15, 1 × MLL-LOC100128568, 1 × LOC387646-MLL and 1 × MLL-partial tandem duplication (PTD). The exclusion of these 32 patients did not interfere with the general conclusions made in this study.

Age distribution according to clinical subtypes

We first analyzed our cohort according to the age at diagnosis. As displayed in Figure 1, the age distribution is quite similar to the expected age distribution known from different cancer registries. Acute lymphocytic leukemia (ALL) incidence has a peak at the age of 2–3 years, and then decreases with age and increases again in older adults. Acute myeloid leukemia (AML) patients display a small peak at 2 years, decline and then steadily increases with age. For the purpose of our study, we separated our cohort into an ‘infant acute leukemia group’ (0–12 months; n=558: 440 ALL, 105 AML, 13 N/A), a ‘pediatric acute leukemia group’ (13 months–18 years; n=416: 205 ALL, 202 AML, 9 N/A) and an ‘adult acute leukemia patient’ group (>18 years; n=616: 333 ALL, 272 AML, 11 N/A). As shown in Figure 1, we also added information about therapy-induced leukemia (TIL; n=77). Thirty-three patients could not be simply categorized into ‘ALL’ or ‘AML’ because they received other diagnoses (MLL=18; myelodysplastic syndrome=5, primary myelofibrosis=1; lymphoma=2) or because we had simply no informations from the corresponding center (unknown disease type=7).

Figure 1
figure 1

Age distribution of investigated patients. The age distribution of all analyzed patients (n=1690) is summarized. (Upper part) Diagram displaying ALL and AML patients. Age at diagnosis was for infants (0–1 year), pediatric (1–18 years) and adult patients (>18 years). The number of ALL, AML and other patients is listed below. We also added the information about TIL patients, the number of complex MLL rearrangements (CL) and specified the ‘Non-ALL’ and ‘Non-AML’ patients (MLL, myelodysplastic syndrome (MDS), primary myelofibrosis (PMF) and unknown) in more detail for each age group. The precise number of patient cases summarized on the right.

Full size image

Identification of MLL rearrangements and their distribution in clinical subgroups

The most frequent MLL rearrangements in these six subgroups were summarized in Figure 2. Infant ALL patients (n=440) displayed 216 t(4;11)(q21;q23) involving the AFF1/AF4 gene, 73 t(9;11)(p22;q23) involving the MLLT3/AF9 gene, 96 t(11;19)(q23;p13.3) involving the MLLT1/ENL gene, 22 t(10;11)(p12;q23) involving the MLLT10/AF10 gene, 1 t(6;11)(q27;q23) involving the MLLT4/AF6 gene, 12 t(1;11)(p32;q23) involving the EPS15 gene and 20 other MLL rearrangements (9p13.3, 9p22, AFF4/AF5, DCP1A/SACM1L, AFF3/LAF4 (2 × ), BTBD18, N/A (9 × ), PICALM, PRPF19, EEFSEC and TRNC18).

Figure 2
figure 2

Classification of patients according to age classes and disease type. (Top) Frequency of most frequent TPGs in the investigated patient cohort of MLL-rearranged acute leukemia patients (n=1590). This patient cohort was divided into ALL (left) and AML patients (right). Gene names are written in black, and percentages are indicated as white numbers. Thirty-three patients could not be classified into the ALL or the AML disease types, respectively. (Middle) TPG frequencies for the infant, pediatric and adult patient group. (Bottom) Subdivision of all three age groups into ALL and AML patients. Negative numbers confer again to the number of patients who were neither classified to the ‘ALL’ nor to the ‘AML’ subgroup.

Full size image

Infant AML patients (n=105) displayed 2 t(4;11)(q21;q23) involving the AFF1/AF4 gene, 23 t(9;11)(p22;q23) involving the MLLT3/AF9 gene, 1 t(11;19)(q23;p13.3) involving the MLLT1/ENL gene, 28 t(10;11)(p12;q23) involving the MLLT10/AF10 gene, 18 t(11;19)(q23;p13.1) involving the ELL gene, 3 t(6;11)(q27;q23) involving the MLLT4/AF6 gene, 1 t(1;11)(p32;q23) involving the EPS15 gene and 29 other MLL rearrangements (11q24, ABI1, ABI2, MLLT11/AF1Q (7 × ), FLNA (2 × ), FNBP1, GAS7, KIAA1524, MYO1F (3 × ), N/A (3 × ), NEBL, NRIP3, PICALM, SEPT6 (3 × ) and SEPT9 (2 × )).

Pediatric ALL patients (n=205) displayed 97 t(4;11)(q21;q23) involving the AFF1/AF4 gene, 37 t(9;11)(p22;q23) involving the MLLT3/AF9 gene, 40 t(11;19)(q23;p13.3) involving the MLLT1/ENL gene, 4 t(10;11)(p12;q23) involving the MLLT10/AF10 gene, 5 t(6;11)(q27;q23) involving the MLLT4/AF6 gene, 4 t(1;11)(p32;q23) involving the EPS15 gene and 18 other MLL rearrangements (1p32, 21q22, MLLT6/AF17, BCL9L, FOXO3 (2 × ), AFF3/LAF4 (3 × ), MAML2 (2 × ), N/A (2 × ), PICALM, RUNDC3B, SEPT5, SEPT11 and TRNC18).

Pediatric AML patients (n=202) displayed 2 t(4;11)(q21;q23) involving the AFF1/AF4 gene, 73 t(9;11)(p22;q23) involving the MLLT3/AF9 gene, 10 t(11;19)(q23;p13.3) involving the MLLT1/ENL gene, 40 t(10;11)(p12;q23) involving the MLLT10/AF10 gene, 19 t(11;19)(q23;p13.1) involving the ELL gene, 2 MLL PTDs, 19 t(6;11)(q27;q23) involving the MLLT4/AF6 gene, 3 t(1;11)(p32;q23) involving the EPS15 gene and 34 other MLL rearrangements (11q23.3, ABI1 (2 × ), ACACA, ACTN4, MLLT6/AF17 (2 × ), MLLT11/AF1Q (4 × ), ARHGEF17, BUD13, CASC5, LAMC3, NA (3 × ), SEPT2, SEPT5, SEPT6 (6 × ), SEPT9 (5 × ), SEPT11, TET1 and VAV1).

Adult ALL patients (n=333) displayed 274 t(4;11)(q21;q23) involving the AFF1/AF4 gene, 6 t(9;11)(p22;q23) involving the MLLT3/AF9 gene, 37 t(11;19)(q23;p13.3) involving the MLLT1/ENL gene, 1 t(10;11)(p12;q23) involving the MLLT10/AF10 gene, 1 t(11;19)(q23;p13.1) involving the ELL gene, 1 MLL PTD, 6 t(6;11)(q27;q23) involving the MLLT4/AF6 gene, 1 t(1;11)(p32;q23) involving the EPS15 gene, and 6 other MLL rearrangements (11q23 (2 × ), ACTN4, CEP164 and TET1 (2 × )).

Adult AML patients (n=272) displayed 3 t(4;11)(q21;q23) involving the AFF1/AF4 gene, 71 t(9;11)(p22;q23) involving the MLLT3/AF9 gene, 12 t(11;19)(q23;p13.3) involving the MLLT1/ENL gene, 20 t(10;11)(p12;q23) involving the MLLT10/AF10 gene, 29 t(11;19)(q23;p13.1) involving the ELL gene, 64 MLL PTDs, 33 t(6;11)(q27;q23) involving the MLLT4/AF6 gene, 4 t(1;11)(p32;q23) involving the EPS15 gene and 36 other MLL rearrangements (MLLT6/AF17 (7 × ), MLLT11/AF1Q (2 × ), AKAP13, AP2A2, ARHGEF12, C2CD3, CASP8AP2, CBL, DCPS, GMPS, CEP170B (2 × ), ME2, MYH11, NA, PDS5A, PICALM, SEPT5, SEPT6 (2 × ), SEPT9 (5 × ), SMAP1, TET1 (2 × ) and TOP3A). All these data are summarized in Table 1.

Table 1 Overview about all investigated TPGs
Full size table

On the basis of the above distribution, about 95% of all ALL patients (n=978) were characterized by the fusion genes MLL-AFF1/AF4 (∼60.0%), MLL-MLLT1/ENL (∼17.7%), MLL-MLLT3/AF9 (∼11.9%), MLL-MLLT10/AF10 (∼2.8%), MLL-EPS15 (∼1.7%) and MLL-MLLT4/AF6 (∼1.2%), respectively. About 84% of all AML patients (n=579) were characterized by the fusion genes MLL-MLLT3/AF9 (∼28.8%), MLL-MLLT10/AF10 (∼15.2%), MLL-ELL (∼11.4%), MLL PTDs (∼11.4%), MLL-MLLT4/AF6 (∼9.5%), MLL-MLLT1/ENL (∼4.0%), MLL-SEPT6 (∼1.9%) and MLL-MLLT6/AF17 (∼1.6%), respectively. This updates recently published data on the frequency and distribution of different MLL fusion partner genes.15, 16, 17

Breakpoint distribution according to clinical subtypes

We also investigated the distribution of chromosomal breakpoints within the MLL breakpoint cluster region in all investigated clinical subgroups. Briefly, the breakpoint cluster region is localizing between MLL exon 9 and MLL intron 11, where the majority of patients had their individual breakpoints (n=1530). Only sixty patients (3.8%) had their breakpoint outside of the major breakpoint cluster region (see Supplementary Table S1).

Of interest, recently published clinical studies put a new focus on chromosomal breakpoint localization: the distribution of chromosomal breakpoints within the MLL breakpoint cluster region was correlated with the outcome of MLL-rearranged leukemia patients.18 Basically, the outcome of leukemia patients with breakpoint in MLL intron 11 was worse compared to those patients with upstream breakpoints. A rational explanation for this observation came from the PHD1–3 domain, which is encoded by MLL exons 11–16. This domain confers oligomerization19 and was described to bind to the CYP33/PPIE protein.20, 21 In addition, the PHD3 domain binds either to CYP33/PPIE or to methylated lysine-4 residues of histone H3.22 Binding of PHD3 to H3K4me2/3 peptides is greatly enhanced by the adjacent bromo-domain,23 but CYP33/PPIE represents a prolyl-peptidyl isomerase and performs a cis–trans isomerization of the proline-1665 residue. This cis-to-trans conversion is mutual exclusive with H3K4me2/3 binding by the PHD3 domain. By contrast, a CYP33/PPIE-bound PHD3 enables binding to BMI1 and associated repressor proteins (HDAC/CBX4/KDM5B), and thus switches the human MLL protein from a transcriptional activator/maintenance factor to a transcriptional repressor. It is worth noting that the adjacent bromo-domain binds to ASB2 and triggers the degradation of MLL.24 Similarly, a recent publication demonstrated that the PHD2 domain also binds another E3 ligase, named CDC34, which controls again the steady-state stability of the MLL protein.25

Breakpoints upstream of MLL exon 11 will not alter the domain structure and the associated functions of the PHD1–3 domain, whereas breakpoints within MLL exon 11 or intron 11 will definitively destroy this cysteine–histidine-rich domain, most likely because of an alternative protein fold.18 This will have several effects on the functions of the resulting fusion proteins, like for example, losing the oligomerization capacity, an increased fusion protein stability or losing the ability to switch into a transcriptional repressor (CYP33→BMI1/HDAC/CBX4/KDM5B).26 As this should impact cancer biology and clinical behavior, we started to analyze the breakpoint distribution for all clinical subgroups and compared them with the mean distribution observed for all 1590 patients. We decided not to use a random distribution of breakpoints because this will be based only on the length of each DNA region, but will not take into account that the specific chromatin features of MLL intron 11 that is highly sensitive against cytotoxic drugs, exhibits a DNase1 hypersensitive site,27 an apoptotic cleavage site,28 an RNA polymerase II binding site29 and several topoisomerase II binding sites.30

For our analyses, we subdivided the MLL breakpoint cluster region into three subregions: (A) exon 9–intron 9=1761 bp; (B) exon 10–intron 10=679 bp; and (C) intron 11–intron 12=4929 bp. The observed ‘mean breakpoint frequencies’ for these three regions were A=38.5%, B=19.5% and C=38.7% for all 1530 patients listed in Supplementary Table S1.

As shown in Figure 3, we first subcategorized all patient cases according to their origin. We had 70 samples from North and South American states, 1403 samples from European countries and 117 cases from Russia, Asian countries or the Australian continent. When analyzing the breakpoint frequencies for A–C, it became obvious that the majority of patients in Europe display a breakpoint distribution that was nearly identical to the mean breakpoint frequencies mentioned above. The South American patient group was very young and displayed a nonsignificant tendency to MLL intron 11 breakpoints (43.5% vs 37.4%), whereas the Russian/Asian/Australian group displayed a shift towards breakpoints localizing within MLL intron 11 (50.43% vs 37.4%, P=0.138). This could neither be attributed to the mean age nor to a higher rate for secondary malignancies (6% vs 5% in Europe). Of interest, all 77 cases of our cohort that were classified as therapy-induced leukemia (TIL) displayed a breakpoint distribution of A=33.8%, B=9.5% and C=54.1%. Thus, even when a controlled exposition to drugs was causing an MLL rearrangement, only a maximum of 54% MLL intron 11 breaks could be reached. As this is the first description of such a phenomenon and we are missing demographic controls, we cannot draw any conclusions about a putative environmental or maternal exposition during pregnancy that would explain such a shift towards MLL intron 11 recombinations. However, when we analyzed this phenomenon in more detail (see Supplementary Table S2), we realized some remarkable differences in certain countries that are even gender specific. Currently, we have no explanations for the observed differences, but future research may help to unravel this phenomenon.

Figure 3
figure 3

World distribution of patients. (Top) Worldmap grossly dividing the investigated patients into three distinct subgroups: American, European and Asian countries. The number of investigated patients is shown and the contribution of individual countries is given in patient numbers. Each country is indicated by its international country code. (Below) Information about the patient cohort. Mean age, age range and the amount of infants (I), pediatric (P) and adult patients (A) is indicated. In addition, we added the amount of therapy-induced malignancies in number and percentage. The breakpoint distribution for each subgroup within MLL exon 9/intron 9, MLL exon 10/intron 10 and MLL exon 11/intron 11 is displayed. Red mark in MLL intron 11: fragile site within MLL that is sensible to exogenous drug exposure.

Full size image

Another observation concerning the breakpoints localization became obvious, when we analyzed breakpoint distributions together with TPGs. As shown in Supplementary Table S3, recombinations affecting MLLT4/AF6 and MLLT10/AF10 display a tendency for MLL intron 9 breaks rather than MLL intron 11 breaks (MLLT4/AF6, P<0.0001; MLLT10/AF10, P=0.006). This was quite different for AFF1/AF4 and MLLT1/ENL recombinations where MLL intron 11 breaks seem to be favored (P⩽0.0001). As already described above, the biological properties of the MLL PHD1–3 domain depends on the MLL breakpoint. Thus, all fusions occurring within MLL introns 9 and 10 will result in fusion proteins that are still able to oligomerize and to be controlled in its steady-state abundance like the wild-type MLL protein. Vice versa, recombination within MLL intron 11 will result in fusion proteins that could neither be degraded efficiently nor can be switched into transcriptional repressor proteins.

These findings also suggest that oligomerization capacity or binding to certain PHD domain-interacting proteins may be quite important for the oncogenic function exerted by MLL fusion proteins. In addition, the breakpoint distribution in infant and adult patients changes significantly: infants display a higher rate of MLL intron 11 breakpoints (P<0.0001), whereas adults display a higher rate of MLL intron 9 breakpoints (P=0.009). These findings could not be attributed to the number of cases with secondary malignancies (TIL) or any other parameter, which we listed. These data underscore the importance of the precise breakpoint localization that may—dependent on the involved fusion partner gene—influence even the outcome of patients.18

Novel TPGs

Apart from the many new MLL fusion genes that have already been discovered at the DCAL and published in the past years (see Supplementary Table S4; n=26), we present additional eight novel TPGs: RUNDC3B (Run domain-containing protein 3B; 483 amino acids), AP2A2 (adaptor protein complex AP-2 subunit α-2; 939 amino acids), PRPF19 (pre-mRNA processing factor; 504 amino acids); BUD13 (619 amino acids), CEP164 (centrosomal protein; 1460 amino acids), AKAP13 (A kinase-anchoring protein (PKA associated), ARHGEF13; 2813 amino acids), MYH11 (myosin heavy chain 11; 1938 amino acids) and ME2 (malic enzyme 2, NAD(+)-dependent, mitochondrial (malate to pyruvate conversion); 584 amino acids).

The RUNDC3B protein has been described to bind to RAP2,31 a RAS adaptor protein, which has distinct roles in cell adhesion and cell migration. AP2A2 interacts with the mutant form of Huntingtin and alters the kinetic of aggregate formation, thereby functioning as chaperone.32 PRPF19, also named PRP19 or SNEV, was described to be part of large protein complexes involved in pre-mRNA processing,33 DNA repair,34 regulation of proteasomal degradation35 and was also described as ‘senescence evasion factor’.36 For BUD13 no functional data are available. CEP164 is a centrosomal protein that binds to XPA and is required for UV-dependent DNA repair.37 Upon DNA damage, CEP164 becomes phosphorylated by ATM/ATR at the serine-186 residue.38 AKAP13, also known as AKAP-Lbc, represents a Rho-GEF that is regulated by LC3/MAP1LC3A, an important protein for autophagy.39 It has been described to be involved into the signal pathway from TLR2 to NFKB140 and to enhance the cAMP-controlled activation of ERK1/2.41 MYH11 is a smooth muscle myosin gene that has been identified through chromosomal rearrangements with CBFB. These inv(16) AML patients express the CBFB–MYH11 fusion protein that is highly oncogenic.42 Finally, ME2 is a nuclear-encoded mitochondrial enzyme that converts malate into pyruvate.

The MLL recombinome

Within the past 22 years, many genetic aberrations involving the human MLL gene located on chromosome 11 band q23 have been described. Seventy-nine TPGs out of 121 are now characterized at the molecular level (see Supplementary Table S4 and Table 1). Forty-five MLL fusion genes have been described by others, whereas 34 TPGs have been first identified at the Frankfurt DCAL. Additional seven loci are presented here, where neither a direct fusion partner gene nor a ‘spliced fusion’ could be identified. Spliced fusions have been described in cases where the 5′-portion of the MLL gene (exons 1–9) is fused with the upstream region of another intact gene. In most of these cases, the last MLL exon splices to the second exon of this downstream located gene. Examples for this type of mechanism have already been described,15 but will also be discussed below. Finally, additional 35 genetic loci were identified by cytogenetics but not further characterized. All yet characterized TPGs and the appropriate citation references were summarized in Supplementary Table S4.

Genetic alterations resulting in genetic rearrangements of the human MLL gene

In general, human MLL rearrangements are initiated by a DNA damage situation, which induces DNA repair via the non-homologous-end-joining DNA repair pathway.43, 44 Genetic recombinations involving the human MLL gene are predominantly the result of ‘reciprocal chromosomal translocations’ (n=51; see Figure 4). On the basis of our analyses and the literature, reciprocal recombinations lead to fusions of the 5′-MLL gene portion with the following TPGs: ABI1, ABI2, ACTN4, AFF1/AF4, AKAP13, ARHGAP26, ARHGEF17, ASAH3, CASC5/AF15Q14, CASP8AP2, CEP170B, CREBBP, DAB2IP, DCP1A/SACM1L, EEFSEC/SELB, ELL, EP300, EPS15, FOXO3, FOXO4, FRYL, GAS7, GMPS, GPHN, KIAA1524, LAMC3, LASP1, LPP, MAPRE1, ME2, MLLT1/ENL, MLLT3/AF9, MLLT4/AF6, MLLT6/AF17, MLLT11/AF1Q, MYO1F, MYH11, NCKIPSD, NEBL, PDS5A, RUNDC3B, SACM1L, SEPT2, SEPT5/PNUTL, SEPT9, SEPT11, SH3GL1, SMAP1, TET1/LCX, TNRC18 and TOP3A, respectively.

Figure 4
figure 4

General recombination mechanism and associated TPGs. (Top) Genes are categorized either by reciprocal chromosomal translocation (rCTL; n=51), spliced fusion (Spl; n=3), inversions at 11p/q (Inv; n=9), insertions (Ins1 and Ins2; n=12) or 11q deletions (Del; n=4). (Bottom) All identified recombination events, arranged according to the number of DNA double-strand breaks (DSBs) necessary to explain the recombination event. Green: Chromosome 11; red and orange: partner chromosomes involved in the recombination process. Green vertical bars: MLL; red, orange, blue and pink vertical bars: partner genes involved in recombination events; derivative 11 chromosomes is always depicted by ‘Der’. Black and white horizontal lines: recombination sites on wild-type and derivative chromosomes. rCTL: reciprocal chromosomal translocation; Del/Inv: deletion/inversion; 3 W-CTL: three-way chromosomal translocation; CTL+Δ: chromosomal translocation including deletion(s); Ins1: chromosomal fragment including portions of the MLL gene is inserted into a partner chromosome; Ins2: chromosomal fragment including portions of a partner gene is inserted into the MLL gene; cCTL: complex chromosomal translocations, for example, by chromothripsis.

Full size image

Gene-internal PTDs of specific MLL gene portions (duplication of MLL gene segments coding either for introns 2–9, 2–11, 4–9, 4–11 or 3–8) are frequently observed in AML patients.45 MLL PTDs mediate dimerization of the MLL N terminus, a process that seems to be sufficient to mediate leukemogenic transformation.46 We have observed MLL PTDs in 2 patients within the group of pediatric AML, 1 patient within the group of adult ALL and 65 patients within the group of adult AML. This demonstrates that MLL PTDs are predominantly detected in adult AML patients, in line with previously published data.47

MLL recombinations involving only chromosome 11 are based on two independent DNA strand breaks that are accompanied either by inversions or deletions on 11p or 11q (Inv, Del). Several recombinations have been characterized that belong to these two groups. MLL fusions to AP2A2, BTBD18, BUD13, C2CD3, LOC100131626, MAML2, NRIP3, PICALM and PRPF19 are based on the inversion of a chromatin portion of 11p or 11q, leading to reciprocal MLL fusions. By contrast, a deletion on chromosome 11 fuses the 5′-portion of MLL directly to another gene located further downstream (ARHGEF12, BCL9L, CBL and CEP164). In few cases, we observed that the 3′-truncated MLL is located upstream of another, intact gene. In that case, we could demonstrate an ‘MLL spliced fusion’, which means that the last exon of the MLL gene splices directly to the second exon of the further downstream gene. This has been observed for the MLL-DCPS fusion. Beside the above-mentioned DCPS gene, other genes have been identified that can transcriptionally fuse to 5′-MLL sequences. These were ZFYVE19, and also the MLL fusion partners like AFF1/AF4, CT45A2, ELL, EPS15, MLLT3/AF9, MLLT4/AF6, MYO1F and SEPT5. In case of MLLT1/ENL, about 50% of all recombination events were spliced fusions,48 and for MLL-EPS15 fusions about 30%. Spliced fusions to AFF1/AF4, CT45A2, DCPS, ELL, MLLT3/AF9, MLLT4/AF6, MYO1F, SEPT5, ZFYVE19 and SEPT5 represent very rare events.

Beside reciprocal chromosomal translocations of MLL, MLL PTDs and 11p/q rearrangements (Del and Inv), additional genetic rearrangements were identified in the genomic DNA of analyzed leukemia samples. While the previous rearrangements are based on two independent DNA strand breaks, all other genetic events observed for the MLL gene represent more complex rearrangements with at least three or more DNA double-strand breaks. In these cases, the expected reciprocal MLL fusion gene cannot be detected, because other sequences will be fused to the 3′-portion of the MLL gene.

Complex MLL rearrangements are best represented by ‘three-way chromosomal translocations’ involving three independent chromosomes and resulting in three different fusion genes. More complex is a mechanism that we referred to ‘chromosomal fragment insertions’. Either a fragment of chromosome 11 (including portions of the MLL gene) is inserted into another chromosome (Ins1), or vice versa, a fragment of another chromosome (including portions of a TPG) is inserted into the breakpoint cluster region of the MLL gene (Ins2). An insertion mechanism is required in those cases where the transcriptional orientation of a given TPG is not identical to the transcriptional orientation of the MLL gene. The MLL gene is transcribed in telomeric direction. TPGs with a transcriptional orientation in direction to the centromer are predominantly recombining with MLL by such a chromatin insertion mechanism. These genes are ACACA, AFF3/LAF4, AFF4/AF5, CENPK/FKSG14, FLNA, FNBP1, LOC100128568, MLLT10/AF10, SARNP, SEPT6, SORBS2/ARGBP2 and VAV1. In all these events at least three independent fusion genes will be generated. The most prominent gene frequently involved in the latter mechanism is the MLLT10/AF10 gene (see below).

Finally, even more complex rearrangements may occur when ‘chromothripsis’ comes into play. Chromothripsis has been identified as novel mechanism that generates many fusion alleles in a single event upon a single-cell division (for a review see Holland and Cleveland49).

Reciprocal MLL fusions

From two recent papers it became clear that reciprocal MLL fusion proteins may have an important role for cancer development.50, 51 Therefore, we also put emphasis on the analyses of complex MLL rearrangements. These 182 patient cases had three-way or four-way translocations resulting in more than two fusion alleles. From these 182 cases, 63 were identified to carry a single 3′-MLL gene portion that was not fused to any upstream gene (only non-coding loci were identified). By contrast, 119 reciprocal gene fusions were identified from which 80% were out-of-frame fusions. Only 24 reciprocal MLL fusion genes with in-frame fused exons were identified, being capable of expressing the C-terminal portion of the MLL protein under the control of promoters that derive from reciprocal fusion partner genes (n=24; ACER1, ADARB2, APBB1IP, ATG16L2, CEP164 (2 × ), DENND4A, FLJ46266, GNA12, GPSN2, LOC10013227, LRRTM4, , MYO18A, , N-PAC, NFKB1, NKAIN2, PIUP4K2A, RABGAP1L, RNF115, SCAF8, SEPT8, SEPT5, TRIP4, UVRAG and WNK2). In all other cases (n=158), the 3′-MLL gene portion was fused either to no gene (n=63; 1p36, 1q25, 3 × 1q32, 2p12, 2p13, 2p16, 2 × 2p21, 2q11.2, 3p23.3, 4p14, 2 × 4q12, 4q13, 6 × 4q21, 4q22, 4q27, 2 × 4q28, 5q23, 6p21, 6q27, 7p14, 7q22, 8p21, 9p13, 9p21, 9p23, 10p12, 10p15, 11p11, 11p15, 11q12, 11q13, 2 × 11q14, 11q21, 3 × 11q22, 9 × 11q23, 12p13, 15q13, 17q11.2, 19q12, 20q11.2 and 2 × 22q13) or to genes in an out-of-frame or a head-to-head manner (n=119; ADSS, ANTXR2, ARCN1, ARHGAP12, BMP2K, BTN3A1, BUD13, C18orf25, CACNA1B, CACNB2, CCDC33, CDK14, CMAH, CRLF1, CRTAC1, CUGBP1, DHX16, DLG2, DNAH6, DNAJA1, DNAJC1, DOCK5, DSCAML1, DSCAML1, ELF2, EPYC, ETV6, FCHSD2, FXYD2, FXYD6, GRIA4, GRIP1, GTDC1, HELQ, HK1, IKZF1, KDM2A, 2 × KIAA0999, KIAA1239, LMO2, LOC100506746, LOC390877, LOC441179, LPXN, LRBA, MALAT1, MCL1, MDM1, MED1, MEF2A, MEF2C, MMP13, MPZL2, MPZL3, NCAM1, NDUFS3, NRG3, NT5C2, PARP14, PBRM1, PBX1, PDE6C, PHLDB1, PITPNA, PIWIL4, RDH5, RNF25, RPS3, SCGB1D1, SCN3B, SEC14L1, SFRS4, SGK1, SLC43A3, SNAPC3, SORL1, 2 × SVIL, TCF12, TIMM44, TLN1, TMEM123, TMEM135, TNRC6B, TNRC6C, TNXB, TPTE2P5, TUBGCP2, UBASH3B, UBE4A, UNC84A, USP20, WDTC1 and ZNF57).

As summarized in Supplementary Table S5, a total of 20 different genes were identified that were involved in these complex rearrangements (ABI1 (1/3), MLLT10/AF10 (41/120), MLLT6/AF17 (1/10), MLLT11/AF1Q (4/13), AFF1/AF4 (49/600), AFF4/AF5 (1/1), MLLT4/AF6 (6/67), MLLT3/ AF9 (25/291), ELL (4/68), MLLT1/ENL (16/199), EPS15 (2/26), AFF3/LAF4 (2/2), LOC100131626 (1/1), MYO1F (2/3), PICALM (1/4), SEPT6 (3/11), SEPT9 (1/12), TNRC18 (1/1) VAV1 (1/1) and Xq26 (1/1)). The 3′-portion of these TPGs were regularly fused to the 5′-portion of MLL, whereas the above-mentioned 182 loci or genes were fused to the 3′-portion of the MLL gene. The latter fusions are termed ‘reciprocal TPGs’ and are summarized in Supplementary Table S5. In all cases where the 3′-portion of the MLL gene was fused either to a chromosomal locus (non-coding) or in an out-of-frame manner to another gene, one would argue that no transcript is being made. However, the 3′-portion of the MLL is by itself sufficient to produce its own mRNA (starting at the MLL intron 11 to exon12 borderline), which can be translated into the MLL* protein.29 This MLL* protein starts at a bona fide AUG start codon encoded by MLL exon 18, which results in a protein beginning within the MLL BD domain and ending at the end of the SET domain. The MLL* protein is processed by Taspase1 and results in a 97 kDa MLL*-N and an MLL-C protein fragment. This shorter version of MLL (∼235 kDa) loses all functions of the N-terminal portion, whereas functions of the C-terminal portion are retained (for example, H3K4 HMT activity).

Additional 19 MLL rearrangements have been characterized where we could not identify the direct MLL fusion partner gene. However, in all 19 cases we were able to isolate the reciprocal MLL fusion alleles (1q25, 1q32, 7q22, 9p21, 11p11, 11q21, 11q23, CRTAC1, DNAJA1, DSCAML1, KDM2A, RNF115, RNF25, SEPT5, SORL1, USP20, WDTC1 and ZNF57). Only 2 of these 19 cases displayed an in-frame fusion to the 3′-MLL portion (RNF115-MLL and SEPT5-MLL), whereas all the others had solely the intact 3-portion of MLL left to express the MLL* protein (see Supplementary Table S5).

Discussion

Here, we present an update of the ‘MLL recombinome’ associated with different hematologic malignancies, and in particular with acute leukemia (ALL and AML). All our analyses were performed by using small amounts of genomic DNA that were isolated from bone marrow or peripheral blood samples (n=1622) of leukemia patients. In some cases, we analyzed cDNA from a given patient to validate the presence of MLL spliced fusions, or to investigate alternative splice products generated from the investigated MLL fusion genes. The results of this study allow to draw several conclusions.

The applied long-distance inverse-PCR technique allowed to identify direct and reciprocal MLL fusions, MLL gene-internal duplications, chromosome 11 inversions, chromosomal 11 deletions and the insertion of chromosome 11 material into other chromosomes, or vice versa, the insertion of chromatin material of other chromosomes into the MLL gene (see Figure 4). Moreover, we successfully extended our knowledge by analyzing more cases with complex MLL rearrangements. During the latter analyses, a large collection of reciprocal MLL fusions was identified. About 15% represent in-frame fusions that can be readily expressed into a reciprocal fusion protein. All other characterized reciprocal MLL alleles represented out-of-frame fusions with either a chromosomal locus or a reciprocal TPG, but even these events allow to transcribe and express a 5′-truncated MLL protein, termed MLL*.29 This shorter version of MLL has no ability to bind Menin1, LEDGF or MYB, but still carries all enzymatic functions necessary to carry out H4K16 acetylations by the associated MOF protein or H3K4 methylation by the SET domain complex.

The analysis of 1622 MLL fusion alleles led to the discovery of 34 novel TPGs in the past 10 years, of which 26 have already been described (see Supplementary Table S4). Eight TPGs are completely new and have not been published yet. Taken together with 45 MLL fusions that have been described by others (see Supplementary Table S4), we can present today a total of 79 ‘direct MLL fusions’ that have been characterized at the molecular level. All these MLL fusions provide a rich source for future analyses of oncogenic MLL protein variants.

According to our data, the seven most frequent rearrangements of the MLL gene occur either with TPGs like AFF1/AF4, MLLT3/AF9, MLLT1/ENL, MLLT10/AF10, ELL, MLLT4/AF6 or derive from gene-internal duplications (MLL-PTDs). Their occurrence differed significantly in the cohorts of infant, pediatric and adult leukemia patients. We also observed tendencies that correlate specific gene fusions with sex or age at diagnosis. Examples were that MLLT3/AF9 (P=0.080), MLLT10/AF10 (P=0.019) and MLL-PTDs (P=0.065) occur more frequently in the male group of patients, whereas the female patients were more affected by MLL-AFF1/AF4 fusions (P=0.015). The most striking finding was that breakpoint distributions differ significantly when concerning distinct TPGs and age groups. It is well known that breakpoints in infants occur more frequently in MLL intron 11. We could validate this finding for MLL-AFF1/AF4 and MLL-MLLT1/ENL fusions, but observed a completely contrary situation in case of MLL-MLLT10/AF10 fusions. Quite surprising was the breakpoint distribution for MLL-AF6 fusions that displayed a clear preference for MLL intron 9 recombinations. Again, these deviations from the observed mean breakpoint distribution are an argument for differences in the biology of the resulting fusion proteins with respect to oligomerization or factor binding dependency. This has to be investigated in more detail in the future to understand these observations.

An important translational aspect of this study is the establishment of patient-specific DNA sequences that can be used for monitoring MRD by quantitative PCR techniques. Owing to the fact that a given MLL fusion allele is genetically stable and a monoallelic marker for each tumor cell, a more reliable quantification and tracing of residual tumor cells becomes possible. For each of these 1622 acute leukemia patients at least one MLL fusion allele was identified and characterized by sequencing. Several prospective studies were already initiated and first published data verified the reliability of these genomic markers for MRD monitoring.4 Therefore, the use of these MRD markers will contribute in the future to a better stratification of leukemia patients, which will help to further improve the outcome.

The analysis of the MLL recombinome allows to classify MLL fusion partner genes into functional categories. As discussed above, only very few TPGs are recurrently identified in different individuals, and moreover, with a significant frequency. On the basis of this study, these TPGs are AFF1/AF4, MLLT3/AF9, MLLT1/ENL, MLLT10/AF10 and MLLT4/AF6. At least for the AFF1/AF4, MLLT3/AF9, MLLT1/ENL and MLLT10/AF10 protein exists a functional correlation, as all these proteins are organized within a protein complex (or different subcomplexes) that affect transcriptional elongation. AF4 is the docking platform for AF9 or ENL, which both interact (via MLLT10/AF10) to DOT1L.52, 53 DOT1L enable methylation of lysine-79 residues of histone H3 proteins, a prerequisite for the maintenance of RNA transcription.54, 55 AF4 binds with its N-terminal portion to the P-TEFb kinase that phosphorylates the largest subunit of RNA polymerase II, DSIF, the NELF complex and UBE2A. This converts RNA POL A into POL E and allows gene transcription.56 As a result, increased and extended H3K79 methylation signatures seem to accompany the presence of several fusion proteins (MLL-AFF1/AF4, AFF1/AF4-MLL, MLL-MLLT3/AF9, MLL-MLLT1/ENL, MLL-MLLT10/AF10 and MLL-MLLT4/AF6),57 whereas an additional increase in H3K4 methylation was only demonstrated by the presence of the reciprocal AFF1/AF4-MLL56 that causes pro-B ALL in C57Bl6 mice50 and was shown to cooperate with the RUNX1 protein.58 Thus, all the major MLL fusions share a common pathway, which is not only functionally related but offers new and interesting venues to develop new drugs against this leukemias, for example, by the development of DOT1L inhibitors.59 This shared pathway and the effects of certain MLL fusion protein on basic transcription and on the epigenetic layer are summarized in Figure 5. The fusion proteins MLL-MLLT1/ENL, MLL-MLLT3/AF9 and MLL-MLLT10/AF10 recruit thereby the AFF1/AF4 complex, whereas the reciprocal AFF1/AF4-MLL fusion protein is able to perform exactly the same actions on RNA polymerase II and DOT1L. Thus, future therapies addressing either the inhibition of DOT1L, P-TEFb or blocking the interaction of the MLL N terminus with MENIN1/LEDGF/MYB are promising new ways to address these leukemias. In addition, the inhibition of Taspase1 would help to inactivate the AFF1/AF4-MLL fusion protein, as the uncleaved fusion protein is rapidly degraded by SIAH1 and SIAH2.60

Figure 5
figure 5

Common pathways of the most frequent MLL fusions. The four most frequent MLL fusions, MLL-ENL, MLL-AF9, MLL-AF10 and AF4-MLL, are either interacting directly with the AF4 complex or are mimicking the AF4 complex in case of AF4-MLL. The crucial components within the AF4 complex are the P-TEFb kinase and the H3K79 HMT DOT1L protein. Hyperactive AF4 or AF4-MLL is strongly enhances the transcriptional processes. In addition, changes in the steady-state AF4 complex stability is causing extended H3K79me2/3 signatures. Future inhibitory strategies are indicated in red.

Full size image

In summary, MLL rearrangements are associated with poor outcome in pediatric and adult acute leukemia. As outlined above, the systematic analysis of the MLL recombinome allows one to draw conclusions on certain aspects of the hematomalignant transformation processes. We also present additional information as Supplementary data files (see Supplementary Tables S6–8), which contain general information about the investigated patient cohort, the analyzed T-ALL cases (n=36) and the TIL cases (n=77). Our efforts to analyze the MLL recombinome will be continued and provided as free-of-charge service to any collaborators.