Aneuploidy occurred at low frequency. Whole chromosome gains occurred in 4/12 tumors; one tumor had a partial deletion of chromosomes 18 and 19, but no other very large-scale (i.e., >10 Mb) CNAs were noted. Comparatively, CNAs occurred with high frequency (Figure 1), with amplifications much more common than deletions (Supplementary Table 1 and Supplementary Table 2). Excluding copy numbers affected by aneuploidy, 1,830 genes were amplified, and 378 genes were deleted across the 12 tumors. Tumors carried a mean of 152 amplifications and 31.5 deletions. A large fraction of CNAs were recurrent. We identified 168 unique recurring amplifications, spanning 439 genes, and 25 unique recurring deletions, spanning 35 genes. Median amplicon size was 91,543 bp, with a minimum amplicon size of 14,799 bp. Median size of deleted region was 99,798 bp with minimum size of 20,510 bp.

We previously reported B-cell differentiation block at the pro-B stage in mice transplanted with Prdm14-transduced bone marrow [12]. We evaluated tumors for deletions of genes involved in B-cell development and differentiation. Five of twelve tumors had deletions in critical B-cell developmental genes with tumor suppressor function, including Ebf1, Pax5, and Ikzf1 (Figure 2a–c). Gene deletion of B-cell developmental genes occurred only in precursor-B lineage tumors, except in one tumor (MPr14-111) which did not have either B- or T-cell rearrangement, yet had the immunophenotype of a CLP-like tumor [12]. Numerous tumors have deletion within the immunoglobulin heavy chain and T-cell receptor region, consistent with rearrangement.

All tumors had at least amplification of one known oncogene or deletion of one tumor suppressor gene (Table 2). Given the expansion of abnormal lymphoid progenitors prior to leukemia onset, we expected to find derangement of apoptosis control genes. Four of twelve tumors had deletions of Cdkn2a, which encodes the INK4-ARF master apoptosis control proteins; gene expression was relatively low in all tumors with gene deletion, but expression across tumors does not appear to be solely dependent on gene deletion (Figure 3a,c). Four of twelve tumors had amplification of Tbx2, a gene involved in suppression of senescence. Though gene amplification appears to be heterozygous (mean copy number change of 3.02), gene expression is massively expanded in these tumors, with 45–101 fold change in gene expression in tumors with amplification (Figure 3b,d) compared to controls and tumors without amplification. Of note, one control appears to have gene amplification but, unlike the tumors, does not have increased gene expression.

Seven of twelve tumors had amplification of Kif11 (Eg5), a necessary component of the mitotic spindle required for proper chromosomal segregation, and whose amplification leads to the development of lymphomas and solid tumors in mice, usually associated with tetraploidy [23]. All tumors with erythroblastosis, an unusual feature in lymphoblastic leukemia and lymphoma, carried Eg5 amplification, but not all tumors with Eg5 amplification demonstrated this phenotype.

Five of twelve tumors have CNAs at known common fragile sites, genomic regions susceptible to DNA damage under replication stress [24]. Deletion of Fhit was detected in one tumor, and amplification of Wwox and Grid2 was seen in two tumors each. Additional recurrent gene deletions occurred in genes spanning large (>200,000 kB) regions of DNA, including Macrod2, Pcdh15, and Shroom3.

Leukemias were classified into subtypes based on presence of IgH or T-cell receptor rearrangements, and CNAs were analyzed by class. Principal component analysis demonstrated that leukemias did not group according to their condition, demonstrating no consistent global genetic changes that were class-defining. Although all classes of malignancies exhibited similar numbers of CNAs, we noted substantially fewer known leukemogenic CNAs in T-lymphoblastic leukemias than in other types (Table 2). As expected, B-cell development gene deletions (i.e., Pax5, Ikzf1, Ebf1) did not occur in T-lymphoblastic leukemias but did occur in other types (precursor B, CLP-like, mixed lineage). Deletions in common fragile sites, CNAs in Cdkn2a or Tbx2, and chromosomal rearrangements were also not seen in the T-lymphoblastic leukemias, although Eg5 amplification was present. In comparing PRDM14-induced T-ALL mutations to previously published candidate driver mutations in murine T-lymphoblastic lymphoma identified using a Sleeping Beauty transposon mutagenesis screen [25], little overlap was noted, except for amplification of Chst11 (tumor MPr14-148) and Picalm (tumor MPr14-258). Mixed/unclassified lineage leukemias demonstrated similar chromosomal changes as precursor-B leukemias, except for two tumors with amplification of Lmo1, a gene implicated in T-ALL leukemogenesis [26,27].

PRDM14-binding loci in murine ES cells have been previously characterized using ChIP-Seq pulldown of FLAG-PRDM14 [16]. We cross-referenced these known binding sites with genes amplified and deleted in PRDM14-induced tumors. Five of 35 genes (Ebf1, Sox2ot, Shroom3, Slc10a7, and Ncam1 [which encodes CD56]) with recurrent deletions had PRDM14 binding sites within the gene region, and four of five had multiple binding sites. PRDM14 binds also within intronic sequence of Pax5 [16], which is deleted in tumor 147 (Figure 2b). Though PRDM14 does not bind at a preponderance of amplified loci, it does bind at numerous amplified loci previously characterized in oncogenesis, including Kif11, Lmo1, Notch1, and Lpp.

Rearrangements of the MLL gene with numerous partners in human leukemias have been characterized [28]. Though we were limited from performing karyotypic analysis of our tumors due to tissue availability, we did note substantial amplification of MLL rearrangement partners. Eight of twelve tumors demonstrated amplification of these genes, including Lpp, Mllt3, Mllt10, Ell, and Eps15 (Table 2).

We previously described abnormal gene expression seen in an expanded population of common lymphoid progenitors (Il7ra⁺lin⁻Kit⁺Sca1⁺) that predated the onset of leukemia in our mice [12]. We performed gene ontology analysis from this experiment to evaluate for enrichment of genes involved in DNA repair that may be up- or downregulated. All statistically significant (p-value corrected for multiple testing <0.05) gene changes were evaluated, regardless of fold change (FC). Enrichment for genes involved in DNA repair was noted (p = 2.8E-2, Benjamini = 3.9E-1). Fourteen genes involved in DNA repair—Xrcc6, Slk, Uhrf1, Usp1, Apbb1, Ddb1, Rad23b, Blm, Fancd2, Lig3, Brip1, Cinp, Ercc3, and Rtel1—had statistically significant decreased expression (Supplementary Table 3). In particular, Xrcc6 (FC = −3.31), Fancd2 (FC = −1.27), Blm (FC = −1.26), and Brip1 (FC = −1.23) are involved in maintaining chromosomal stability [29,30,31]. Of the ten upregulated genes—Ube2b, Ube2a, Setx, Chaf1b, Ercc5, Mbd4, Poll, Mgmt, Msh5, and Cep164—involved in DNA repair, only Cep164 (FC = 2.0) is directly involved in maintenance of chromosome stability [32]. None of the other increased genes showed fold change above 1.4, and their role in DNA repair is restricted to nucleotide excision repair or protein ubiquitinization after radiation exposure [33]. None of the upregulated genes participate in homologous recombination repair (HRR) or non-homologous end-joining (NHEJ). Correlation with previously published PRDM14 binding within mouse ES cells demonstrated no PRDM14 binding within 100 kB of any of these loci [16].

Meiosis-specific genes (gene ontology analysis, p = 1.9E-1, Benjamini = 8.1E-1), including Rsph1, Spo11, Zfp318, Msh5, Clgn, and Ube2b are also upregulated in the abnormal pre-leukemic cells (Supplementary Table 3). Two genes with a known role in normal homologous recombination in meiosis are upregulated. Msh5, noted above, also is required for crossing-over in meiosis-specific DNA recombination [34]. Spo11 induces double stranded breaks during meiotic recombination [35] and is upregulated 2.3 fold.

PRDM14-induced lymphoblastic leukemia/lymphomas show heterogeneity of genomic aberrations, similar to that seen in human disease. Human ALL is characterized not just by genetic translocations but by sporadic deletions in genes necessary for B-cell development and differentiation (Pax5, Ebf1, Ikzf1) or apoptosis (Cdkn2a), as are seen frequently in our tumors. Unlike human ALL, however, PRDM14-induced leukemias bear many more gene deletions and particularly amplifications per individual. In their analysis of 228 ALL patients samples, Mullighan et al. reported an average of 3.83 deletions per individual ALL specimen, and only rare focal amplifications, arguing against genomic instability as a mechanism of cancer initiation in most childhood lymphoblastic leukemias [1]. Murine PRDM14-induced leukemia/lymphomas have a mean of 152 amplifications and 31.5 deletions. The high number of detected CNAs did not appear to be an artifact of overly permissive CNA analysis; frequent amplicons (mean 18.4 per tumor) were detected even using excessively strict (copy number >3 or <1) criteria. This level of CNAs is unusual for oncogene-induced murine tumors [8].

DNA damage appears to result from replication stress and chromosomal instability. The high frequency of common fragile site CNAs (5/12 tumors) plus additional CNAs within large genes suggests significant endogenous replication stress. DNA damage at common fragile sites has been shown to precede CNAs at other locations when comparing pre-neoplastic tissue to malignant cells [36]. The high number of CNAs in our tumors, along with the decreased expression of genes required for DNA repair, suggests that chromosomal instability also plays a role in our model of leukemogenesis. Decreased expression of Xrcc6, which is greater than 3-fold reduced in pre-leukemic cells, appears to be a primary mechanism of DNA repair failure. Xrcc6 encodes the Ku70 subunit of the Ku heterodimer, an essential component of the DNA-dependent protein kinase (DNA-PK) complex required for initiation of NHEJ in double stranded break repair [29,37].

Failure to complete NHEJ alone does not explain the degree of DNA damage, particularly regarding amplifications. Pluripotency reprogramming normally induces high fidelity DNA repair mechanisms, including increased expression of numerous HRR and DNA repair genes [22]. Unlike in ES or iPS cells, however, aberrant Prdm14-expressing lymphoid progenitors fail to upregulate HRR or NHEJ genes. Only one gene, Cep164, has increased expression in response to aberrant somatic Prdm14 expression, as opposed to the more global changes seen in iPS or ES cells relative to more differentiated counterparts; additionally, this gene does not participate in direct DNA repair but contributes to cell cycle arrest in response to DNA damage [32]. Thus, the combined effect of decreased expression of DNA repair enzymes plus failure to upregulate DNA repair mechanisms in response to replication stress appears to be the critical mechanism of eventual widespread DNA damage.

Knowing that PRDM14-induced lymphoid progenitors have differentiation block at the pro-B stage [12], and knowing now that putative leukemia precursor cells have impaired ability to repair double-stranded breaks, a more complete model of Prdm14-related cancer initiation emerges (Figure 4), particularly as related to B-lineage or mixed B/T leukemias. PRDM14 expression attempts to reprogram differentiating hematopoietic precursors. The replication stress contributed by activation of pluripotency, the deactivation of genes involved in chromosomal stability (Xrcc6/Ku70), the failure to activate DNA repair mechanisms, and the activation of homologous recombination factors lead to excessive DNA damage. Cells fail to complete the normal B-cell differentiation program, perhaps due to impairment of NHEJ in V(D)J recombination or deletion of B-cell development genes. Furthermore, stem cell reprogramming does not occur properly, likely because other pluripotency factors are missing [15]. Later events include activation of oncogenes (Lmo1, Notch1) sometimes but not necessarily via amplification; we previously demonstrated that Myc, for instance, has increased expression in tumors but not in preleukemia, and is not amplified [12].

Tumors appear to have arisen only after the mutational burden and involvement of oncogenes and tumor suppressors circumvented highly active apoptosis pathways. Notably, our previous gene expression studies of pre-leukemic progenitor cells noted upregulation of multiple p53 pathway genes, including p21 [12], suggesting that preleukemic cells appear to have intact apoptosis mechanisms. Ultimately, however, senescence mechanisms appear to be defeated in our tumors by genomic alteration. Six of twelve tumors demonstrate either amplification of Tbx2 or deletion of the tumor suppressor Cdkn2a. It is deleted here (33%) at a similar rate as seen in human disease (25%) [1]. In human lymphoblastic leukemia, but not in solid tumors, its mechanism of deletion involves aberrant RAG targeting, as RSSs flank CDKN2A breakpoints [5]. TBX2 inhibits CDKN2A by repressing its promoter, preventing expression that would otherwise be induced by Myc or Ras-pathway genes [38]. TBX2 also binds the promoter and represses expression of p21, a key mediator of senescence [39]. In our model, the abrogation of senescence and concomitant failure of DNA repair appears to be a critical mechanism of Prdm14-induced tumorigenesis.

Curiously, this model does not completely apply to TCR-rearranged leukemias without IgH rearrangements seen in this study. Though similar degree of genomic derangement is seen in these tumors, few CNAs in these tumors are in known oncogenes or tumor suppressors, or in candidate tumor initiating genes. Kif11 (Eg5) is amplified in these tumors, but whether this represents a driver or passenger mutation is unknown, given the profound aneuploidy seen in EG5-induced tumors that is not seen these PRDM14-induced leukemias [23]. The mechanism of PRDM14-induced T-leukemogenesis remains a subject for further study.

Of the amplified loci with known PRDM14 binding sites, many of these involve genes implicated in oncogenesis. LMO1 is amplified in a subset of human T-lymphoblastic leukemias [27], and its gene product works in concert with SCL to create conditions allowing NOTCH1 activation and T-lymphomagenesis [26]. A majority of tumors also have Kif11 (Eg5) amplification, which is known to induce tetraploid tumors in mice [23]. The frequency of these and other oncogene amplifications at PRDM14 binding loci, as well as the presence of recurrent deletions at or near PRDM14 binding sites, raises the question of whether PRDM14 binding contributes to replication stress by directly interfering with normal DNA replication mechanisms in somatic cells.

All mouse experiments were carried out under the approval of the Institutional Animal Care and Use Committee at Baylor College of Medicine (BCM). Mice were housed in the barrier facility at BCM, under the care of the Center for Comparative Medicine, which is accredited by the Association for Assessment and Accreditation of Laboratory Animal Care International.

The protocol for transducing bone marrow with Prdm14-containing vector was described previously [12]. Briefly, donor CD45.2 C57BL/6J mice were injected with 5-fluorouracil five days prior to bone marrow harvest. Extracted marrow was transduced with a GFP-labeled murine stem cell virus-based vector (MIGR1) containing Prdm14 (MIGR1-Prdm14) or empty vector (EV). Transduced marrow was transplanted into lethally irradiated CD45.1 C57BL/6J mice. Mice were aged until development of systemic illness; mice sacrificed for illness were noted to have lymphoblastic leukemia/lymphoma as described [12]. Four weeks after transplantation, a fraction of mice were harvested, and marrow cell suspensions underwent cell sorting to isolate cells with a common lymphoid progenitor signature (CLPs; Il7ra⁺lin⁻Kit⁺Sca1⁺; no additional sorting for GFP) as described [12].

Twelve leukemic lymph nodes and spleens from MIGR1-Prdm14 transduced mice were selected for aCGH analysis (Table 1). Three spleens from EV-transduced mice were selected as controls. Genomic DNA was extracted from frozen, unsorted tumor samples using DNeasy Blood and Tissue Kit (Qiagen, Valencia, CA). Sample quality check and chip hybridization were performed by the BCM Genomic and RNA Profiling Core Facility. DNA underwent amino-allyl labeling with Cy3 (control) and Cy5 (samples) and was hybridized to Mouse aCGH SurePrint G3 (1 × 1 M) slides (Agilent, Santa Clara, CA, USA). Two control specimens also underwent Cy5 labeling for comparison to the third control specimen. Array was scanned using GeneTAC UC-4 Microarray Analyzer (Digilab, Holliston, MA, USA). Data were analyzed and visualized using Partek Genomic Suite (Partek, Chesterfield, MO, USA). Results were preprocessed using quantile normalization and were adjusted for identical distribution. Deletions and amplifications were evaluated by genomic segmentation, comparing copy number in samples versus control. The minimum genomic markers in any segment is 10, the p-value threshold for two neighboring regions having significantly different means is 0.001, and the minimum signal-to-noise ratio for each transition is 0.3. To correct for admixture of normal tissue within tumor tissue, and allowing for the possibility of several subclones of cells within the leukemia population, cutoffs of copy number greater than 2.5 or less than 1.5 were used to call amplification and deletion, respectively. As a control for this, a control of greater than three chromosomes or less than one failed to detect differences in sex chromosome genes.

Gene expression array was performed on Il7ra⁺lin⁻Kit⁺Sca1⁺ cells as described previously [12]. Gene ontology analysis was performed using functional annotation clustering in the Database for Annotation, Visualization, and Integrative Discovery (DAVID) [33,40].

Total RNA was isolated from frozen control EV spleens and tumors using TRIzol Reagent (Life Technologies, Carlsbad, CA, USA). RNA was reverse transcribed using the SuperScript III First-Strand Synthesis System for RT-PCR kit (Life Technologies). Resulting cDNA was amplified using real-time PCR with Power SYBR Green PCR Master Mix (Life Technologies) and gene-specific primers (Supplementary Table 4). Amplification and data analysis were conducted on a Rotor-Gene Q machine and software (Qiagen, Valencia, CA, USA). Relative gene expression was calculated with the 2^−ΔΔCt method [41]. Expression of ribosomal protein L19 (Rpl19) was used as an endogenous control, and all values are relative to control EV spleen. Gene expression levels in tumors with amplification were compared to levels in tumors without amplification and in control EV spleen using a one-way ANOVA with Bonferroni correction for multiple comparisons. Statistical analysis and visualization of PCR results were conducted using Prism (GraphPad, La Jolla, CA, USA).