Patient Derived Xenograft (PDX) Models Recapitulate the Genomic-Driver Composition of Acute Leukemia Samples

Xenotransplantation of primary AML samples into immunodeficient mice (PDX models) represents a unique opportunity for pre-clinical testing on a group of primary human samples that possess defined genomic lesions. However, given recent recognition that multiple genomically distinct sub-clones can exist in AML, there is a risk that there may be selection for sub-clones from the transplanted sample that might not fully represent the patient’s disease.

We transplanted 160 (70 T-ALL 56 AML, 32 B-ALL, 2 MLL) patient samples of which 120 engrafted into at least 1 irradiated NSG mouse. 45 AML samples engrafted with a median latency 107+/-41 days. Transplantation of 6 PDX AML samples resulted in immunophenotypically identical disease within 87+/-35 days. 2 MLL samples engrafted in 100% of mice with a median latency of 103+/-13 days. 25 B-ALL samples engrafted with a median latency of 95+/-44 days. Secondary transplantation of 3 PDX B-ALL samples resulted in engraftment of leukemia cells with an identical immunophenotype in 100% of transplanted mice within 52+/-3 days. 48 T-ALL samples engrafted in at least one mouse within 50 days. Secondary transplantation of a single T-ALL PDX sample resulted in 100% engraftment within 31+/-10 days.

Genomic DNA and total RNA were isolated from 150 (AML: 16Pt+33PDX; MLL 2Pt+6PDX; B-ALL 17Pt+38PDX; T-ALL 19Pt+19PDX) samples. Adaptor ligated sequencing libraries were captured by solution hybridization using baitsets for 405 cancer-related genes and selected introns for 31 genes frequently rearranged for DNA-seq, and 405 cancer-related and 265 genes frequently rearranged for RNA-seq. All libraries were sequenced averaging >500x for DNA and >6M total pairs for RNA (HiSeq). We detected on average 23+/-12 including a mean 5+/-4 known pathogenic variants such as CDKN2A/B deletion (20/13); FLT3 (SNV & -ITD) and NOTCH (11 ea); WT1 and TP53 (10 ea); NRAS (9); PTPN11 (7); NPM1c, PTEN, and KRAS (6) DNMT3A, IDH1/2, and ASXL1 (5 ea); FBXW7, CEBPA, and TET2 (4 ea); PHF6 and NF1 (3 ea); IKZF1, ATM, and JAK2 (2 ea). Analyses of fusion RNA molecules detected known fusions: MLL-AF4 (4); MLL-AF9 (2), CRLF2-P2RY8, ETV6-RUNX1 or TEL-AML1, PBX1-TCF3 (2 ea); MLL-AF10, MLL-ELL, MLL-EP300, MLL-PTD, BCR-ABL, BCL2-IGK, MYH11-CBFB, along with novel fusions: TCF3-OAZ1, RB1-RCBTB2, PAX5-FLI1, and PAX5-MSI2.

The mutations found in the 54 patient samples were consistently identified in the 96 PDX, however some cases showing variation in allele frequency between diagnostic and engrafted samples. Collectively, all 1420 and 288 disease relevant variant allele frequency (VAF) correlated significantly between patient and PDX samples (R²=0.55, R²=0.43), respectively. We then assessed VAF changes from diagnostic to PDX sample as a measure of clonal concordance. Diagnostic and PDX sample were considered discordant if at least one disease relevant VAF demonstrated significant variation between these samples, accounted for small variability of infrequent variances considering SD of sequencing detection. 31 samples were scored as concordant and 23 as discordant which were similarly distributed between disease lineages and did not correlate with diseases status, future relapse or overall survival.

Using the same rules we further accessed concordance only between PDX samples in 23 cases when patient samples were transplanted into multiple mice. All 10 groups of PDX samples that were concordant with patient samples were also concordant within the groups. 5 groups of PDX samples that were discordant with patient samples were concordant within groups. 8 groups of PDX samples that were discordant with patient samples were also discordant within their groups. Overall 15 samples produced concordant engraftment in mice and 8 samples produced discordant engraftment. We hypothesized that specific genomic lesions in the 8 groups might underline this discordance. Mutations of FLT3, RAS, TP53, PTPN11 and NOTCH1 correlated with clonal discordance.

These findings show that the leukemias that are engrafted in mice mirror the genomic diversity of primary leukemia samples, and that the majority of PDX samples have a genotype similar to that observed in the clinical isolate. More importantly, our data demonstrate the feasibility of developing a large, genetically annotated bank of PDX leukemia models that can be used to test and credential novel therapeutics that target driver mutations in different leukemia subsets.