Novel molecular and computational methods improve the accuracy of insertion site analysis in Sleeping Beauty-induced tumors

Abstract

The recent development of the Sleeping Beauty (SB) system has led to the development of novel mouse models of cancer. Unlike spontaneous models, SB causes cancer through the action of mutagenic transposons that are mobilized in the genomes of somatic cells to induce mutations in cancer genes. While previous methods have successfully identified many transposon-tagged mutations in SB-induced tumors, limitations in DNA sequencing technology have prevented a comprehensive analysis of large tumor cohorts. Here we describe a novel method for producing genetic profiles of SB-induced tumors using Illumina sequencing. This method has dramatically increased the number of transposon-induced mutations identified in each tumor sample to reveal a level of genetic complexity much greater than previously appreciated. In addition, Illumina sequencing has allowed us to more precisely determine the depth of sequencing required to obtain a reproducible signature of transposon-induced mutations within tumor samples. The use of Illumina sequencing to characterize SB-induced tumors should significantly reduce sampling error that undoubtedly occurs using previous sequencing methods. As a consequence, the improved accuracy and precision provided by this method will allow candidate cancer genes to be identified with greater confidence. Overall, this method will facilitate ongoing efforts to decipher the genetic complexity of the human cancer genome by providing more accurate comparative information from Sleeping Beauty models of cancer.

Figure 1. Dynamic filtering to remove background transposon insertion events.

(A) Read distributions from two independent tumor samples are shown along with the calculated cutoff points using three independent methods: negative binomial (NB), 1% of the top (i.e. most abundant) site and 0.1% of total reads (B) An experiment to simulate a variety of sequence read depths shows that the cutoff method used by the analysis pipeline is influenced by read depth.

Figure 2. Illumina-based LM-PCR analysis consistently identifies transposon insertions in SB-induced tumors.

A total of 62 T-cell lymphomas induced by SB mutagenesis using a ligation-mediated PCR approach were analyzed. Two technical replicates were performed to assess the consistency of the results (A). The results of the technical replicates were compared to assess the reproducibility of the approach. Both the rank (B) and abundance (C) of insertion sites showed a strong positive correlation between the replicate runs.

Figure 3. Determining optimal read depth in SB-induced tumors.

(A) An experiment was performed to simulate various read depths in 30 Vav-SB (left) and 32 CD4-SB (right) tumors. The average number of clonal transposon insertion sites was determined from 20 independent iterations of each read depth simulation. (B) The consistency of transposon insertion site identification varied with sequence depth. For example, only 10% of transposon insertion sites were identified in 100% simulations at a simulated read depth of 1,000.

Figure 4. Comparison of three independent methods to identify CISs within SB-induced tumors.

We compared the performance of Monte Carlo simulation (MC), Gaussian Kernel Convolution (GKC) and gene-centric common insertion site analysis (gCIS) in identifying candidate cancer genes in both Vav-SB (A) and CD4-SB (B) tumors. In addition, the number in parentheses indicates the number of total genes in each region that have evidence as human cancer genes in either the COSMIC or CGC databases. In addition, we compared the results of MC and gCIS analysis using data generated by 454 or Illumina sequencing of the same tumor samples in both Vav-SB (C) and CD4-SB (D) lymphoma models.

Figure 5. Determining the affect of sequence depth on the accuracy of CIS identification.

Read depth simulations were performed as previously described. However, gCIS analysis was performed on each of 20 iterations at all simulated read depths. In addition, the values generated from analysis of the actual 454 data obtained for the same tumor samples are shown. (A) The total number of gCIS genes identified in at least one of the 20 iterations is indicated along with the average number of gCIS genes at each simulated sequence depth. (B) The gCIS results obtained by analyzing all sequence data in both tumor models were used as reference data sets. The accuracy (% of genes found in reference set) and sensitivity (% of genes in reference set that were detected) were determined. For example, only 41% of the reference gCIS genes in the Vav-SB model were found at a depth of 1,000 reads per sample (i.e. 41% sensitivity). However, 90% of gCIS genes identified at this read depth were found in the reference data set (i.e. 90% accuracy).

Figure 6. The affect of varying read depth on the genomic distribution of CIS genes.

The genomic position of each CIS gene is indicated as a vertical bar. The color of the bar indicates the minimum number of reads per sample required to consistently identify the CIS gene at the indicated position. The height of the bar indicates the mutation frequency within the each tumor model as determined by analysis of all sequence data. A large number of false-positive CISs were identified in one or more iterations of the simulation to approximate a read depth of 5,000 (white bars).