The genotoxic potential of retroviral vectors is strongly modulated by vector design and integration site selection in a mouse model of HSC gene therapy (original) (raw)
Vector construction and testing in tumor-prone HSC transplantation. The panel of chimeric and parental vectors tested is shown in Figure 1A. We challenged the previously reported safety of LV by introducing the strong spleen focus–forming virus (SF) retroviral enhancer/promoter (30) in the U3 region of the LTR (LV.SF.LTR) and comparing this vector to a γRV carrying SF LTRs (RV.SF.LTR) (31). Similarly, we tested the oncogenic potential of a γRV with SIN LTRs carrying the moderately active human phosphoglycerate kinase (PGK) promoter in internal position (SIN.RV.PGK) (13). To address the position dependence of strong enhancer/promoters in genotoxicity, we placed the SF sequence in an internal position within the vector (SIN.LV.SF) and compared it with its SF.LTR counterpart. Finally, to assess the impact of promoter strength, we compared SIN LV with internal SF and PGK promoter. All vectors expressed GFP and were pseudotyped by vesicular stomatitis virus G protein (VSV.G).
Transduction of Cdkn2a–/– HSPC by chimeric vectors and tumor development in transplanted mice. (A) Scheme of the proviral forms of the vectors tested. SIN, SIN LTR with deletion of the U3 region; SFFV, enhancer/promoter of the SF U3 LTR; PGK, promoter of the human phosphoglycerate kinase gene; SD and SA, viral splice donor/acceptor sites; cPPT, central polypurine tract; PRE, posttranscriptional regulatory element from the woodchuck hepatitis virus. Transgene transcripts are indicated by arrows. (B) Percentage of GFP+ (mean ± SD) _lin_–_Cdkn2a_–/– cells transduced in vitro with the indicated vectors 6 days after transduction. Number of experiments indicated on top. The average VCN per cell measured by Q-PCR 14 days after transduction is indicated inside the bar. (C) Percentage of GFP+ cells (mean ± SD) in the blood of transplanted mice at 6–8 weeks after transplant. Number of mice indicated on top. (D) Representative H&E-stained sections of BM (left panel) and gut (right panel) from a myeloid tumor in the RV.SF.LTR group. Original magnification, ×20. Scale bar: 100 μm. (E) VCN distribution in tumor-infiltrated (VCNtum) tissue (BM, spleen or thymus) for each mouse analyzed (dots) among the different treatment groups. Horizontal line represents the average VCN for each group.
For each vector, Cdkn2a–/– BM–derived lineage marker–negative (_lin_–) cells were transduced or not (mock) using the same culture conditions, a vector dose of 1 × 108 HeLa transducing units (TU)/ml (MOI = 100), and transplanted after a total culture time of 96 hours. Upon transduction, GFP+ cells ranged from 84% to 95% by FACS for all vectors (Figure 1B). Average vector copy number per cell (VCNin vitro) was measured by quantitative PCR (Q-PCR) in a sample of cells cultured in vitro for 2 weeks after transduction to exclude nonintegrated vector DNA. VCNin vitro was 11.3 ± 5.1 (n = 3) for LV.SF.LTR; 3.4 ± 0.2 with RV.SF.LTR; 15.5 ± 4.4 for SIN.LV.SF (n = 3); and 13 ± 4 for SIN.RV.PGK (n = 4). Because RV.SF.LTR transduction yielded a much lower VCNin vitro than LV.SF.LTR, we added another condition using LV.SF.LTR at a lower MOI (MOI = 10; 1 × 107 TU/ml) to obtain more comparable transduction levels. This resulted in VCNin vitro 1.6 ± 0.1 and 22% or more GFP+ cells (note that the percentage of GFP+ cells by FACS underestimates the transduction frequency of this vector because of low MFI).
Lethally irradiated wild-type FVB mice were transplanted with 7.5 × 105 vector-treated cells (LV.SF.LTR, n = 29; SIN.LV.SF, n = 39; RV.SF.LTR, n = 12; SIN.RV.PGK, n = 34; LV.SF.LTR, MOI = 10, n = 22; mock-transduced, n = 32, for a total of 168 mice in 7 different experiments). At 6 to 8 weeks after transplant, all mice showed normal frequency of myeloid (CD11b+), B (CD19+), and T (CD3+) lymphoid cells in the blood, and the percentage of GFP+ cells was consistent with the levels of transduction observed in vitro among all lineages (range 60%–90% GFP+ for all groups or 20% for LV.SF.LTR, MOI = 10) (Figure 1C and Supplemental Figure 1; supplemental material available online with this article; doi:10.1172/JCI37630DS1).
As expected, from 60 to 400 days after transplant, all mice in all experimental groups developed hematopoietic malignancies (7). The diagnosis was based on a blinded histopathology examination of BM, spleen, thymus, liver, kidney, lung, brain, gut, and lymph nodes and by FACS analysis of BM, blood, spleen, and thymus used to evaluate the proportion of myeloid and lymphoid cells and the relative frequency of GFP+ cells (Supplemental Table 1 and Supplemental Figure 2). There was an increased frequency of myeloid tumors in the LV.SF.LTR and RV.SF.LTR groups as compared with the mock and other groups. Statistical significance, however, was only reached by LV.SF.LTR when both the MOI = 100 and MOI = 10 groups were merged to increase the sample size (31/49 vs. 7/20 for the mock; P = 0.038, Fisher’s exact test) (Table 1). Moreover, the guts from the LV.SF.LTR and RV.SF.LTR treatment groups had a significantly increased occurrence of malignant myeloid infiltration (42% and 44%, respectively, vs. 0%–12% range of all the other groups; P = 7 × 10–4 and 8.4 × 10–3 vs. mock, respectively, Fisher’s exact test) (Table 1 and Figure 1D). Skewing of tumor phenotype and affected tissues may indicate an effect of vector treatment on the spontaneous oncogenesis of the mouse model.
Tumor phenotype incidence in each group of transplanted mice
Tumor-infiltrated tissues were systematically analyzed by Q-PCR to measure the relative amounts of donor-derived and recipient cells and the VCN in donor cells. Cdnk2a–/– cells constituted more than 70% of the affected tissue, indicating the donor origin of tumors. Donor cells in the tumor-infiltrated tissues, obtained from each group, had an average VCN of 9.9 ± 6.2 for LV.SF.LTR, MOI = 100, (n = 29) and 1.2 ± 1.2, MOI = 10, (n = 20); 3.0 ± 1.4 for RV.SF.LTR (n = 10); 11.4 ± 7.4 for SIN.LV.SF (n = 32); and 8.6 ± 4.1 for SIN.RV.PGK (n = 20) (Figure 1E and Supplemental Table 1).
Assessing the oncogenic risk of vector treatment. The mock-transduced control group had a median survival time corresponding to a donor cell age of 257 days, consistent with the median survival of Cdkn2a–/– mice (32, 33) and virtually identical to that in our previously published results (7), showing that cell manipulation and transplant procedures per se do not accelerate tumor onset and that our assay is highly reproducible. Survival in each experimental group was analyzed by Kaplan-Meier curves (Figure 2, A and B). Mice transplanted with HSPCs transduced at the higher dose of LV.SF.LTR (LV.SF.LTR, MOI = 100) died significantly earlier than the mock-transduced controls (median survival, 187 days vs. mock 248 days; P < 0.0001, log-rank Mantel-Cox test). Mice treated with the lower vector dose (LV.SF.LTR, MOI = 10) displayed a median survival of 211 days (Figure 2A). The median survival time was 194.5 days for RV.SF.LTR mice (Figure 2A), 227.5 days for SIN.RV.PGK mice, and 238 days for SIN.LV.SF mice (Figure 2B).
Survival curves related to vector treatment. (A and B) Kaplan-Meier survival curves of mice transplanted with cells transduced with SF.LTR vectors (A) and SIN.LTR vectors (B). For each panel, the survival curve of the mock group is shown (n = number of transplanted mice). Survival of the LV.SF.LTR, MOI = 100, group was significantly shorter than that of the mock group (P < 0.0001; Mantel-Cox log-rank test). (C) Survival probability over time was calculated for each group using the estimated log-logistic parameter (shown in Table 2) and compared with that of the mock group. The survival probability of the LV.SF.LTR, MOI = 100, and the RV.SF.LTR groups was significantly lower than that of the mock group (P values are indicated). A sample of the transduced cells was kept in vitro for 2 weeks after transduction to measure the average VCN (shown for reference).
In order to perform a risk assessment analysis that accounted for the accelerated time of death revealed in the descriptive statistics, we used an accelerated failure time model to estimate the risk of death and the survival probability for each treatment. Within this class of models, the log-logistic distribution provided the best fit to our experimental data (Supplemental Statistical Methods). The impact of the treatment (represented by the log-logistic parameter) and the survival probability during time were estimated for each treatment group and compared with that of the mock (Table 2 and Figure 2C). For the LV.SF.LTR, MOI = 100, as well as for the RV.SF.LTR groups, the survival probability was significantly lower than that of the mock (P = 1.8 × 10–5 and P = 2.8 × 10–2, respectively). On the other hand, treatment with LV.SF.LTR, MOI = 10, did not significantly reduce the survival probability as compared with that of the mock, indicating that the genotoxicity of LV.SF.LTR at this dose was below the detection limit of our in vivo assay. Notably, mice in the LV.SF.LTR, MOI = 10, and RV.SF.LTR groups were transplanted with cells carrying similar VCNin vitro (2 and 3, respectively), indicating a higher genotoxicity of RV.SF.LTR. The same analysis performed on the SIN.RV.PGK and SIN.LV.SF groups showed no significant impact of these vectors on the survival probability in our assay, even if both groups had high VCNin vitro.
log-logistic accelerated failure time model estimates the impact on survival of vector treatment in each group
Because transduction with each vector resulted in a different integration load even when using the same MOI (MOI = 100), we adopted the VCN in tumor (VCNtum) as a measure of dosage to perform VCN-matched comparisons between different vector treatment groups. Mice were stratified in groups having a VCNtum ranging from 1 to 6 (VCNtum1–6) or above 6 (VCNtum>6). The stratification criterion adopted allowed us to compare a relevant number of mice with a similar vector load (Table 3). log-rank Mantel-Cox test on the Kaplan-Meier curves showed that the mice in the LV.SF.LTR-VCNtum>6 group died significantly earlier than those in the LV.SF.LTR-VCNtum1–6 group (P = 1 × 10–3) and the mock group (P < 1 × 10–4) (Figure 3A). The survival of the RV.SF.LTR-VCNtum1–6 group was also significantly reduced with respect to the mock group (_P_ = 2 × 10–4) and the VCN-matched LV.SF.LTR-VCNtum1–6 group (_P_ = 2 × 10–2). Indeed, the LV.SF.LTR-VCNtum1–6 group, which comprised 8 mice from the MOI = 100 and 13 from the MOI = 10 transduction groups, had a survival not significantly different from that of the mock group. The SIN.RV.PGK and SIN.LV.SF groups did not show any significant acceleration with respect to the mock group when stratified according to VCNtum>6 (Figure 3B). Similar results were obtained by applying the log-logistic model to determine the survival probability of each stratified vector treatment group compared with the mock group (Supplemental Figure 3, A–C). These results showed that vector genotoxicity is dependent on the presence of active LTRs and on the VCN of in vitro–cultured cells or tumors. However, the negative impact on survival varies according to the type of vector and VCN.
Survival curves and risk assessment related to vector treatment and dose. (A and B) Kaplan-Meier survival curves of mice treated with the SF.LTR (A) and SIN LTR vectors (B) stratified by the VCNtum. (C) Percentage hazard of death over time of mice depending on the vector used and a fixed VCNtum of 1 (solid lines) or 10 (dashed lines).
Vector treatment groups stratified by mice that developed tumors with VCNs ranging from 1 to 6 and VCNs greater than 6
In order to quantitatively assess the relative impact of VCNtum for each vector, we used the VCNtum of each mouse as a covariate in the accelerated failure time model. We observed that VCNtum acts linearly on the log-logistic hazard (Supplemental Statistical Methods). VCNtum affected negatively and significantly the risk of death only in interaction with the LV.SF.LTR, MOI = 100 (log-logistic parameter = –0.019; P = 6.51 × 10–5) and RV.SF.LTR (log-logistic parameter = –0.065; P = 6.58 × 10–3) but not with the LV.SF.LTR, MOI = 10, or any SIN.LTR vector (Table 4). The lack of a statistically significant impact of VCNtum for LV.SF.LTR, MOI = 10, suggests that genotoxicity was too low at this dosage to be measured in a reliable manner (Supplemental Figure 3D). This approach allowed estimation of the relative risk associated with different vectors at set dose levels. We thus plotted the risk of death at a fixed VCNtum of 1 or 10 for the genotoxic vectors (Figure 3C). RV.SF.LTR showed the highest risk, whereas LV.SF.LTR required a 10-fold higher integration load (VCNtum) to reach the same risk of γRV.
log-logistic accelerated failure time model estimates the impact on survival of VCN per cell in tumors and vector treatment in each group
Vector integration analyses. To gain functional evidence that LV.SF.LTR-driven oncogenesis in our Cdkn2a–/– model was mediated by insertional mutagenesis and to gain more insight into the low genotoxicity profile of SIN.RV.PGK and SIN.LV.SF, we compared genes targeted by integration of these vectors in tumors and in the cells used for transplant after 2 weeks of culture (in vitro). DNA from tumor-infiltrated BM (36 LV.SF.LTR, 27 SIN.LV.SF, and 23 SIN.RV.PGK mice) was subjected to a low-sensitivity linear amplification–mediated PCR (LAM-PCR) protocol aimed at identifying provirus-genomic junctions from predominant clone(s) in mixed populations (4, 7). Standard LAM-PCR protocol was used for the in vitro–cultured cells (34). We univocally mapped a total of 529 vector integration sites on the mouse genome (UCSC Mouse Genome Browser, February 2006 release) divided into 6 data sets: LV.SF.LTR, 100 sites from tumors and 70 in vitro; SIN.LV.SF, 80 from tumors and 90 in vitro; SIN.RV.PGK, 54 from tumors and 135 in vitro. The nearest gene (known to Entrez Gene or Ensembl) was then identified by bioinformatics analysis (Supplemental Table 2). Redundant integrations were excluded from calculations (total nonredundant integrations in tumors: LV.SF.LTR = 93 and SIN.LV.SF=78).
In the LV.SF.LTR and SIN.LV.SF data sets, the distribution of vector integrations displayed a pronounced tendency to integrate within genes (70%) without preference for transcription start sites (TSS), a pattern similar to that previously reported for other LVs (14, 16, 17, 23) (Supplemental Figure 4). On the other hand, 40% of SIN.RV.PGK integrations were located within genes, and 32% of the integrations clustered within ± 5 kb from the TSS.
Each data set was searched for matches to retroviral or Sleeping Beauty (SB) transposon common integration site (CIS) genes contained in the Retrovirus Tagged Cancer Gene Database (RTCGD) (35) and the frequency compared with the expected random frequency (Figure 4A, Supplemental Table 3A, and Supplemental Table 4). The SIN.RV.PGK had a pronounced tendency to target retroviral CIS in vitro and in tumors (P < 0.0001 vs. expected random; χ2 test) but not SB CIS. Several of the SIN.RV.PGK integrations targeting CIS genes mapped in the same narrow region previously targeted by retroviruses in the RTCGD (Supplemental Figure 5). On the other hand, the frequency of SIN.LV.SF integration near retroviral or SB CIS genes in vitro or in tumors was not significantly different from the expected random frequency. In contrast to the pattern observed for both SIN.LVs, LV.SF.LTR targeted both retroviral and SB CIS genes in vitro and in tumors at significantly high frequency (P < 0.0001 vs. expected random; χ2 test).
Vector integration site analysis in cells before transplant and in tumors. (A) Percentage of retroviral, SB transposon, and total (all) RTCGD CIS genes targeted in vitro and in tumors by each vector, as indicated. The expected random frequency was calculated as fraction of all mouse genes (25,613 genes). Significant overrepresentation versus the expected frequency (P < 0.05; χ2 test) is indicated by asterisks. *P < 0.05; **P < 0.01; ***P < 0.001. (B) Percentage of vector integrations (as in A) targeting the indicated GO classes in vitro and in tumors. Significantly overrepresented classes are indicated by asterisks. Fisher’s exact test. (C) Significance of overrepresented functional pathways of the IPA software is shown as –log10_P_ value. The significance threshold of P < 0.05 is indicated. Multiple comparison error correction by Bonferroni’s method decreases the significance level to P < 0.00026 (63 gene classes for 3 vectors = 189 gene comparisons, α level of 0.05). Prolif., Proliferation; Post-Translat., Post-Translational; Interact., Interaction. (D) Percentages of genes belonging to the indicated IPA functional pathways were compared for each data set. Statistically significant differences (P < 0.05, Fisher’s exact test) between the vectors in the same condition (bracket) or between the in vitro and tumor data set for the same vector (arrow) are indicated. Arrows from left to right indicate a significant enrichment from in vitro to tumors of the given gene class.
The new data sets were then evaluated with DAVID-EASE (36) and Ingenuity Pathways Analysis (IPA) software for overrepresentation of Gene Ontology (GO) classes and signaling or disease pathways (Figure 4, B and C). Analysis was performed at high stringency and the results limited to the significantly overrepresented classes with an increase of 3-fold or greater with respect to the expected random frequency (DAVID-EASE; Supplemental Tables 3 and 4) or with 3 or more genes in at least 1 data set (IPA; Supplemental Table 5) and validated by Bonferroni’s correction for multiple comparison error.
For LV.SF.LTR, the GO class Phosphorylation and Kinase Activity was overrepresented in vitro, while the GO classes Regulation of Apoptosis, Mitotic Cell Cycle, and B Cell Differentiation and the IPA classes Cancer, Cell Cycle, and Cell Death were all strongly overrepresented in tumors. As shown by P value ranking, the latter were the most significant overrepresentations found among all data sets, consistently with the observed oncogenic effect of LV.SF.LTR.
For SIN.LV.SF, the only overrepresented gene classes were the GO Chromatin Modification and the IPA Molecular Transport in vitro and the GO Helicases, Protein Phosphatase, and Intracellular Protein Transport in tumors.
For SIN.RV.PGK, the IPA Cancer and Post-Translational Modification classes were strongly overrepresented in vitro together with the GO classes Chromatin Modification and GTPase Activator, whereas the only overrepresented gene classes in tumors were the GO Protein Transport and Localization and the IPA Gene Expression classes.
We then determined the targeting frequency for the overrepresented IPA classes of each data set and performed 2-tailed Fisher’s exact test for comparing vectors and conditions (Figure 4D). In tumors, LV.SF.LTR integrations at Cancer and Hematological Disease genes were significantly enriched from the in vitro data set and were more frequent than observed for the SIN.LV.SF and SIN.RV.PGK. Interestingly, SIN.RV.PGK integrations at Cell Cycle, Cell Death, and Cell Growth and Proliferation genes were found at significantly reduced frequency in tumors compared with the in vitro data set.
Mechanism of insertional mutagenesis by LV.SF.LTR. In order to characterize the oncogenic mechanism of LV.SF.LTR, we measured the transcription level of genes near the vector integration site in early occurring tumors. When possible, tumors selected for analysis were transplanted into secondary mice to obtain biological replicates. For each gene near the integration site, we compared the average expression level in tumors carrying an integrated vector near or within that gene to the expression levels in tumors of identical phenotype but with different or no integration (Figure 5 and Supplemental Table 6). Q–RT-PCR was performed on cDNA from tumor-infiltrated BM and/or spleen tissue of 12 mice to test the expression of 18 genes surrounding 11 integration sites contained in 6 different tumors.
Gene expression analysis at LV.SF.LTR integration sites in tumors. (A–C) Expression of the indicated genes was measured by Q–RT-PCR on tumor-infiltrated BM or spleen cDNA (see also Supplemental Table 6). Expression data for primary and serially transplanted tumors with an integrated vector near the tested gene (INT) and phenotype-matched tumors with integrated vector in different sites or without integrations (No INT) are plotted. Each point is the fold change relative to matched-type tumor-infiltrated BM or spleen from the mock group (control level = 1); the horizontal bar represents the average. P value of the Mann-Whitney test comparison between the samples is indicated. P < 0.05 is considered significant. Genomic region targeted by the vector (vector position and orientation are represented by arrows) is shown below each set of expression data. Genes above the thick horizontal bar (chromosome) are transcribed from left to right; those below the chromosome are transcribed in the opposite direction. (A) Tgtp, which encodes for an interferon-inducible T cell–specific GTPase and whose TSS maps 530 bp from the vector integration, was overexpressed in both tumor-infiltrated BM and spleen of 2 primary and 4 secondary transplanted mice bearing the same integration; the expression of other genes surrounding the integration was not altered (see details in Supplemental Table 6). (B) Another integration from the same groups of mice mapped within the Sos1 (37) oncogene, leading to its significant overexpression. (C) Vector integration occurred within the Eps15 (38) oncogene, leading to its overexpression in tumors of 1 primary and 2 secondary transplanted mice.
Two primary lymphoid tumors and 4 secondary transplants (2 from each primary tumor; tumors originated from different mice transplanted with the same in vitro–transduced cell populations) shared 4 integration sites. Among 9 genes tested that surrounded these integrations, Tgtp (Figure 5A) and Sos1 (Figure 5B) were strongly and significantly overexpressed with respect to the controls (24.7 ± 3-fold increase and 6.3 ± 3.8-fold increase, respectively; P < 0.001, n = 5 vs. 9). These findings were confirmed in the spleen (9.5 ± 5.4-fold increase, P < 0.001; and 5.2 ± 3.3-fold increase, P < 0.01, respectively; n = 6 vs. 12).
In another lymphoid tumor and its 2 secondary transplants, Eps15 showed a 5.4 ± 2-fold increase compared with the controls (P = 0.013, n = 3 vs. 11) among 5 genes near the vector integration (Figure 5C). All other genes tested were not significantly different from the controls. Of note, SOS1 and EPS15 have been implicated as oncogenes in human cancer development (37, 38).
In a myeloid tumor, 1 LV.SF.LTR integration mapped within intron 11 of Braf, a genomic region targeted several times by transposon integrations in sarcomas of _Arf_–/– mice (39) (Figure 6A). In this tumor, we detected a chimeric LV-Braf transcript that contained LV LTR and leader sequence up to the splice donor motif fused to the start of exon 13 and the remaining coding sequence of Braf (Figure 6B). This transcript must originate from the LV 5′ LTR by splicing out the genomic sequence spanning from the LV splice donor to the acceptor site of exon 13 (Figure 6C). The putative protein encoded by this transcript is a truncated Braf molecule with constitutive kinase activity similar to that previously reported upon transposon integration within the same region and that has been directly implicated in cell transformation (39).
Oncogenic LV/Braf chimeric transcripts in an LV.SF.LTR tumor. (A) Genomic position of an LV.SF.LTR integration in a myeloid tumor targeting intron 11 of Braf. Chromosome (Chr) number and coordinates are indicated on top. The genomic interval covering exons 11 to 14 (gray boxes) is depicted. The position of the LV.SF.LTR integration (black box; LTR direction is indicated by the gray arrow) clusters with 20 SB integrations from sarcomas (39) in a narrow 4-kb region within introns 11 and 12. (B) RT-PCR using primers complementary to LV LTR and exon 22 of Braf on cDNA from the tumor described in A amplified a 1500-bp product. RT+, tumor cDNA; RT–, tumor RNA processed without reverse transcriptase; M molecular size markers. (C) The sequence of the RT-PCR product in B aligns to LV and to Braf exons. Black bars, amplified cDNA sequence; dashed lines, splicing events; F and R arrows, primers used for cDNA amplification; 3′UTR, 3′ untranslated region of Braf; SD, LV 5′ splice donor site. The cDNA sequence was LV specific up to the splice donor site (HIV) fused to the correct splice junction of exon 13 of Braf (boxed); exon 12 appears to be skipped. The first putative starting ATG codon in exon 13 is in the correct frame to produce a truncated Braf protein (indicated).
In 1 myeloid and 1 lymphoid tumor, the Nsd1 gene was targeted by 2 independent integrations 2,373-bp apart and in opposite orientation from each other (introns 5 and 6). The levels of expression of Nsd1 appeared to be reduced to about 40% in both tumors bearing the integration (n = 2 vs. 5). Interestingly, NSD1 haploinsufficiency is the major cause of Sotos syndrome and is associated with malignant tumor formation (40).
Overall, in each of the 6 tumors tested, we found at least 1 integration that resulted in either oncogene overexpression, generation of aberrant transcripts encoding a truncated constitutively active oncogenic protein, or putative haploinsufficiency of a tumor suppressor gene. These genotoxic events recapitulate those previously described for γ-retroviruses or transposon-driven oncogenesis.