Tumor cells escape suicide gene therapy by genetic and epigenetic instability

Cell death can be induced when proteins that have a toxic or proapoptotic function or that activate precursors of toxic drugs are expressed. Genes encoding such proteins are termed suicide genes.^1-5  Retroviral vectors with their preference for transduction of dividing cells are promising tools for suicide gene transfer into tumor cells.^1,2,6  Several approaches have been proposed to improve the efficiency of gene transfer in vivo and to ensure selective expression of suicide genes in malignant cells.^1-10 

Less well investigated are potential mechanisms of escape from retroviral suicide gene delivery, potentially even occurring after transduction with intact transgenes. Such studies are an important part of the preclinical evaluation of suicide gene therapy, especially since the introduction of suicide genes has been proposed to increase the safety of retroviral vector–mediated gene transfer into postnatal stem cells. In this case, the rationale is that activation of the suicide principle may allow the deletion of a malignant clone that may arise in consequence of protooncogene activation by random vector insertion. Given the demonstration of malignant complications of retroviral vector–mediated gene transfer both in mice and in humans,^11,12  there is an urgent need to evaluate the potential efficiency of such a preventive expression of suicide genes. This “proactive” scenario differs from the “reactive” introduction of suicide genes into pre-existing tumors in 2 major respects. Firstly, the suicide gene would be expected to be present in the same chromosomal position of any cell of the malignant clone, and secondly, the malignant clone would undergo massive clonal expansion before the application of the suicide medication.

Our hypothesis was that alterations in transgene structure or function would promote escape from negative selection. Such alterations do not appear unlikely, since genetic and epigenetic instability is characteristic for malignant tumors^13-16  and independently encountered in experimental transgenesis.¹⁷  To test the type and incidence of such escape mechanisms, we established an experimental tumor, using retrovirally transduced EL-4 lymphoma cells.¹⁸  Our model identified 3 potential mechanisms of escape from negative selection after suicide gene delivery: gene deletion by postinsertional recombination of sequences between the long terminal repeats (LTRs) of the retroviral vector, deletion of the chromosome containing the suicide gene, and transgene silencing. Our observations have significant implications for the design of proactive and reactive scenarios of suicide gene therapy.

C57Bl/6J-Ly5.1 mice obtained from The Jackson Laboratory (Bar Harbor, ME) were maintained pathogen-free at the Heinrich-Pette-Institute (HPI) animal facility and all animal experiments were approved by the regional committee. The day before tumor cell inoculation, diameters of the hind legs (5 mm) were determined using a caliper. Animals were then challenged with 1 × 10⁵ congeneic (Ly5.2) EL-4 tumor cells via subcutaneous injection into the right hind flank and analyzed for tumor induction 3 to 4 times weekly. For this purpose, the diameter measurements with a caliper of the cell-injected hind leg (right) were compared with the uninjected one (negative control). At beginning of visible and palpable tumor growth (day 10 after transplantation), mice were treated twice daily with ganciclovir (GCV; Roche, Grenzach-Whylen, Germany; 2 × 50 mg/kg resuspended in sterile filtrated 1 × phosphate-buffered saline [PBS]) for 7 consecutive days intraperitoneally or intralesionally. After tumor remission, mice were analyzed for up to 6 months for potential tumor relapse.

Vector plasmid pMO3N expressing the neomycin-resistance cDNA (neoR) under control of the LTR of Moloney murine leukemia virus (MLV) has been described.¹⁹  The suicide gene vector plasmid pMO3TIN was constructed by inserting the herpes simplex virus–thymidine kinase (HSV-TK) cDNA as a NotI-HindIII fragment along with the internal ribosomal entry site (IRES) of poliovirus (HindIII-NotI fragment) into the NotI site of pMO3N, upstream of neoR. Phoenix-gag-pol packaging cells and human HT-1080 fibrosarcoma cells (American Type Culture Collection [ATCC], Manassas, VA; no. CCL-121) were grown in Dulbecco modified Eagle medium (DMEM), mouse EL-4 lymphoma cells (ATCC; no. TIB-39) in Roswell Park Memorial Institute 1640 (RPMI). All media were supplemented with 10% fetal calf serum (FCS), 2 mM glutamine, 1 mM sodium pyruvate, and 100 U/mL penicillin/streptomycin.

The day before transfection of Phoenix-gp retroviral packaging cells, 5 × 10⁶ cells were plated in a 9-cm Petri dish. For transfection, the medium was exchanged and 25 μM chloroquine (Sigma, München, Germany) was added. Retroviral vector DNA (3-5 μg) was transfected using the calcium phosphate precipitation method. To generate vesicular stomatitis virus-G (VSV-G)–pseudotyped particles, 2 μg M57 (MLV gag-pol expression plasmid) and 2 to 3 μg M4 (VSV-G expression plasmid) were transfected in addition. Medium was changed after 6 to 8 hours. Supernatants containing viral particles were collected 24 to 84 hours after transfection, filtered through a 0.45-μm filter (Millipore, Schwalbach, Germany), and used to transduce 2 × 10⁵ target cells (EL-4). Transduction was assisted by addition of 4 μg/mL protamine sulfate and centrifugation for 60 minutes at 2000 rpm (200g) and 25°C to 32°C. Vector titers were determined by transducing predefined numbers of HT-1080 target cells with serial dilutions of supernatant and analyzing the number of surviving cell clones after geneticin transfer unit (GTU) analysis.

Genomic DNA was isolated from vector-transduced cell clones prior to transplantation into mice and after isolation of tumor cells from relapsed mice, respectively, using the DNazol Genomic DNA Isolation Reagent (Molecular Research Center, Cincinnati, OH) according to the manufacturer's instructions. DNA was digested using HindIII or SacI as restriction enzymes allowing to investigate copy number or integrity of integrated vectors, respectively. Southern blots were performed according to standard protocols. Nylon membranes were hybridized with a radioactively labeled ([alpha ³²P] deoxycytosine triphosphate [dCTP]; Hartmann Analytic, Braunschweig, Germany) HSV1-TK full-length probe (NotI-HindIII vector fragment).

Each polymerase chain reaction (PCR) was performed in a final volume of 50 μL containing 5 μLof 10 × Taq buffer (Gibco BRL Life Technologies, Eggenstein, Germany), 5 μL 15 mmol MgCl₂ (Gibco BRL Life Technologies), 0.2 μL 10 mM deoxynucleoside triphosphates (dNTPs; Geneo BioProducts GmbH, Hamburg, Germany), 5 μL 10 ×Yellow Sub (Geneo Bioproducts GmbH), 0.2 μL (5 U/μL) Taq polymerase (Gibco BRL Life Technologies), 1 μL 16 μmol for each primer (LTR1, 5′-TGT TTC CAG GGT GCC CCA AGG-3′; LTR2, 5′-CAC TCA GAG GAG ACC CTC CC-3′),²⁰  and 1 μL (1 μg/μL) genomic DNA as a template (initial denaturation step for 3 minutes at 94°C; 36 cycles of denaturation for 30 seconds at 94°C, annealing for 15 seconds at 52.4°C, and elongation for 15 seconds at 72°C). The resulting LTR-specific PCR product (248 bp) was run on an agarose gel (1.5 g agarose per 100 mL) stained with ethidium bromide. As a control for PCR, additional reactions with murine Xist gene–specific primers (X2-R, 5′-GAA GTG AAT TGA AGT TTT GGT CTA G-3′; X2-L, 5′-GGG ACC TAA CTG TTG GCT TTA TCA G-3′; generous gift from Gerald Schumann, HPI) were performed resulting in a 202-bp PCR product (firstly, 2 cycles each of denaturation for 4 minutes at 94°C, annealing for 1 minute at 54.1°C, elongation for 2 minutes at 72°C; secondly, 30 cycles each of denaturation for 1 minute at 94°C, annealing for 1 minute at 54.1°C, elongation for 1 minute at 72°C; and final product extension for 3 minutes at 72°C).

LM-PCR was performed essentially according to a published protocol by Schmidt et al.²¹  Briefly, genomic DNA isolated from retroviral vector MO3TIN–transduced EL-4 cells was digested with 5 U of restriction enzyme RsaI (New England Biolabs, Frankfurt, Germany) per μg of DNA for 2 hours at 37°C. After standard ethanol precipitation of the heat-inactivated reaction primer, extension with a vector LTR-specific 5′-biotinylated primer (rvLTRI; TIB MOLBIOL, Berlin, Germany) was performed (phycoerythrin [PE], 95°C for 5 minutes, 64°C for 30 minutes, 72°C for 15 minutes). Purified biotinylated PCR products were extracted via streptavidin-labeled magnetic beads (Dynal, Hamburg, Germany). The captured DNA was washed twice with 100 μLofH₂O and resuspended in 7 μLH₂O for overnight ligation at 16°C with 100 pmol of the annealed linker cassette (consisting of oligonucleotides NZ117, 5′-GAC CCG GGA GAT CTG AAT TCA GTG GCA CAG CAG TTA GG-3′; and NZ118, 5′-p CCT AAC TGC TGT GCC ACT GAA TTC AGA TCT CCC G-3′; sequence kindly provided by Nora Zingler, HPI) in the presence of 6 U of T4 DNA ligase (Promega, Mannheim, Germany). A wash step with H₂O preceded first-round PCR using primer combination rvLTRII and OCI (initial denaturation step for 5 minutes at 95°C; 30 cycles of denaturation for 1 minute at 95°C, annealing for 45 seconds at 58°C, and elongation for 1.5 minutes at 72°C; and final product extension for 10 minutes at 72°C). Nested PCR with rvLTRIII and OCII primers was performed under cycling conditions identical to those used for first PCR. Amplified products were separated by preparative agarose gel electrophoresis (2 g agarose per 100 mL), purified using Qiagen Gelelectrophoresis Isolation Kit (Qiagen, Hilden, Germany), and sequenced (SequiServe, Vaterstetten, Germany).

For chromosome preparation, cultured EL-4 cells were treated with colcemid for 1 hour in a concentration of 0.035 μg/mL, incubated in 0.075 M KCl for 20 minutes at 37°C, and fixed in freshly prepared methanol–acetic acid (3:1) at room temperature. Cell suspension was dropped onto glass slides in a climate chamber (Polymehr, Kassel, Germany) at 22°C and 48% humidity. Metaphase chromosomes were denatured in 70% formamide and 2 × standard saline citrate (SSC) for 1.5 minutes and dehydrated. The mouse SKY probe mixture (Applied Spectral Imaging, Migdal Ha'Emek, Israel) was denatured at 80°C for 7 minutes and preannealed at 37°C for 1 hour. Hybridization took place for 48 to 72 hours at 37°C in a humidified chamber. Signal detection procedure was carried out according to the manufacturer's instructions (Applied Spectral Imaging). Chromosomes were counterstained with DAPI (4,6 diamidino-2-phenylindole) and mounted with antifade solution. For image acquisition, the SpectraCube system (Applied Spectral Imaging) coupled to an epifluorescence microscope (Carl Zeiss Jena GmbH, Jena, Germany) was used. Chromosome analysis was performed using the SKYView software package (Applied Spectral Imaging).

On the basis of the measured mean diameter of the injected hind legs, we calculated the mean number of EL-4 cells present in the tumors at the beginning of GCV treatment (day 10 after transplantation). With 1 × 10⁵ originally injected cells and an in vivo doubling time of 38.4 hours (see “Results”, Figure 1B), this corresponded to approximately 6 doublings. On the basis of the actual sizes of tumors, the numbers of EL-4 cells in each tumor at the beginning of GCV treatment were determined. This was possible by calculating the doublings needed to reach a particular tumor size (eg, ∼8 doublings in a tumor of 8.5-mm diameter) and apply this value (n) to the formula f_(x) = y × 2ⁿ, where y is the number of originally transplanted EL-4 cells (1 × 10⁵). Taking into account the time period needed from beginning of GCV treatment to relapse, we were thus able to compute the numbers of GCV-resistant cells existent at the start of treatment. To obtain the mutation frequencies for a particular resistance mechanism in a given clone, the number of GCV-resistant cells existent at the start of treatment was divided by the total number of tumor cells present at this time point. Statistical analysis was performed with SAS software (release 8.02; SAS Institute, Cary, NC).

The experimental setup is shown in Figure 1A. The retroviral vector MO3TIN, expressing the genes for HSV-TK and neoR linked by an IRES of poliovirus, was produced as cell-free supernatant of safety-modified packaging cells and titrated on human HT-1080 cells (determining GTU). For tumor induction experiments, we chose EL-4 lymphoma cells that revealed no evidence of replicating retrovirus (data not shown) and induced a malignant tumor when injected into congeneic immunocompetent C57Bl/6 mice. EL-4 cells were transduced in vitro with a multiplicity of infection less than 1, resulting in a single transgene copy in the majority of transduced cells. This mimics potential limitations encountered in retroviral transgene delivery into tumor cells in vivo and also represents the desired outcome of retroviral gene transfer in the genetic modification of somatic stem cells.²²  MO3TIN-transduced cells were selected in vitro using G418 for 14 days, using concentrations of 1.2 mg/mL. This ensured complete elimination of untransduced cells, as previous dose-finding experiments revealed that concentrations of 0.6 to 0.8 mg/mL were sufficient to kill untransduced EL-4 cells.²³ 

Subcutaneous transplantation of uncloned, neoR-resistant EL-4 cells reproducibly resulted in rapid tumor formation in C57Bl/6 mice. The in vivo doubling time was in the range of 40 hours (Figure 1B).²⁴  Thus, transgene expression did not give rise to immunologic rejection of EL-4 cells.¹⁸  Subsequent experiments addressed whether local or systemic treatment with GCV could reverse the EL-4 tumors. In the case of relapse, detailed analyses were performed to identify the underlying mechanisms (Figure 1A).

At the beginning of visible tumor growth (day 10, equivalent to ∼6 population doublings), mice were injected twice daily for 7 days with GCV (50 mg/kg body weight), given either intraperitoneally or intralesionally. Both regimens resulted in remission of established lesions. However, prolonged observation (for 4 weeks) of mice that received transplants of uncloned putative MO3TIN-transduced EL-4 cells showed local relapse with an incidence of 100%. When material from preliminary tests was analyzed by PCR and Southern blot, the transgene could not be detected in relapsed tumors (data not shown).²³  As survival of untransduced EL-4 cells during G418 selection in vitro was unlikely due to the stringency of G418 selection, this indicated genetic instability occurring during in vivo passage, resulting in loss (partial) of the transgenic allele.

To fully exclude that relapse occurred from residual cells surviving G418 selection and to analyze potential clonal variability in escape mechanisms, 6 of 9 independent G418-resistant MO3TIN-transduced EL-4 clones with single vector insertions (confirmed by Southern blot; data not shown)²³  were used in further investigations. To verify intact TK function prior to tumor inoculation, cells from each of the 6 clones were treated with GCV. More than 95% cell death was observed in all cultures. Thus, the TK gene was intact and functionally expressed in the vast majority of independent clones. This indicated that retroviral transduction of truncated versions of TK was not a frequent event in our experimental setting, in line with recent findings.¹⁰  However, despite high rates of initial cell killing, GCV-resistant subclones from 3 of the analyzed clones could be recovered, indicating a low but detectable mutation frequency to GCV resistance in vitro. Control G418-resistant EL-4 cells transduced with a similar retroviral vector lacking TK (MO3N) remained resistant to GCV toxicity in culture, as wild-type EL-4 cells.

Six clones containing an intact MO3TIN transgene as well as 2 uncloned G418-selected mass cultures (transduced with either MO3TIN or MO3N) were inoculated into 5 mice per group. Rapid tumor outgrowth was observed in animals that received transplants of MO3N-transduced cells with or without GCV administration. Again, initial tumor remission was observed after administration of GCV in all animals that received transplants of MO3TIN-transduced cells. However, relapses occurred in all animal groups that received transplants of either cloned or uncloned MO3TIN cells. Three of the 5 recipients of cells derived from the MO3TIN mass culture showed onset of relapse already at the end of the GCV treatment period, indicative of a relatively high frequency of escape mutants. Between the different MO3TIN clones, the incidence and kinetics of relapse showed a strong variability (Figure 2; Table 1).

Clone no. 2 showed the greatest incidence of relapse (5/5) and the greatest extent of subclonal variability (Figure 2). For clone nos. 1, 3, 4, and 5, the experiment was repeated. Among these 4 clones, no. 5 showed the lowest incidence of relapse (1/10 animals) and no. 4 the highest (9/10 animals; Table 1). The average incidence of tumor relapses from clonal tumors was 48.89% (95% confidence limit, 0.3370-0.6423). By using Fisher exact test we found a highly significant association between the incidence of tumor relapses and the respective clone (P = .0001), suggesting an important role of the vector insertion site.

To analyze the mechanisms of escape from GCV selection, we isolated relapsed tumor cells. In contrast to their clonal progenitors, all relapsed cells were resistant to GCV in vitro, ruling out that the drug regimens were inefficient in dose, penetration, or timing. Based on Southern blot analysis (data not shown),²³  relapsed tumor cells originating from MO3TIN-transduced mass cultures showed a monoclonal or oligoclonal pattern with respect to the vector insertion site. In relapsed tumors originating from defined clones, neither neoR nor TK transcripts were detected by reverse transcriptase–PCR (RT-PCR; data not shown). For a more detailed analysis, we used a PCR to detect residual LTR sequences of the vector²⁰  and checked all relapsed clones using this PCR in addition to the Southern blot that only detected the presence of TK. Southern blot (data not shown)²³  and PCR data (Figure 3) revealed that the GCV-resistant subclones of clone no. 4 (high incidence of relapse with relatively early manifestation) still showed the presence of an intact transgene. In contrast, in all other relapsed tumors we observed either a partial loss of transgene (ie, Southern blot TK negative, PCR LTR positive, as in the single GCV-resistant subclone of no. 5 and the GCV-resistant subclones nos. 2-1, 2-5, and 3-4) or a complete loss of transgene (ie, both LTR-PCR and TK-Southern blot negative as in all other GCV-resistant subclones of nos. 1, 2, 3, and 6).

To address how the complete loss of a retrovirally transduced transgene could be explained, we analyzed the karyotypes of clone no. 1 and a relapsed subclone thereof (no. 1-3) recovered after in vivo GCV treatment. Highly sensitive SKY analysis of subclone no. 1-3 showed the loss of regions A1 to B2/3 from chromosome 6 of one defined derivative fusion chromosome [Der(19)T(6B;19D)] (Figure 4; Table 2). The same chromosomal loss was observed (SKY data not shown) in a GCV-resistant subclone recovered after treating clone no. 1 with GCV in vitro (see “Relapse after GCV (“suicide”) treatment even occurs in clonal tumors”).

For each of the 6 clones, the insertion site of the retroviral vector (Table 3) was identified by LM-PCR, sequence analysis, and databank search.²¹  None of the loci coincided with potential or established oncogenes characterized by studies with murine replication-competent viruses.²⁶  Therefore, it is unlikely that insertional activation of oncogenes or disruption of tumor suppressor genes were the cause of the differential resistance patterns of the clones. In clone no. 1, sequence data were consistent with integration in the above-mentioned fusion chromosome (region B3). Therefore, we conclude that loss of the retrovirally delivered transgene cassette by chromosomal deletion and/or chromosomal loss was the underlying mechanism of escape from suicide gene therapy in the tumor subclones of clone no. 1 (eg, no. 1-3) and possibly other relapsed tumors of clone nos. 2, 3, and 6, where no vector sequences could be detected by Southern blot and PCR analysis.

As mentioned in “Genetic analysis of escape mutants,” the PCR analysis detected LTR sequences in other subclones that showed no TK transgene in the Southern blot (eg, nos. 2-1 and 5-2). Such a result was consistent with postinsertional recombination as the mechanism of escape, possibly involving the LTR sequences.^27,28  Interestingly, the strongest interanimal variability with respect to the latency of relapse was observed in clones showing 2 different mechanisms of escape in their subclones, either chromosomal deletion and/or chromosomal loss, or postinsertional recombination (Table 1; Figure 2). Statistical analyses (Fisher exact test) revealed a highly significant association between the investigated resistance mechanism against GCV and the various clones (P = .000 006).

The earliest relapse, indicating a relatively high mutation frequency, was observed with clone no. 4, in which silencing of expression of the intact transgene occurred in all subclones analyzed (Table 1). The known tumor doubling time, the size of the tumor at the beginning of GCV exposure, and the latency of relapse allowed us to determine the mutation frequency to GCV resistance per cell (Table 3). As it remained unknown which fraction of the GCV-resistant cells might have been eliminated by bystander effects or a potential immune response, the actual mutation frequencies may even have been higher. Thus, the mutation rate of the clone escaping by gene silencing was at least 9 × 10^–2. Possibly depending on the insertion site, postinsertional recombination occurred with frequencies ranging from 3 × 10^–2 to 2 × 10^–5 and chromosomal deletion and/or chromosomal loss with frequencies ranging from 2 × 10^–3 to 8 × 10^–5. If similar mutation rates would occur in the scenario of “proactive” suicide gene delivery prior to clonal expansion in vivo, any of these mechanisms would produce a relatively high number of potential suicide-resistant cells.

The present study reveals that both genetic and epigenetic instability limit the efficiency of retroviral suicide gene therapy of malignant tumors. Besides epigenetic gene silencing, we identified 2 genetic mechanisms of escape: postinsertional recombination, an event that can be triggered by the sequence repetitions in the LTRs,^27,28  and deletion of neighboring chromosomal sequences along with the retroviral insertion site. Escape from suicide gene therapy was observed in all uncloned and cloned tumor populations analyzed. The type and incidence of escape mutants were probably related to the insertion site of the vector, as independent clones differed with respect to the mechanisms of relapse (P = .000 006). Importantly, as killing of mutant bystander cells^1,4  and systemic immunity^6,29,30  are typically extending the efficiency of the GCV-mediated suicide to neighboring cells not actively expressing TK, the incidence of escape variants may have been even higher than detected in our study (range, 3 × 10^–2 to 8 × 10^–5 depending on clone and mechanism).

Besides the dependence on the transgene insertion site, the incidence and type of escape variants is probably influenced by additional factors, such as the type of genetic alterations underlying tumor development, the clonal expansion of tumor cells following gene transfer prior to activation of the suicide mechanism, and the type of vector system used. Additional possibilities for escape mechanisms (not investigated by us) exist in transgene recombinations occurring prior to incorporation into vector particles¹⁰  or at the level of gene transfer, in the response of tumor cells to the suicide mechanism, and in insufficient drug delivery.

It is likely that higher transgene copy numbers per single cells, or the combined transfer of different suicide transgenes,⁶  will decrease the risk for genetic or epigenetic escape from negative selection. However, increased transgene copy numbers of integrating vectors may be difficult to achieve or result in a higher risk for healthy tissues, given the current limitations in tumor cell targeting in vivo.²²  More promising options to reduce the incidence of escape variants in reactive scenarios of suicide gene therapy are an early application of gene transfer in younger stages of tumor development and activation of the suicide mechanism shortly after (retroviral or episomal) gene transfer, before variants with genetic or epigenetic instability develop.

When considering the proactive use of suicide genes to delete malignant clones with insertional up-regulation of protooncogenes,^11,12,22  the escape mechanisms of gene silencing or chromosomal deletion/chromosomal loss could also involve the activated protooncogene. If this represented a necessary signal alteration for tumor cell survival, such clones would be spontaneously extinguished shortly after their occurrence. However, postinsertional recombination may preserve the activated protooncogene while deleting the suicide gene, given that those cis-active elements of the LTR, which were involved in the activation of a protooncogene, remain intact. Postinsertional recombination may also maintain the potential disruption of a tumor suppressor gene. If, however, tumor progression became independent of the initial insertional lesion, all 3 escape mechanisms may foster the survival of mutants.

The experimental system established here may be modified to address the relevance of these different obstacles and develop possible solutions. In this context, it will also be interesting to test the impact of the vector backbone on the incidence of escape from suicide gene therapy. Lentiviral vectors as opposed to the MLV-based vectors studied here not only show a stronger trend to insert into transcribed regions of cellular genes³¹  but also are typically constructed with self-inactivating LTRs.³²  Both features may reduce the incidence of suicide escape mechanisms. Specifically, insertion into active genes may lower the rate of gene silencing, and the shorter regions of sequence identity found in self-inactivating LTRs may prevent postinsertional recombinations within the integrated transgene.

Our ongoing experiments in the mouse model have revealed the risk of leukemia induction by combinatorial protooncogene activations from independent retroviral insertions in single hematopoietic cells.³³  For the proactive use of suicide genes in experimental conditions where cells will need to be transduced with more than one transgene copy to overcome a therapeutic threshold, our data suggest a positive conclusion. In such a case, the incidence of escape would be expected to be too low to allow the survival of mutants with 2 or more spontaneous recombination or silencing events.

The authors wish to thank Gökhan Arman-Kalcek, Michaela Rodenburg, and Martin Forster (HPI) for experimental assistance. We are very grateful for statistical analysis kindly performed by Dr Christel Weiss (Department of Biostatistics, Medical School Mannheim, University of Heidelberg, Germany).