Abstract
The telomerase reverse transcriptase (TERT) gene is responsible for telomere maintenance in germline and stem cells, and is re-expressed in 90% of human cancers. Contrary to common concepts, CpG methylation in the TERT promoter (TERTp), was correlated with TERT mRNA expression. Furthermore, two hotspot mutations in TERTp, dubbed C228T and C250T, have been revealed to assist binding of transcription factor ETS/TCF and subsequent TERT expression. This study aimed to elucidate the combined contribution of epigenetic (promoter methylation and higher-order chromatin structure) and genetic (promoter mutations) mechanisms in regulating TERT gene expression in healthy skin and in melanoma cell lines (n=61). We unexpectedly observed that the methylation of TERTp was as high in a subset of healthy skin cells, mainly keratinocytes, as in cutaneous melanoma cell lines. In spite of the high promoter methylation fraction in wild-type (WT) samples, TERT mRNA was only expressed in the melanoma cell lines with high methylation or intermediate methylation in combination with TERT mutations. TERTp methylation was positively correlated with chromatin accessibility and expression in 8 melanoma cell lines. Cooperation between epigenetic and genetic mechanisms were best observed in heterozygous mutant cell lines as chromosome accessibility preferentially concerned the mutant allele. Combined, these results suggest a complex model in which TERT expression requires either a widely open chromatin state throughout the promoter in TERTp-WT samples due to high methylation or a combination of moderate methylation fraction/chromatin accessibility in the presence of the C228T/C250T mutations.
Author summary PvdV and RvD formulated research goals and aims and supervised the overall progress. Wet-lab experiments, preparation of the manuscript and statistical analysis were performed by CS and CR. CS designed the novel assays. RN was involved in the experimental setup. RvD, NG and PvdV were responsible for funding acquisition. CR, RN, NG, RvD and PvdV critically reviewed the manuscript.
Introduction
Approximately 90% of all human cancers share a transcriptional alteration: reactivation of the telomerase reverse transcriptase (TERT) gene [1, 2]. TERT encodes the catalytic subunit of the ribonucleoprotein telomerase and is capable of extending the repetitive, non-coding DNA sequence on terminal ends of chromosomes, the telomeres. As the single-stranded 5’ ends of chromosomes are shortened with each cellular division, telomeres prevent loss of coding chromosomal DNA [3-6]. Telomerase is only transcribed in a subset of stem cells in growing or renewing tissues, but through reactivation of telomerase expression, cells can extend telomeres or prevent telomeres shrinkage. This is termed telomere maintenance, which is one of the hallmarks of cancer, and allows subsequent indefinite proliferation and immortalization [3, 6-8].
Since the MYC oncogene has firstly been identified to activate telomerase, a variety of epigenetic or genetic mechanisms in the gene body or TERT promoter (TERTp) have followed, such as CpG methylation, histone modifications, mutations, germline genetic variations, structural variations, DNA amplification or chromosomal rearrangements [3, 5, 7].
A widely investigated mechanism that could induce TERT reactivation is the presence of mutations in the gene promoter [7, 9]. Horn and Huang et al. identified two mutually exclusive TERTp point mutations that are correlated to TERT mRNA expression by creating binding motifs for the transcription factor E26 transformation-specific/ternary complex factor (ETS/TCF) [7, 9]. These mutations, chr5:1,295,228 C>T (−124 bp from the transcription start site) and chr5:1,295,250 C>T in hg19 (−146 bp from TSS), henceforth respectively dubbed C228T and C250T, were first identified in melanoma. Furthermore, these mutations showed high prevalence in and were correlated with poor prognosis of cutaneous melanomas [4, 5, 10-12].
An additional mechanism by which a gene can be made accessible to transcription factors, facilitating gene expression, is hypomethylation of promoter CpG islands, a hallmark of euchromatin [13, 14]. Methylation located in the gene body, however, shows a positive correlation with active gene expression [15]. In stark contrast to most genes, TERTp hypermethylation may also allow gene expression since transcriptional repressors rely on unmethylated promoter CpGs, such as CCCTC-binding factor (CTCF)/cohesin complex or MAZ [16-18]. As such, in combination with transcription factor binding, dissociation of the repressor may result in TERT expression [3, 16, 19, 20]. Castelo-Branco et al. proposed that methylation of a specific CpG site in TERTp, cg11625005 (position 1,295,737 in hg19) was associated with paediatric brain tumours progression and poor prognosis [20]. This finding was later supported by the study from Barthel et al., in which the CpG methylation was found to be correlated with TERT expression in samples lacking somatic TERT alterations and to be generally absent in normal samples adjacent to tumour tissue [3].
Chromatin organisation, its plasticity and dynamics at TERTp region have been reported as relevant players in regulation of gene expression by influencing the binding of transcription factors [21, 22]. Cancer cells are positively selected to escape the native repressive chromatin environment in order to allow TERT transcription [23].
In the present study, we aim to elucidate the interaction of genetic and epigenetic mechanisms in regulation of TERTp. We approach this by using novel droplet digital PCR (ddPCR)-based assays [24]. Human-derived benign skin cells (keratinocytes, dermal fibroblasts, melanocytes, skin biopsy samples and naevi) and melanoma cell lines were analyzed. The TERTp mutational status was assessed along with the absolute presence of methylation in the TERTp at a CpG-specific resolution. The effect of chromatin accessibility in TERT expression was evaluated in a subset of cultured melanoma cell lines.
Results
NGS-based deep bisulfite sequencing and development of a ddPCR assay to assess TERTp methylation fraction
We first aimed to quantitatively measure the TERTp methylation at a CpG-specific resolution in primary skin samples and melanoma cell lines. DNA of 44 primary skin biopsy samples and melanoma cell lines was bisulfite-converted (BC) and analysed using NGS-based deep bisulfite sequencing to assess the methylation fraction (MF) in a region of TERTp encompassing 31 CpG sites. The TERTp MF was high in some healthy skin samples, such as normal skin (∼30%), naevi (∼30%) and cultured keratinocytes (∼50%). In the latter group, in fact, the MF was as high as in cutaneous melanoma cell lines (Fig 1).
In order to validate the TERTp MF obtained through NGS in a quantitative manner, we have developed a ddPCR assay using methylation-sensitive restriction enzymes (MSREs) HgaI and AvaI, which recognise the CpG on position 1,295,737 (cg11625005) and 1,295,731 in hg19, respectively (Fig 2). Castelo-Branco et al. showed that methylation of the cg11625005 in TERTp, was associated with tumour progression and poor prognosis of childhood brain tumours [20]. Barthel et al. affirmed a correlation between methylation and TERT expression in samples lacking somatic TERT alterations and a lower methylation level in normal samples [3]. Indeed, in our study, the MF of fibroblasts was as low as that of the unmethylated control DNA, whereas that of the keratinocytes was higher than most of the cutaneous melanoma cell lines (Fig 2B). The MF of cg11625005 (position 1,295,737) obtained through NGS and by ddPCR were highly correlated (R2=0.8166, P < 0.001) (Fig 2C). The MF of 1,295,731 assessed through ddPCR even yielded a stronger correlation (R2=0.9580, P < 0.001) (Fig 2D).
Absence of correlation between methylation fraction and TERT expression
Cancer cells are commonly characterised by hypermethylation of promoter CpG islands resulting in repression of tumour suppressor genes. However, in TERT, promoter hypermethylation was found to be associated with higher expression, since CTCF repressors of TERT transcription do not bind methylated sequences [3, 16, 17, 19]. In our sample cohort, there was no correlation between TERT methylation of cg11625005 and mRNA expression (n=34, Fig 3 and an overview in Fig 7C).
Evaluation of TERTp mutations in a collection of skin samples and melanoma cell lines
Besides promoter methylation, somatic mutations are also known to be correlated with TERTp reactivation. Therefore, we characterised the TERTp mutational status of the sample cohort. Sanger sequencing on one naevus, fresh skin and cutaneous melanoma cell lines 518A2, 607B, A375, 94.07 and 93.08 revealed melanoma-associated TERT C250T and C228T mutations (Fig 4A). Aiming to use the ddPCR method to evaluate the mutational load of the samples, the TERT C250T and C228T mutation assays were validated in three samples of which the mutation was identified in sequencing analysis, 518A2, 607B and A375 (Fig 4B). Following the test runs, the C228T and C250T assays were used on the extended sample cohort (n=61) (S5 Table and Fig 7D). All TERTp-mutated samples were cutaneous melanoma cell lines, however OCM8 and 94.13 cutaneous cell lines tested wild-type. The C250T mutation was not present in combination with the C228T mutation in any sample, confirming that the mutations are mutually exclusive.
Absence of correlation between mutational status and TERT expression
As the presence of mutations in the gene promoter induces TERT reactivation, we assessed the correlation between mutational status with TERT mRNA expression (n=34). When WT and mutated samples (either C228T or C250T) were compared, regardless of origin of the tissue, no significant differences for TERT mRNA expression were found (Fig 5). Moreover, TERT expression was exclusive to the melanoma cell lines, either with or without TERTp mutations (Fig 7C).
TERT expression is correlated to chromatin accessibility
In contrast to most genes, methylation of the TERTp positively correlates with its mRNA expression [3, 16, 17, 19]. Although we were not able to confirm this finding, we investigated whether besides promoter methylation, other mechanisms could contribute to chromatin accessibility to transcription factors affecting TERTp regulation. Therefore, we analysed chromatin state in a subset of melanoma cell lines (cutaneous, 518A2, 607B, 94.07, A375, 93.08 and OCM8; and uveal, OMM2.5 and Mel270) by ddPCR methodology instead of qPCR for an accurate quantification. The positive control gene GAPDH, a housekeeping gene that is generally expressed in all conditions, and thus 100% accessible, was used. The accessibility in the region around cg11625005 shows a high variability, being over 90% in uveal cell lines while being intermediate to low in cutaneous melanoma cell lines (Fig 6A and an overview in Fig 7E and S6 Table). When comparing the accessibility around cg11625005 to the methylation fraction of this CpG, a significant positive correlation was observed (R2= 0.89, P<0.001) (Fig 6B). Another positive correlation (R2=0.59, P<0.05) was found when comparing the accessibility of the same region to the normalised TERT mRNA expression levels in these samples (Fig 6C). In actuality, in this subset of 8 cell lines, the TERTp methylation and gene expression show a statistically significant (P-value<0.05) positive correlation (Fig 6D). The 3 cell lines with higher MF are those with the highest chromatin accessibility (OMM2.5, Mel270 and OCM8). Remarkably, these are also the cell lines with WT-TERTp, in which the chromatin accessibility was significantly higher than in the mutated subgroup (Fig 6E).
In addition, we investigated whether the TERT accessibility originated from the mutant or the wildtype allele. For this purpose, we assessed the fractional abundance of mutated allele, in the subgroup of 4 TERTp-mutated cutaneous cell lines before and after nuclease digestion. 607B cell line was not included since it is homozygous for the mutation and not informative. Assuming that the nuclease digests DNA in open and accessible chromatin regions, the observed decrease in mutation fractional abundance after digestion (Fig 6F) in all 4 cell lines suggest that mutated alleles were preferably digested over WT alleles.
Discussion
By using advanced quantification methods, we investigated the epigenetic and genetic regulation of TERTp in benign and malignant skin cells. Innovative ddPCR-based assays were developed and validated to assess TERT promoter methylation and chromatin accessibility. These methods overcome fallible bisulfite-conversion and avoid semi-quantitative qPCR and provide absolute quantification even in samples that are challenged by DNA concentration and integrity.
The methylation fraction assessed by both NGS and ddPCR was high in a variety of normal samples, of which mainly keratinocytes exceeded levels of cutaneous melanoma cell lines. This is in contrast with previous investigations on brain tumours and skin melanoma that observed a general absence of cg11625005 methylation in normal cells [3, 20]. In our study, methylation of cg11625005 at position 1,295,737 did not stand out across the CpGs in TERTp but seemed to be affected along with other CpG’s in the surrounding region in all samples (Fig 7B). This result suggests that context-related methylation around cg11625005 is biologically relevant in opposition to methylation of one specific CpG. Consistent with previous findings, the methylation of most samples gradually increased in the 5’ direction and decreased near the transcription start site (TSS) of the TERT gene (Fig 7B) [19, 25]. Regardless of the methylation status, human-derived benign cells did not express TERT indicating that other epigenetic mechanisms are involved (Fig 8). In contrast, analysis of tumour cell lines revealed a wide variety of promoter methylation levels (5%-100% MF). TERT expression was found in all tumour cell lines with or without TERTp mutation.
A plethora of histone modifications result in chromatin remodelling that may change accessibility of the TERTp to transcription factors, such as ETS/TCF [7]. Therefore, we explored the higher-order chromatin state and its interaction with methylation levels and mRNA expression in 6 cutaneous and 2 uveal melanoma cell lines. We found that the gene accessibility around cg11625005 showed a positive correlation with the methylation and TERT mRNA expression in these samples.
We next investigated whether both wildtype and mutant TERT alleles were equally affected by higher order chromatin organization and assessed the mutational fraction upon digestion with nuclease, assuming that the nuclease only digests DNA in open and accessible chromatin regions. We could infer that, mutated alleles are more accessible, possibly favouring the binding of transcription factors, such as ETS/TCF, and consequently TERT expression of the mutant allele (Fig 8). The 94.07 cell line is an exception to the rule that still supports the dominant role of higher order chromatin organization since both alleles were equally resistant to nuclease digestion and presented with very low methylation fraction, explaining the lowest TERT expression levels among all cell lines. Our results are in line with the study from Stern et al. and Huang et al., where the authors found that active mutant allele allows monoallelic TERT expression [25, 26].
Another remarkable observation in our study is that in WT TERT-expressing uveal melanoma cell lines, the methylation of the whole region surrounding cg11625005 is close to 100% with a significantly higher chromatin accessibility compared to TERTp-mutated cell lines with moderate methylation. In these cases, of TERTp-WT samples that show gene expression, we were not only able to confirm but also expand previous results, in which TERTp methylation carries out a non-canonical role, leading to transcriptional activation (Fig 8).
We conclude that ddPCR is a highly sensitive and quantifiable technique that can reliably assess methylation fractions and mutational status even in CG-rich sequences such as TERT gene. Further investigation in primary melanoma is needed to assess whether TERT methylation is predictive of worse prognosis and at which methylation fraction this phenomenon occurs [25]. Thereafter, quantification of TERT methylation might be used for the assessment of patient prognosis, as it is readily applicable in the clinic. Although TERT is one of the most affected genes in cancer, with its noncoding mutations cooperating with promoter methylation, further investigation must be conducted to fully understand all epigenetic mechanisms that collectively reactivate TERT.
Material and Methods
Samples, DNA extraction and PCR
Tissue samples were derived from anonymous patients and consisted of 11 normal skin samples, 6 frozen naevi, and low-passage cultured samples: 5 fibroblasts, 6 melanocytes and 8 keratinocytes. Primary human fibroblasts and keratinocytes were isolated from surplus human breast skin as described before [27]. Keratinocytes were used at passage 2, while fibroblasts were used at passage 3-5. The low-passage cultured fibroblasts, keratinocytes and melanocytes were a kind gift from A. El Ghalbzouri and JJ Out-Luiting [27].
We also included 19 cutaneous and 6 uveal melanoma cell lines [28]. The batch thus consisted of 39 primary skin type samples and 25 melanoma cell lines, totalling 61 samples (Table 1). The study was approved by the Leiden University Medical Center institutional ethical committee (05-036) and was conducted according to the Declaration of Helsinki Principles.
DNA was isolated using the QIAamp DNA Blood Mini Kit and the DNeasy Blood & Tissue Kit (both from Qiagen, Hilden, Germany).
Conventional PCR was performed using the PCR-sequencing kit (Thermo Fisher Scientific, Waltham, MA, USA), containing 10X reaction buffer, MgCl2 (50mM), dNTP mix (10nM, Fermentas/Thermo Fisher Scientific), primer mix (900nM each), PlatinumX Taq enzyme (2.5U), 50ng DNA and Aqua B. Braun RNase-free water. A PCR for CG-rich sequences was performed on 50ng DNA using the PCRX Enhancer System (Thermo Fisher Scientific), containing 10X PCRX amplification buffer, MgSO4 (50mM), dNTP mix (10nM), primer mix (900nM each), PlatinumX Taq enzyme (2.5U) and Aqua B. Braun RNase-free water. The samples were amplified in C1000 Touch Thermal Cycler (Bio-Rad Laboratories, Inc., Hercules, CA, USA).
Promoter methylation determination
Bisulfite conversion and next-generation sequencing (NGS)-based deep bisulfite sequencing
DNA was bisulfite-converted (BC) using the EZ DNA Methylation™ Kit (Zymo Research, Irvine, CA, USA) according to the manufacturer protocol (version 1.2.2). BC samples were amplified using the PCRX Enhancer System in the program: 1 cycle of 95°C for 3 minutes, 8 cycles of 95°C for 30 seconds, 58°C for 30 seconds, reducing 1°C/cycle, and 68°C for 1 minute, then 36 cycles of 95°C and 53°C for 30 seconds each, and 68°C for 1 minute, followed by 1 cycle of 68°C for 3 minutes. Tailed primers were used for amplification (900nM each; S1 Table). Samples were sequenced through next-generation sequencing (NGS), MiSeq, 2×300bp paired-end, at Leiden Genome Technology Centre (LGTC).
Novel design of a ddPCR assay using methylation-sensitive restriction enzymes (MSREs) to determine TERTp methylation fraction
The methylation fraction (MF) of the CpG (cg11625005) in position 1,295,737 was determined by an in-house designed ddPCR assay in combination with HgaI methylation-sensitive restriction enzyme (MSRE) that cleaves this CpG when unmethylated, as described by Nell et al. [24]. 100ng DNA sample was incubated with HgaI (2U/μl) and appurtenant 10X NEBuffer 1.1 (both from New England Biolabs, Bioké, Leiden, The Netherlands) for 60 minutes at 37°C and 65°C for 20 minutes. To assess the MF of a CpG adjacent to cg11625005, located in 1,295,731, the MSRE AvaI (10U/μl; New England Biolabs) was employed, which recognises this CpG and cleaves it when unmethylated. Incubation of the DNA samples with AvaI was performed with 10X CutSmart buffer for 15 minutes at 37°C and subsequently 65°C for 20 minutes. For ddPCR reaction, 60ng DNA digested or undigested by HgaI, 2x ddPCR SuperMix for Probes (no dUTP), primers (900nM each), a FAM-labelled in-house-designed probe for the CpG site of interest (250nM, Sigma, St. Louis, MO, USA), and 20X HEX-labelled CNV TERT reference primer/probe (Bio-Rad) for total TERT amplicon count. The primer and probe sequences are presented in S2 Table. The amplification protocol used: 1 cycle of 95°C for 10 minutes, 40 cycles of 94°C for 30 seconds and 60°C for 1 minutes, and 1 cycle of 98°C for 10 minutes, all at ramp rate 2°C/s. Droplets were analysed through a QX200 droplet reader (Bio-Rad) using QuantaSoft software version 1.7.4 (Bio-Rad). Raw data was uploaded in online digital PCR management and analysis application Roodcom WebAnalysis (version 1.4.2, https://www.roodcom.nl/webanalysis/) [24], in which the MF was calculated by dividing the CNV of the digested sample with that of the paired undigested sample.
Assessment of mutational status
Sanger sequencing
The presence of the C228T and C250T TERTp mutations in some samples was evaluated by conventional Sanger sequencing. DNA samples were amplified through the PCRX Enhancer System (Thermo Fisher Scientific) using primers (Sigma-Aldrich) and amplification program described by McEvoy et al. [29].
Mutation analysis using commercial TERT C250T and C228T mutation assays
For most of the samples, the TERTp mutations were detected by the ddPCR technique according to protocol described by Corless et al. [30], using the TERT C250T_113 Assay and C228T_113 Assay (unique assay ID dHsaEXD46675715 and dHsaEXD72405942, respectively; Bio-Rad). Both assays include FAM-labelled probes for the C250T and C228T mutations respectively, HEX-labelled wild-type (WT) probes, and primers for a 113-bp amplicon that encompasses the mutational sites. The ddPCR reaction mix comprised 1X ddPCR Supermix for Probes (No dUTP), Betaine (0.5M; 5M stock), EDTA (80mM; 0.5M stock, pH 8.0, Thermo Fisher Scientific), CviQI restriction enzyme (RE; 2.5U; 10U/μl stock, New England BioLabs), the TERT assay, and 50ng DNA. Droplets were generated in QX200 AutoDG system (Bio-Rad) and amplified in T100 Thermal Cycler (Bio-Rad) according to the recommended cycling conditions and analysed through a QX200 droplet reader (Bio-Rad) using QuantaSoft software version 1.7.4.0917 (Bio-Rad).
Chromatin accessibility
Cell culture and treatment to assess chromatin states
Cutaneous melanoma cell lines A375, 518A2, 607B, 94.07, 93.08, OMM2.5, Mel270 and OCM8 were cultured for 22 days in 9-cm Cellstar® cell culture dishes (Greiner Bio-One GmbH, Frickenhausen, Germany) with Dulbecco’s modified eagle medium (DMEM; Sigma-Aldrich) supplemented with 10% FCS, Penicillin (100U/ml), and Streptomycin (100μg/ml; both from Lonza, Verviers, Belgium) until roughly 95% confluent. Then, different densities (10,000, 20,000, 40,000 and 80,000 cells) of the above-mentioned cell lines were seeded in duplicate into a 48-well plate (Corning Costar, Sigma-Aldrich) required for the EpiQ chromatin assay. The EpiQ™ Chromatin Analysis Kit (Bio-Rad) was performed according to manufacturer’s instructions. Briefly, after 2 days each cell line was 85%-95% confluent. The cells were permeabilised and treated with EpiQ chromatin digestion buffer with or without nuclease for 1 hour at 37°C. Following incubation with EpiQ stop buffer for 10 minutes at 37°C, the DNA samples were purified using alcohol and DNA low- and high-stringency wash solutions. The genomic DNA was eluted in DNA elution solution.
Novel design of a ddPCR assay to assess chromatin opening state
The analysis was performed using ddPCR rather than qPCR, to achieve quantifiable results using GAPDH expression as positive control. The reaction mix consisted of 2x ddPCR Supermix for Probes (No dUTP, Bio-Rad), 20x HEX-labelled CNV TERT reference primer/probe (Bio-Rad), 50ng DNA, and primers (900nM each) and FAM-labelled probes (250nM) for GAPDH, or the methylation region around cg11625005 (S3 Table). Samples were amplified according to the program of the CNV TERT reference primer/probe as described. Gene accessibility was quantified by the digestion fraction between the digested and undigested samples, subtracted from 1.
RNA isolation, cDNA synthesis and quantitative real-time PCR
RNA was obtained using the FavorPrep Tissue Total RNA Extraction Mini Kit (Favorgen Biotech, Vienna, Austria) according to manufacturer’s instructions for animal cells. cDNA was synthesised through the iScript™ cDNA Synthesis Kit (Bio-Rad) according to recommended protocol. TERT mRNA expression was assessed by qPCR performed with 3.5ng DNA, IQ SYBR Green Supermix (2x; Bio-Rad), and 0.5μM PCR primers (Sigma-Aldrich; S4 Table) in a Real-Time PCR Detection System CFX96 (Bio-Rad) and normalised to reference gene expression (RPS11, TBP and CPSF6, S4 Table). Data was analysed through the ΔΔCT method in Bio-Rad CFX manager software (version 3.1, Bio-Rad).
Statistical Analysis
MF obtained using ddPCR was calculated with 95% confidence interval through RoodCom WebAnalysis (version 1.4.2). Significant testing of linear regression and multiple comparisons in correlation plots was performed through GraphPad Prism (version 8 for Windows, GraphPad Software, CA, USA).
Funding
This project has received funding from the European Union’s Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie grant agreement No. 641458. R.Nell is supported by the European Union’s Horizon 2020 research and innovation program under grant agreement No 667787 (UM Cure 2020 project).
Competing interests
The authors report no conflict of interest.
Supporting information
S1 Table. Tailed primers used for amplification of 325-bp region in bisulfite-converted samples.
S2 Table. Primers and probe sequences to amplify the 106-bp amplicon in a novel design of a ddPCR assay to determine the methylation fraction.
S3 Table. Primers and probe sequences to amplify the 231-bp region encompassing 31 CpG sites around the cg11625005 in a novel ddPCR assay to assess the chromatin state.
S4 Table. Primer and probe sequences for TERT expression in qPCR.
S5 Table. Overview of the methylation fraction (measured by ddPCR and NGS), mutational status and TERT mRNA expression of our sample cohort (n=61).
S6 Table. Overview of the methylation fraction (measured by ddPCR and NGS), mutational status and TERT mRNA expression and chromatin accessibility in the subset of melanoma cell lines present of our cohort (n=25).
Acknowledgements
We thank Mieke Versluis, Wim Zoutman, AG Jochemsen and Mijke Visser for useful discussions. We would like to thank Coby Out and Tim van Groningen for the assistance with cell culturing.