EBV DNA is lost upon explantation of NPC tumor cells at different rates
In the absence of selection, EBV-infected epithelial cells lose the virus over time. This represents a major technical limitation to experimental EBV virology, which has plagued the establishment of EBV-positive adherent cell lines and slowed investigations into EBV-associated epithelial cell cancers. Yet, it resembles the natural course of events after explantation of EBV-positive tumor cells from EBV-associated epithelial cancers, such as NPC, and provides a general model system to study large (≥100 kbp) episome loss from mammalian cells.
Here, we employ the HONE-1 cell system to understand the mechanism of the loss of EBV episomal DNA in NPC tumor cells. Previously, viral load was measured using a single probe directed against a short segment of the episome by either DNA hybridization or PCR. These studies showed that the EBV episome was lost upon the adaptation of primary NPC tumor cells to growth in an artificial medium.24,25
The experimental design using only a single probe for the viral genome could not determine whether the viral episome was being lost in 1 piece or whether it was being lost through cumulative deletions. This could theoretically be accomplished by successive subgenomic deletions first leading to defective genomes, which are then lost only after an essential gene function becomes impaired. To decide between these 2 scenarios, we use a real-time QPCR-based EBV genome array that contains 1 primer pair for each viral orf.
Cells from cultures of HONE-1 parental and HONE-1 clone 13 cells were completely karyogramed. Supplemental Figures 1a and 1b
show karyotypes of the 2 HONE-1 cell lines. The 2 NPC cell lines were almost identical, near-tetraploid, with nearly all of the same structural abnormalities, including the same marker chromosomes. The only difference between the HONE-1 cell lines were the presence of 3 normal 16s in the parental cell line and 4 normal 16s in HONE-1 clone 13 cells, in the parental cell line, which was not seen in HONE-1 clone 13 cells and a twelfth marker in HONE-1 clone 13 cells not seen in the parental cell line. Both cell lines showed some cell-to-cell heterogeneity. Complete ISCN descriptions are shown in Supplemental Table II
To confirm prior reports on HONE-1 cells, we used samples previously prepared from cells frozen at different passage levels. Our definition of a passage indicates the number of times the cells were trypzinized and seeded into new flasks. We extracted DNA and conducted PCR using primer pairs directed against 2 different EBV genes (LMP-1 and EBNA-1). To determine whether the HONE-1 cells contained the EBV genome, we performed PCR with primers specific for LMP-1. Primers directed against GAPDH were used as control for DNA purification. In this and subsequent experiments only a single band was observed after PCR analysis attesting to the specificity of all primers ( and data not shown). Two independent cultures of parental HONE-1 cells at Passage 8 contained EBV DNA, whereas in 2 later passages, 12 and 21, the amount of the EBV DNA was below the detection limit of the assay. Based on EBER-ISH, 2% of cells were EBV positive at Passage 9 (data not shown). Of note, even though late passage parental HONE-1 and HONE-1 clone 13 no longer contain the EBV genome or any pieces of it (see later), the cells continue to grow in culture indefinitely and also are tumorigenic in nude mice (unpublished observation). This result corroborates the known phenotype of EBV episome loss in NPC explants.
Figure 1 (a) Ethidium bromide stained 2% agarose gel of the products of real-time QPCR using primers specific for GAPDH (labeled G) or EBV Lmp-1 (labeled L). The input was DNA from Hone-1 parental cells at Passage 8 (labeled p.8) from duplicate cultures, Passage (more ...)
Using primers specific for the human mitochondrial region, which were validated for forensic PCR, we amplified 2 regions of the mitochondrial genome from every passage of cells used in our experiments, sequenced the resulting PCR products and looked for evidence of single nucleotide sequence polymorphism (SNP) (Supplemental Table III
). There were no SNPs in Region 1. Region 2 showed 9 SNPs, which were the result of a double peek in 1 sample, namely clone 13 Passage 2. All other sequences were identical. This demonstrates that all the cells analyzed herein represent clonal populations that stemmed from the same donor.
Evidence for biphasic EBV episomal loss from primary NPC explant cultures
We improved upon prior studies by using real-time QPCR with primers directed against the EBNA1 coding region. This determined the total rate of episome loss, since EBNA1 is required for latent oriP-dependent replication and partitioning. Any cell that does not express EBNA-1 loses all EBV episomes, whereas even minimal EBV episomes are maintained as long as EBNA-1 protein is present. Similar overall DNA amounts were used as input. This was evidenced by measuring the amount of the cellular gene for 18S ribosomal RNA, which yielded a mean CT18S
of 19.81 with a SD of 1.92. The 95% confidence interval (95
CI) was 18.96–20.69 across n
= 20 samples. CT values generated from real-time QPCR represent a logarithmic measure of target copy number, wherein a lower CT value represents a higher level of target. A SD of 1.92 corresponds to 21.94
= 3.7-fold. To adjust for variability in DNA input levels, we normalized CT values for EBNA-1 (CTEBNA1
) to host DNA as follows: dCT = CTEBNA1
(). All reactions were conducted in duplicate. Technical replicates, which reflect the overall pipetting and detection error, differed on average by 0.89 CT units (95
CI: 0.00–1.77, n
= 10) for CT18S
. The reproducibility of duplicates was not affected by input DNA levels. This can be shown by a lack of correlation between variability and mean levels.26
The regression coefficient r2
of difference vs
. mean was 0.0338, demonstrating that we operated within the linear range of the assay. To obtain biological replicates, we determined the DNA content for HONE-1 clone 13 cells at Passage 4 and immediately thereafter at Passage 5. This yielded identical results for viral DNA levels (dCTpassage 4
= 16.12 and dCTpassage 5
= 16.67, respectively, ) and attests to the reproducibility of our DNA extraction procedure.
We used cells of 2 different NPC cultures. HONE-1 clone 13 is a unique clonal derivative of the original HONE-1 explant culture. It has been shown to retain the EBV genome in the absence of selection at early passages and without rapid loss of the EBV genome for some time.11,12
It shows a constant, slow rate of EBV genome loss (), which correlates linearly with passage number (r2
= 0.688 with slope of m
= 0.38 ± 0.15, n
= 5). By contrast, the uncloned HONE-1 cell population, representing the parental explant population, showed a biphasic rate of episomal loss (): first, a typical, rapid phase of EBV episomal loss (as previously observed27
), followed by a “slow”, more gradual rate of loss, which correlates linearly with passage number (r2
= 0.865 with a slope of m
= 0.50 ± 0.20, n
= 3). The rate of episome loss in the latter phase of parental HONE-1 cells was not significantly different from the rate of loss of HONE-1 clone13 cells. After Passage 21, the EBV DNA signal was below the limit of detection in both cultures. The biphasic behavior of the uncloned HONE-1 population is typical for primary NPC explant cultures and was our first indication of 2 pathways for EBV episome loss: the first, rapid and almost catastrophic; the second, gradual and correlated with passage number in culture.
We hypothesized that the 2 quantitative different phases of episome loss correspond to 2 qualitatively different mechanisms of episome loss and that catastrophic loss precedes gradual degradation. The parental, uncloned cell population starts out with a higher average EBV copy number as reflected in a lower dCT18S at Passage 8. This is followed by rapid episome loss as reflected in an increased dCT18S at later passages (). At around Passage 10 both HONE-1 clone 13 and parental cell populations exhibit equivalent average copy numbers, and from that point on changed at a similar rate.
These data support the idea of a finite carrying capacity for long-term episomal maintenance in latently infected cells. Although clonal variation can result in initial high variability of episome copy number (Passage < 10), once the cells become culture-adapted (Passage > 10) both NPC cultures carry the same average EBV episome number and lose the episome at a similar constant rate. We do not know the molecular mechanism behind this phenotype. One could speculate, however, that a host factor, such as a replication licensing protein, becomes rate limiting as the cells adapt from the tumor microenvironment to single cell growth in culture.
Different pathways can account for loss of episomes
We envision 2 scenarios, depicted in , to explain the loss of EBV episomes: (i) the entire episome is either propagated to the next generation or lost as a whole; (ii) intraepisomal mutation, deletion and/or recombination takes place generating 2 partial episomes. Only the part of the genome that contains all essential cis-elements is propagated to the next generation. These scenarios are not mutually exclusive.
Figure 3 Genome coverage with 75 EBV orf-specific primer pairs. Plotted is the number of matches for a given primer on the vertical axis vs. the position of the primer on the horizontal axis. Primers used are shown in gray diamonds. Here, the number of matches (more ...)
These scenarios lead to different genome configurations over time, which we measured by quantifying the level of all different EBV genes (primer pairs p1 …n, n = 75), each located within a different EBV orf in relation to each other. Each of these loci was quantified independently using real-time QPCR at Passages 0, 4, 8, 22 after explantation from the tumor.
If the entire episome is lost as a whole (), then all orfs are lost at the same rate. At each passage, each orf will be present at the same copy number within the culture. Alternatively, if the episome shrinks by recombining out nonessential orfs (), individual orfs will be lost at different rates. At each passage a given orf may or may not be present at the same copy number as all other orfs. The further any 2 orfs are apart from each other, the more likely a recombination event can occur. If EBNA, oriP and Qp were in the same location, the chance of losing a particular orf by recombination would increase linearly as its distance increases from the oriP. In EBV, however, the 3 essential elements EBNA, oriP and Qp are located in different places on the circular episome. Therefore, a linear distance relationship is not expected (). The recombination distance function for circular episomes with multiple essential elements is not trivial and was approximated as described in the Material and methods section.
Figure 2 Possible scenarios for EBV episomal loss: (a) Loss of the entire episome at once (all markers) and (b) progressive loss of nonessential regions (adjacent markers) by successive recombination events. Any cell that does not retain EBNA1, oriP and Qp loses (more ...)
An important aspect of this model is that the pattern of orf loss relative to each other is not affected if the episome replicates in between host cell divisions28
or if not. Traditionally, the rate of single gene loss is calculated by normalizing the levels for each single orf (as measured by a real-time QPCR primer pair pi
) at each passage to cellular DNA (18S RNA gene locus) dCT(pi
) = CT(pi
) − CT18s
and plotting dCT(pi
) against passage number as in . This assumed rate of loss to be constant. In contrast, in our novel array approach, we normalize CT for each primer pi
at each passage tk
to the median of all 75 primers for that sample dCT(pi
) = CT(pi
) − CT(pmedian
). This yields the relative copy number of 1 orf to all others. If an orf is recombined out or lost by deletion, its relative copy number will decrease. This is reflected by an increased dCT(pi
) value, as more cycles are needed to detect residual cells within the flask that still carry an intact episome. Importantly, this normalization method neither requires that the rate of viral episome loss be constant nor that all primers have equivalent efficiency.15,29
Our model further assumes that any fragment that is separated from the oriP is irretrievably lost, and that any time a recombination/cell division event separates EBNA-1, oriP and Qp, all viral genome fragments are lost during subsequent cell division, i.e., there is only 1 latent origin per genome.
Of note, at each passage only a fraction of cells is transferred to the next flask, such as only 10% in a “1:10 split”. This explains why pieces of EBV, which are not able to replicate in between passages, become diluted out with increasing passage number. The only exception here would be pieces that were integrated into the host chromosome genome. Since both NPC cultures lost all EBV genes at the last passage, we conclude that this was not the case.
A real-time QPCR array for EBV allows comparative gene copy number determination
We used our real-time QPCR-based microarray for EBV18,19
to experimentally distinguish between the 2 possible scenarios. shows the distribution of primers across the EBV genome. The actual primers used in the array (gray diamonds) show 100% sequence identity. Therefore, the number of matches is equal to the total primer length. Using blastN we also computed the next best match for each primer on the EBV genome, which could result in missprimed amplification products. The next best match for each primer on the EBV genome is shown in open circles. There are at least seven [95
CI: 7.82, …, 8.53 (n
= 160)] nucleotide mismatches between the correct match (gray diamonds) and the next best match (open circles) making it unlikely that under our QPCR conditions (Tm
= 62°, extension time 60 sec) a given primer would anneal anywhere else but at its cognate site. Exceptions were the BWRF1 primers, which can anneal at multiple locations in the BWRF1 repeats. Except for the internal repeat (IR) region, our array covers the entire genome evenly. The mean distance between adjacent primers was 1714 bp (95
CI: 1394, …, 2033 bp for n
= 75 primers pairs). Hence, this array has a resolution of ~2,000 bp.
We used this genome array to determine EBV genome copy number (). All HONE-1 clone 13 cell time points (Passages 2, 4, 5, 10, 19) were measured in biological duplicates, HONE-1 parental cells at time point Passage 8 were measured in biological duplicate, and Passages 12 and 21 represent single measurements. All measurements were normalized to total DNA using primers specific for GAPDH. plots dCTgapdh
for each EBV orf in HONE-1 clone 13 cells at successive passages5,10,19
on the vertical axis relative to dCTgapdh
for each EBV orf at Passage 2 on the horizontal axis.
Figure 4 (a–c) Plot of log relative levels (dCT(GAPDH)) for HONE-1 clone 13 cells passage 5, 10 and 19 on the vertical and Passage 2 on the horizontal axis. Shown is the mean of biological duplicates. (d) Tree-view representation of all CT data. Yellow (more ...)
With increasing passage number most EBV-derived signals were lost, as evidenced by an increase in dCT. Because 40 was the maximal cycle number in our QPCR protocol, the maximum value of dCT was 40-CTgapdh, which is ~8 units. At Passage 5 (), we were able to detect almost all orfs. Except for a few outliers relative abundance (dCT) correlated linearly with the abundance at Passage 2 as indicated by the red regression line. At Passage 10 (), many orfs were no longer detectable as indicated by increased dCT. At Passage 19 (), only orfs corresponding to genome positions 584 and 166807 (LMP2A), 147521, 161974 (BALF2), 149133 (BVRF2), 143664 (BXLF1), 133162 (BDLF1), 111336 (BKRF4), 89805,89476 (BLLF2, BLLF1b), 87423, 84523 (BSLF1, BSRF1), 79673 (BaRF1), 72113 (BOLF1), 63798 (BPLF1), 57556 (BFLF1), 9672 (BCRF1) were still detectable, and all others were not. This is indicated by a horizontal dCT line, which no longer correlates with CTs from Passage 2. OriP maps to 7,421–9,538, Qp to ~85,000 and EBNA-1 to ~96,000 (). Hence, within the resolution of our analysis, these essential latent loci were selectively retained in HONE-1 clone 13 cells.
To independently confirm this observation, we applied unsupervised cluster analysis to the data (). Again we found a dramatic difference between parental HONE-1 and HONE-1 clone 13 cells. The parental cell population lost the entire EBV genome at once, as indicated by the abrupt color change (yellow to blue) for almost all primers between p8 and p12. By contrast, HONE-1 clone 13 cells show a gradual loss of signal over time, corroborating the pair wise regression analyses (). Some genes were lost between p2 and p6. Others (middle section) were lost between p5 and p19. Signal fluctuation in between passages did occur. However, most of these fluctuations were within the margin of error and were only overemphasized in this particular representation, which is based on raw data rather than averages (). BWRF1, BYRF1 and BLRF3 signals were uniformly positive and therefore not used for comparison. These results confirm that HONE-1 clone 13 cells behave differently than the uncloned, parental HONE-1 population. The data suggests that HONE-1 clone 13 cells lose the viral episome by gradual fragmentation, resulting in defective episomes that retain the essential cis-elements longer than orfs with no known role in episome maintenance.
HONE-1 clone 13 cells lose the EBV episome through successive deletions and recombination
To explore further the scenario of accumulative deletion/recombination (), we tested the hypothesis that adjacent orfs are retained or lost together during passage in culture. To do so, we established a statistical model based upon the relationship between orf map position and relative copy number of individual orfs for each passage. Using HONE-1 clone 13 cell Passages 2, 5, 10 and 19, we calculated the differences in signal (ddCT) between 1 orf and its immediate neighbor (+1), or its second (+2) and third (+3) adjacent orf as such: ddCT(pi) = dCT(pi+1) − dCT(pi), ddCT(pi) = dCT(pi+2) − dCT(pi), ddCT(pi) = dCT(pi+3) − dCT(pi). If 2 orfs within a given pair are present, signal difference ddCT(pi) is minimal for cells of the same passage number. If 1 orf within a given pair is lost then the signal difference ddCT(pi) is maximal. The further apart 2 orfs are, the more likely it is that a random recombination/deletion event removes 1 orf, but not the other. This is the principle of gene mapping by recombination frequency. By contrast, if the entire episome is only lost or maintained as a whole, then the signal difference ddCT(pi) between any 2 orfs will remain constant regardless of their relative location on the viral episome.
We computed Z scores of these ddCT(pi), which allowed us to evaluate relative orf retention at different passages independent of the total copy number. We established pairwise correlations for all possible combinations (data not shown). Except at Passage 19, where most of the orf signals were lost, significant overall correlations were evident, as even if individual orfs are lost by deletion, the majority of orfs remain linked to each other on the episome and therefore are present at similar levels. However, the further apart (i + 1, 2, 3) any 2 orfs are located on the EBV episome, the less their relative levels correlate with each other ().
The squared Pearson correlation coefficients comparing Passage 2 (i + 1) standardized ddCTs to all other passages exemplify this approach (). For the same input DNA, there exists a perfect correlation (r2 = 1.00) of all primer pair distances to itself.
The i + 1 distances (immediately adjacent markers) were significantly correlated at Passage 5 (r2 = 0.54) to Passage 2 and Passage 10 (r2 = 0.43) to Passage 2, but not at Passage 19 (r2 = 0.1), since at this point most of the episomes were lost. If viral episomes can only be lost as a whole, we would expect similar correlation coefficients for more distant markers (i + 2) and (i + 3) as well. This, however, was not the case.
There was less of a correlation when comparing i + 1 and i + 2 differences at Passage 2 (r2 = 0.27), i.e., for the same DNA than between different Passages 2 and 5 (r2 = 0.54) for immediate neighbors (i + 1). Even less of a correlation was observed comparing i + 1 to i + 3 differences (rightmost group). This gradual decrease in correlation would not be expected if the entire episome were lost at the same time (Model A). This gradual decrease in correlations would also not be expected for completely random data as shown in . Only a piecemeal loss of episomal genetic material is consistent with this result, which proves that more distantly located orfs are more likely to be separated by recombination or deletion.
The parental cells lost all orf signals at once and thus there were no significant correlations (data not shown). Hence, the HONE-1 clone 13 phenotype is unlikely to be due to a systematic bias, but reflects the underlying biological process of episomal loss.