Charting Latency Transcripts in Kaposi's Sarcoma-Associated Herpesvirus by Whole-Genome Real-Time Quantitative PCR

Cell lines and induction conditions.

BCBL-1 cells were cultured in RPMI 1860, 25 mM HEPES (pH 7.55), 15% fetal bovine serum, 0.05 mM 2-mercaptoethanol, 1 mM sodium pyruvate, 2 mM l-glutamine, 0.05 μg of penicillin/ml, and 50 U of streptomycin/ml (all reagents from Life Technologies, Rockville, Md.) at 37°C and in 5% CO₂. Cells were split every 5 days to 2 × 10⁵ cells/ml, and a fresh aliquot was thawed after 30 passages. Where indicated, the cells were reseeded to 2 × 10⁵ cells/ml and induced 24 h later with 40 ng of TPA/ml (×10,000 stock in dimethyl sulfoxide; Calbiochem, San Diego, Calif.), 1 μM ionomycin (×2,000 stock in ethanol; Sigma, St. Louis, Mo.), or 0.8 mM sodium butyrate (×1,000 stock in ethanol; Sigma) for 24 h. Afterward the cells were collected by centrifugation (200 × g for 10 min at 4°C in a Hereaus Megafuge) and resuspended in an equal amount of fresh medium. Cells were collected at the indicated times after induction by centrifugation, washed once in phosphate-buffered saline, pelleted, and resuspended in 500 μl of RNAzol (Tel-Test, Inc., Friendswood, Tex.) per 10⁷ cells.

RNA isolation and RT.

RNA was isolated by using RNAzol according to the supplier's protocol. Poly(A) mRNA was prepared by using dT-beads (Qiagen, Inc., Valencia, Calif.) according to the manufacturer's recommendations, and 500 ng of mRNA was reverse transcribed in a 20-μl reaction with 100 U of Moloney murine leukemia virus reverse transcriptase (Life Technologies), 2 mM deoxynucleoside triphosphates, 2.5 mM MgCl₂, 1 U of RNasin (all from Applied Biosystems, Foster City, Calif.), and 0.5 μg of random hexanucleotide primers (Amersham Pharmacia Biotech, Piscataway, N.J.). The RT reaction was sequentially incubated at 42°C for 45 min, 52°C for 30 min, and 70°C for 10 min. The reaction was stopped by heating it to 95°C for 5 min, 0.5 μl of RNase H (Life Technologies) was added, and the reaction was incubated at 37°C for an additional 30 min. Afterward, the cDNA pool was diluted 25-fold with diethyl pyrocarbonate-treated distilled H₂O and stored at −80°C.

Primer design.

The predicted KSHV ORFs were extracted from the complete KSHV sequence as described by Russo et al. (60). Quantitative real-time PCR primer pairs were designed by using Prime Express 1.5 (Applied Biosystems) to comply with TaqMan criteria (5), namely, a T_m of 59 ± 2°C for each primer, a maximal T_m difference for both primers of ≤2°C, a GC content of 20 to 80%, no GC clamp, a primer length of between 9 and 40 nucleotides, fewer than four repeated G residues per primer, no hairpins with a stem size of ≥4, and a amplicon size 50 to 150 bp. Where applicable, TaqMan probes were designed to have a T_m to be greater than 10°C compared to the flanking primer pair and no G at the 5′ end. The top 100 potential primer pairs for each KSHV ORF were visually inspected, and the highest ranking primer-probe combination was selected, unless a primer-probe combination with similar characteristics but closer to the 3′ end of the ORF could be identified. In cases where the predicted ORFs overlap (orfK3 exon 1 and orf70; orf17 and orf18; orf19, orf20, and orf21; as well as spliced forms of orfK8), primers were selected outside the region of overlap. In the absence of a complete transcript map for this virus, we cannot, however, exclude the possibility that some primer combinations were located in regions in which the 3′ untranslated region (3′UTR) or the 5′UTR segments overlap. Primer-probe combinations for known spliced transcripts were designed by using relaxed criteria in order to span splice junctions. The primer sequences are available online (http://w3.ouhsc.edu/mi/frame_faculty/dittmer.html).

Real-time quantitative PCR.

The set of forward primers was synthesized on a separate plate then the reverse primers were synthesized by using a MerMade 96-well synthesizer (BioAutomation Corp., Plano, Tex.) and stored in distilled H₂O at −80°C at a concentration of 100 pmol/μl (100 μM). Primer length and purity was verified by using a P/ACE MDQ Capillary Electrophoresis System (Beckman Coulter, Inc., Fullerton, Calif.). Forward and reverse primers were combined from the master plate to yield enough primer pairs for 20 assays at a time, and their concentrations were adjusted to 1 pmol/μl (1 μM). The final PCR contained 2.5 μl of primer mix (final concentration, 166 nM), 7.5 μl of 2× SYBR PCR mix (Applied Biosystems), and 5 μl of sample. For TaqMan PCR, FAM-TAMRA-labeled probe (Applied Biosystems) was stored at 100 μM for long-term at −80°C and added to the primer mix to yield a 1 μM stock and a 166 nM final concentration. In this case the PCR contained 2.5 μl of primer-probe mix (final concentration, 166 nM), 7.5 μl of 2× TaqMan PCR mix (Applied Biosystems), and 5 μl of sample. PCR was set up in a segregated, locked, “white” room (using designated pipettes and filtered tips), in which only primers and PCR mix was handled. Here all surfaces and tools were treated with 10% bleach monthly and exposed to ceiling UV lights overnight. Designated gowns, gloves, and face masks were required for all work. Samples were prepared and added to the PCR mix in a second, “gray” room, in which no cloned KSHV plasmids were present, by using a designated set of pipettes and filtered tips. Real-time PCR was performed by using an ABI PRIZM5700 or ABI PRIZM7700 machine (Applied Biosystems) and universal cycle conditions (2 min at 50°C and 10 min at 95°C and then 40 cycles of 15 s at 95°C and 1 min at 60°C) (26). C_T values were determined by automated threshold analysis by using the ABI PRIZM software. Here, the threshold was set to five times the standard deviation (SD) of the non-template control. Dissociation curves were recorded after each run, and the amplified products were routinely analyzed by agarose gel electrophoresis.

Calculations.

In order to evaluate the significance of the real-time quantitative PCR array data, some theoretical observations seem in order. (i) Prior to reaching saturation (due to exhaustion of primers and nucleotides, loss of polymerase activity, etc.), PCR amplification proceeds exponentially and can be described by N_i = N₀ × (1 + k)ⁱ, where N₀ represents the number of molecules in the original sample and N_i is the number of mRNA molecules at cycle i (i = 0 . . . 40). During the exponential phase the amplification efficiency k (0 ≤ k ≤ 1) of a given primer pair is constant. Real-time quantitative PCR visualizes the progressive amplification at each cycle and measures the amount of product under exponential conditions, exclusively. (ii) The amount of PCR product at each cycle is quantified by the incorporation of a fluorescent dye (SYBRgreen) or the cleavage of a TaqMan oligonucleotide (26). By either method the fluorescence intensity Rn has a logarithmic dependence on fluorophor concentration, yielding Rn = log(N_i) = log[N₀ × (1 + k)ⁱ]. Real-time quantitative PCR compares two samples with the target concentrations N_a and N_b by recording the cycle numbers (C_T) for a and b at which the amplification product yields enough fluorescence to cross an operator determined threshold, T (set at five times the SD of the non-template control). Consequently, Rn_a = Rn_b and log[N_a × (1 + k)^a] = log[N_b × (1 + k)^b] or log(N_a) − log(N_b) = log_{(1 + k)}b − log_{(1 + k)}a = log_{(1 + k)}(b − a) (for i = 0, N_i = 0 = N₀ × (1 + k)⁰, i.e., N_{i = 0} = N₀). Ideally, k = 1 and (1 + k) = 2, i.e., at each cycle two reactions products are produced per target molecule. This leads to N_i = N₀ × (1 + 1)ⁱ = N₀ × 2ⁱ. Assuming that log = log₂, N_a/N_b = 2^{(b − a)}, where N_a/N_b represents the fold difference in mRNA levels of two samples, C_T = a and C_T = b. The algorithm used here was designed to identify primer pairs, which under universal cycling conditions (94°C for 30 s and 60°C for 60 s [40 cycles]) have an amplification frequency of 1.8 ≤(1 + k) ≤ 2. This is achieved by restricting the amplicon length to 100 ± 50 bp and the primer T_m to 60 ± 1°C for each primer. Hence, it is possible to extract the relative ratio of abundance in two samples based on this calculation. (iii) By comparison, solution hybridization-based DNA arrays have similar characteristics, since the color intensity ratio in a fluorescent Cy3/Cy5 DNA array also exhibits a logarithmic dependence on the amount of hybridized probe. This is simply the nature of fluorescent absorption (7). Analogous to the amplification efficiency k for PCR, a hybridization-efficiency K₀ applies to DNA arrays, which is a function of the length and base composition of the particular cDNA fragment at a given hybridization temperature. For the purpose of this study, we therefore applied similar, rank-based statistics and cluster algorithms (17) to compare the relative ratios of the mRNA levels between different samples as determined by real-time quantitative RT-PCR.

Genome scan by real-time quantitative RT-PCR.

To demonstrate that the PCR signal is a measure of the amount of reverse-transcribed mRNA rather than contaminating genomic DNA, we compared RT-positive to RT-negative cDNA pools from TPA-induced BCBL-1 cells. Equal amounts of total poly(A) mRNA (as determined by absorption spectroscopy at 260 nm) served as starting material. Figure 3A plots the unmanipulated C_T values for each KSHV-specific primer pair obtained under either condition (as mentioned above, C_T values represent the amount of target on a log₂ scale). In this case, primer pairs were sorted based on the C_T of the RT-positive reaction rather than their coordinates on the KSHV genome. For each individual primer pair, RT-negative samples contained much less target (as indicated by the higher C_T value) than the RT-positive samples. In all subsequent experiments we adjusted the amount of input cDNA so that no amplification could be observed with RT-negative samples. This demonstrates that our assay measures KSHV mRNA levels and not contaminating viral DNA.

Figure 3 also introduces the first layer of our quantitative analysis. C_T values are not a linear measure of target quantity, since they depend logarithmically on the amplification efficiency k for each primer pair, as well as the initial target copy number N₀. Although absolute copy numbers can be determined through dilution of an external standard, this is not practical for a 96-primer pair array. Nor is it necessary for the purpose of a genome scan experiment, which is—just like hybridization-based DNA arrays—concerned only with the differences between different treatments. Those were calculated as follows: dC_T = C_T (mock) − C_T (treatment) for each primer pair.

At no point in this study did we attempt to compare C_T values between two different primer pairs. However, we calculated the mean difference over all primer pairs between the RT-positive and RT-negative sample. Figure 3B shows the relative frequency of dC_T = C_T (RT^neg) − C_T (RT^pos). The dC_Ts are normally distributed with mean difference of 13.25 and an SD of ±2.66. Based on an amplification efficiency of 1.8 < (1 + k) < 2 (2^13.25 = 9,742 and 1.8^13.25 = 2,412, respectively), the fraction of contaminating DNA accounted for no more than 0.05% of the signal. Figure 3B also superimposes the normal distribution based on these parameters, which fits the experimental data. The fact that C_T values have a logarithmic dependence on the mRNA concentration is extremely beneficial for our statistical analysis, since the data are inherently variance stabilized (17, 23).

What is the measurement error and linear range in these experiments? To gain an understanding of the limitations of the real-time quantitative RT-PCR array, two kinds of replicate experiments were conducted. (i) We previously published the dilution profiles for several primer pairs (29). This emphasized the extraordinarily robustness of real-time quantitative PCR, since the C_T values were linear dependent on input cDNA over more than 4 orders of magnitude, with as little as 10³ BCBL-1 cell equivalents as input. We have obtained similar performance data for the primer pairs used here (data not shown). All data in this particular study are based upon input mRNA amounts, which were poly(dT) purified from ≥10⁴ BCBL-1 cells per primer pair (or 10⁶, i.e., 1 ml of culture per 96-well array). Hence, we were in the middle of the linear performance range of the assay. (ii) To determine the combined handling and instrument error, we performed triplicate measurements with a 22-primer-pair subset. Figure 3C compares the standard deviation (SD) values for each triplicate measurement (on the horizontal axis) to the mean C_T value. No correlation between the dC_T and the SD is evident (R < 0.4 by linear regression analysis). This establishes that the experimental error is independent of the target concentration N₀ or amplification efficiency k of a given primer pair.

Biological variation accounts for the overwhelming majority of error in most genome array applications. To determine the contribution of changes in culture conditions to our array analysis, we prepared mRNA from mock-treated BCBL-1 cells cultured for 24, 48, or 72 h and determined the abundance of KSHV mRNAs by using the real-time RT-PCR array. Initially, we normalized each RT reaction according to the total amount of input poly(A) mRNA. Under these circumstances the level of GAPDH mRNA did not change significantly over time (C_Tgapdh = 12.71 at 24 h, C_Tgapdh = 13.58 at 48 h, and C_Tgapdh = 12.05 at 72 h). Since GAPDH mRNA was also the most abundant mRNA species, we normalized all individual C_T values according to dC_Ttest = C_Ttest − C_Tgapdh. Figure 4 shows the resulting dC_T values for each of the KSHV primer pairs at indicated times in culture. As a result of this normalization, higher dC_T values represent lower target concentrations relative to the level of GAPDH mRNA. Primer pairs were sorted according to dC_T values at t = 72 h. The dC_T values at 24 and 48 h are virtually identical for every mRNA. In contrast, all values were increased after 72 h in the same culture medium, i.e., expression levels were lower due the cells entering saturation phase. Interestingly, this result was observed after we adjusted for fluctuations in GAPDH mRNA levels. This implies that the KSHV transcripts levels were depleted in the stationary phase relative to GAPDH. An exception were the mRNA levels for K3, K5, orf55, and K11 which were identical at all three time points (SD ≤ 0.5). RT-PCR was able to detect a signal for lytic KSHV mRNAs even in the absence of induction, since on average 3% of the BCBL-1 cells undergo spontaneous reactivation in culture (56). If we assume a total input of 10⁶ cells, that amounts to ≥10⁴ cells, which is well within the sensitivity of RT-PCR. To describe this result in quantitative terms, the normal distributions for the relative differences, ddC_T = dC_T(t = 72) − dC_T(t = 24, t = 48), were calculated for each primer pair and used to construct frequency histograms. The mean ddCT for 48 and 24 h for all 96 primer pairs is ddC_T = −0.30 ± 1.4 compared to ddC_T = 2.44 ± 1.6 for 72 and 24 h and ddC_T = 2.74 ± 1.3 for 72 and 48 h. The difference between the 72 h and 24 h time points was significant to P ≤ 5 × 10⁻⁷ based on a one-way analysis of variance (ANOVA), whereas the differences in KSHV mRNA levels between 48 and 24 h were insignificant. The data presented henceforth were obtained from BCBL-1 cells, which were under experimental conditions for 48 h or less. The data in Fig. 4 were used to establish the average margin of error for further studies, i.e., the minimal biologically meaningful difference, which is significant to dC_T ≥ 1.50 or 2^1.5 ≥3-fold (5 × the mean ddC_T = 0.30 for the 48- and 24-h replicate measurement of n = 96 primers).

Identification of type one latency mRNAs in the KSHV genome.

We and others previously identified a cluster of KSHV latent mRNAs, based upon in situ hybridization of KS tumor biopsies (13, 15, 54, 70). These experiments showed that the LANA/orf73, v-cyclin/orf72, and v-FLIP/orf71 mRNAs—unlike that of the latent mRNA for kaposin—were not induced upon reactivation of BCBL-1 cells. The mRNAs encoding LANA, c-cyc, and v-FLIP are coterminal, and the common promoter for these mRNAs, likewise, was resistant to TPA stimulation (29). Another latent mRNA, encoding LANA-2/vIRF-3, has recently been described, but conflicting data have been reported with regard to its expression profile (37, 58). We therefore applied the KSHV real-time quantitative PCR array to distinguish between the hypotheses (i) that LANA-1 was the only type one latent mRNA in the KSHV genome, (ii) that LANA-2 also belonged to this class, and (iii) that there might be additional type one latent messages.

First, poly(A) mRNA was isolated either from TPA-treated or mock-treated BCBL-1 cells at 48 h after induction and subjected the real-time quantitative RT-PCR. Figure 5A shows the outcome of such an experiment. In this representation dC_T values (normalized relative to GAPDH) are used as x and y coordinates and are plotted for every primer pair of the KSHV array. Hence, GAPDH marks the origin at 0,0. If two dC_T values are identical under either condition, they specify a point on the 45° axis (slope of 1). If a given mRNA is induced by TPA (lower dC_T), the point is above, and if it is inhibited, the point is below the 45° axis. orfK8.1 illustrates this projection: mock-treated BCBL-1 cells yielded a dC_T = 7.5, which is indicated on the y axis. TPA-treated BCBL-1 cells yielded a dC_T = 2.5, which is indicated on the x axis. As before, a lower dC_T corresponds to higher target concentrations. Assuming an amplification efficiency of (1 + k) = 2, the induction can be approximated to 2^{(7.5 − 2.5 = 5)} = 64-fold, which agrees with existing Northern blot and protein data (36, 52, 76). Almost every KSHV mRNA is induced at 48 h after TPA treatment, as evidenced by the location of their corresponding dC_T above the 45° line. Several primer pairs in the array are specific for the two latent mRNAs encoding LANA, v-cyclin, and v-FLIP. All fall on the 45° axis, as indicated by the slope of regression m = 0.958 with R² = 0.967. The “LANA” primer pairs are located within ORF LANA or in the LANA 5′UTR, primer pair “v-FLIP” is located within ORF v-FLIP, primer pair “cyc” is located within ORF v-cyclin, and primer pair “cyc-spl” spans the v-cyclin splice site. The different amplicons do not have identical dC_Ts, because the dC_T values depend on (i) the level of either the tricistronic 5,300-nucleotide mRNA encoding LANA, v-cyclin, and v-FLIP or the bicistronic 1,700-nucleotide mRNA encoding v-cyclin and v-FLIP, as well as (ii) the individual amplification efficiencies k for each primer pair. The dC_T values for LANA-2/vIRF-3 (two primer pairs) also fall into this group. Hence, LANA-2/vIRF-3 mRNA is another latency type one mRNA in the KSHV genome (as represented on our array). In addition to LANA and LANA-2/vIRF-3, there are other primer pairs on the 45° line, as well some primer pairs the target mRNA of which seemed inhibited by TPA (below the 45° line). However, these mRNAs increased at later time points after induction as discussed below.

To gain a better perspective of the significance of these observations, we calculated the ddC_T as dC_Tgapdh (mock) − dC_Tgapdh (TPA). The result is graphed in Fig. 5B. ddC_T for GAPDH is located at 0; a positive ddC_T indicates induction, and a negative ddC_T indicates inhibition. The ddC_T for GAPDH marks the 25th percentile of KSHV mRNAs on the array; in other words, 75% of the KSHV primer pairs are induced relative to GAPDH. For instance, orfK8.1 represents one of the most highly altered KSHV mRNAs. In contrast, each primer pair that measures the abundance of a latent mRNA has (i) a negative ddC_T, (ii) differs by <1 ddC_T unit from GAPDH, and (iii) is located in the lower 25th percentile of all data. Other mRNAs with a ddC_T value in the lower 25th percentile were the ORFs 19, 21, 40, 67, 68, K5, K9, and K4. However, their ddC_T values increased at later time points (see below). This demonstrates that LANA and LANA-2 are resistant to TPA induction, which is a novel, unique property among KSHV mRNAs.

To further corroborate this result, we diluted the input cDNA pool 10-fold and subjected it again to real-time quantitative PCR analysis. A similar picture ensued (data not shown) verifying that relative induction ratios did not depend upon total mRNA levels. We then calculated the difference ddC_T = dC_Tgapdh (mock) − dC_Tgapdh (TPA) for each mRNA. Next, ddC_T values were rank ordered and split into two groups: group 1 contained the results of primer pairs for the predicted latency type one genes (n = 8, since multiple primer pairs for the each latency type one mRNA were used), and group 2 contained all other KSHV primer pairs (n = 85). Actin, c-myc, and GAPDH were excluded. Next, the ANOVA based on ranks was calculated, which showed that both groups differed significantly (P ≤ 0.0004). This demonstrates that latency type one mRNAs are a separately regulated class of KSHV mRNAs and that both LANA and LANA-2/vIRF-3 belong to this class.

It is important to bear in mind that the magnitude of induction is dependent on the amount of mRNA in the mock control, as well as the amount in the induced sample. About 3% of BCBL-1 cells express a high enough level of lytic proteins to be detected by immunofluorescence in the absence of stimulation (56, 76). The nut-1 message is a point in case. Its levels are increased 10- to 100-fold by TPA, depending on the experimental conditions and the method of observation (28, 69, 74). Since nut-1 mRNA is also the most abundant lytic mRNA, based on absolute copy number, it is easily detected in mock-treated PEL cells (63, 75). Yet, in situ hybridization showed that it is present in only a few, lytically infected, KS and PEL tumor cells (68). As determined by real-time quantitative RT-PCR, overall nut-1 mRNA levels were as abundant as GAPDH mRNA levels and, after TPA induction, nut-1 mRNA was more abundant than GAPDH mRNA. Similarly, some early promoters might be leaky in cells even though these cells never complete the KSHV lytic cycle, resulting in a significant level of mRNA in mock-treated BSBL-1 cultures. This complicates any interpretation based solely on a single time point.

To verify that LANA-2's resistance to lytic reactivation held true over time, we performed a time course experiment. Again, BCBL-1 cells were induced by TPA, and mRNA was isolated at various times after induction and quantified by real-time RT-PCR. Figure 6 shows the outcome of the experiment. In this case, dC_T values (normalized to GAPDH) were sorted based on the 24-h-mock-treated values and plotted for each primer pair. In contrast to mock treatment (see Fig. 4), dC_T values (and the underlying mRNA levels) were clearly different as early as 24 h after TPA induction compared to the mock-treated control at 24 h. As expected, the real-time RT-PCR amplification for two cellular genes actin and c-myc (underlined in Fig. 6) yielded dC_T values which were independent of TPA (c-myc is induced within 30 min of serum stimulation in 3T3 cells [35] but returns to steady-state levels thereafter). The primer pairs that measured mRNAs that did not respond to TPA in a consistent manner were also determined (indicated by asterisks in Fig. 6). In the extreme, these were the result of a pipetting or instrument error for that particular well and time point. In other cases, they represented mRNAs that were induced with delayed kinetics or which were unusually abundant in the mock-treated sample. In contrast, the arrows in Fig. 6 denote the results of primer pairs for latency type one messages, for which all three independent datum points—24, 48, and 72 h after TPA treatment—were nearly identical. Moreover, their dC_T values were the same or higher (indicating a lower mRNA abundance) in TPA-treated BCBL-1 cells as for mock-treated BCBL-1 cells. This group of mRNAs was qualitatively and quantitatively different from the rest of the KSHV messages.

Figure 7 shows a conventional cluster analysis of the data. All latency type one mRNAs (LANA, LANA-2, v-cyc, and v-FLIP) group together, and the cellular genes actin, GAPDH, and myc are also part of this cluster. In contrast, immediate-early lytic mRNAs (orf50/Rta and 57/Mta), spliced forms for k-bZIP and gamma-2 mRNAs (orfK8.1 and orf29), are located elsewhere. The spliced v-cyclin mRNA signal is segregated from the other latency mRNAs, since its levels are higher in mock-treated than in TPA-treated samples. This effect is most pronounced at 72 h, suggesting that spliced v-cyclin mRNA might be rapidly degraded in lytically replicating cells. Cluster analysis revealed that orfK1 and orf75 also group with the latency type one mRNAs, whereas both mRNAs were clearly induced, as judged by a single 48-h time point comparison. It is conceivable that, by chance, a summation over the principal components (time points) for K1 and orf75 gives the same rank as a latent pattern (K1 is one of the mRNAs for which the clustering result of the two previous array analyses is discordant [28, 50]). This is one of the theoretical limitations of any cluster analysis. Since multiple mRNAs are transcribed across the K1 locus and are regulated differently (62), array analysis for this ORF is less reliable. By and large every single KSHV mRNA ordered according to real-time quantitative RT-PCR analysis follows the transcription pattern that was previously established by using the KSHV cDNA arrays (28, 50). This demonstrates that the mRNAs for LANA, v-cyclin, v-FLIP, and LANA-2 were not induced by TPA at any time after TPA induction. No other mRNA that can be measured with our array shared this property. This supports a model in which latency type one mRNA levels are regulated differently than the rest of the KSHV genome. They are resistant to the reactivation-induced genome-wide induction.