Slow, stochastic transgene repression with properties of a timer
© Wang et al.; licensee BioMed Central Ltd. 2006
Received: 13 February 2006
Accepted: 8 May 2006
Published: 9 June 2006
When gene expression varies unpredictably between genetically identical organisms, this is sometimes ascribed as stochastic. With the prevalence of retroviral vectors, stochastic repression is often observed and can complicate the interpretation of outcomes. But it may also faithfully reflect characteristics of sites in the genome.
We created and identified several cell clones in which, within a given cell, retroviral transcription of a transgene was repressed heritably and essentially irreversibly. This repression was relatively slow; total repression in all cells took months. We observed the dynamics of repression and found that they were ergodic, that is, tending with a probability to a final state independent of previous conditions. Different positions of the transgene in the genome demonstrated different dynamics. At a position on mouse chromosome 9, repression abided by near perfect first-order kinetics and was highly reproducible, even under conditions where the number of cell generations per day varied.
We propose that such a cell division independent 'off' mechanism could play a role in endogenous gene expression, potentially providing an epigenetically based timer for extended periods.
When a single outcome is not a certainty, but instead chosen seemingly randomly from two or more possible states, that process is often termed 'stochastic'. Stochasticity is also used to explain phenotypic differences between cells of a genetically identical population. Examples include how individual cells of Escerichia coli  or yeast  in the same culture produce differing amounts of a protein; yeast express either the a or α mating type locus ; olfactory neurons each express a different odorant receptor ; mature T cells choose to express either CD4 or CD8, but not both ; B cells express one functional immunoglobulin allele while excluding the other [6, 7]. Ronai et al. have demonstrated that within a clonal population of cells, epigenetic differences at the immunoglobulin locus can lead to distinct expression states that can be inherited from generation to generation [8–10]. Also, using transgenic reporter constructs, Walters et al. studied the effect of enhancers on genetic variegation that results from slow gene repression [11, 12]. Weinberger et al.  have shown that the fluctuations in amounts of the viral protein Tat can lead to different expression states of green fluorescent protein (GFP) expressed from an HIV-based vector . They demonstrated that Tat is a decisive component in a positive feedback loop, and that stochastic and variable expression of Tat affects whether GFP is expressed at a high or low state. Such phenotypic bifurcation may also explain proviral latency .
The assumption that a mechanism is stochastic can be reasonable, and in biology many stochastic models abound [2, 14–18]. But in biology, final outcomes are also often instructed and so the issue of stochasticity is not always clear. While phenotypic outcomes might appear random, if one tallies enough events, the ensemble of events should reveal that outcomes are probabilistic. We describe a system to characterize the repression of a transgene in a mammalian cell line. Using this system, we demonstrate that slow repression can abide by first-order decay kinetics over long time periods. Here, we focus not on the fluctuation of expression due to stochasticity, but describe how predictable dynamics of repression can be determined by a stochastic decision.
Vector integration sites in clones
Gene repression as a state function
Since our aim was to study gene inactivation, cultures were not initially selected with antibiotics. Also, because we waited two days before isolating fluorescent clones, we avoided those clones where GFP inactivation was rapid (that is, occurring in less than two days). This method facilitated the study of gene repression that is observable over longer periods of time. By using flow cytometry to measure GFP fluorescence, we were able to track gene repression of a population on a cell-by-cell basis. In total, we created and tracked 93 clones (Figure 1b; Additional data file 4) that differed in GFP expression. Differences in integration sites likely account for these differences. Of the 93 clones, 45 clones were transduced with GFP and 48 clones were transduced with GFP followed by the Ig enhancer. Between those with and without the additional enhancer, we observed little discernable difference in gene inactivation. Nearly all of the GFP expression profiles had multiple peaks (for example, bi- or tri- modal) and this meant that fluorescence behaved as a state property, with decreases in fluorescence corresponding to transition from a high to low expression state (Figure 1b; Additional data file 4). Out of 93 clones, there were only three (clones 9, 11, and 13) in which the decrease in fluorescence appeared continuous, and all were decreases from an initially low level.
After 32 days in culture, 26 of 93 clones had repressed GFP expression to the point that more than 90% of the cell population produced a relative fluorescence less than 100 (Figure 1c). In contrast, 4 of 93 clones had populations where less than 10% of the cells had fluorescence less than 100. In 38 of the clones, 10% to 50% of the cells in the culture produced fluorescence less than 100.
Stochastic repression of gene transcription
After the first experiment (Figures 1b, and Figure 2b, top), cells with high fluorescence were isolated by fluorescence-activated cell sorting (FACS) and reintroduced to culture. Again, GFP was similarly repressed (Figure 2b, middle). From this second experiment, cells with high fluorescence again were re-isolated and monitored (Figure 2b, bottom), this time for a longer period. Again, the dynamics of repression were similar to the previous experiments. It appeared that the re-isolated populations were not affected by time spent in culture and had no memory of previous experiments. This property was convenient from a practical standpoint - for subsequent experiments, we could reset cultures to a common starting point where all cells expressed high, unrepressed GFP.
Heritability and irreversibility of gene repression
For clones 6 and 18, we asked whether the repressed transcription states were stably inherited. When we isolated cells with repressed GFP levels (low or no observable fluorescence) and put these cells back into culture, we saw no high GFP-expressing cells reemerge (Figure 3b). Here we had passaged the cells by splitting the cultures by a third daily (that is, for a 3 ml culture, 2 ml of fresh medium was combined with 1 ml of the old culture) for 27 days. Because cultures reached a steady-state number of cells (stationary phase) each day, the amount that the culture was split corresponded to the number of cell generations per day. By approximating the average number of generations per day to be equal to -ln(split fraction)/ln(2), we determined that repressed expression was inherited through at least 43 generations. This meant that the observed GFP dynamics were due to repression alone and not a combination of repression and de-repression.
Kinetics of gene repression
We queried whether different proliferation conditions could affect gene repression. Since we needed to maintain healthy viable cells over months, inhibitors of cell cycle to control growth rate were not an experimental option for us. Instead, cultures were 'split' (that is, diluted) daily with different fractions of new medium, ranging from a half to a sixth. For cultures that reached steady-state cell concentrations each day following medium replenishment, we calculated that splitting by a half, a third, a quarter, a fifth, and a sixth corresponded to an average of 1.0, 1.6, 2.0, 2.3, and 2.6 generations, respectively, per day. In a previous study, by staining with carboxyfluorescein diacetate succinimidyl ester (CFSE) we have established that these different culture conditions do vary the number of cell generations per day . In addition, the 18-81 cell line was well suited for these experiments because they are highly non-adherent and divide rapidly. In all conditions we studied here, the cell medium never reached a pH under 7 (as observed by phenol red indicator), exhibited no visible signs of cell stress - lack of vacuoles, consistent cell shape and membrane integrity, and so on - and viable cell staining determined the presence of less than 2% dead cells. We also noted that clone 5 cells could not divide fast enough to keep pace with a culture dilution rate of one-fifth and clone 6 could not do so when diluted by a sixth. Thus, here we have analyzed only cultures where cell numbers were steadily maintained. Although our method to vary the number of cell generations worked well, it is indirect, and our results should be interpreted with this in mind.
R-squared correlation coefficients for GFP repression data fit to zero-, first-, or second-order kinetics
Dynamics of zero order would suggest that gene repression was instructed by some factor unrelated to the cells in the culture, or that there existed some intricate quorum-sensing among cells, so as to maintain a constant repression rate despite the fact that the pool of 'available' cells (that is, with unrepressed GFP) was decreasing with time. As there was no clear case where the gene repression was closer to zero order than first order, we conclude that such active regulation of repression was unlikely.
For clones 6 and 18, the decrease in cells producing fluorescence fit rather well to first-order kinetics (Figure 4b). Clone 6 in particular demonstrated a good fit to first-order kinetics and its R-squared values (Table 2) were 0.996 or greater in all experiments. However, clones 6 and 18 behaved differently when passaged with different splitting regimens. For clone 18, the rate of fluorescence loss increased as the cultures were replenished by greater fractions of medium (Figure 4c, right). Since here greater passaging fractions necessitated a greater frequency of cell division, it is possible that clone 18 stochastically represses GFP in a cell division-dependent manner. Such a mechanism might be similar to the cell cycle-mediated silencing that has been reported in yeast . In contrast, for clone 6 the repression rate of GFP was for the most part unaffected by its splitting schedule and largely independent of number of cell generations (Figure 4c, middle). Unlike clone 18, the loss of fluorescent cells appeared to depend on time (days) alone. Because of its good fit to first-order kinetics, the rate constant for clone 6 is a direct measurement of the probability of repression. With a rate of 0.04 days-1, this meant that a cell highly expressing GFP had a 1 in 25 chance each day to become a cell with repressed GFP.
Cis-acting site dependence of repression kinetics
Methylation and histone deacetylation
We sought and found evidence that gene repression, which might appear sporadic, can lead to a highly reproducible outcome. We could demonstrate that the stochastic repression that accumulates over long periods of time can be described in terms of a probability. As much of our daily research involves the practice of cell culturing, we were impressed by the fact that a dynamic read-out could be largely independent of when and how cells were fed over a period of 80 days, and so we assign some importance to this observation. For a phenotype linked to the timed repression (or activation) of a single gene, an epigenetically based, stochastic timer might be well suited to dose the production of secreted factors, since the effective concentration of this factor depends on the collective secretion from a population of cells. Thus, it is possible that such a mechanism could schedule the dosage of hormones and drive the development of an embryo or child. Furthermore, because our study was performed using retroviral constructs, our observations may be immediately relevant to applications involving transgene expression with retroviruses, for example, retrovirus-mediated gene therapy.
In biology, the term 'stochastic' has been used with varying connotation and degrees of stringency; in all cases it means that there is an element of probability that is involved in a decision. If mice from a litter of congenic mice have different phenotypes, one might say that there is stochastic gene expression. Stochasticity has also been used to characterize natural fluctuations (up and down) in gene expression and this idea of stochasticity falls in line with the idea of an inherent level of 'noise.' Here, we have not studied 'noise', though no doubt this is also playing a role in our system. In an attempt to distinguish ourselves terminology-wise, we have tried to establish the idea of stochastic dynamics.
In conclusion, we investigated a system that behaves with near perfect first-order decay and by the inherent properties of a first-order system this strongly established that a stochastic decision was made during the gene repression. We showed that there can be locations in the genome where the repression is considerably less sensitive to, and perhaps even independent of, cell division frequency. We hope that our study will bolster the idea that in any biological system, as long as there is no instructional program, feed-back, or cellular quorum sensing, the observation of first-order gene repression is indicative of a stochastic mechanism.
Materials and methods
Constructs were based on the MoMLV retroviral vector contained in p102.21 (kindly provided by JB Lorens, Rigel Pharma, South San Francisco, USA). They contain enhanced green fluorescent protein (EGFP; Clontech, Mountatin View, CA, USA), or enhanced yellow fluorescence protein (EYFP; Clontech), followed by an internal ribosome entry site (IRES) and puromycin resistance gene. In plasmid pGFP-Eμ-IRES-Puro, EGFP was followed by the mouse Ig μ intron sequence, defined by a 1-kb Xba I fragment from the major intron. Plasmid pGFP-IRES-Puro contained no Ig enhancers.
Retroviral vectors were packaged using PhoenixEco cells (ATCC SD 3444), and 18-81 mouse cells were infected with the vectors. All cultures were grown at 37°C and 5% CO2 in RPMI 1640 media with 10% fetal calf serum. Cells expressing GFP were isolated by FACS and clones were expanded from single cells. When used for expanding clones and monitoring gene repression, culture medium contained no puromycin. When selection was necessary, the medium contained 1.5 μg/ml puromycin. To vary the number of cell generations per day, cells were 'split' by different fractions. The passaging fractions and volumes (ml) in terms of culture plus pre-warmed fresh medium are: one-half, 1.5 + 1.5; one-third, 1.0 + 2.0; one-quarter, 0.75 + 2.25; one-fifth, 0.6 + 2.4; one-sixth, 0.5 + 2.5. In experiments where the fraction is not noted, cells were split by one-third. Cultures were split at the same time each day. Under these conditions, all cultures attained a saturation concentration of approximately 3 × 106 cells/ml after one day. GFP expression was measured by flow cytometry, and for all analyses the relative fluorescence of cells without GFP was set to 3.0. Each flow cytometry measurement consisted of data from at least 5 × 103 cells. The inhibitors of gene repression, azacytidine and trichostatin A, were added to the cultures in concentrations up to 100 μM and 10 nM, respectively.
DNA from cells was isolated, and the GFP genes were PCR-amplified using Pfu polymerase (Stratagene, La Jolla, CA, USA). The PCR products were incubated with Taq polymerase to add deoxyadenosine overhangs and then cloned into pCR2.1-TOPO (Invitrogen, Carlsbad, CA, USA). The plasmids were then amplified in E. coli and sequenced.
Expression of GFP and mouse beta-actin was measured using quantitative RT-PCR. RNA was extracted from cells and cDNA produced by standard application of reverse transcriptase and random hexamer primers. PCR was performed using the ABI PRISM® 7700 machine (Applied Biosystems, Foster City, CA, USA). GFP expression was normalized to the amount of expressed mouse beta-actin. Primers and probes were as follows: GFP, 5'-CCACATGAAGCAGCACGACT-3', 5'-TGCGCTCCTGGACGTAGC-3', 5'- [6-FAM]-TTCAAGTCCGCCATGCCCGAA- [TAMRA]-3'; beta-actin, 5'-AGGTCATCACTATTGGCAACGA-3', 5'-CACTTCATGATGGAATTGA-ATGTAGTT-3', 5'- [6-FAM]-TGCCACAGGATTCCATACCCAAGAAGG- [TAMRA]-3'.
Identification of reporter integration sites
Integration sites of the vectors were PCR-amplified using Taq polymerase. The first amplification step of the PCR used a biotinylated primer homologous to the vector and a non-specific primer, degenerate at the 3' end and constant at the 5' end. Biotinylated PCR product was recovered by use of streptavidin-coupled magnetic beads (Invitrogen). The genomic DNA flanking the 3' long terminal repeat (LTR) of the provirus was then amplified from the biotinylated PCR product. This secondary PCR utilized a nested LTR-specific primer and a primer homologous to the constant section of the non-specific primer of the first PCR. PCR products were cloned into pCR2.1-TOPO, amplified in E. coli, and then sequenced. The final PCR product contained the virus-genome junction and the junction sequence was verified for each integration site.
Additional data files
The following additional data are available with the online version of this paper. Additional data file 1 is a movie of GFP fluorescence of clone 5 over 84 days. Additional data file 2 is a movie of GFP fluorescence of clone 6 over 84 days. Additional data file 3 is a movie of GFP fluorescence of clone 18 over 84 days. Additional data file 4 contains additional flow cytometry data. Expression of GFP fluorescence as profiled by flow cytometry in clones 25 to 96. Clones 25 to 48 and 73 to 96 were transduced with GFP plus an adjacent Ig enhancer. Clones 1 to 24 (Figure 1b) and 49 to 72 were transduced with GFP without an Ig enhancer (Figure 1a). X-axis, relative fluorescence; Z-axis, normalized cell number; Y-axis, clones 25 to 39, (from front to back) after 13, 22, 32, and 42 days in culture; clones 40 to 48, after 22, 32, and 42 days; clones 49 to 96, after 13, 22, and 32 days; upper left corners show the clone identification number followed by the fraction of the population with fluorescence less than 100 after 32 days.
We thank Bruce Wang (Picobella Co.) for assistance in characterizing integration sites, and Frank Stahl, Dave Reichmuth and Sridharan Raghavan for thoughtful discussion. This study was supported by a grant (AG20684) from the National Institutes of Health. C.L.W. was supported by the NIH Immunology Training Grant (NIH/NIAID T32 AI07334).
- Elowitz MB, Levine AJ, Siggia ED, Swain PS: Stochastic gene expression in a single cell. Science. 2002, 297: 1183-1186. 10.1126/science.1070919.PubMedView ArticleGoogle Scholar
- Raser JM, O'Shea EK: Control of stochasticity in eukaryotic gene expression. Science. 2004, 304: 1811-1814. 10.1126/science.1098641.PubMedPubMed CentralView ArticleGoogle Scholar
- Sussel L, Vannier D, Shore D: Epigenetic switching of transcriptional states: cis- and trans-acting factors affecting establishment of silencing at the HMR locus in Saccharomyces cerevisiae. Mol Cell Biol. 1993, 13: 3919-3928.PubMedPubMed CentralView ArticleGoogle Scholar
- Serizawa S, Miyamichi K, Nakatani H, Suzuki M, Saito M, Yoshihara Y, Sakano H: Negative feedback regulation ensures the one receptor-one olfactory neuron rule in mouse. Science. 2003, 302: 2088-2094. 10.1126/science.1089122.PubMedView ArticleGoogle Scholar
- Leung RK, Thomson K, Gallimore A, Jones E, Van den Broek M, Sierro S, Alsheikhly AR, McMichael A, Rahemtulla A: Deletion of the CD4 silencer element supports a stochastic mechanism of thymocyte lineage commitment. Nat Immunol. 2001, 2: 1167-1173. 10.1038/ni733.PubMedView ArticleGoogle Scholar
- Nadel B, Cazenave PA, Sanchez P: Murine lambda gene rearrangements: the stochastic model prevails over the ordered model. EMBO J. 1990, 9: 435-440.PubMedPubMed CentralGoogle Scholar
- Coleclough C, Perry RP, Karjalainen K, Weigert M: Aberrant rearrangements contribute significantly to the allelic exclusion of immunoglobulin gene expression. Nature. 1981, 290: 372-378. 10.1038/290372a0.PubMedView ArticleGoogle Scholar
- Ronai D, Berru M, Shulman MJ: The epigenetic stability of the locus control region-deficient IgH locus in mouse hybridoma cells is a clonally varying, heritable feature. Genetics. 2004, 167: 411-421. 10.1534/genetics.167.1.411.PubMedPubMed CentralView ArticleGoogle Scholar
- Ronai D, Berru M, Shulman MJ: Positive and negative transcriptional states of a variegating immunoglobulin heavy chain (IgH) locus are maintained by a cis-acting epigenetic mechanism. J Immunol. 2002, 169: 6919-6927.PubMedView ArticleGoogle Scholar
- Ronai D, Berru M, Shulman MJ: Variegated expression of the endogenous immunoglobulin heavy-chain gene in the absence of the intronic locus control region. Mol Cell Biol. 1999, 19: 7031-7040.PubMedPubMed CentralView ArticleGoogle Scholar
- Walters MC, Fiering S, Eidemiller J, Magis W, Groudine M, Martin DI: Enhancers increase the probability but not the level of gene expression. Proc Natl Acad Sci USA. 1995, 92: 7125-7129. 10.1073/pnas.92.15.7125.PubMedPubMed CentralView ArticleGoogle Scholar
- Walters MC, Magis W, Fiering S, Eidemiller J, Scalzo D, Groudine M, Martin DI: Transcriptional enhancers act in cis to suppress position-effect variegation. Genes Dev. 1996, 10: 185-195.PubMedView ArticleGoogle Scholar
- Weinberger LS, Burnett JC, Toettcher JE, Arkin AP, Schaffer DV: Stochastic gene expression in a lentiviral positive-feedback loop: HIV-1 Tat fluctuations drive phenotypic diversity. Cell. 2005, 122: 169-182. 10.1016/j.cell.2005.06.006.PubMedView ArticleGoogle Scholar
- Arkin A, Ross J, McAdams HH: Stochastic kinetic analysis of developmental pathway bifurcation in phage lambda-infected Escherichia coli cells. Genetics. 1998, 149: 1633-1648.PubMedPubMed CentralGoogle Scholar
- McAdams HH, Arkin A: Stochastic mechanisms in gene expression. Proc Natl Acad Sci USA. 1997, 94: 814-819. 10.1073/pnas.94.3.814.PubMedPubMed CentralView ArticleGoogle Scholar
- Carrier TA, Keasling JD: Investigating autocatalytic gene expression systems through mechanistic modeling. J Theor Biol. 1999, 201: 25-36. 10.1006/jtbi.1999.1010.PubMedView ArticleGoogle Scholar
- Kuznetsov VA, Knott GD, Bonner RF: General statistics of stochastic process of gene expression in eukaryotic cells. Genetics. 2002, 161: 1321-1332.PubMedPubMed CentralGoogle Scholar
- Cook DL, Gerber AN, Tapscott SJ: Modeling stochastic gene expression: implications for haploinsufficiency. Proc Natl Acad Sci USA. 1998, 95: 15641-15646. 10.1073/pnas.95.26.15641.PubMedPubMed CentralView ArticleGoogle Scholar
- Wabl M, Burrows PD, von Gabain A, Steinberg C: Hypermutation at the immunoglobulin heavy chain locus in a pre-B-cell line. Proc Natl Acad Sci USA. 1985, 82: 479-482. 10.1073/pnas.82.2.479.PubMedPubMed CentralView ArticleGoogle Scholar
- Peters A, Storb U: Somatic hypermutation of immunoglobulin genes is linked to transcription initiation. Immunity. 1996, 4: 57-65. 10.1016/S1074-7613(00)80298-8.PubMedView ArticleGoogle Scholar
- Wang CL, Harper RA, Wabl M: Genome-wide somatic hypermutation. Proc Natl Acad Sci USA. 2004, 101: 7352-7356. 10.1073/pnas.0402009101.PubMedPubMed CentralView ArticleGoogle Scholar
- Wang CL, Wabl M: Hypermutation rate normalized by chronological time. J Immunol. 2005, 174: 5650-5654.PubMedView ArticleGoogle Scholar
- Lau A, Blitzblau H, Bell SP: Cell-cycle control of the establishment of mating-type silencing in S. cerevisiae. Genes Dev. 2002, 16: 2935-2945. 10.1101/gad.764102.PubMedPubMed CentralView ArticleGoogle Scholar
- Fryer CJ, Archer TK: Chromatin remodelling by the glucocorticoid receptor requires the BRG1 complex. Nature. 1998, 393: 88-91. 10.1038/30032.PubMedView ArticleGoogle Scholar
- Huang M, Qian F, Hu Y, Ang C, Li Z, Wen Z: Chromatin-remodelling factor BRG1 selectively activates a subset of interferon-alpha-inducible genes. Nat Cell Biol. 2002, 4: 774-781. 10.1038/ncb855.PubMedView ArticleGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.