Intraclonal protein expression heterogeneity in recombinant CHO cells

Abstract

Therapeutic glycoproteins have played a major role in the commercial success of biotechnology in the post-genomic era. But isolating recombinant mammalian cell lines for large-scale production remains costly and time-consuming, due to substantial variation and unpredictable stability of expression amongst transfected cells, requiring extensive clone screening to identify suitable high producers. Streamlining this process is of considerable interest to industry yet the underlying phenomena are still not well understood. Here we examine an antibody-expressing Chinese hamster ovary (CHO) clone at single-cell resolution using flow cytometry and vectors, which couple light and heavy chain transcription to fluorescent markers. Expression variation has traditionally been attributed to genetic heterogeneity arising from random genomic integration of vector DNA. It follows that single cell cloning should yield a homogeneous cell population. We show, in fact, that expression in a clone can be surprisingly heterogeneous (standard deviation 50 to 70% of the mean), approaching the level of variation in mixed transfectant pools, and each antibody chain varies in tandem. Phenotypic variation is fully developed within just 18 days of cloning, yet is not entirely explained by measurement noise, cell size, or the cell cycle. By monitoring the dynamic response of subpopulations and subclones, we show that cells also undergo slow stochastic fluctuations in expression (half-life 2 to 11 generations). Non-genetic diversity may therefore play a greater role in clonal variation than previously thought. This also has unexpected implications for expression stability. Stochastic gene expression noise and selection bias lead to perturbations from steady state at the time of cloning. The resulting transient response as clones reestablish their expression distribution is not ordinarily accounted for but can contribute to declines in median expression over timescales of up to 50 days. Noise minimization may therefore be a novel strategy to reduce apparent expression instability and simplify cell line selection.

Figure 1. Expression heterogeneity in a clone.

A) Bicistronic antibody expression constructs designed to screen IgG4 kappa light chain (LC) and gamma heavy chain (HC) transcription using fluorescent reporter proteins (EGFP and EYFP) translated from the same mRNA by an attenuated encephalomyocarditis virus (EMCV) internal ribosomal entry site (aIRES). A metal-responsive promoter drives transcription (Methods). A hybrid synthetic intron situated immediately upstream of the aIRES improves efficiency of 3′ pre-mRNA processing . Features other than fluorescent proteins and immunoglobulin chains are identical between constructs. B) Bivariate distribution of reporter protein fluorescence in cells from a dual-expressing clone (5H6, cell-specific antibody secretion rate ∼2 pg/cell-day) measured by flow cytometry (main panel). Note split linear-log axes. Spectral overlap and autofluorescence were compensated using single-color controls and untransfected cells (adjacent panels, see Methods). Histogram counts on each axis in main panel are univariate distributions of EGFP and EYFP fluorescence in the clone, with coefficients of variation (CV = s.d./mean) of 0.7. R ² in main panel is for linear fit to double-positive cells (fit not shown). 10,000 events shown in each panel. Fluorescence in arbitrary units (A.U.).

Figure 2. Variation of expression with cell size and cell cycle.

A) Fluorescence and cell volume measurements on live cells (Clone 5H6) by imaging flow cytometry. Cell volume inferred by calculation from brightfield projected area (Methods). EGFP and EYFP fluorescence collected in a single channel. Linear fit (red line) and R ² are shown in main panel. Contours are percentiles (5%). Histogram counts on each axis are shown fitted to a log-normal distribution (red curves). Fluorescence in arbitrary units (A.U.). B) Cell image field (montage) illustrating subset of events from (A) with image centers aligned to match corresponding graph coordinates. False-color overlay of bright field and fluorescence channels. Scale bar, 20 µm. Note: brightness is perceived per unit projected area, not per unit volume as plotted, making larger cells appear brighter in cross-section due to depth of field. C) Fixed cells (Clone 5H6) stained with propidium iodide for DNA content and measured by conventional flow cytometry. EGFP fluorescence corrected for cell volume (estimated by FSC-A, see Methods). EYFP similar (not shown). Linear fit (red line) and R ² are shown in main panel. Contours are percentiles (5%). Histogram counts on upper axis indicate cell cycle phases (G0/G1, S, G2/M). Fluorescence in arbitrary units (A.U.).

Figure 3. Measurement noise and dynamic response of sorted subpopulations.

A) High (red) and low (blue) subpopulations (top and bottom 5% of expressing cells, respectively) were sorted from the clone (5H6) by FACS along with a control population (gray, all expressing cells, including high and low). Fluorescence in arbitrary units (A.U.). B,C) EGFP fluorescence distributions of sorted subpopulations were monitored over time (in the presence of selection). Average deviations due to cell variation () and measurement noise () in the high subpopulation are indicated by red and black arrows, respectively. Measurement noise was estimated to be ∼30% of total variation (Methods). The low subpopulation and EYFP channel yielded similar estimates (not shown). Data is from two independent sorting runs. Fluorescence in arbitrary units (A.U.). D) Median fluorescence for each subpopulation was normalized to the control and plotted as a function of time (open circles). Relaxation half times () were estimated by fitting a first order exponential decay (lines). (high sort) = 3 days (∼5 generations); (low sort) = 7 days (∼11 generations). EYFP similar (not shown).

Figure 4. Dynamic response of high-expressing subclones.

Five high-expressing subclones were isolated from parental clone 5H6 by stringent FACS sorting. First, the top fluorescing 0.05% of the double-positive population was sorted and recultured for 10 days. Then the top 0.05% of this enriched subpopulation was cloned by single cell deposition into a 96-well plate and the brightest five clonal colonies (5/26) were selected (overall stringency roughly 1 in 20,000,000). A control 96-well plate sorted from the center of the double-positive population yielded 50 clonal colonies, but none of comparable brightness to the selected subclones. Expression dynamics in the parental clone, A) and the five chosen subclones, B–F), were monitored during long-term culture (in the presence of selection) by flow cytometry (median EGFP fluorescence, closed symbols) and ELISA (cell specific antibody secretion rate, pg/cell-day, open symbols). Data is presented in terms of double positive cells, normalized to the parental clone, and fitted to a first order exponential decay (lines). See Methods for details. Horizontal gray lines indicate the level of the parental clone (normalized expression = 1). Error bars are standard errors. for subclones 5H6-GC2, -GC7, -GE5, -GF10, and -GG8 were 4 days (∼4 generations), 6 days (∼7 generations), 3 days (∼4 generations), 7 days (∼7 generations) and 2 days (∼2 generations), respectively. EYFP similar (not shown).

Figure 5. In-situ fluorescence and morphology of sorted subclones.

Fluorescence and phase contrast images of the parental clone (5H6) and subclones (5H6-GC2, -GC7, -GE5, -GF10, -GG8) in adherent culture 24 days after subcloning. Scale bars, 100 µm.