Skip to main content

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

Quantifying cellular capacity identifies gene expression designs with reduced burden

This article has been updated

Abstract

Heterologous gene expression can be a significant burden for cells. Here we describe an in vivo monitor that tracks changes in the capacity of Escherichia coli in real time and can be used to assay the burden imposed by synthetic constructs and their parts. We identify construct designs with reduced burden that predictably outperformed less efficient designs, despite having equivalent output.

This is a preview of subscription content

Access options

Buy article

Get time limited or full article access on ReadCube.

$32.00

All prices are NET prices.

Figure 1: The capacity available for E. coli gene expression can be indirectly measured to quantify burden.
Figure 2: Relationships among capacity, growth rate and construct output for a combinatorial library of constructs expressing VioB-mCherry fusion protein.

Change history

  • 25 June 2015

    In the version of the Supplementary Software originally posted online, some files needed to run the software were missing. These files have been included as of 25 June 2015.

References

  1. Heinemann, M. & Panke, S. Bioinformatics 22, 2790–2799 (2006).

    CAS  Article  Google Scholar 

  2. Arpino, J.A. et al. Microbiology 159, 1236–1253 (2013).

    CAS  Article  Google Scholar 

  3. Brophy, J.A. & Voigt, C.A. Nat. Methods 11, 508–520 (2014).

    CAS  Article  Google Scholar 

  4. Hoffmann, F. & Rinas, U. Adv. Biochem. Eng. Biotechnol. 89, 73–92 (2004).

    CAS  PubMed  Google Scholar 

  5. Shachrai, I., Zaslaver, A., Alon, U. & Dekel, E. Mol. Cell 38, 758–767 (2010).

    CAS  Article  Google Scholar 

  6. Gorochowski, T.E., van den Berg, E., Kerkman, R., Roubos, J.A. & Bovenberg, R.A. ACS Synth. Biol. 3, 129–139 (2014).

    CAS  Article  Google Scholar 

  7. Cardinale, S. & Arkin, A.P. Biotechnol. J. 7, 856–866 (2012).

    CAS  Article  Google Scholar 

  8. Sleight, S.C. & Sauro, H.M. ACS Synth. Biol. 2, 519–528 (2013).

    CAS  Article  Google Scholar 

  9. Moser, F. et al. ACS Synth. Biol. 1, 555–564 (2012).

    CAS  Article  Google Scholar 

  10. Tan, C., Marguet, P. & You, L. Nat. Chem. Biol. 5, 842–848 (2009).

    CAS  Article  Google Scholar 

  11. Cardinale, S., Joachimiak, M.P. & Arkin, A.P. Cell Rep. 4, 231–237 (2013).

    CAS  Article  Google Scholar 

  12. Siegal-Gaskins, D., Tuza, Z.A., Kim, J., Noireaux, V. & Murray, R.M. ACS Synth. Biol. 3, 416–425 (2014).

    CAS  Article  Google Scholar 

  13. Klumpp, S., Zhang, Z. & Hwa, T. Cell 139, 1366–1375 (2009).

    Article  Google Scholar 

  14. Scott, M., Gunderson, C.W., Mateescu, E.M., Zhang, Z. & Hwa, T. Science 330, 1099–1102 (2010).

    CAS  Article  Google Scholar 

  15. Carrera, J., Rodrigo, G., Singh, V., Kirov, B. & Jaramillo, A. Biotechnol. J. 6, 773–783 (2011).

    CAS  Article  Google Scholar 

  16. Gerosa, L., Kochanowski, K., Heinemann, M. & Sauer, U. Mol. Syst. Biol. 9, 658 (2013).

    Article  Google Scholar 

  17. Vind, J., Sorensen, M.A., Rasmussen, M.D. & Pedersen, S. J. Mol. Biol. 231, 678–688 (1993).

    CAS  Article  Google Scholar 

  18. Algar, R., Ellis, T. & Stan, G.-B. In Proc. 53rd IEEE Conference on Decision and Control (IEEE, 2014).

  19. Hersch, S.J., Elgamal, S., Katz, A., Ibba, M. & Navarre, W.W. J. Biol. Chem. 289, 28160–28171 (2014).

    CAS  Article  Google Scholar 

  20. Chappell, J. et al. Biotechnol. J. 8, 1379–1395 (2013).

    CAS  Article  Google Scholar 

  21. Daniel, R., Rubens, J.R., Sarpeshkar, R. & Lu, T.K. Nature 497, 619–623 (2013).

    CAS  Article  Google Scholar 

  22. Moon, T.S., Lou, C., Tamsir, A., Stanton, B.C. & Voigt, C.A. Nature 491, 249–253 (2012).

    CAS  Article  Google Scholar 

  23. Farzadfard, F. & Lu, T.K. Science 346, 1256272 (2014).

    Article  Google Scholar 

  24. Gardner, T.S., Cantor, C.R. & Collins, J.J. Nature 403, 339–342 (2000).

    CAS  Article  Google Scholar 

  25. Haldimann, A. & Wanner, B.L. J. Bacteriol. 183, 6384–6393 (2001).

    CAS  Article  Google Scholar 

  26. Pédelacq, J.D., Cabantous, S., Tran, T., Terwilliger, T.C. & Waldo, G.S. Nat. Biotechnol. 24, 79–88 (2006).

    Article  Google Scholar 

  27. Shetty, R.P., Endy, D. & Knight, T.F. Jr. J. Biol. Eng. 2, 5 (2008).

    Article  Google Scholar 

  28. Canton, B., Labno, A. & Endy, D. Nat. Biotechnol. 26, 787–793 (2008).

    CAS  Article  Google Scholar 

  29. Gibson, D.G. et al. Nat. Methods 6, 343–345 (2009).

    CAS  Article  Google Scholar 

  30. Lynch, S.A. & Gallivan, J.P. Nucleic Acids Res. 37, 184–192 (2009).

    CAS  Article  Google Scholar 

  31. Caspi, R. et al. Nucleic Acids Res. 42, D459–D471 (2014).

    CAS  Article  Google Scholar 

  32. Arai, R., Ueda, H., Kitayama, A., Kamiya, N. & Nagamune, T. Protein Eng. 14, 529–532 (2001).

    CAS  Article  Google Scholar 

  33. Salis, H.M. Methods Enzymol. 498, 19–42 (2011).

    CAS  Article  Google Scholar 

  34. Quan, J. & Tian, J. Nat. Protoc. 6, 242–251 (2011).

    CAS  Article  Google Scholar 

  35. Welch, M. et al. PLoS ONE 4, e7002 (2009).

    Article  Google Scholar 

  36. Plotkin, J.B. & Kudla, G. Nat. Rev. Genet. 12, 32–42 (2011).

    CAS  Article  Google Scholar 

  37. Mitarai, N., Sneppen, K. & Pedersen, S. J. Mol. Biol. 382, 236–245 (2008).

    CAS  Article  Google Scholar 

  38. Bonomo, J. & Gill, R. Biotechnol. Bioeng. 90, 116–126 (2005).

    CAS  Article  Google Scholar 

  39. Li, G.-W., Oh, E. & Weissman, J. Nature 484, 538–541 (2012).

    CAS  Article  Google Scholar 

  40. Li, G.-W., Burkhardt, D., Gross, C. & Weissman, J. Cell 157, 624–635 (2014).

    CAS  Article  Google Scholar 

  41. Navon, S. & Pilpel, Y. Genome Biol. 12, R12 (2011).

    CAS  Article  Google Scholar 

  42. Goodman, D.B., Church, G.M. & Kosuri, S. Science 342, 475–479 (2013).

    CAS  Article  Google Scholar 

  43. Grote, A. et al. Nucleic Acids Res. 33, W526–W531 (2005).

    CAS  Article  Google Scholar 

  44. Puigbo, P., Guzman, E., Romeu, A. & Garcia-Vallve, S. Nucleic Acids Res. 35, W126–W131 (2007).

    Article  Google Scholar 

Download references

Acknowledgements

The authors thank J.J. Collins, M. Scott, D. Goodman, A. Wipat, D. Siegal-Gaskins, J. Lucks, J. Chappell, C. Hirst, K. Royle and colleagues at the Centre for Synthetic Biology and Innovation for thoughts and advice during this project. This work was supported by grants from the UK Engineering and Physical Research Council (EP/G036004/1, EP/J021849/1 and EP/J02175X/1).

Author information

Authors and Affiliations

Authors

Contributions

F.C., R.A., G.-B.S. and T.E. designed the research; F.C. and R.A. performed the experiments; R.A. and G.-B.S. developed and performed model simulations; F.C., G.-B.S. and T.E. analyzed data; F.C., G.-B.S. and T.E. wrote the paper.

Corresponding author

Correspondence to Tom Ellis.

Ethics declarations

Competing interests

The authors declare no competing financial interests.

Integrated supplementary information

Supplementary Figure 1 Design of the capacity monitor synthetic cassette and schematic of its integration into the E. coli genome.

(a) The synthesized capacity monitor cassette consists of a designed synthetic constitutive promoter (BBa_J23100), a designed synthetic RBS, a codon-optimized superfolder GFP sequence (sGFP) and a designed synthetic terminator part (BBa_B1002). (b) For genomic integration into E. coli, the synthesized cassette is inserted into integration plasmids that target either the λ or the φ80 genomic loci. (c) The capacity monitor cassette and the downstream KanR selectable marker integrate into the λ locus of the E. coli genome via site-specific recombination with the recombinase provided in trans by helper plasmid pINT-ts.

Supplementary Figure 2 Relationship between capacity and growth rate during exponential growth.

Data from Figure 1a are shown for DH10B cells with the integrated capacity monitor containing no plasmid (DH10B::GFP), containing the uninduced Lux operon plasmid (+ Lux Uninduced), and containing the Lux operon plasmid induced after 120 min of growth (+ Lux Induced 120 m). The mean capacity (a), OD600 (b) and growth rate (c) for each are compared over 270 min of an exponential-growth cycle in M9 fructose media in a shaking microwell plate. Induction at 120 min is indicated by thick red arrows. Measurements at 1 and 2 h after induction are indicated by dashed lines and are equivalent to measurements taken in the burden assay for Figure 1c and Supplementary Figure 3. Error bars represent the standard deviation of three technical repeats. Scatter plots of capacity versus growth rate of DH10B::GFP (d), Lux uninduced (e) and Lux induced 120 m (f) were determined using the mean values from a and c. During the exponential phase, all uninduced cells exhibited a linear relationship between capacity and growth rate, with both steadily increasing over time, in agreement with previous work13 (time is denoted by the arrow direction). This relationship was disturbed by entry into stationary phase (blue lines) or induction of expression (red lines). Entry into stationary phase caused a decrease in growth rate that preceded a change in capacity. Conversely, induction of expression (thick red arrows) caused a decrease in capacity that preceded a change in growth rate. A return to a linear relationship between growth and capacity was not seen until 2 h after induction of Lux operon expression, and the decrease in capacity during the first hour after induction determined the reset point for the linear relationship.

Supplementary Figure 3 Extended data from Figure 1c.

Burden imposed on MG1655 and DH10B E. coli by the different parts of the Lux, Luc and sCG inducible expression constructs and by three different plasmid backbones. Measured capacity 1 h after induction and growth rates 1 and 2 h after induction are shown. Green bars indicate L-arabinose–induced samples, and gray bars represent uninduced samples. Error bars represent the standard error of three biological repeats. Plasmids used here were pSB1A2 (high-copy pUC origin AmpR), pSB1C3 (high-copy pUC origin CmR) and pSB4C5 (low-copy pSC101 origin CmR). Constructs were all on pSB1C3 plasmids.

Supplementary Figure 4 Burden imposed by alternative expression constructs.

(a) The burden of the alternatives to AraC-pBAD that gave expression after induction was assayed by measuring the capacity and growth of TetR, LuxR and LacI inducer constructs in DH10B and MG1655 cells with the capacity monitor. Results are discussed in Supplementary Note 1. No gene or operon was placed downstream of the constructs, so measurements gave the burden of the induction system with (green bars) or without (gray bars) inducer only. Plasmid backbones used were pSB1A2 (high-copy pUC origin AmpR) and pSB1C3 (high-copy pUC origin CmR). Promoters used to express the regulating transcription factors were pLac, pTet and p105 (synthetic promoter BBa_J23105). In their respective constructs, these acted as constitutive promoters, with the promoter downstream of the regulator (pTet, pLux and pLac) being regulated. Graphs show average measured capacity 1 h after induction. Error bars show standard deviation of three independent repeats. (b) As discussed in Supplementary Note 2, the TetR-pTet inducer system was compared to an RNA-only equivalent, a theophylline-activated riboswitch30, in DH10B and MG1655 cells with the capacity monitor (top panel) to assess the burden of RNA-based and protein-based regulation. For pLac-Riboswitch and pLac-TetR-pTet, no gene was placed downstream, so measurements gave the burden of the induction system with (green bars) or without (gray) inducer only. For pLac-Riboswitch-mRFP and pLac-TetR-pTet-mRFP, the mRFP gene was placed downstream, allowing for determination of construct output and confirmation that the induction system worked (bottom panel). Graphs show average measured capacity (GFP production rate) and output (RFP production rate) 1 h after induction. Error bars show standard deviation of three independent repeats. Plasmid backbone used was pSB1A2 (high-copy pUC origin AmpR). Underlined text highlights functionally equivalent regulation units.

Supplementary Figure 5 The combinatorial library for different levels of inducible expression of a VioB-mCherry fusion protein.

(a) Schematic of the construct. In the absence of L-arabinose, AraC represses VioB-mCherry production by binding to the pBAD promoter; when L-arabinose is present, it binds to AraC, making it an activator of gene expression from pBAD. The components of the construct can be changed to tune gene expression. Two versions of the pBAD promoter were used, one with a wild-type core promoter sequence (wt pBad) and one with point mutations that increased output by approximately 1.5-fold (stronger pBad). The RBS region was varied using two synthetic 5’ UTR sequences designed by the RBS Calculator to give strong and weak translation initiation rates. Codon optimization was varied by the use of either an optimized version of the VioB-mCherry coding sequence (CDS) synthesized for efficient E. coli expression or a de-optimized version that introduced a potential translation bottleneck19,37 (see Online Methods). (b) Schematic of the library. Construct copy number was varied using two different chloramphenicol-selected plasmid backbones, one with a high-copy pUC19-derived pMB1 origin of replication (pSB1C3) and the other with a medium-copy p15A replication origin (pLys).

Supplementary Figure 6 The capacity, growth rate and construct output produced by the VioB-mCherry combinatorial library 1 hour after induction in MG1655 cells in exponential growth.

Bold text indicates the use of strong core promoter and RBS sequences, and regular text indicates weaker versions. Colored bars indicate induced samples, and gray bars represent uninduced samples. Capacity, output and growth rate were measured as for Figure 2a. Error bars represent the standard error of three biological repeats.

Supplementary Figure 7 Comparison between two strong alternative RBSs.

Comparison of construct output and DH10B capacity and growth rates 1 h after induction are shown for VioB-mCherry constructs with strong RBSs (H1, H2, M1, M2 and H5) and modified versions of these constructs (H1*, H2*, M1*, M2* and H5*) with a different VioB-mCherry 5’ UTR sequence encoding an alternative strong RBS sequence.

Supplementary Figure 8 Simulated output and capacity at steady state determined by a translational resource model.

mRNA numbers and RBS strengths were varied to match the relative changes between experimental constructs H1–H4 and M1–M4 (Fig. 2a).

Supplementary Figure 9 Simulated output and capacity at steady state as determined by a translational resource model.

mRNA numbers and RBS strengths were varied to match the relative changes between experimental constructs H1–H4 and M1–M4. Shown is the simulated effect of varying RBS strengths for two versions of a synthetic construct, one with an optimized coding sequence (CDS) and one with a de-optimized CDS. At weak RBS strengths, CDS optimization had no effect on output, as it was not limiting (see also Supplementary Fig. 10). At strong RBS strengths, only the optimized CDS was able to attain greater output, whereas with a de-optimized CDS less output was seen for a similar loss in capacity, matching experimental observations presented in Figure 2a (compare H2 and H5 for strong RBS, and H4 and H6 for weak RBS). The output from the simulated construct is shown in red shades with the right axis, and the simulated output of a capacity monitor is shown in green shades with the left axis. Simulation settings were the same as described for Figure 2d, with 400 mRNAs used for the construct. The simulated elongation rate throughout the optimized CDS was set to 1 for all 100 translation elongation steps. For the de-optimized CDS, the simulated translation elongation rate was set to 1 for steps 1–79 and 90–100 but to 0.5 for steps 80–89, mimicking a translation bottleneck due to rare codons or another sequence-specific effect.

Supplementary Figure 10 Heat map of simulated construct outputs and simulated expression efficiency for long CDS.

Heat map of simulated construct outputs for codon-optimized (a) and de-optimized (b) long CDS constructs alongside heat maps for codon-optimized (c) and de-optimized (d) constructs. Locations of simulations mapping to all constructs assayed in Figure 2a are shown. Simulated circuit output is the simulated production rate of the circuit protein at steady state. Simulated expression efficiency was calculated as the product of circuit output and the number of free ribosomes available at steady state.

Supplementary Figure 11 Expression analysis of H3 and M1 constructs after overnight induction.

(a) Optical density of DH10B and MG1655 E. coli hosting the H3 and M1 expression constructs (Fig. 2e) when grown under induction conditions for 24 h in shake-flasks. Growth was performed at 37 °C in 50 ml media in 500-ml baffled flasks with orbital shaking. 200 μl of culture was removed every 2 h between hours 16 and 24 after induction, and optical density was determined by measuring OD600 while simultaneously measuring RFP for Figure 2e. Error bars represent the standard error of three independent repeat experiments done on consecutive days. (b) Flow cytometry analysis of the GFP (x-axis: FL1-H) and VioB-mCherry (RFP, y-axis: FL2-A) content of DH10B and MG1655 E. coli hosting the H3 and M1 expression constructs 16 h after induction (see Fig. 2e). Results are shown for three independent experiments done on consecutive days. Populations where escape mutants no longer expressed VioB-mCherry are highlighted in red.

Supplementary Figure 12 Analysis of escape mutant populations following 24 h of growth in the presence of induction.

H3 and M1 constructs in capacity monitor–containing MG1655 and DH10B cells were grown with shaking at 37 °C from individual colonies for 24 h in 5 ml of supplemented M9 with 0.4% fructose, 0.2% L-arabinose, 34 μg/ml chloramphenicol and L-arabinose inducer. For H3 in each strain, 5 colonies were grown, and for M1 in each strain, 30 colonies were grown. (a) The number of colonies that reached high density after 24 h of growth and the fluorescence of these. (b) For each strain, two H3 and four M1 samples that reached high density were plasmid-prepped, and the plasmids were sent for DNA sequencing to determine construct integrity.

Supplementary Figure 13 Comparison of simulation predictions and actual obtained results for the capacity and output of scFv constructs expressed in E. coli DH10B cells with the capacity monitor.

See Supplementary Note 4 for a discussion of these results. (a) The model was used to simulate the steady-state burden of expressing a coding sequence (CDS) shorter than the long CDS presented in Figure 2d. Heat maps for the design space for short CDS constructs are shown in Supplementary Figure 14. The effects of not having codon optimization and switching to a weaker RBS were simulated and compared to the existing simulation results (from Supplementary Figs. 8 and 10) performed for equivalent long-CDS constructs. “Un-optimized” describes a CDS with some poor codons interspersed throughout the sequence. “De-optimized” describes intentional introduction of many poor codons at one point in the sequence in order to create a potential translational bottleneck. (b) The capacity and estimated construct output produced by short-CDS (846 bp) scFv constructs F1, F2 and F5 1 h after induction in DH10B cells in exponential growth. Existing data from Figure 2a for equivalent long-CDS VioB-mCherry constructs H2, H5 and H4 are also shown for comparison. Bold text indicates the use of strong RBS sequences. Colored bars indicate induced samples, and gray bars indicate uninduced samples. Capacity and growth rate were measured as for Figure 2a. scFv output was estimated from the band intensity of the western blot shown in Supplementary Figure 15 (n = 1). Error bars represent the standard deviation of three independent repeats

Supplementary Figure 14 Heat map of simulated construct outputs and simulated expression efficiency for short CDS.

Heat map of simulated construct outputs for codon-optimized (a) and un-optimized (b) short CDS constructs alongside heat maps of simulated expression efficiency for codon-optimized (c) and un-optimized (d) constructs. Locations of simulations mapping to all constructs assayed in Supplementary Figure 13 are shown. “Simulated circuit output” is the simulated production rate of the circuit protein at steady state. “Simulated expression efficiency” was calculated as the product of the circuit output and the number of free ribosomes available at steady state.

Supplementary Figure 15 Construct output, capacity and growth rate of scFv constructs with alternative optimized sequences.

(a) Western blot of DH10B E. coli with capacity monitor expressing constructs F1, F2, F3, F4 and F5 with (+) and without (-) construct induction (n = 1). Proteins were harvested from cell culture 4 h after induction. Blot shows expression of His-tagged scFv constructs (32 kDa) as well as the antibody positive control (Ab Ctrl) and 10–170-kDa size marker (M) separated by SDS-PAGE (12%). scFv expression was estimated by subtracting the band intensity compared from the surrounding background intensity using ImageJ. Intensity measurements are given on the gel in black text below each band. Size marker molecular weights (kDa) are given in white text. (b) Capacity and growth rate produced by F1, F2, F3 and F4 constructs with different codon optimization 1 h after induction in DH10B cells in exponential growth. Output as estimated in a is also shown. This figure is discussed in Supplementary Note 5. Bold text indicates the use of strong RBS sequences. Green bars indicate induced samples, and gray bars indicate uninduced samples. Capacity and growth rate were measured as for Figure 2a. Error bars show the standard deviation of three independent repeats.

Supplementary information

Supplementary Text and Figures

Supplementary Figures 1–15, Supplementary Tables 1–6 and Supplementary Notes 1–5 (PDF 4672 kb)

Supplementary Software

Supplementary Software (ZIP 53 kb)

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Ceroni, F., Algar, R., Stan, GB. et al. Quantifying cellular capacity identifies gene expression designs with reduced burden. Nat Methods 12, 415–418 (2015). https://doi.org/10.1038/nmeth.3339

Download citation

  • Received:

  • Accepted:

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1038/nmeth.3339

Further reading

Search

Quick links

Nature Briefing

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing