The COMET toolkit for composing customizable genetic programs in mammalian cells

Abstract

Engineering mammalian cells to carry out sophisticated and customizable genetic programs requires a toolkit of multiple orthogonal and well-characterized transcription factors (TFs). To address this need, we develop the COmposable Mammalian Elements of Transcription (COMET)-an ensemble of TFs and promoters that enable the design and tuning of gene expression to an extent not, to the best of our knowledge, previously possible. COMET currently comprises 44 activating and 12 inhibitory zinc-finger TFs and 83 cognate promoters, combined in a framework that readily accommodates new parts. This system can tune gene expression over three orders of magnitude, provides chemically inducible control of TF activity, and enables single-layer Boolean logic. We also develop a mathematical model that provides mechanistic insights into COMET performance characteristics. Altogether, COMET enables the design and construction of customizable genetic programs in mammalian cells.

Fig. 1. Investigation of COMET promoter design rules.

a The schematic shows the modular, tunable features of COMET TFs and promoters. b Five ZFa with different ZF domains all induced reporter expression (one-tailed Welch’s t-test: *p < 0.05, **p < 0.01, ***p < 0.001). c Increasing the number of ZF binding sites increased the level of gene expression in the presence of ZFa (ANOVA p < 0.001) but not without ZFa (ANOVA p = 0.24). Reporter expression increased significantly from 6 to 8 and from 8 to 10 binding sites but not on either side of this range (Tukey’s HSD test with α = 0.05). d Moving the ZF binding site array farther upstream of the TATA box reduced reporter expression (two-factor ANOVA p < 0.001), and arrays with more binding sites showed more substantial decreases in reporter expression. e Compaction of ZFa binding sites enhanced ZFa-induced reporter expression, for an equivalent number of ZF binding sites (one-tailed Welch’s t-test, p = 0.002), and across compact promoters, ZFa-induced reporter expression increased with the number of binding sites (ANOVA p < 0.001). Reporter expression increased significantly from 2 to 3, 3 to 4, 5 to 6, and 8 to 12 binding sites (Tukey’s HSD test with α = 0.05). Experiments were conducted in biologic triplicate, and data were analyzed as described in Methods. Error bars represent the S.E.M. Source data are provided in the Source Data file.

Fig. 2. A model for COMET-mediated gene regulation.

This figure summarizes the process of model development, refinement, and fitting. a The COMET model (model outputs are represented by the lines on each plot) explains experimentally observed trends (circles) for reporter expression as a function of ZFa dose and promoter features. This model uses a fitted response function for ZFa-induced gene expression (discussed in b–e) and simulates a cell population to account for variation in gene expression (Supplementary Fig. 3); lines depict the average outcome for the population. The experiment was conducted in biologic triplicate, and data were analyzed as described in Methods. Error bars represent the S.E.M. b We started with a detailed model of transcriptional activation in which reporter expression depends on TF concentration, a metric related to TF-DNA-binding affinity (w), TF-DNA-binding cooperativity (n = 1 for non-cooperative, n > 1 for cooperative), RNAPII recruitment cooperativity by each multiple-TF configuration at a promoter (ρ = 0 for non-cooperative, ρ > 0 for cooperative), and maximum promoter activation by each configuration (0 ≤ α ≤ 1). c This model yielded four types of landscapes (i–iv) under different assumptions, and two representative examples of each type are shown. COMET most closely resembles (iii). d, e A model that represents ZFa-induced reporter expression by a response function was used to fit the data in (a) (the workflow for parameter estimation is depicted in e). The terms in this concise model can be related to terms in the mechanistic model. Landscapes in (c,d) are simulations of a single cell (homogenous model), and those in (a) are simulated mean values for a heterogeneous population. The outputs of this final fitted model are represented alongside experimental data in (a). Source data are provided in the Source Data file.

Fig. 3. Characterizing an expanded panel of ZFa.

a Nineteen ZFa were paired with cognate x6-C promoters, and all significantly induced gene expression (one-tailed Welch’s t-test all p < 0.02). b ZFa-induced gene expression increased with the number of binding sites, on compact promoters, for ZF2 (ANOVA p < 0.001) and ZF3 (ANOVA p < 0.001). c Investigating the orthogonality between the 12 strongest ZFa using x6-C promoters. Abbreviations: V (Vector control, no ZFa gene), C (Control reporter, no ZF binding sites). Experiments were conducted in biologic triplicate, and data were analyzed as described in Methods. Error bars represent the S.E.M. Source data are provided in the Source Data file.

Fig. 4. Tuning transcription through ZF mutants and AD variants.

a The cartoon illustrates arginine-to-alanine (R-to-A) mutations in the ZF domain, which decrease the DNA-binding affinity. b Left: ZF mutations modulate the steepness and the maximum of the ZFa dose response profile. Circles represent experimental data and solid lines represent fitted response function models. Right: correlation between m and w parameters across mutants. The regression line is m = 7.3 × 10²w + 8.6, and the shaded region is the 95% confidence interval (one-tailed permutation test p < 0.001). c The cartoon depicts evaluated ADs. d Effects of AD on inducible reporter expression with different promoters. Gene expression varied with the choice of promoter (two-factor ANOVA p < 0.001) and choice of AD (p < 0.001), and an interaction was observed between these two variables (p < 0.001). e Combined effects of AD variants and ZF mutations were identified. Gene expression was affected by both the ZF mutations (two-factor ANOVA p < 0.001) and the AD (p < 0.001), with an interaction seen between these two variables (p < 0.001). Each mutant ZFa induced more reporter expression with VP64 than with VP16 (one-tailed Welch’s t-test, all p < 0.05) and with VPR than VP64 (one-tailed Welch’s t-test, all p < 0.01). All VPR-ZFa induced similar expression regardless of the use of a WT or mutant ZF (Tukey’s HSD test with α = 0.05). f The choice of AD affects the steepness and the maximum of the dose response. Circles represent experimental data and solid lines represent fitted response function models. g The cartoon summarizes expected trends in output gene expression that result from tuning each modular feature of the ZFa and promoters. These design choices can produce either a vertical shift or diagonal shift in response profiles with respect to the number of binding sites and the dose of ZFa. Experiments were conducted in biologic triplicate, and data were analyzed as described in Methods. Error bars depict S.E.M. Source data are provided in the Source Data file.

Fig. 5. Transcriptional inhibition.

a The schematic depicts two types of inhibitors that were evaluated. A ZF1/ZF2x6-C hybrid promoter is activated by ZF1a and inhibited by ZF1i or ZF2i. b ZFi and ZFi-DsRed differentially inhibit transcription (one-tailed Welch’s t-test: **p < 0.01, ***p < 0.001). c The cartoon summarizes the proposed conceptual model of ZFi-mediated inhibition. Within each cell, a promoter can occupy states with different configurations of ZFa and ZFi. Several example states are shown for three conditions of increasing dose or strength of inhibitor (i.e., DNA-binding affinity) relative to activator. d ZFi and ZFi-DsRed differ from standard competitive inhibitors. Predictions for competitive inhibition alone, for various promoter configurations, are depicted with solid lines (Methods). COMET inhibitors track the dotted lines, which represent fits to the dual mechanism model, except in the case of ZFi paired with x6-C, which tracks the competitive inhibition-only prediction. Each condition uses ZF1a at a dose of 40 ng. X-axes are scaled linearly from 0 to 10 ng and logarithmically above 10 ng. e Measured and predicted reporter expression were compared for a panel of ZFi mutants. Each condition uses ZF1a(RAAR) at a dose of 40 ng and the ZF1x6-C compact promoter. Experiments were conducted in biologic triplicate, and data were analyzed as described in Methods. Error bars represent the S.E.M. Source data are provided in the Source Data file.

Fig. 6. Characterization of promoter design rules in the genome.

The cartoons summarize the systems used to evaluate promoter performance characteristics across three contexts: (a) multiple plasmid transient transfection, (b) single plasmid transient transfection, and c single-copy stable integration at a genomic safe harbor locus. The promoters included here comprise 1, 3, 6, or 12 ZF1 binding sites positioned using spaced or compact architectures upstream of the YB_TATA minimal promoter driving an mKate2 reporter gene. Constitutive EBFP2 was used as a transfection control in the transient transfection context and as a marker for genomic locus activity in the stable context. Bar graphs and histograms show reporter expression for EBFP2-expressing cells. In all contexts, ZFa-induced gene expression increased with the number of binding sites on spaced and compact promoters (ANOVA p < 0.00001). To profile the range of inducible expression conferred by each promoter, stable cell lines and transiently transfected cells were characterized using two distinct sets of flow cytometry settings (voltages), each of which was independently calibrated to yield comparable absolute fluorescence units (bar graphs). Experiments were conducted in biologic triplicate, and data were analyzed as described in Methods. Error bars represent the S.E.M. Source data are provided in the Source Data file.

Fig. 7. Engineering small molecule-responsive TFs.

a The cartoon illustrates chemically responsive control of gene expression using rapamycin-inducible ZFa (RaZFa). b The effects of promoter architecture and AD on RaZFa performance were evaluated. For all RaZFa on both promoters, reporter expression was significantly higher with rapamycin than DMSO (one-tailed Welch’s t-test, all p < 0.05). Fold induction is shown above the rapamycin case for relevant conditions. c Gene expression in the absence of rapamycin was affected by VP16-FRB dose (two-factor ANOVA p < 0.001) and FKBP-ZF dose (p < 0.001), with no interaction between these variables (p = 0.14). Reporter expression after rapamycin addition was affected by VP16-FRB dose (two-factor ANOVA p < 0.001) and FKBP-ZF dose (p < 0.001) with a significant interaction between these variables (p < 0.001). d Effects of subcellular localization tags: N = nuclear, x = no localization, C = cytoplasmic. For VP64-based RaZFa, gene expression in the absence of rapamycin was affected by AD-FRB localization (two-factor ANOVA p = 0.01) and FKBP-ZF localization (p < 0.001), with no interaction between these variables (p = 0.39). For VP64-based RaZFa, gene expression after rapamycin addition was not affected by AD-FRB localization (two-factor ANOVA p = 0.26) but was affected by FKBP-ZF localization (p = 0.02), with an interaction (p = 0.001). For VPR-based RaZFa, gene expression in the absence of rapamycin was affected by AD-FRB localization (two-factor ANOVA p < 0.001) and FKBP-ZF localization (p < 0.001), with an interaction (p = 0.03). For VPR-based RaZFa, gene expression in the presence of rapamycin was affected by AD-FRB localization (two-factor ANOVA p < 0.001) but not by FKBP-ZF localization (p = 0.29), with no interaction (p > 0.05). Experiments in (c, d) use a ZF1x6-C promoter. Experiments were conducted in biologic triplicate, and data were analyzed as described in Methods. Error bars represent the S.E.M. Source data are provided in the Source Data file.

Fig. 8. Composing Boolean logic.

a The cartoon summarizes a strategy for single-layer, promoter-based logic gates with ZF-TFs. We hypothesized that AND gate promoters could be designed by using multiple repeats of a paired ZF3/ZF2 motif. Full occupancy of this promoter by both ZF2a and ZF3a mimics a fully occupied x6-C promoter, and partial occupancy (with either ZFa alone) mimics an x3-S promoter. Thus, there is a large increase in gene expression when the promoter is occupied by two types of ZFa compared to one type. b Candidate two-input AND gates were constructed using one to four repeats of paired binding sites in the promoter. AND gate behavior is considered significant if reporter expression with both ZFa is greater than the sum of reporter expression with each ZFa individually (one-tailed Welch’s t-test: *p < 0.05, **p < 0.01). c Two-input dose response for the AND gate with three repeats of paired binding sites. The landscape is shaded from green to purple to facilitate visualization in the z-axis direction. d A theoretical model of COMET AND behavior is compared with other models of transcriptional AND gates; the latter vary in whether activators have multiplicative cooperativity (ρ) and whether maximum activation (α) is equivalent for TFs individually and together (Methods). e A three-input AND gate was constructed using two repeats of a triplet binding site motif. AND gate behavior is considered significant if reporter expression with all three ZFa is greater than the sum of reporter expression with each ZFa individually, and also greater than the sum from each of the three combinations with two co-expressed ZFa and the other ZFa individually (one-tailed Welch’s t-test, **p < 0.01 for all four of these tests). f A four-input gate was constructed using the binding site arrangement shown. Experiments were conducted in biologic triplicate, and data were analyzed as described in Methods. Error bars represent the S.E.M. Source data are provided in the Source Data file.