In silico design of context-responsive mammalian promoters with user-defined functionality

Abstract

Comprehensive de novo-design of complex mammalian promoters is restricted by unpredictable combinatorial interactions between constituent transcription factor regulatory elements (TFREs). In this study, we show that modular binding sites that do not function cooperatively can be identified by analyzing host cell transcription factor expression profiles, and subsequently testing cognate TFRE activities in varying homotypic and heterotypic promoter architectures. TFREs that displayed position-insensitive, additive function within a specific expression context could be rationally combined together in silico to create promoters with highly predictable activities. As TFRE order and spacing did not affect the performance of these TFRE-combinations, compositions could be specifically arranged to preclude the formation of undesirable sequence features. This facilitated simple in silico-design of promoters with context-required, user-defined functionalities. To demonstrate this, we de novo-created promoters for biopharmaceutical production in CHO cells that exhibited precisely designed activity dynamics and long-term expression-stability, without causing observable retroactive effects on cellular performance. The design process described can be utilized for applications requiring context-responsive, customizable promoter function, particularly where co-expression of synthetic TFs is not suitable. Although the synthetic promoter structure utilized does not closely resemble native mammalian architectures, our findings also provide additional support for a flexible billboard model of promoter regulation.

Figure 1.

Identification of transcriptionally active transcription factor regulatory elements (TFREs) that bind relatively abundant host-cell components. (A) RNA-seq analysis of CHO cell transcriptomes determined the relative expression level of host-cell transcription factors (TFs). Points represent the average expression level of each TF in three discrete CHO cell lines, sampled at exponential and stationary phases of culture (n = 6). Inset graph shows the expression level of 67 TFs that exhibit high (ranked in the top 30% of TF mRNA expression levels) and stable expression across different CHO cell lines and growth phases (maximum fold change < 1.5). FPKM = fragments per kilobase of transcript per million fragments mapped. (B) Cognate binding sites of TFs with appropriate expression dynamics were identified and cloned in series (6× copies) upstream of a minimal CMV core promoter in secreted alkaline phosphatase (SEAP)-reporter vectors. CHO cells were transiently transfected with each homotypic TFRE-reporter and SEAP activity was measured 24 h post-transfection. Data are expressed as a percentage of the production exhibited by the strongest homotypic promoter. Bars represent the mean ± SD of three independent experiments (n = 3, each performed in triplicate). (C) An example homotypic promoter nucleotide sequence is shown. AARE sites are in bold, 6 bp spacer sequences are underlined, and the CMV core promoter is italicized.

Figure 2.

Heterotypic assemblies of modular transcription factor regulatory element (TFRE)-blocks exhibit transcriptional activities spanning over two orders of magnitude. (A) TFREs that were transcriptionally active in homotypic architectures (see Figure 1) were combined together in varying combinations to construct libraries of heterotypic promoters. Multiple constructs were created within each library, where the order, orientation, spatial positioning and copy number of composite-TFREs was varied (the total number of discrete promoters created within each library is shown). (B) Heterotypic elements were inserted upstream of a minimal CMV core promoter in secreted alkaline phosphatase (SEAP)-reporter vectors and transiently transfected into CHO cells. SEAP expression was quantified 24 h post-transfection. Data are expressed as a percentage of the production exhibited by the strongest heterotypic promoter. SEAP production from the control hCMV-IE1-SEAP reporter is shown as the black line. Each bar represents the mean of two transfections; for each promoter, <10% variation in SEAP production was observed. qPCR analysis of SEAP transcript abundance confirmed that relative protein activities in cell culture supernatants were linearly correlated with SEAP mRNA levels (see Supplementary Figure S2).

Figure 3.

The function of modular transcription factor regulatory element (TFRE)-blocks in heterotypic promoter architectures is independent of binding site order and spacing. (A) The function of discrete TFREs within heterotypic elements can be influenced by multiple ‘rules’. We modelled heterotypic promoter activities (see Figure 2) using TFRE copy numbers as the only predictor variable. (B) The linear regression model's predictive power was analyzed using leave-one-out and five-fold cross validations (CV). (C) The relative transcriptional activity of a single copy of each modular TFRE-block within heterotypic promoters was determined by analyzing the model coefficients. TFRE regression coefficients were multiplied by 8.9 to obtain normalized TFRE activities.

Figure 4.

Synthetically designed promoters exhibit predictable activities in vitro. (A) Promoters with identical transcription factor regulatory element (TFRE)-compositions, but varying TFRE-orders, were designed in silico. The in vitro activity of each synthetic element was predicted using our model of heterotypic promoter activities (see Figure 3). (B) Synthetic promoters were chemically synthesized, inserted upstream of a minimal CMV core element in secreted alkaline phosphatase (SEAP)-reporter vectors, and transiently transfected into CHO cells. SEAP expression was quantified 24 h post-transfection. Data are expressed as a percentage of the production exhibited by the strongest in vitro-constructed heterotypic promoter, IVC1 (equivalent to 100 relative promoter units (RPU); see Figure 2). Values represent the mean ± SD of three independent experiments (n = 3, each performed in triplicate).

Figure 5.

In silico designed sequences exhibit custom-defined functionalities in vitro. (A) Millions of transcription factor regulatory element (TFRE)-combinations were constructed and tested in silico using our model of heterotypic element activities (see Figure 3). Selection criteria were applied to identify combinations optimal for the context of biopharmaceutical production in CHO cells. Constituent TFREs within each promoter were then specifically arranged to prevent occurrences of sequence features that can contribute to promoter silencing. (B) Synthetic promoters with varying designed activities were chemically synthesized, inserted upstream of a minimal CMV core element in secreted alkaline phosphatase (SEAP)-reporter vectors, and transiently transfected into CHO cells. SEAP expression was quantified 24 h post-transfection. Data are expressed as a percentage of the production exhibited by the strongest promoter. (C) CHO cells were stably transfected with synthetic promoter-reporter plasmids coexpressing a glutamine synthetase selection marker gene. Three distinct stable pools were created for each reporter-plasmid, where recombinant vectors randomly integrate into varying chromosomal locations to create heterogeneous cell populations. Following selection in medium containing methionine sulfoximine, promoter activities were measured during a 7-day batch-production process. SEAP titer and mRNA abundance were determined in mid-exponential and stationary phases of growth. Data are expressed as a percentage of the expression exhibited by the strongest promoter. (D) Stable pools were subcultured in selective medium for sixty generations. SEAP expression was quantified at the end of a 7-day batch-production process. Data are expressed as a percentage of the production exhibited by each pool at generation fifteen. Values represent the mean ± SD of three independent experiments (n = 3, each performed in triplicate).