Gene syntax defines supercoiling-mediated transcriptional feedback

Abstract

Gene syntax-the order and arrangement of genes and their regulatory elements-shapes the dynamic coordination of both natural and synthetic gene circuits. Transcription at one locus profoundly impacts the transcription of nearby adjacent genes, but the molecular basis of this effect remains poorly understood. Here, using integrated reporter circuits in human cells, we show that supercoiling-mediated feedback regulates expression of adjacent genes in a syntax-specific manner. Using Region Capture Micro-C, we measure induction-dependent formation of supercoiled plectonemes and syntax-specific chromatin structures in human induced pluripotent stem cells. Using syntax as a design parameter, we built compact gene circuits, tuning the mean, variance, and stoichiometries of expression across diverse delivery methods and cell types. Integrating supercoiling-mediated feedback into models of gene regulation will expand our understanding of native systems and enhance the design of synthetic gene circuits.

Figure 1:

Supercoiling-mediated feedback couples transcription and genome folding of adjacent genes. a) Supercoiling modifies the energy required for polymerases to bind and locally melt DNA, leading to increased or decreased polymerase initiation. This biophysical feedback loop (purple arrows) generalizes across integration method and cell type, opening new engineering capabilities. b) Using Ensembl annotations for the human genome, gene extents were identified using the maximum extent of all annotated exons. For each pair of adjacent genes, the relative orientation and intergene spacing was computed, split into equal-sized quantile bins, and summarized by orientation. c) Due to accumulated positive supercoiling at the downstream promoter, expression from an upstream gene is predicted to decrease expression of a downstream gene. Two-gene constructs expressing fluorescent proteins from PGK promoters were integrated using PiggyBac in HEK293T cells. For a representative biological replicate, the distribution of the reporter is shown as a function of position in the circuit. In the hiPSCs, the adjacent gene is expressed from the weak PGK promoter and the reporter gene is expressed from the strong EF1a promoter.

Figure 2:

Transcription induces syntax-specific coupling of expression of adjacent genes. a) We integrated two-gene systems consisting of an dox-inducible gene (TRE) and a constitutively expressed gene (EF1a). The resulting cell lines were flow sorted to single cells and expanded as monoclonal populations. b) The geometric mean reporter expression, normalized to the uninduced condition, is shown as a function of dox concentration for the three different syntaxes. Geometric mean and associated 95% confidence interval shown over merged distributions from three wells. c)-d) Full reporter protein and mRNA distributions are shown in the uninduced case and the second-highest dox induction state. Fold change of the geometric means are shown in black. e) Dox was sequentially introduced and removed in order to measure the turn-on and turn-off dynamics of the integrated systems. The systems respond reversibly to the presence of dox. Geometric mean and associated 95% confidence interval shown over merged distributions from three wells.

Figure 3:

Transcription induces syntax-specific chromatin structures across synthetic gene circuits and the surrounding locus. a) The resulting cell lines were split into two conditions, with half induced with doxycycline (dox) for 72 hours prior to harvesting. The cells were processed following the Region Capture Micro-C protocol in order to measure population chromatin structure around the region of integration. b) - g) Region Capture Micro-C data was binned at 500 bp and 2 kb resolution and iteratively balanced within the capture region. b) For the divergent cell line, only relatively small differences differentiate the induced and uninduced conditions across the ~700 kb region around the integration site. c) For each condition, the fold change in contact probability was computed. The resulting distribution was binned by region. In the divergent case, the inter-TAD region shows reduced contact probability upon induction. d) Examining a corner dot representing a loop between the integration region and the first intron of CLYBL, the induced divergent condition shows two corner-dots, suggesting an induction-dependent formation of a double-loop structure. This double loop does not appear in the downstream tandem syntax. e) For the divergent cell line, the surrounding 10 kb region around the integration site is shown at 500 bp resolution. f) The off-diagonal score is shown across the entire capture region. No other region shows strong increases in off-diagonal score upon induction. g) Local plectoneme formation can be quantified using the off-diagonal score. As opposed to a “corner dot” structure which indicates a loop domain, plectonemes should show contacts along the matrix off-diagonal. Examining the region immediately around the integration location, this off-diagonal score remains low in the uninduced case. However, upon induction, we see a strong increase in contact probability along the central off-diagonals and as measured by the off-diagonal score.

Figure 4:

Syntax-based tuning optimizes circuit expression and biologic production without part substitution. a) The light and heavy chains are expressed from two-gene constructs integrated at the Rogi2 locus in HEK293T cells. Antibody titer, as measured via sandwich ELISA, differs across syntaxes. Points depict N=3–6 biological replicates. Statistics are two-sided student t-tests. **** : p < 0.0001 b) Two-gene circuits, consisting of a constitutive gene and an inducible gene, were lentivirally transduced into HEK293T cells alongside a rtTA-expression lentivirus. Joint distributions in the absence (gray) or presence (colored) of 1 μg/mL dox are shown. Percentages refer to the proportion of cells expressing (top) or not expressing (bottom) the inducible gene. c) The geometric mean of the inducible gene (gray) is shown as a function of dox concentration. Gray shading represents the 95% confidence interval across four biological replicates. d) The stoichiometric ratio between the inducible and constitutive genes is shown as a function of dox concentration. Strong coupling reduces the change in this ratio. Colored shading represents the 95% confidence interval across four biological replicates.

Figure 5:

Syntax augments performance of compact gene circuits across cell types. a) A dox-inducible circuit relies on expression of a synthetic activator (rtTA) to activate the TRE promoter; an “all-in-one” circuit places both the activator—co-expressed with a fluorescent protein—and the inducible gene of interest in the same cassette. The performance of the system is determined by the interplay between the biochemical coupling and the supercoiling-dependent biophysical coupling. b) Representative microscopy images showing the expression of the constitutive activator (left) and the inducible gene (right) for the circuit in a) transduced into hiPSCs and induced with 300 ng/mL dox. Scale bar represents 50 microns. c) Joint distributions of the activator and inducible gene expression are shown for each syntax. Dashed lines depict expression gates set by the untransduced population. Percentages refer to the proportion of double-positive cells in the induced case. d) Geometric mean expression of the inducible gene and constitutive activator are shown for uninduced (gray) and induced (colored) populations. e) The joint distributions of the double-positive populations for induced cells in c) can be decomposed into intrinsic (off-diagonal) and extrinsic (on-diagonal) noise. f) The intrinsic noise is shown for HEK293T cells lentivirally transduced with the cis-inducible circuit in a) or the trans-inducible circuit from fig. 4b. Noise is calculated for double-positive populations of induced cells. g) The cHS4 core or full insulator sequence was placed in the intergenic region of the all-in-one inducible circuit. The no insulator condition (none, black) is the same as in d). Geometric mean expression of the inducible gene and constitutive activator are shown for hiPSCs lentivirally transduced with these circuits and induced with 300 ng/mL dox. Points represent the mean ± standard error for n = 3 biological replicates. Statistics are two-sided student t-tests. n.s.: p > 0.05, *: p < 0.05, **: p < 0.01, ***: p < 0.001, ****: p < 0.0001