Thread by @GallowayLabMIT

How does transcription of proximal genes influence the expression of their neighbors? How do we engineer compact circuit designs for specific profiles of expression that enable robust cellular engineering? Today @CellReports we provide some answers (1 /n) tinyurl.com/24vtfbyw

We integrated DNA biophysics with classic #synbio and stochastic simulations to explore how to optimize the expression levels, variance, and dynamic response of transgenic systems. We found that supercoiling-mediated feedback significantly influences these behaviors! (2/n)

As RNAP transcribes, it generates a wave of overwound DNA (+ supercoiling) ahead & leaves a wake of underwound DNA (- supercoiling). Accumulated supercoiling both reshapes the RNAP binding energy landscape and can stall polymerases, creating supercoiling-mediated feedback! ( 3/n)

What’s special about supercoiling-mediated feedback? It is FAST, really fast! Because this feedback operates at the speed of supercoiling generation, we have very rapid feedback that impacts subsequent RNAP binding and transcriptional events, coupling adjacent genes. ( 4/n)

BUT *how genes are coupled* depends on the specific syntax with which genes are encoded. What’s syntax? Syntax is the relative order and orientation of genes (convergent, divergent, etc). Both order and orientation affect expression level, variance, and dynamics. (5/n)

Additionally, expression context matters bc because it sets the boundary conditions. For plasmids, supercoiling radiates around the circle, potentially canceling out. Integrated constructs lack this propagation. This results in different syntax-specific expression profiles.(6/n)

What does supercoiling accumulation look like for different syntaxes? Positive supercoiling accumulates between convergent genes, reducing RNAP loading (yellow), whereas negative supercoiling accumulates between divergent genes, facilitating RNAP loading at both promoters. (7/n)

Can we just introduce spacing to negate the effects of supercoiling? Not if you want truly compact designs. Looking at inter-gene spacing from 500 bp to 10 kb, we observe extremely small effects from spacing (note the flatness of the colored traces over the range). (8/n)

To ensure we develop experimentally-testable predictions, we examined two-gene systems in which the reporter gene is constitutively expressed (colored) & the adjacent gene is inducible (gray). Changes in the reporter arise from induced transcription of the proximal gene.(9/n)

We examined how variance in expression changes when the adjacent gene is induced. Notably, we see big syntax-specific differences in noise. (10/n)

In addition to looking at single traces, we can examine the ensembles and compute the cross-correlation to identify correlated and periodic coupling between genes. With positive cross-correlation for divergent and anticorrelated for convergent at zero offset. (11/n)

Interestingly, syntax modulates burst frequency of the reporter while minimally impacting burst size. (12/n)

The result that stunned us was the effect in tandem. Tandem designs might be the most common design in synbio and have been presumed to be a neutral “default.” Induction of the upstream gene substantially reduces activity of the downstream gene. (13/n)

Summarizing these behaviors, we find that convergent induces either-or behavior, divergent induces correlated bursting, and tandem displays upstream dominance in which the upstream gene regularly transcribes leading to low frequency bursts of the downstream gene. (14/n)

But wait a second, what about histones? Eukaroytic chromatin is more than just naked DNA. How might those affect your model? Also what about other structures like R-loops and G-quadruplexes which may form more favorably in under or overly wound DNA? Thank you for asking! (15/n)

Our original torque/energy model is species-agnostic in order to predict behavior across diverse organisms. However, nucleosomes and DNA structures may impact regulation. To model these extra behaviors, we can adjust our energy and torque functions. (16/n)

First up, nucleosomes! Nucleosomes "store" negative supercoiling and are modeled as a buffer against positive supercoiling. We find that nucleosome buffering does not significantly impact the steady state expression profile and had a mild effect on bursting. (16/n)

When we instead penalize RNAP binding at extreme hypernegative and hyperpositive supercoiling densities to generically model structures like R-loops, we observe globally reduced expression and smaller syntax-specific differences in expression profiles. (17/n)

Notably, the extreme-supercoiling penalty reduces burst size across syntaxes. (18/n)

These different models generally preserve the syntax-specific patterns of expression observed in our base model and mostly just scale transcriptional activity. Thus, our model may serve as a useful, extensible reference model across prokaryotic and eukaryotic organisms. (19/n)

So, how might supercoiling-mediated feedback influence biochemically-coupled biological networks? For instance, what happens when you build a classic dual-repressor toggle switch with different syntaxes? How does supercoiling-mediated feedback impact toggle behavior? (20 /n)

The convergent syntax reinforces bistability, whereas divergent syntax under equal induces collapses to monostability. Together these data indicate that the selection of circuit syntax can reinforce or destabilize circuit designs via supercoiling-mediated feedback. ( 21/n)

Finally, we wanted to ask if supercoiling-mediated feedback could explain the transcriptional dynamics of native systems with adjacent coupled genes. (22/n)

We were inspired by the work of Zinani (doi.org/10.1038/s41586…) where they found that proper transcriptional coordination between hes1 and her7 could be maintained only when retained at the same locus. (23/n)

Simulating this networked for paired and unpaired mutants, we found that supercoiling-mediated feedback supported robust transcriptional coordination, showing strong periodic behavior as essential and as observed in vivo. (24/n)

So, we see that supercoiling-mediated feedback rapidly couples and tunes transcription, enabling regulatory behaviors that are entirely inaccessible to other forms of regulation and which can combine to reinforce precise temporal coordination across multiple genes. (24/n)

We’re not the first to think about supercoiling impacting circuits. For recent inspiring work in bacterial see: doi.org/10.1016/j.cels…
For measurements in bacteria and yeast: elifesciences.org/articles/67236
For endogenous networks in yeast: doi.org/10.1101/2022.0…
(25/n)

For related models of how transcription influences neighboring genes: doi.org/10.1093/nar/gk…
doi.org/10.1016/j.bpj.…
(26/n)

Finally, thanks the team at @CellReports, who were delightful to work with and especially to @sfmatheson and @NatalieCainSci! This work was also supported by @NIGMS and we appreciate their pivotal support of this work. Thanks to the mammalian #synbio community for feedback!(27/n)

Also grateful to @ChemE_MIT and my lab which believed in and supported this work when it was just a crazy research statement and some ppt slides. (27/n)

And of course, I’m very proud of the tremendous creativity and diligence from Christopher Johnstone (meson.us) who did all the essential work and put up with my near-endless questions and suggestions. Excelsior! (28/n)

And if you made it this far, I’m guessing you watch the Marvel credits for a reason…Do we think our predictions will hold in cells? We do 😉 Stay tuned! (29/n)