Preliminary Thermodynamic Mimic R2 Findings

  • 2
  • Article
  • Updated 4 years ago
Here is a short list of findings from the R2 Mimic results.

1. A few sequences from the Switch Cloud Lab: The Next Generation 4 target, appear to be working FMN riboswitches, based on a single mimic sequence per WT sequence.

2. The revised mimic search algorithm generated a significant number of mimics where there was little to no base pairing between the residues in the 6x5 aptamer loop. This disproves the hypothesis that the v. 2 script may be better than the v. 1 script based on a model where the partition function of a sequence may be an important determining factor in mimic fitness. It may be better to focus on heuristic score terms, such as the base pairs and content of the mimic, rather than on a computational model focusing on the predicted free energy.

3. For each of the Bistable 3 and Hair Trigger targets, 20 identical mimic sequences were submitted. The hypothesis had been, that identical thermodynamic mimics in different sequences, would have different results. This was proved to be correct, and could lead to opportunities for "standardizing" certain sequences to be used when a defined mimic situation is present.

4. Mimics that consist solely of R-U base pairs are highly volatile, and do not give similar results to mimics with G-C base pairs. It could be that the thermodynamic properties of these mimics are accurately defined, but the kinetic nature is not accounted for. If the R-U base pairs cannot "kinetically lock" the structure of the RNA once it adopts a specific secondary structure, then the RNA may not be able to stick in the ON state secondary structure.

5. The Lines Pseudo-Titration Experiment was both a success and failure. It was a failure, in the sense that out of 75 valid thermodynamic mimics, ranging from -1 kcal/mol to -7 kcal/mol, only a handful of mimics actually caused a conformational shift in the RNA structure, which was the result hypothesized to occur. It was a success, because the pseudo titrations may prove that the RNA sequence cannot work as a riboswitch, and build a foundation for a method to screen potential riboswitch sequence candidates.
Photo of Brourd

Brourd

  • 437 Posts
  • 79 Reply Likes

Posted 5 years ago

  • 2
Photo of rhiju

rhiju, Researcher

  • 403 Posts
  • 122 Reply Likes
Will look forward to more analysis if the team is preparing a wiki. Do you have enough sequences for 'The Next Generation 4' for us to consider testing by actual FMN titrations?

If not, given what you've learned about use of mimics, could you design one more Paper Lab round to capture 16-32 solutions (perhaps on 3-4 targets) that you think might be worth running through FMN titrations?
Photo of Brourd

Brourd

  • 437 Posts
  • 79 Reply Likes
Potentially, we may be able to come up with a list of 8 or so sequences.

The four sequences currently being used for the Bistable 3 and Hair Trigger targets.

The "Lines" target sequence,

and 3-6 sequences from the novel Switch Cloud Lab: The Next Generation 4 targets.

As for another round, that would probably require tweaking the mimic generation/fitness algorithms, and we do not have any players qualified to do so at the current time.
Photo of rhiju

rhiju, Researcher

  • 403 Posts
  • 122 Reply Likes
Couple more questions:

First, do you think you can predict the Kd for all eight of these? If so, we'll 'pull the trigger' on testing those predictions through actual FMN titrations.

Second, are any of these solutions to switch puzzles that were previously unsolved? 
Having at least one of those would make for the most exciting publication and capture a lot of attention.

P.S. As devs are moving forward with the RNA-on-chip technology, we see that use of mimics may be critical for players to troubleshoot their designs. So we are very excited about these results from the mimic project; they are foundational for advances over the next 2-3 years of eterna.
Photo of Brourd

Brourd

  • 437 Posts
  • 79 Reply Likes
I'm not entirely sure if I can predict the individual Kd values for individual RNA designs, or at least, I am not too sure if I know how to do the math (I'll get back to you on that).

We do have a hypothesis that we can predict the approximate concentration of FMN required to cause the RNA to switch structures, using the Kd defined by the Das lab for the free energy calculations in Eterna. However, only three of those eight sequences actually have the data necessary to make the predictions.

"Second, are any of these solutions to switch puzzles that were previously unsolved?
Having at least one of those would make for the most exciting publication and capture a lot of attention."


Could you elaborate on what you mean by "unsolved"?

If it is related to the pseudo-titrations by mimics, we could run this experiment with any number of sequences that have not been probed with FMN before. The main issue is that it requires a significant number of mimic sequences (35+ or so for now). We could probably limit that number to less than 15, but the quality of the prediction may decrease. However, it would not hurt to try, and it would also allow us to test a far greater number of sequences at any one time.
Photo of rhiju

rhiju, Researcher

  • 403 Posts
  • 122 Reply Likes
Regarding Kd, I meant approximate concentration of FMN required to switch the RNAs -- it sounds like you do have a way to estimate those.

Regarding 'unsolved', I meant a switch puzzle for which our initial cloud lab results +/- FMN did not show an acceptable solution. My question is: do we have solutions to some of these puzzles taking advantage of the mimic strategy that we can then validate through actual FMN  binding tests?
Photo of Brourd

Brourd

  • 437 Posts
  • 79 Reply Likes
'Solved; or 'unsolved' are difficult terms to define with riboswitches. If you mean the specific +/- FMN concentration from Round 70. 200 μm, then no, our current sequences have focused on designs that were either a complete success or a failure as a switch, using identical targets.

A number of the targets may also lead to solutions that are "impossible" with such a limited concentration of FMN.

However, we did do have a predicted solution, for the necessary concentration to cause a conformational shift in the ensembles of two R70 RNA sequences that could not be solved with 200 μm FMN.

I will try to get screenshots of some tables that illustrate the idea.
Photo of rhiju

rhiju, Researcher

  • 403 Posts
  • 122 Reply Likes
Thanks for the clarification -- really helpful!
Photo of Brourd

Brourd

  • 437 Posts
  • 79 Reply Likes
Here is an image of the mimic titration table for the Bistable 3 - Mod of Eli by mat747 sequence.

Bistable 3

The table and values are based on experimental data from the first round of the thermodynamic mimics, with each sequence being synthesized twice. The hypothesis would be, that we can predict the association of FMN to an RNA sequence based on the number of residues that "switch" when the solution has a specific mimic bonus, which would then be converted to concentration FMN.

So, for this particular sequence, we would hypothesize, based on the sequencing data, that at around -5 kcal/mol, or close to a concentration of 12.48 mM, close to 80% or so of the residues in the RNA will switch reactivities as we would want.

Granted, this is based on all 10 of the residues that should shift reactivities in that particular sequence. There is a second measurement/prediction based solely on the four residues that are a part of a helix coming off of the FMN binding site, but those values are a bit more volatile.

This experimental set up could be used for any sequence, allowing us to predict the necessary concentration that would allow FMN to fully associate with a sequence and cause it to switch secondary structures.
Photo of salish99

salish99

  • 295 Posts
  • 58 Reply Likes
Thanks for this overview, brourd.
Photo of salish99

salish99

  • 295 Posts
  • 58 Reply Likes
nice
Photo of rhiju

rhiju, Researcher

  • 403 Posts
  • 122 Reply Likes
that's beautiful...
Photo of Hyphema

Hyphema

  • 91 Posts
  • 25 Reply Likes
Very interesting Brourd. So do you have any other particular sequences you would like to try using the pseudo-titration method to "predict" the FMN concentration. Before actually testing the theory using FMN.

On a side note, is there a past switch lab we have done that was "semi-successful" to demonstrate switching in a real FMN environment that can be used to "reverse" test your pseudo titration theory using mimics in that sequence? That may also help us uncover what FMN concentration we need to increase the percentage of switching. ( I hope I made sense. : )
Photo of Brourd

Brourd

  • 437 Posts
  • 79 Reply Likes
There are two other sequences that we have this "pseudo-titration" data for.

One was a "semi-successful" sequence as you called it, and also a sequence that we had a rather interesting result with the pseudo-titration protocol.

Unfortunately, it is one of those questions we do not know how to answer.

Would it be better for the Das Lab or Eterna to come up with a random set of FMN targets, and then run blind design and sequencing tests alongside regular titration experiments? After this, we would compare both and determine if the predicted mimic pseudo-titration chemical mapping data is similar in precision and accuracy to the FMN titration chemical mapping data.

Or, do we instead search for previous RNA sequences with titration data for FMN, and then use our mimic protocol on those and compare? This second option would be the simplest experiment to run if we wanted to gain a better understanding of the mimic sequences, without the expense required for new FMN titration experiments.
Photo of rhiju

rhiju, Researcher

  • 403 Posts
  • 122 Reply Likes
The coolest thing from my point of view would be if we can use the mimic strategy to screen designs that solve switch puzzles at 200 uM FMN, and then we test those predictions; ideally we'd then test switch designs involving a diffrent aptamer as well (say, theophylline).

This would demonstrate that we can design switches with a general experimental pipeline even without having to do experiments adding the small molecule. Sometimes we think of eterna's expeirmental pipeline as an 'experimental supercomputer'. the mimic strategy is like a simulation of small molecule binding within this supercomputing framework.
Photo of Brourd

Brourd

  • 437 Posts
  • 79 Reply Likes
That is most certainly the final goal, and it would be cool if we had a method and set of designs to immediately test this.
Photo of Hyphema

Hyphema

  • 91 Posts
  • 25 Reply Likes
Brourd said.... Or, do we instead search for previous RNA sequences with titration data for FMN, and then use our mimic protocol on those and compare? This second option would be the simplest experiment to run if we wanted to gain a better understanding of the mimic sequences, without the expense required for new FMN titration experiments.

Sorry, I don't think I was very clear. I meant do we have any "semi successful " past switch lab designs where FMN was used that could be tested with the mimic protocol. Just as you are suggesting, as a better and more simple means of testing mimic sequences. I was wondering what designs in what lab do you think are worthy to test.