By the way. I wonder if a RNA switch, when it shifts shape, does so in s sudden jump? I don’t think it can glide as not all bases in the strands are complementary.
I remember someone ask if it was the presence of the FMN molecule that induced the switch. Is it because it is somehow energetic favorable for the switch binding up with the molecule?
Do the initiation of the switch happen first in the end of the switch that gets in touch with the molecule?
Great question, Eli!
A useful way to think about RNA structures and switches is through statistical mechanics. Basically, even at room temperature, any RNA sequence can fold into lots of different structures. The most stable structures, which have the lowest free energy, are the ones that the RNA most frequently folds into.
When you play EteRNA, the “current” state of the RNA, is the structure with the lowest free energy - the most common one. The RNA molecules you have designed may be in that structure 99.9% of the time, or it may be much less - the proportion of molecules in each of the different structures depends on the free energy differences between each structure.
When you look at the SHAPE switch data, sometimes you see that a dark band became a little fainter but didn’t totally disappear when we added FMN - that’s a perfect example of how adding the FMN made some (but not all!) of the molecules switch structure. Perhaps half of the molecules switched, and could react with SHAPE, and half them did not switch and couldn’t react - so we see half the intensity of the band.
Now to get back to your question. Even before you add the FMN, the FMN binding structure is already “available” to the RNA - it just may be occupied at a very, very low level. But when you add the FMN, the molecules in that state are stabilized and get stuck there! Molecules are very dynamic and are continuously moving around and changing their structures slightly - so eventually, as molecules pass through the FMN-binding state, many of them get stuck there. This can happen very quickly or very slowly, depending on the RNA.
The details of how molecules “re-fold” as I am describing here is another tricky question, and still a very active area of research! EteRNA players might one day tackle these important kinetic issues (how fast molecules switch states), but right now we are concentrating on really understanding what the most stable states are for each sequence and how to stabilize the state we are interested in.
Hi Tom
Thanks for an awesome answer!
I understands it now as the RNA switch sometimes is in closed position, other times in open and ready to bind state, no matter if there is a FMN molecule around to bind up with. So it binds when FMN comes, because it is already ready. I thought the RNA switch hid it’s binding position away until needed.
It do makes sense that it switches around between differnet positions. That way some RNA switches are always ready for action, when the opportunity comes.
I guess I wonder now, why the switch has to switch at all, as all it needs to be able to bind up, is the second molecule-binding shape.
Calling the RNA a “switch” refers to the fact that its most stable state is different when FMN is around as opposed to when it FMN is not around.
When we design a switch puzzle in EteRNA, we are making it so that the most stable state (the target state) is different between experiments where FMN is present and when it is not. Even if the FMN binding state is technically available when FMN is not around, the non-binding target state is the most common one. Then when we add FMN, and FMN makes the FMN-binding state more stable, the binding state becomes the most common one (the new target state when FMN is around).
We can see this switch in the SHAPE data - when really dark bands totally disappear when we add FMN, it shows that a lot of the RNA molecules really are changing shape!
Thanks for yet another great answer. I understand this much better now.
I have a forum thread and a couple of Google docs (linked from the thread) that go over the switching idea in the context of barrier energies.
On one hand you can think of it as a nucleotide refolding via a series of small steps from one form to the other. On the other hand you can recognize that *all* (or most) of the intermediate shapes are likely in solution at any given time, based on the relative free-energies of the shapes. Rather than a single nucleotide transforming in a series of successive steps, various shapes are constantly (re)folding and transforming from one shape to another to another, and perhaps back again.
After a while, the in vitro system is in dynamic equilibrium, with roughly as many shapes folding into any given shape over a time interval as out of it. Adding the FMN to the solution, *shifts* the stable point as now some molecules can gain an energy bonus not only from their shape, but by binding with the FMN molecule. That change in the equilibrium state triggers a “switch” in the relative concentrations of the various shapes.
Think of it this way: imagine a rubber sheet with dimples and bulges in a frame (an analog for a set of RNA shapes with various free energies). If you pour water (RNA strands) over it there will be more depth of water over the dimples (more negative free-energy shapes) and less over the bulges (higher-free-energy shapes), but the water (RNA) can still flow around (transform/refold) from one spot (shape) to another. Now push down in one spot making it deeper there. Water will flow around to fill the void until it achieves a new equilibrium with more water over the new deeper dimple. This is like adding FMN to help favor one shape by giving it an energy bonus.
There may be some shapes that are more “stable” than others in that it takes an input of “activation energy” to bump them out of this shape and these may exist in slightly greater abundance for a while, like water flowing down a hillside collects in small ponds.
So over time, any given RNA strand may assume multiple shapes, and at any one time, multiple stands will be in differing shapes, so that there are fractional concentrations of each shape in the solution. The RNAshapes program/web-service shows this fairly well with percentage concentration estimates. RNAsubopt is another program for showing multiple foldings of the same strand; it gives free-energy estimates, but not probabilities.
Folding into a single RNA shape seems to be slightly easier as you can try to find a sequence that strongly prefers one shape with few near-misses energy wise. For switch puzzles, at least so far, there seems to be a slight benefit to having sequences where there are tens of thousands of possible shapes within 9kcal of the minimal free energy. Perhaps that’s because these sequences seem to have more likelihood of unfolding/refolding.
Of course, I’m not a micro-biologist, nor RNA specialist. This explanation is all based on what I have gleaned from the writings and tools of others plus some thought experiments.
@jandersonlee – this is an awesome explanation.
it would be interesting (and unexpected) if the kinetic analysis based on energy barriers led to more robust switches with the current eterna puzzles, which should not depend strongly on having fast kinetics. if there is, let’s keep discussing. one thing we can try is to heat/cool the RNAs with and without FMN present to help anneal in the favored structure.
I’d be very interested in how the RNAsubopt/barriers analysis correlates with the data for the more recent puzzles (and the series of more to come…).
To continue the analogy, the barrier energy would be analogous to the water level needed to connect the two dimples for two different shapes.
I’ll run the numbers on the synthesized sequences from the last lab and see if any correlations seem to show up.
Well I ran some barrier numbers for the last lab and put them in a Google Doc spreadsheet.
The top lines show a correlation between the lab score and a few different metrics. One line, labeled C76, shows the correlation between score and metrics for sequences scoring 76 or better, and the other for sequences scoring 59 or better.
In the higher scoring region (76 and above), there appears to be little correlation between the metrics and the score. In the full range, there is some seeming correlation on a few factors:
0.03 vs -0.42 on ALT SID: Suggests while there is a broad range allowable on this metric (up to around 2000 at least) values in the tens of thousands may have a detrimental effect on lab score. What is ALT SID? It is the alternative shape id: the ranking of the free-energy alternative (aptamer loop) shape without the energy bonus added in comparison with other sorted/rank-ordered alternative folding shapes as ranked by RNAsubopt.
-0.11 vs -0.33 on ALT PLAIN ENERGY: ALT PLAIN ENERGY (APE) is the free energy of the intermediate form that serves as an energy barrier between the plain (unfolded) and ALT (aptamer loop) shapes as estimated by custom software. That this correlation is negative suggest that a higher free-energy of the barrier shape may lead to a slight decrease in score. While submission 1543935 was able to score 85 points with an ALT PLAIN barrier energy of 3.8 kcal, 1473508 and 1440295 scored much lower with APE values of 5.1 and 5.4 respectively. Still, correlation is not causation, and it could be other factors like the higher ALT SID values on these candidates that made the difference.
0.02 vs -0.24 for 4.9 kcal: This metric measure the number of various folding shapes estimated to exist within -4.9 kcal of the minimum energy by RNAsubopt. A wide range is seen for designs scoring between 76 and 95 points (from 216 to 5941). A couple of outliers of 12248 and 6575 on the same two low scoring designs mentioned above probably are responsible for the higher correlation value on the larger range. Once again, correlation is not proof of causation.
A few other metrics also correlate in the low -0.18 to -0.23 range. A couple of cases where the unbonded shape has a high free energy (-0.5 and -0.6 for PRIM ENERGY) suggests that perhaps there may be a limit to how high a free energy one can have on the unbonded shape without paying a penalty; however a design with a score of 85 and an estimated PRIM ENERGY of -0.2 suggests this this penalty is either not certain or may be mitigated in some other way. Could it be because the one design had an energy difference between the two shapes of 4.0 kcal, while the other designs were 4.7 and 6.0 kcal? Yet two other designs scored in the 80s with energy deltas of 4.6 and 4.7 kcal.
Any correlations could simply be the effect of the so-called law of small numbers, where seeming patterns can often be found in small sample sizes that do not scale up when the sample size is increased. All in all, no strong, obvious correlations were seen. Perhaps it is better to keep ALT SID under 2000, but there were also poorly scoring designs that satisfied this criteria. All of the designs scoring over 90 had an APE of 1.3 or less, but the sample size was just six and so did many other designs that did not score as well. Thus at this point there is no strong evidence to confirm any strong causal relationship between barrier energies, shape ranking, or similar metrics with lab score, but also too small a sample to deny the possibility of a subtle causal relationship.
I was listening to the podcast MicrobeWorld’s Meet the scientist, when I found this fascinating quote by Susan Golden:
For somebody who does biochemistry and molecular biology, like me, we take for granted that everything that goes on in a cell is shape. But that is not neccessarily obvious to people who don’t do this kind of thing for a living. But shape is actually what drives the interaction of all kinds of molecules inside a cell, you know there is a lock and key, things fit together. And that makes something happen. When you have the lock and the key, something turns.
I think we gamers are getting there too. 
Thanks for reporting a thorough and statistically careful analysis. The statistics are going to be getting better, of course, as we get more switches, especially with different shapes. (We are also hoping to slightly accelerate the experimental schedule.) So it might be fun to revisit this analysis in a couple months.
I hope Eli won’t mind too much if I hijack her thread, but on the other hand, it not off-topic, and it’s partly her fault if I spent so much time writing this paper.
So, this is simply a call for comments and feedback, thanks in advance.
I doubt Eli will object to you jumping in on the thread, but *he* might object to being called a *she*.
oops… I should ask next time I have a doubt… thanks
hmm, looks like I can’t edit the previous post… I’m left with the only option of apologizing, sorry Eli
Apology accepted Thanks for the huge addition to your document. I look forward to see what this will bring to our switch design analysis in the time to come.