Friday, May 12, 2023

diaPASEF -- what are all these new methods and what do they do?


It took about 7 months of doing diaPASEF every day, plus attending multiple talks (including two really great ones last week by Vadim Demichev and Matt Willetts, respectively) before I felt confident finishing this post. And I'm largely doing it to straighten out my own thoughts before I dive into altering the provided methods on my TIMSTOF SCP (which I strongly believe can be improved but it is such a miserable pain in the butt to generate super high quality actual single human cell digests to optimize on that you don't want to just jump in on it. AND diluting a perfectly prepared human cancer cell line digest and optimizing on that isn't something I find helpful. In fact I find the results more misleading than anything else. We didn't pick K562 as a proteome digest standard because it is difficult to work with...sort of the opposite...and no cell I've worked with has been anywhere near as convenient...) 

There have been a flurry of new methods for diaPASEF recently and I probably can't get to all of them and remain employed. 

I'm going to talk about regular plain old diaPASEF (first described here)

and py-DIAD (described here), but maybe more importantly Syncho-PASEF which py-DIAD enables and was described here

and Slice-PASEF (described here)

First of all, diaPASEF is just DIA where you allow the TIMS to provide some extra level of selectivity. On the first level it really isn't all that different from using FAIMS or Selexion in conjunction with DIA. You get reduced background thanks to the IMS aspect of things so you get a better S/N ratio on your ions.

HOWEVER. That isn't all of it. While peptide m/z isn't the only factor that affects how an ion moves in collisional cross section space (CCS, which is a function of the 1/k0 [ pronounced "won over kay knot"] value the raspberry pi that powers the TIMSTOFs generate) it does play considerably into the CCS.

For example, when you look at the Mobilogram (pronounced "Mo Bill O. Graham") you see how m/z reflectes on the 1/k0 value. 

Here we see two distinct populations of ions. The top smear is likely the +1 ions and you probably don't want those for proteomics. The bottom sideways smear is the multiply charged stuff. If you look at something with an m/z of 800-900 where I've drawn two perfectly parallel lines, it seems like the vast majority of the ions of interest are between 1.2 and 1.0. There are definitely some ions above and below that, but most are in the middle. diaPASEF is going to exploit this likelihood with some obvious positive effects that don't take too much imagination to come up with, but with some detriments that we'll need to get to in a second. (Yes, I know 100 Da is weird for DIA, but it was hard for me to draw the lines any better in the Snipping tool, for illustration purposes only, more below) 

One thing we need to think about is what the TIMSTOFs do quickly and what they do less quickly. If you leverage these properly you're going to get the best results and -- best I can tell, that's the main thing all of these new methods are doing. 

Quads are super fast, but that speed decreases as the window they need to scan through increases in size. (Quad speed is generally described in amu/second. Which is why QQQ marketing will tell you how many SRMs you can get/second but you can only get that speed if you use a uselessly small quad isolation window and virtually no ion accumulation / dwell time.) TOF is super fast until you start averaging scans. TIMS ramping is pretty fast, depending on how you do it, but typically not as fast as either of the above. And ion accumulation is, by definition, slow and the more ions you want to accumulate the slower it is. Now what you need to do is think about leveraging these things in the most efficient ways. Some of this is actually done behind the scenes by the Raspberry Pi, something that I think the Slice-PASEF team might have found a pleasant surprise. 

Back to the image above for diaPASEF. 

In regular old DIA/SWATH with no PASEF we'd just accumulate from 800-850 (900 as drawn) m/z and fragment everything. But the TIMS isolation gives you something cool that you can do here on top of it. What if you took that 800-900 m/z and you did it twice, once with only the ions with 1/k0 of 1.2 - 1.1 and a second time with 1.1 - 1.0 as I've again drawn with amazing aptitude with just a mouse and my nondominant hand below? 

Now you've both removed the background ions AND cut the complexity of your DIA window by a factor of two. Since the resolution of the ion mobility is actually pretty high, you could cut that window effectively into more windows to further decrease the complexity.

However, the TIMS isolation and accumulation for each packet takes time, so you've got to keep track of that. Worth noting, the PA in PASEF is parallel accumulation. You actually have the capability of accumulating multiple "packets" of ions at a time. There are limits to that in regards to what ions can be accumulated at the same time and how many of them there can be. So you have to keep those limitations in mind. I won't get into that here, not because I don't understand it -- because I also don't want to.

In regular old DIA you'd just use more narrow m/z isolation windows to increase your selectivity of ions and you can do that here, but more windows equals more time for both this as well. The trick is finding the ideal compromises, but you can see how this has advantages, I hope. The default methods do seem to use 3 IMS windows to cut up each m/z DIA windows. Cool, right? 

Let's get to the problems, though. 

What about the ions that are outside of the normal-ish m/z 1/k0 isolation windows? Hmmm...what peptides probably behave less predictably than every other tryptic peptide....oh....the PTMs...?

Yeah. The PTMs. This is where the first py-DIAD paper steps in.

py-DIAD is an offline tool that allows you to draw your ion mobility windows in funny ways. It helps me to think that it is sort of like variable window SWATH but you also get control over your 1/k0 windows. 

Drawing funny windows (by having the python script here) allows you to get your main body of ions you're interested in while still ignoring the (small) plus one ions and picking up the PTMs that migrate outside of your ion mobility cloud. Do you under-represent these things in regular old diaPASEF? Possibly, it is hard to tell in these data, they EGF stimulated some cells which made phosphorylation go craaaaazy but they get about 3x more phosphopeptides by tinkering with the windows to optimize them for phosphos. Cool work for sure.

Now that you've got py-DIAD and you can do whatever you want with your diaPASEF windows that opens up a bunch of other possibilities. And this is where Synchro-PASEF comes in. In this, the goal is to get ULTRA SELECTIVE so that you get the vast majority of the +2 charged peptides in the cleanest way possible. 

Remember what I typed above? That quads are super fast by the AMU? (P.S. Try a full scan on a QQQ. SLOOOOOW.) Also if you jump from one quad isolation window to another that also requires time. So Synchro-PASEF keeps that in mind and continuously ramps the quad. Fast by AMU, so then it cuts very small m/z /amu windows across the ion beam in a continuous order. The end result is that you scan right through the high density area where the vast majority of your +2 peptides are using tiny DIA windows. Super selectivity, by still being really fast. 

Stolen from the paper. So here you've got two different experimental goals. 1) Catch everything even if it is weird (PTMs) 2) Catch all the normal things with incredibly high efficiency! 

That's it for this group for now. Let's move into something I was tremendously excited about, but I've struggled to implement. Slice-PASEF (I keep thinking it is Splice-PASEF and I can't ever find it).

Vadim is concerned about DIA duty cycle. In his definition, a 100% duty cycle would be where every precursor is fragmented all the time. If you DIA two windows then you're at 50% duty cycle and 10 windows puts you at 10%. What he and his team tried to do is maximize duty cycle by swapping the original diaPASEF windows sideways. 

Compare this supplemental figure from the Meier et al., paper supplemental with the SlicePASEF workflow

See? Totally sideways. I tried to find a cool gif, but everything I google for sideways gives me a really boring old movie. The IMS windows are smaller, but the m/z windows are bigger. Where does more selectivity come from? It seems like it depends on the sample, but for lower concentration samples it seems like Slicing improves IDs. It does, however, appear to largely depend on more focused libraries to really shine. I think there might be some camps on this sort of thing when we discuss it in DDA, but in DIA we're continually focusing on the peptides that are the easiest to see. Classically, you're building your DIA windows off of DDA data, so you're biasing your results toward the easiest to see peptides. The phenomenal results from this workflow depend largely upon moving down your concentration. For example, if you only use in your library for your 1 ng injection peptides that were detected in a 100ng injection, then that makes sense. But I think if we did that in DDA people would get really mad about it. 

I'm not sure this was the best received article in the world and I suspect there are still people who would find a link to it annoying. The truth, though, is that if you're doing DIA you're probably doing this to some level. 

To improve the selectivity of SlicePASEF you do cut up the wider windows into smaller ones, but that increases you duty cycle. Pros and cons.

Okay. This is as far as will get on this but it made it clearer to me to write it. There are more and maybe I'll get to them.

Thanks for people who were unfortunate enough to read this, I need to drop two additional links here. 

midiaPASEF --

SpeedyPASEF --

1 comment:

  1. Nobody has any clue what any of these methods actually do. By the time anyone could investigate what they might do, they will no longer be in use. A more complicated instrument is probably already being manufactured that will demand even more complicated methods that only the marketing department can explain. If you can understand the gist of what is going on, it is by definition NOT cutting edge. The bamboozleome is not to be trifled with!