Proteomics Show 102 - Dr. Jan Mulder and BRAINS

 

Lazy post day! Reviewer comments (....some a little past the due date...sorry....) on ....I shouldn't type how many papers....it'll make me a little stressed out...... on my desktop.....

I am legitimately loving recording this new season. Thank you US HUPO and these amazing guests we have lined up. Did you know a brain...just....unfolds....? I did not know this. With the podcast actually now racking up more listens than this blog, I should probably advertise this on the other thing.... But this took 6 minutes and most of that was trying to decide on a gif. 

Why do immunopeptidomics anyway? And what comes after?

 

Do you wonder why we don't just do immunopeptidomics by genomics technologies? Besides the obvious fact that it's impossible? Or just wonder what happens after you've spent a really long time working on the crappiest peptides you've ever tried to fragment?  

Then this is the review for you (and you)! 

I'm convinced! Illumina Protein Prep might be a game changer!

Brazenly borrowed from this whitepaper. 

If I have a super power as a person or a scientist, it is that I'm very okay with being wrong. It helps that it happens all the time and the fact that I have friends and a domestic partner who are way way way smarter than me. I'm used to be the dumbest person in the room and I can just discover that I'm wrong.

And boy - was I wrong about this new Illumina Protein Prep thing. 

I thought it was just a repackaging of SomaScan, a product that has had the strangest propensity for avoiding the very simple experiment that would make me stop making fun of it. After a decade I was starting to think that 1) They were doing it just to get on my nerves or 2) They had done it - and aptamer off binding could not be used to estimate a protein concentration in a complex mixture in any meaningful way (translation - it doesn't work). 

But Illumina has been killing it for years and years! We have petabytes full of Illumina short read sequencing data all over the world. Sure, you could argue they missed the long read sequencing bandwagon and that is a little weird. But a behemoth of an organization like that has the money and the people to avoid becoming complacent.

So when Illumina acquired whatever SomaScan had changed their name to that month, you had to think "wait. maybe there IS something to it!" 

And here I sit while turning a TOF after a power outage that caused me to miss the last day of a conference. Embarrassed and corrected.

The A problem with aptamers is that they are only linear within an EXTREMELY narrow linear dynamic range. If sample A has x target and sample B has 2x target, you can basically see that difference. If sample C has 10x target, you're probably okay, but you're at the end of the dynamic range. If sample D has 1,000,000,0000x more protein, you get about the same value as sample C. More on that and other problems with aptamers here. 

This new product is so much more than the original product it was based on - because after you have your aptamer readout you NOW do NGS sequencing on tags on those aptamers. And then you do the quantification off of the NGS readout! By counting the reads! And we all know that there is no better way of doing quantification than counting things. And if there is, it's probably counting an indirect measurement of an indirect measurement. Wait. Didn't we do something like that before? 

Okay, but that doesn't fix the linear dynamic range issue of the original measurement. But now you've got rock solid absolutely amazing quan on those narrow measurements, right? 

And this is where I change my mind about this whole thing! 

This group took a good hard look at precision and accuracy in a pile of different ways to do RNASeq, with a special emphasis on low input techniques like scSeq and scNSeq, but lots of work on the bulk as well.

The CVs ARE AMAZING.

Less than 1! Across the board! Okay, fancy mass spec people, tell me how many times that you've reported a CV <1 across an entire dataset. I'd love to say that I only report out proteins with less than 10%, but we use a 20% CV cutoff.

Oh...fuuuuuuuuuuuuuuuck..... they mean CV%, right? Not CV 1 = 100%??

Oh. So...a CV of 1 is a CV% of 100%. Right. So I'm going to puke. Hey! And the new TIMSTOF water pumps reset their temperature after a power outage. That's cool. So..I have more time since I have to set my water cooler temperature to the temp written on it in sharpie (25C) and I assume wait for this thing to re-equilibrate...

Okay, so maybe we need to look at these numbers a little more. 

It's hard to see but there is a red line which is a CV of 0.1 or  CV% of 10. As you might notice. They don't often get very close to those numbers. Now, we could argue this is cheating. The maximum number of cells analyzed in each study was used to generate a pseudobulk metric. So this is averaging thousands or tens of thousands of cells. What we need is - yeah! 

This paper - 

Which features a super duper method for improving RNASeq reproducibility in measurements! 

And - ACROSS A GENE they get to 

Around 22 to 24 CV%. Ouch. 

This is where it gets way weirder. My TOF is finally back so I need to go do work, but do you think they're attaching a huge gene to each aptamer? Or do you think they're attaching a single short oligo? I'm no expert, but I suspect it's the former and this is like global proteomics CV% on a single peptide compared to across a protein. The numbers get better when you've got a higher sequence coverage.

I'll be honest, I started out this post as an April Fool's joke, but it turned out that I learned a lot. 

I'm not going to change the title, though. I think that this product will change the game and I don't think it's going to be in a great way. On paper this product looks like it will still not be able return quantitative protein values, and it looks like when it does, the variability in metrics will be worse than the product it is based on due to the difficulty in reproducing the output data consistently. 

We'll see, though. If you are using this product, or have access to it and you want to do the easy and obvious experiment to show me I'm wrong and this works, please reach out. In the meantime I'll still tell every conference audience and every classroom I'm in front of that there is zero evidence that this stuff can quantify a protein. 

On site at ABRF 2026!

 

This is the first time ABRF has happened in a city that I live in! Man, once upon a time there were so many proteomics people that we had competing initiatives. We were standardizing methods and standardizing standards and comparing software and it was all a lot of fun and it all kind of stopped. But you know what you can talk about anywhere you want to? 

Single Cell Proteomics! Dr. Kyle Swovick, photo taken from the front row, organized a session and, for a small conference with 3 competing sessions at a meeting with sessions with riveting titles like "asset management" I'd call it a hit. Definitely better attendance than last year's session I organized that was called "Yo, Justin Walley and I are totally going to do SCP in core facilities, hold our beers". This was sort of a victory lap for me because we now have paying customers in our core at Pitt for SCP and the first data delivered received this response, that I don't mind at all.

Kyle's core at Rochester also does it! It's a real thing! Kyle talked about what a pain in the butt it was to reproduce methods from the literature that largely forgot important details and I called the authors of a preprint that said "5,000 proteins per cell" in the title "a bunch of assholes" on video for a second year in a row! 

And we had great questions and interactions.

Was that all the proteomics? Nope! It sure wasn't! 

Someone who isn't an ABRF member nominated John Yates for the lifetime achievement in BioAnalysis award, and they were like "wait. didn't we give that to him 15 years ago? Holy shit. Why haven't we given him this award already??" So they did.

And I got to be like 4 rows back for John's history of protein mass spectrometry. Super funny with xeroxes of papers (younger scientists, we used to go to the library and find papers in actual journals and photocopy them) and a lot of self-effacing humor and reflection. And cuttings from papers of the past that seem either predictive or ridiculous in the context of today. Or both. 

Super fun thing I was very glad to catch.

The rest of ABRF? Hmmm....there are some cool new LCM solutions I'm investigating, and some cell sorters. And a great convention center! 

Maybe I'll write more about it when it's over. 

As an aside - 

Conference travel in the US is really really hard right now. Last weekend at BWI TSA screening was taking over 5 (FIVE) hours. I've flown out of BWI easily 300 times in my life. It's a small airport. 30 minutes is a big deal for TSA there. I agreed to speak at an awesome university a 7.5 hour drive from my house. And I don't think I can get there faster in a plane right now, but it'll cost me about the same in gasoline. ...yaaay...

Transfer single cell peptides all over the place with NO peptide loss at all!

 

4 people have sent me this new preprint and I've had to go "...oh...it's in my draft's folder..." but the power went out in my building while I was at a conference and it's sort of a mess and I'm grumpy, so Imma post this. 

So...sometimes I see these papers where someone does something like - 

1) Get better results than anyone in the world has ever gotten with that instrument (or at all) 

2) And when you do that...if you don't make the data publicly available, I have to think....

Let's do background first! 

THE number one challenge in single cell proteomics (or any low input stuff) is informatics, probably, but right there with it is just protein and peptide loss through plain old surface adhesion stuff.

If you put a 200 picograms of peptide diluted in a 4 microliter droplet in a 384 well plate you'll lose about 30 picograms of that material to that plate surface (say..125uL well volume). It's just gone and DDM will help, you you aren't getting that back. Pick that up in a DDM coated pipette tip and you'll lose another 20 picograms. If you take that droplet and you add enough liquid to it to allow it to effectively vortex it and then thoroughly vortex it  - you'll lose a whole lot more! Peptides stick to plastics.

Even when we're dealing with hepatocytes at 400 picogram of protein (single nuclei) we're probably not getting 200 picogram to the instrument because with our workflow. Loss loss loss. And you can buy a more sensitive instrument for $1M or you can cut your losses by 30% and get the same thing.

In this study - this group transferred a single cell peptide load over and over again! Like 4 times! (four, 四, cuatro) and then - got more proteins per cell than anyone has ever reported on that same instrument with nanopots (single transfer, but exceptionally low loss during prep) or with single well digest, pickup, load, on that same instrument. They just pipetted the cells from here to there and used magnetic nanoparticles and then desalted and then transferred it again and then still got huge numbers. That's a big big claim and you generally expect big evidence to support something like that. But...when there isn't any....it's hard to say or think nice things. 

The Proteomics Show is back! Check out chapter one of "All the Parts!"

 

In this US HUPO sponsored season (thank you!) we're being serious scientists who are asking the big questions of bigtime experts of specific tissues and organs. Neely kicks it off in this week's episode with "...What...IS...hair...?" And one we just recorded features a guest who speaks to our level when referencing the activity of adrenaline as - and I quote - "Yo, get down there and dilate, we gotta go!" 

US HUPO is letting us do a big unsupervised season and so far, I'm happy to say I'm enjoying it. Listen to Episode 101 with Glendon Parker, wherever you get podcast things.

How do commercial procedures impact the blood plasma proteome!

Oh. This. Is. Sick.

Matt Foster and I were JUST talking about this at US HUPO. Sometimes blood gets frozen in transit, y'all. And sometimes it probably sits around on some phlebotomist's cart for a whole day and then gets frozen. 

The last paper I could find on this weird stream of consciousness website was almost 10 years old. Wait. How long have I been doing this? That seemed like a recent paper... Whatever. It was before all these fancy nanoparticle thingamabobs showed up.

Actually, I'm procrastinating - here is a rant. I'm convinced that absolutely no one understands how these nanoparticles work. For real. Y'all will give me the "protein corona" like that explains anything at all. 

That's what it said in the first Steve Carr paper and I swear they just made that term up in Boston and everyone just repeats it. Also - how do you copy a GIF on a MacBook? Why is this thing so fast and why are so many things impossible to do on it? 

Okay - but that paper! The paper takes a look at the practical real life concerns in blood plasma proteomics! Like travel time and storage time and freeze thaws! And - obviously old fashioned and new fashioned blood plasma proteomics methods. But everyone has done the latter, even if they can't agree. This one looks at the interesting stuff too! 

Non-reducing proteomics opens up a whole new pile of PTMs under stress!

I don't have my notebook on me but someone who was on The Proteomics Show dropped a knowledge bomb on us about cysteines a couple of years ago. She said something like only 12% of cysteines are involved in those disulfide bridges we're all so worried about. Might have been a he and might have been 2 or 40%. 

So...when this PI is in the lab, he commonly skips the alkylation and reduction steps. Not just because he doesn't know where the reagents are, but also because I spent a couple of years studying drugs that bind cysteines. True story, about 1/4 of the pharmacology faculty at Pitt seem to study similar drugs. 

This group shows that it might be a good idea even beyond cysteine alkylating compounds! 

For real, what are all those other cysteines doing? Just being sulfury? Sulfur isn't the most abundant thing and it only gets incorporated as a nutrient by plants through a painstaking process that starts with a slow oxidation by bacteria. Seems like a lot of work for something that just does a couple of things, right? Evolution is too cheap for many single use materials. 

Super cool paper and definitely worth thinking about when you can't find that DTT thing on your first pass through the -20C? Chemical cabinet? I honestly don't know. 

Trap column optimization for single cell or sub-nanogram proteomics!

 

Okay, so this might be more interesting to me right this second than it normally would be, but I'm very glad to have this group's notes and findings! 

There is a paywall on this, but if you were a little sleuthy you might find an earlier version of it that isn't quite as polished that is not. 

Proteomics of organoids IN OUTER SPACE!!

Y'all know what organoids are, right? 

Some people got together like 20-ish years ago and were like "yo, I wonder if it actually makes sense that all these human cells are growing in 2 dimensions...?" For real, like, they don't grow like that inside a human being, outside of perhaps Lady Cassandra....

So they make cells grow in little balls instead. Still...not...like normal...but if you give people a dose of a drug and measure a response and you work out that same dose for cells growing in 2 dimensions vs growing in the little balls of cells (organoids) the latter is closer to the human.

SO.

What if. 

You (not you, but some NASA affiliated people, unless that IS you. Then you, obviously!)

PUT ORGANOID IN OUTER SPACE???

 (They did single organoid, so I assume it's singular when referencing them?) Sounds funnier, at the very least. 

We send way dumber shit than that to outer space all the time. A ketamine addict who runs a racist social media platform keeps failing to get rockets into space that probably just have pictures he colored of himself inside them. Who knows? Certainly not ol' Space Ketamine Karen.

What was I typing abou...THIS STUDY THAT I SAW PREVIEWED BY ALINE AT US HUPO! 

To make this even more spacy! They ran the organoids on - you guessed it - an Orbitral Trap tandem Time of Flight Mass Spectrometer that we know as the Asstral! 

A massively paralleled ion trap inspired by nature!

 

Okay, so maybe what we actually need is 500 parallel ion traps within our instruments!!! 

This is just an early proof of concept of teeny tiny ion traps operating under the control of individual (or individual clusters?) of GPU (CUDA-type?) cores. I've never seen anything at all like this so it jumped the queue of things I meant to type about this week. 

Higher throughput and coverage than O-Link or Illumina Protein Prep - on a SCIEX?

 

Yikes. Okay, so if there is legitimate instrument competition across the board, I think there should be some seriously competitive pricing in the LCMS space this year. If you're not seeing it, demo something else and let them know about it because - holy cow...

I had a super early version of the 7600. I had it before it could do ZenoPulsing/ ZenoTrapping in DIA. It was a nice instrument and super ridiculoulsy great at targeted proteomics. That PRM thing (mRmHR?) loved Skyline and it processed data super fast. But it wasn't up there with my TIMSTOFs. Not really. You could doulble the gradient length and increase the amount you loaded on column by 2-4x and get the same coverage as the TIMSTOF Flex with the XR catrdidge. 

You sure as hell couldn't get 4,700 protein groups at 500 SPD from 200ng of K562! Make no mistake thsi is far higher throughput than the "next gen" proteomics solutions out there. This is more than 10x more samples/day than Illumina's new protein prep - AND this thing can QUANTIFY proteins. Friendly reminder that Illumina Protein Prep only detects proteins. It does NOT QUANTIFY proteins in any meaningful way. O-Link can provide meaningful quantitative responses, but on a single instrument setup 500 SPD is going to be 2x - 4x the throughput (though I've heard this is improving with 1536 well plate preps or something). 

Something weird on the 7600 was that it had a lot of the optics from the earlier generation QQQs. The 8600 doesn't. It's got the new stuff and on top of the speed they're showing off, this thing can hit solid coverage at single cell equivalents. They even prep single HeLa cells in 1 cell, 3 cell, 10 cell and stuff and it looks good. Also there is some 3 proteome digest stuff that looks pretty solid - obviously better quantitative accuracy at the lower speeds. CVs look decent. If you could nitpick these results you could say that the ion mobility equipped instruments do get cleaner spectra and tighter CVs but - for real - SCIEX seems to have a legitimate competitor for the other company flagships. I was about to make a bunch of highlander jokes following some recent software company acquisitions but it looks like there is still legitimate competition out there in the world! 

Pepyrus - User defined HLA (MHC) peptide libraries!

 

...well...this is frustratingly brilliant....

So why we're all messing around trying to figure out how to do a better job deep learning endogenous peptides, this group decided to just cut out the middle of the plan completely.

This system relies on using E.coli to generate the peptides that you think are there which can be directly informed from your genomics data - to actually make the endogenous peptides that you might be able to see - if they are there. Double dash! 

Then you can have the real spectral libraries for that mutant form that would be really super amazingly cool to see on the surface of that cancer cell. If it is there you see it by mass spec and then you can design a CAR(R)-T thingamabob or immunotherapeutic whatchamacallit to those cells.

For real, just intimidatingly shockingly clever, because it really seems super obvious once you get 3 pages into it. Not that I could have pulled it off (please see thingamabobs above and genetic expressoin in little poop bacteria seems like weird magic these days), but it does seem like someone should have thought of it.

Site specific glycoproteomics of hepatocarcinoma!

 

This is a solid new study with a noteworthy method combination.

The global proteomics was done one a TIMSTOF with diaPASEF and analysis in SpectroNaut. The glycoproteomics was done following some sort of chemical enrichment thing followed by TMT proteomics. It's a big 'ol pile of files.

I was looking for how they would normalize the glycopeptides against the whole protein concentrations and they actually include that part in the method section! Our puppy just scratched our kid, sounds minor, but I've got to stop typing here even though I haven't got to how they used some combination of ETD and HCD to do the glycan and TMT based quan. Worth checking out though! 

Charge detection mass spectrometry of intact HIV Envelope Proteins - with Glycoform information!?!

...well...I didn't know that you could do this....

And - wow - has ChemRXIV gotten a glow up! It used to be the ugliest duckling of the preprinting world, but now it's totally modern and legible! 

For this study something called a Q Exactive HUMR with charge detection was used to work out these assembled protein complexes. Flipping everything I know about Orbitrap intact protein work on it's head, this group ran the charge detection thingy at 200,000 resolution with direct nanoinfusion for possibly as long as 30 minutes. 

I've done a lot of intact mAB work, somehow almost always out of the goodness of my heart. Actually - advice to younger scientists who might be way too nice and willing to help people - If anyone wants you to get a good intact mass of an mAB - especially if they are a startup - payment up front every single time. And if they said they worked with me, definitely reach out so I can tell you about how I took work from a nice CRO or three and got $0.00 for it and I'll suggest you charge them 100x what you're thinking. 

Wait. What was I typing about? OH YEAH!

Okay, so on mABs on a regular Orbitrap I generally run 15,000 resolution, when KRAS will look great at 70,000. Intact mABs are 10x larger than KRAS and those spare molecules in the Orbitrap wreak havoc on something that large. On an Exactive EMR, the biggest thing I ever did was maybe 250kDa and I had to run it at 7,000 resolution or something to get it to resolve

So whatever this charge thingy flips that upside down! 200,000 resolution! That's crazy! And they step through collision energies to get data on this heavily glycosylated complex! Super cool work.  

Could you profile snake evolution by proteomics of their venom?!?

 

I missed this recent study and it's ridiculously cool! 

I'm going to start by just stealing this line in the conclusions by the authors: 

"...From an ecological-evolutionary perspective, a post-transcriptional mechanism that modulates rapid variation of the venom phenotype can potentially confer adaptive advantages in response to environmental changes...."

...right??? because of course it would. 'Cause if you're a snake and you want to spread across a mountainous region where some of your 94th cousins live close to the ocean and you live at 3,000 meters you and your cousins have different things that you need to need to defend your self from, or prepare for lunch, right? 

And is the fastest way to do that adaptation waiting 140,000 years and 220,0000 generations for your DNA to make those changes? Or could post-transciptional mechanisms lead you to make a toxin that can KILL THE MOUNTAIN THINGS IN YOUR WAY? 

Sounds to me like one of those things is more energetically favorable than the others.

This group does some awesome stuff - first looking at the intact protein profiles of the snake venoms they got from different elevations - and they look different by eye, for sure. They also....have different lethalities on different animals...which.... yo, you wouldn't know until you checked, right? 

Bottom up proteomics is done with SDS-PAGE, in-gel digestion and analysis on a Q Exactive Plus.

All the data is up on MASSIVE accessioned as MSV000096598 if you want to take a look at it. 

I tell you what, I get to do a lot of cool stuff in a medical school with some sky bridges to where the operating rooms and MDs are, but I'll never get tired of looking at someone using the same tools we have to do things like question basic evolutionary paradigms with weird stuff like poisons they squeezed out of a snake. Okay, I do get bored when people have the same tools but do a half-assed job and get away with it because they completely invaluable samples, but that's not what you'll find here. 

NIFty -Never Impute Features (thank you)!

 

This was the first poster I hunted down after Day 2 Lightening Talks at US HUPO 2026, and now I can share it with you guys! It's great because another SCP Biorxiv preprint came out this week and my opinions on that that one definitely have to stay in my drafts folder. This is a place for positive commentary, mostly! 

Proteomics has a very strange relationship sometimes with the concept of zero. I get it, dividing by zero is not allowed in Excel, and that's probably a significant part of the problem, so there are piles of smart-ish ways to not have a zero. Realistically, single cells are still at the limits of our detection limits and zeroes are pervasive. Everyone's heard this a million times, but the first good scSeq dataset I got on a drug I was studying was a little more than 90% zeroes. There were a couple transcripts that were detected in less than 10 cells out of 7,000 cells so...for that transcript, in that study it was more than 99.875% zeroes. 

SCP also has batch effect problems that primarily effect lower abundance proteins. Those high abundance proteins can look pretty great from study to study. 

This is me rambling about what I understand about NIFty, which tries to solve both problems by 1) using zero values as if they mean things and 2) looking for classifying data using the proteins that are less variable between batches.

And then there are a bunch of things like this, which an AI just informed me indicates that you are supposed to read the formula in a snobby British accent. Which....makes me think I've pushed the Chipotle support bot a little too hard today. 

You might just want me to stop typing and post the Github so you can check this awesome thing out yourself! 

https://github.com/PayneLab/nifty

Can't tell what species it is peptide level data? Enter PEPTONIZER 2000!

 

Given the sheer multiplicity of peptide level data, we're probably approaching a point where just trying to do metagenomics classification seems silly compared to the metaproteomics stuff, right? 

The way I think of it - 

Yo, is this species A or species B?  I've got 20 bases of relatively fragile oligonucleotides, so  4 bases ^20 is what I'm working off of to make that species ID. That's a big number. Something in the e12s according to some extremely sleepy Excel thing that took 11 seconds and might not be right at all. 

Or. I've got some collagen which can have a half life of over 70 years....and I've got 20 amino acids from that. so...it's 20^20-ish. Change the 4 to 20 in that same excel spreadsheet and now you've got something that could occure by chance in the e26 times. 

Okay, but I know how to get good sequencing data of peptides....what's the next step?

PEPTONIZER 2000! 

Here come the LLM Search Engines, introducing ChatDIA!

 

Okay, so everyone else should be really embarrassed they didn't put together a preprint and get credit for "ChatDIA". I just submitted a one paragraph "paper" to Biorxiv on my lunch break, so ChatDDA is mine in the next 72 hours or so.

That's not to say that this doesn't have merit. This thing appears to fare well against DIA-NN!