The 5th annual Johns Hopkins Mass Spectrometry day 9/17/2026!

 

Tracking predator cues in marine prey!

 

I liked this paper from the very first sentence...

Abstract

Prey rely on chemical cues to detect predators, especially in coastal ecosystems where danger is concealed by turbid waters. Juvenile oysters (Crassostrea virginica) respond to blue crab (Callinectus sapidus) chemical cues by growing stronger shells, while mud crabs (Panopeus herbstii) reduce foraging to avoid detection. 

...because...what...? 

Did you know that? No, I did not. This is the paper, and this isn't what I expected to learn about at lunch today. I expected to be at a liver center seminar but it's so hot in Pittsburgh (a small city in the Appalachian mountains) to safely have school. I tell you what, the climate collapsing isn't going to be boring. Also, kindergarten remote learning is a lot of dancing.

Wait. Okay, how do you set up an experimental design for something like this? 

Oysters were settled onto 4.5 cm × 4.5 cm marble tiles and allowed to grow for 1 week in 1250 L mesocosms with flow-through seawater prior to experimentation. Oyster spat were then transferred to aquaria and repeatedly exposed to 1 mL mixtures of blue crab urine from crabs that subsisted on either an oyster or mud crab diet (n = 3 mixtures in block 1 and n = 9 mixtures in block 2 per predator diet), settled seawater (negative control; n = 3 mixtures in block 1 and n = 9 mixtures in block 2), or a positive control of predator water 

Ummm....okay, that makes sense, but it's sorta funny, right? 

Oh. Okay, I had all sorts of dumb ideas about how they'd ever get the molecules in the sea water. What they do instead is see what oysters make thicker cells in response to crab A or crab B's urine and then they save back some of the urine that they analyze. Then they can track down the molecule that the oysters respond to. They use NMR for the metabolomics. So - no proteomics here - but it is in Journal of Proteome Research so it counts. 

NMN + Sugar = Reduced hypobaric hypoxia!

Despite the weird repository used for the data in this new paper, it's still pretty cool in this context.

Imagine that you're trying to climb Everest and your goal is to have mice carry all your stuff. Maybe you can't afford a human to carry the stuff? Maybe you're just really mean to animals. 

Everything is going great, until about 1/3 of the way up you discover that these mice are getting really lazy. Do you have an oxygen tank for a mouse? NO? I think you're just doing this because you don't like animals. 

Okay, but I wouldn't type all of this because I don't have a solution, right? What if it's a solution you can get at a gas station?!? Check this out. 

Through extensive trial and error of the stuff you can absolutely get behind the counter at shady gas stations, this team came up with the solution. 

Nicotinamide mononucleotide (NMN) and sugar! To prove that this made the mice better at carrying stuff, they chopped a bunch of them up and then did a bunch of multiomics on the resulting tissue. 

Metabolomics and lipidomics were performed on an Exploris 120 and the Proteomics of just the tendons and ligaments were performed on an Orbitrap Eclipse. 

Something interesting is that they used a specific kit for extracting proteins from the ligaments and tendons. I don't know if it's any good, but it is interesting to me that such a thing (and a company I've never heard of) exists. You can find it here. 

The end findings are extended to humans as the mechanism and how to reduce hypobaric hypoxia in us as well. Given the ENORMOUS differences in how humans and mice respond to these conditions...

(for real, mice drop their body temperatures at high elevations, on purpose. Humans generally never change their body temperatures.... human do a funny thing with a word I can't remember where we jack our body pressure way up to compensate for extremely high altitudes, which is part of the reason we need time to compensate and recover at high altitudes. Here's a good link to a whole lot of reasons we have to be cautious interpreting rodent and humans in these kinds of experiments) 

... this could be a stretch, but I like a good over the counter solution to problems, so if anyone ever talks me into running a race in Colorado again I'll try pumping myself more of more NMN and sugar than I normally do just to find out, I guess. 

...iHUPO single cell initiative TIMSTOF Ultra2 methods don't seem to perform super well in my hands.....

 

I'm back at this great iHUPO single cell initiative paper again, I guess. Sorry if you aren't interested, but it's TIMSTOF Ultra2 vs Astral on standards and real single cells, so I'm interested. 

Another thing that caught my interest was how very different the instrument method on the TTU2 is compared to the basically default instrument method we have ran for single cells for about 18 months.

Above you can see our method on the left and the method from the iHUPO paper on the right. The format is a little different since I had to cut the darned table out of a PDF. Which is always a joy. 

But you'll see that we use 8 cycles and we run from something like 0.65 to 1.4 and the iHUPO method runs 5 cycles and from 0.65-1.3. 

I feel like we tried a ton of different methods and ramp times and finally ended up with basically the default method plus 50% more ramp time worked better for everything in our hands. But we optimized on mouse hepatocytes last year, but those are kind of big and also kind of friendly (there are like 550 proteins that you can detect even if the instrument is vastly underperforming, because they make up about 90% of the total protein content. Compare that to human plasma where 1 protein makes up about 90% and you'll see what I mean).

We run a longer ramp time which gives us a full ramp at a speed of 6.37 Hz according to the software, so each precursor gets measured over 6 times/second, right? 

The iHUPO method is faster and has smarter looking windows 

It gets something in the 9.34 Hz and uses bigger windows at the low and high end of the mass range. 

I was super pumped to run something that looks so good on paper, and....it's kind of meh....

To be fair, this is just K562 standard at 200pg. BOOOOOORRRRING, but we get a steep drop off when we go from 40SPD to 80SPD on the EvoSep One. And I've never even tried 120SPD until now. Maybe with a 50% faster method it makes sense? 

This is a terrible way to display data, but I've got like 15 free minutes on a Sunday night to type this up. 

There are only 2 replicates of the 80SPD for each. EvoTips are expensive, yo. But we run the 80SPD standard method all the time. It's...like...our standard method.... and these results look normal. 

At 80SPD the iHUPO method drops off nearly 4k precursors and 450 protein groups? Ouch. 

I ran more of the remaining tips on the instrument in 120SPD because we don't have benchmarks for this and it's not ...as...bad....but ouch. It's still totally worse. 

Obviously, we're using an EvoSep and they used a funny preformed gradient on a Vanquish Neo. And quantitative accuracy measures more than protein numbers. 

BTW, this is just command line DIA-NN running at the end of each file. Nothing special. No MBR, no whatever. 

But if you're looking at this paper and thinking WHOA, the Astral gets better data than the TIMSTOF, I'd take it with a grain of salt. This doesn't look like the best method to me, and instrument comparisons are dumb anyway. 

Did an Abird outperform a FAIMS Duo source in this new study?

 

I recently blogged about this new preprint and while going over it in more detail to see how we could improve our workflows I had to be impressed by one of the findings I didn't mention. 

Two of the Asstral / Astral (spelled different by geography, I guess) systems had the FAIMS 2 Duo system and one of them didn't. Instead, it had the humble Abird system.

Abird has been around a long time and some labs use it on every system and other labs laugh about it seeming like a silly idea to reduce your background signal. 

In a couple of places in this single cell study the one Astral equipped with Abird OUTPERFORMED the 2 Astrals with FAIMS. For some context, the one time I had access to FAIMS on an Orbitrap Exploris 480 the reason we didn't like it was that we couldn't generate enough nitrogen for the silly thing. To make up for it we rolled in big liquid N2 dewars which cost us about $250 every 3 days or so if we used the FAIMS. In one month we could have paid for an Abird. 

I don't know what a Duo costs (I hear they often bundle it in with new systems almost for free, but if you want to add one to an existing system it can be in the HPLC price range), but if you're looking for better proteomics results at an ultra-low load level this seems like a low cost investment. (BTW, I don't know if it works on anything aside from Thermo systems). Excerpt from preprint. 

Wait. How many proteins are in the nucleus?

 

Ummm...okay... so I stand corrected.... Thank you ProteinAtlas! 

https://www.proteinatlas.org/humanproteome/subcellular/nucleoplasm

So...I'm surprised by a whole lot of this. I'd personally expected that when we got around to doing single nuclei proteomics that there wouldn't be a lot there. 

Something like 

HISTONES

and all the AHNAKs

and that would be about it. Sure, some transcription factors, but they're low copy number at the best of times.

And there is a whole lot more going on there which doesn't make much sense to me at all, and that's okay.

So far we've seen two studies with single nuclei proteomics, one published and one that I'm sure will show up any day now.

In this one, Derks et al., used multiplexed DIA reagents on a TIMSTOF SCP and optimized out some carrier channel levels for the nuclei. Across the study I count around 1,800 protein groups in their data, however the actual protein IDs are missing from a lot of the processed data. I assume that's so they can do unbiased clustering downstream. SCP-Viz requires an identifier to perform clustering which makes it harder for me to reanalyze the data, but it's still a big number. Mun et al., 

did a far smaller number of nuclei, but did it with label free diaPASEF. I've got to dig back into their numbers but they were also far larger than I'd expected (see eroneous assumptions above...)

Discovery proteomics with O-Link reveals new HCC markers?

 

When I think of proximity extension assays (PEA) which is either famous because of O-Link - or is? O-Link, I think of purely using it for large population studies or for validating discoveries by high depth mass spectrometry based proteomics.

But...could you use it in place of mass spec proteomics? Because that is basically what this group did here. 

Now...if your protein or proteoform isn't in this panel and that's all you see, you've just wasted a lot of money and time. However, in this case the depth afforded the serum proteomics allows the discovery of some interesting new proteins in a small cohort. Which then seems to be supported when they run a slightly larger pile of  samples. So, in this case it looks like a win! 

Lyse your cells in 100% formic acid to get 40% more membrane proteins??

Okay. Chemically I do not understand why this would work. That's okay, I don't understand how a lot of things work....

They start out by laser capture microdissecting a pretty big area of cells. Something like 500 micron x 500 in 15 micron cuts. And you stop wondering why the cut is that big when you see the digests are loaded on a Q Exactive Classic. Man, I love those old things, but proteomics has moved forward a good bit since 2012. 

They seriously just put 5 microliters of 100% formic acid onto the slices and dry it off by speedvac and then LysC digest for 3 hours and then trypsin digest overnight. Big boost in membrane protein IDs over just DDM lysis alone.

Then they repeat it with smaller sections using an Astral. There might be two rounds of this to get to very small cuts. This ends up letting them see everyone's least favorite super important proteins, those awful SLC things. And it isn't a small list. They pick up several of them. 

If you are also struggling with low input preps under-representing membrane proteins, this might be worth checking out. I suspect you probably have LCMS grade formic acid sitting around somewhere. 

Multi-omics day on-site at Waters corporation!

 

Hey proteomics people! Did you know that Waters makes mass spectrometers? They totally do and, while I'm not at all sure how this happened, I got invited to drop in and meet scientists and see some talks and touch some instruments. 

I'm pretty sure I got to visit because of some impressive numbers out of the P10 instrument which looks like a compelling piece of proteomics hardware. I'm trying to talk them into running some dumb stuff on one of them for me. The day featured a small group of people with very different mass spectrometry interests getting to wander around the demo labs and touch things and see the Cyclic MS thing (that I know Padula has and likes, but I wasn't really sure what it was) and lift the MALDI and DESI sources that you can just move from one instrument to another. Which, I can't remember anyone else doing, but they were just sitting there.  

There were also fun customer talks from this guy https://info.liningtonlab.org/ at Simon Fraser. 

Did you know there was an effort to compile all identifications of all natural products on earth? And maybe if you had all of them together you could look to see if evolution has biases toward certain chemical scaffolds - which, it absolutely looks like it does. Super ridiculously cool talk. He did admit that most of the natural product efforts have probably not been published, but they've got 37,000 of them all pulled together. 

You should 100% check out the Natural Product Atlas here. 

Eric Gier, a grad student at in Facundo's lab at Georgia Tech showed his work using cyclic ion mobility to resolve oxylipins from brain injury models. Apparently which of those isomers is building up has a lot to do with how your brain recovers from impacts. Which makes the 100-150ms he showed he needed to really pull them apart into separate peaks not seem like a very reasonable use of time. 

While trying to figure out my hand writing and misspelling of Eric's last name in my notes, I stumbled across this JASMS paper featuring open source software for integrating data from the ion mobility roller coaster. 

One takeaway from the lab was that Waters instruments are a lot physically larger than I expected. The cyclic ion mobility system I saw was easily Astral size, but the new one is a benchtop. A very big benchtop. But they've clearly got their niches for these hardware. Moving native proteins and isomeric drugs around, purifying them by doing multipass IMS and then kicking out the ones they want for MS/MS? That's all stuff I saw people do and things I can't think of anything else that could do. 

Wait. Are we doing this wrong? An argument for SUPER MASS SPECTROMETRY!

 

This is an interesting one. 

What if it wasn't the most efficient to have proteomics mass spec instruments EVERYWHERE? 

Rather, what if mass spectrometry proteomics was, instead, a fully centralized system where each sample could be measured comprehensively on multi-capacity hardware without all the compromises that come with solo 150kg - 400 kg vacuum chambers? 

In the simulations presented here it seems to make a lot of sense!  

The author makes the argument for splitting the ion beam from a single sample and routing it through to multiple mass analyzers simultaneously. Considering this from the company that can route ions around a lab with lenses that resember a garden hose. This is not an endorsement, but you might not know that the Spion is a thing, and it's crazy. https://www.tracematters.com/technology

The biggest proteomic human atlas yet should reduce our reliance on transcripts to determine specificity!

 

I won't like that picture freaks me out a little, but here we are.

2,800 human samples analyzed at incredible depth. 

You have to be asking yourself that with a dataset this large and comprehensive, clearly we'll know for sure that the correlation is between transcript abundance from GTEX and this, right? 

Uuuuummmm..... yeah, it's...the same....though that 0.105 from the pancreas is....special.... (this is from Supplemental Table 4). There is SO much data to dig through here. 

I found out about the paper (how did I miss it??) when I stumbled onto this. 

Bruker is boosting customer support and hiring.

Today I had the privilege of meeting with some fancy people from all over the place to discuss some concerns I have voiced and/or amplified about Bruker Daltronics support system. My concerns largely centered on what looked like a model I've seen before where an LCMS business is the cash cow (wtf is a cash cow?) for a big company to buy a lot of dumb shit. Who cares if a company uses their profits to buy other little companies? Not me, as long as that doesn't mean you lose all your field applications scientists or field service engineers for the stuff that you've already sold. 

It is possible that I over-reacted. It would absolutely be the first time that has ever happened, however. As anyone who has ever worked with me can attest to 😇. But it's also possible that the mass spectrometry at the company isn't in charge of the whole thing and that sometimes there are corporate level shakeups where you have to do what you are told and deal with the fall-out. 

Shortly after my ASMS possible over-reaction we had one of their top engineers on site followed by the amazing Dr. Josh Beri that most of us know who made sure our Ultra2 was back to pumping out world class data. To be fair I should have probably updated a really negative post yet again and I'll link it back to this one. 

More importantly, I have independently verified that what I heard from said nice fancy people that they are actively building up the LCMS support network. At least one person I got used to emailing when we were stuck will be back in a couple of weeks and there are a pile of open field service jobs at least around the US. I've screenshotted some of those above. Which, given, the fact the economy in the US is rooooouuugh right now, sounds both very positive and hard to do. As an aside, the actual Bruker career page is a silly mess. Support jobs are distributed between the categories of Technical support and sales/support and you have to actually look at "ALL" to find the open applications jobs by manually going through. Interestingly it maxes out at 3 pages. So you might have to go through several iterations to find that there is a job posting on their website. I've seen worse. 

I'm not sure I've seen any openings for field applications scientists while digging around, but there was a specific TIMSOmni Apps job on Linkedin that appears to be closed and possibly filled so I'm optimistic.

New people I've met from Bruker now have been legit as well. There are some serious key hires who are accomplished mass spectrometrists who are currently adjusting to their roles and willing to get on planes to talk to customers even if they're weird bloggers in small cities who were kindof a lot mean to them right when they started their jobs. 

Summary - science business stuff is hard. My TIMSTOF is back up and generating really good data and I am a lot more confident and happy with my customer experience right now than I was a couple of months ago. 

If you are wondering if there are people standing in my office making me type thiiiiiiiii HELP sdlkjdsfkjldasfljk980909098890098fjljdsflkjdsflksfjkfdaslkas;dlkjsdf lkhjasdfljksdf 

That's a joke. I wrote this because I felt like it even though it was very very boring to do. 

Defining quality standards for single cell proteomics OR Astral vs TIMSTOF Ultra2

 

Okay. So what if you sent the same single cells out to a whole bunch of labs that agreed to send the data back?

Obviously no one would want to play unless they had the very best stuff, right? Or maybe you had to have the stuff to play? I don't know but this is super cool. 

Now....I am a little confused about the HPLC setup. It looks like they used a bunch of instruments but there is exactly one NanoLC method in the paper. I know people who like the Neo enough that they have their LC on one computer and their MS on another. So maybe that's what happened here? 

It's a bunch of labs so you do get to see lab to lab variability on the same instruments as well. 

NanoLC dead volume calculator since I can't find the MSBioworks one!

 

This is a nanoLC volume calculator inspired by the AMAZING (but dead?) MSBioWorks phone application I miss a lot and made with the frustrating assistance of some AI thing my university lets me use. It's honestly pretty good at color palettes, even if it's bad at math and confident/arrogant about it.

nanoLC System Volume

To use this, I've assumed some things about your flow path. You'll need to select how many columns and tubing sections. Let me know if I'm not considering modern nanoLC options, yo. It'll autofill some normalish stuff, so you'll need to fill in the inner diameter and length for each. It sums everything into total system volume live.

00 Set Up Your Flow Path

How many columns?

Quick select

How many tubing sections?

Sections 02 and 03 below build themselves to match — enter the inner diameter and length for each capillary as it appears.

Line weight and column shading scale with each segment's volume — thicker lines and denser fill hold more mobile phase.You'll have to believe me because I can't figure out why these fonts are OUT OF CONTROL

Total system volume

0.0 nL

01 Injection Loop 0.0 nL

Loop volume (nL)

Inner diameter (μm)

Length (mm)

Computed volume

0.0 nL

02 Tubing Sections 0.0 nL total

03 Columns 0.0 nL total

04 Additional Dead Volume 0.0 nL total

— Breakdown

Internal Dead Volume Injection loop + pre-valve tubing

0.0 nL

0.0% of total

Operational Dead Volume (Probably the one you care about) Valve → columns → MS inlet

0.0 nL

0.0% of total

Component	Zone	Volume	Share	%

How this is calculated (shoutout to Cheryl Reger for somehow being a legit geometry and algebra2 teacher in one of the single worst school districts in the United States) It's all middle school geometry, yo!

Cylindrical volume — tubing, loop, and columns are treated as cylinders:
V = π × (ID / 2)² × L

Column void volume — the mobile-phase-accessible fraction of a packed bed:
V_void = V_geometric × ε_t   (ε_t ≈ 0.6–0.7 for fully porous particles)

Total system volume — sum of every contributor along the flow path:
V_system = V_loop + ΣV_tubing + ΣV_column + ΣV_fittings

This is the volume relevant to gradient dwell / delay volume and extra-column band broadening — figures depend heavily on real fitting and frit geometry, so treat this as an estimate to refine against a manufacturer's spec sheet. Also...there isn't really a "Zero Dead Volume" union. There has to be a little gap.

AI in proteomics - from protein identification to virtual cells!

 

Man, I need to start being more collaborative so I can get on these gigantic papers that are going to get a billion lazy citations (that's where you use the title and abstract to make a citation without ever reading the paper).

This new one is definitely going to get cited like crazy and since there are people out there who judge you by how many citations you have, each of these people are going to get supercharged CVs! 

You don't have to read it, you can just guess what it's about and when you need a citation number 68 for deep learning you can just pop this one in. 

There are SO MANY obvious uses of AI out there! Check out this month's cover of the Journal of Proteome Research (real, not edited) - it's killing me. 

Wait. So....WHAT IS IN THE "S-TRAP?" It's a lava lamp shaped like a 1980s Bosch automobile fuel filter! AND the EvoTip loading is a syringe that drips Play-Doh(TM) onto the top of an oRsBiTraP? 

I posted this on the socials somewhere else, but it's similar to this used TIMMYTOOFY SCeePea system from Booker Caltronics. 

Objectively, $350k is pretty legit, but you do have to wonder why the decision was made to like not...take a picture of the system.... 

Proteome Discoverer 3.3 has a DIA-NN (commercial license required) node!

 

I was trying to help someone out who, due to the fact the US government is working great, has instruments but can't pay their annual license fee to use the DIA cloud nodes. I was like - how hard could it be to link PD through the node maker things to DIA-NN and then integrate the whole thing.

Up until the .parquet thing appears it looked pretty straight-forward if you're smart, but I'm not and it all looked like it was going to work, but wasn't so I decided to blame it on 1) whatever I was actually supposed to be doing 2) the parquet thing and 3) and the fact that after a certain version, DIA-NN isn't entirely open access. It's free but there are permissions to consider. Yeah, that's totally what it was. One or all of those. 

Turns out there is already a DIA-NN node released in 2026. That was last year? Nope! This year! It's still 2026!

You can watch a webinar about it here by the amazing Michaela Scigelova and that DIA-NN guy. 

Now...obviously...it only accepts .raw files. But why would you run anything else in Thermo's solution for Protein Informatics? Weirdo. 

O-Link identifies ALS phenoconversion markers!

 

There is a LOT to be excited about in this new study. 

This statement at the end of it, is....maybe not one of them.... but they specifically limited access to these data in the IRB documentation, so even if you wonder what personal information could possibly be contained within a targeted antibody panel, it's against the rules to make this publicly available. 

However - here we are with what might be the first win in ALS diagnostics with a proteomics technology in....ever....? 

Surprisingly, it's actually a small cohort. It's less than 70 individuals? But with a pile of blood / plasma draws so it works out to ~550 samples. So....also evidence that O-Link Explore targeted panels can provide statistically valid data without an n>1000? 

The end point isn't a new ALS diagnostic marker. It is, however, a panel of proteins that may allow earlier diagnosis. And maybe if it is caught earlier then there are ways to treat it and prolong lifespan and (as/more imporant) healthspan in this fucked up awful disease. 

Proteomics takes on the brain decay paradox! The what?!?

 

So....ummm...this is a thing....

...a gross thing....

I know a lot of people who worry about stuff like "post-mortem degradation" which is where you might not want to do proteomics on someone who has been dead and puppeted around for a month in the hot sun. 

However - and this is where it gets weird - sometimes the most fragile and easy to homogenize part of the body (the brain) ends up around IN ANCIENT BODIES. What? Gross? YES! 

I'm not sure I'm going to be able to eat after this. Maybe I'll stop. 

Meh. According to these people over 4,000 ANCIENT BRAINS have been recovered from the past. Some as old as 12,000 years??? Which sounds impossible. My postdoc advisor would time me getting brains out of mice because she was sure we wouldn't care about the proteins in the cells in the subventricular thingamabob if we didn't do it fast. 

It turns out that if you have the right conditions the brain doesn't just break down. Why? When? Proteomics and a very icky experimental design to the rescue! 

I can't spend a lot more time on this, but you should check it out. There is some DDA proteomics and some DIA and two different instruments AND there are proteins that last a LONG TIME and these authors work out the biochemical reasons for it! Super cool stuff.