Skip to main content


Screening, Within and Without

So I’m a small-molecule drug discovery person at heart, since I started out as a synthetic organic chemist. Talking about vaccines and antibodies nonstop is a bit of a strain, then, because immunology is full of stuff that’s rather far removed from good ol’ small-molecule ligands. Actually immunology eventually wears out everyone. Even immunologists. It just keeps on going: detail on top of detail, layer upon layer of complex, interconnected, rococo feedback loops and backup systems, repurposed evolutionary holdovers, insanely subdivided cell lineages, all the rest of it.

But in the end, even a medicinal chemist starts to feel at home in some parts of it. Because one of the key events in the adaptive immune system is exactly like one of the key events in so many small-molecule drug research programs: high-throughput screening. I know that I have a lot of people reading this who have never done any such thing, so it’s worth some explanation. Now, in traditional med-chem and its allied fields, HTS has been a way of life for around 30 years now (before that it was low-throughput screening!) It’s had a lot of refinements and extensions, but the fundamental process has not changed one bit: round up some of your biological target (an enzyme, a receptor, an interaction between two proteins, whatever you’ve got), and find some setup where you can get a signal when a compound does what you want to it. Do you want to set off a signal from a receptor? Keep two proteins from coming together? Gum up the works of some enzyme? Work up some assay system where when that desired event happens, you get a signal that come up out of the noise.

It could be (and often is) a flash of fluorescence at a particular wavelength. That idea has had plenty of changes rung on it – maybe you’re looking at two wavelengths at the same time and checking the ratio between them. Or you can set things up where it’s the polarization of that fluorescent light that’s telling you something. Maybe you’re running things in a way where the fluorescence is already going on and your desired stuff will shut it off. You can have fluorescent probe compounds; there are whole careers involved with those things. Then there are of course fluorescent proteins (like the famous Green Fluorescent Protein), and listing the ways that those are used in assays would take us the rest of the week. There’s are physical phenomena that will allow fluorescence to be set off only when two suitable species are close enough together in space, and as you can imagine, these have been turned into workhorse assays.

Luminescence can be used in similar ways. The enzyme luciferase has been exploited six ways from Sunday, with a long list of variations. Any good screening lab has a list of their compounds that have been shown to directly interfere with the enzyme (and can thus hose up your data). We love luciferase so much that we use both the firefly kind and one from a deep-sea critter called a “sea pansy” because they come in both green and red versions (and sometimes you use both of those at the same time, too). That same sort of “only when they come close together” trick can be worked with luminescence, too. No. there are all sorts of tricks you can play with funny wavelengths of colorful light, and all sorts of interesting ways you can set up assays to emit them.

But it doesn’t have to be light. A classic way to do receptor assays (and others) is to have a radioactive compound already stuck to your target. You add some of your test substance and see if it kicks any of the hot stuff off, which you read off after a filtration step: the more radioactive your filtrate, the better your compound bound to the target. You can mix radioactivity and a light-driven readout, with a scintillation proximity assay: a radioligand that gives off a beta-particle lights things up only when it’s next to your target, and not under other conditions.

No matter how you run any of these, the idea is to get such an assay set up so that there’s a high signal-to-noise, low chances for false positives and false negatives, no need for too many picky dispensing or mixing steps to get it running, and (especially) the ability to run at the smallest scale you can handle. I well recall when doing such things in assay plates with 8 wells by 12 (96 total) was considered kind of high-tech, but for a long time now any assay that can’t be shrunk down to 384 wells (16×24) or 1536 (32×48) has been greeted with a weary sigh and a roll of the eyes. There are legions of commercial assay kits and associated dispensers, plate-handling robots, and plate-reading machines that will assist you in getting all this going.

You want this miniaturization, of course, because you want to put the HT in the HTS. A full-deck screen at a big drug company can be a few million compounds if you pull out all the stops (which we don’t so often any more, to be honest), and doing that 96 wells at a time will take you a while. Especially when you consider controls and duplicates, and you’d better if you don’t want to seriously waste your time. No, if you’re going to screen a really impressive set of compounds, you’re going to need all the help you can get. The latter 1990s and early 2000s in the drug business was a period when the high-throughput screeners and the combinatorial chemists tried to outdo each other

And here’s where we get back to coronaviruses, antibodies, and vaccines. Because high-throughput screening is what goes on in your bloodstream constantly, as your adaptive immune system watches for pathogens. Everyone carries around a huge variety of different antibodies, all of which fit into some basic structural templates. No one is quite sure just how many different antibodies a person has, actually, because a real count is just an overwhelming task: the usual guesses are in the tens of billions, hundreds of billions, maybe a trillion different ones, which is a hard number to grasp. Imagine a company with a million compounds in its screening deck (OK), and then try to picture a million such companies (nope, not happening). And it might be a lot more incomprehensible than that – the first link in this paragraph will take you to a paper that estimates that the available diversity for circulating antibodies is on the order of 10 to the 18th, and what’s yet another factor of a million between friends? Given the combinatorial possibilities, it is beyond certain that no two humans have ever had the same repertoire of antibodies, and that no two humans ever will.

These things are floating around in your blood and being displayed on the surface of your various B cells (a mere hundred thousand or so per B cell), just waiting to see if there might someday be something that they bind to. They’re just like a compound collection in a drug company, actually, except they’re not being stored in separate vials or wells, but are rather dumped all together in your bloodstream. And the assay conditions have been long worked out by evolution, with a huge signal/noise: activation of complement or of the various types of effector cells that respond so dramatically to the presence of antibodies bound to an antigen target.

There’s actually a screening method used in drug discovery that’s broadly similar to this: DNA-encoded libraries. That’s where you build up a huge set of small molecules, each of which has its own DNA “bar code” attached to it. You screen these all at once, too – you can hold up a small plastic Eppendorf vial that has (easily) tens of millions of different compounds in it, each with its own DNA identifier. You run the assay in a way that you “pan” for the potent binders, washing the less potent ones out of the system. Then after you finally knock the strong binders off with stronger conditions, you amplify and sequence their DNA barcode regions and get a reading of what small-molecule structures they must have been. That makes it sound relatively straightforward, but doing this right takes a good deal of care and a lot of data analysis of the hits at the end.

But you can see how it works in a roughly similar fashion to the way native antibodies work in our own bloodstreams. You start with a large variety of compounds, all mixed together, and you try them against a given target. When one of them binds strongly, you use some sort of amplification to pick out this rare event which is present in extremely small concentration. With DNA-encoded libraries, it’s getting rid of all the weaker binders by washing them off, and using PCR to make far more copies of the DNA sequences you have left. In the immune system, the amplification is built into the cellular responses, which propagate strongly through the immune system and set off further responses in turn.

The immune system is a lot more impressive, of course. It’s had a long time to get better, under constant whole-organism threats of illness and death and subsequent inability to reproduce. The screening collection in any single human’s bloodstream is far larger than any human efforts have ever reached, so the amplification of any given binding signal has to be a lot more robust. That’s been under serious selection pressure as well, of course: if we pick the wrong compound in a screening effort we will waste time and money, but if the immune system picks the wrong antigen to go off on, it can start attacking your own body’s tissues and kill you. The amount of infrastructure that’s been built up over the millennia to avoid that is pretty intimidating all by itself.

Note that we wouldn’t even be able to do the DEL trick without piggybacking on all that evolutionary work, either. The ligase enzymes that allow us to build up the DNA barcodes and the PCR that lets us amplify the DNA at the end – yeah, we stole all that and repurposed it, and there’s no way that you could ever get things to work without them. But it’s just like the fluorescent proteins mentioned above – molecular biology and chemical biology depend on being able to repurpose the amazing array of tools found in living cells. We’ve added plenty of our own technologies as well. No living organism does anything like electrospray mass spectrometry, Förster resonance energy transfer, NMR, surface plasmon resonance or the like. But combining what we’ve learned of chemistry and physics with what evolution has developed, you can run some pretty fancy systems, which are getting fancier all the time.

But for sheer library size and ability to pick out hits, nothing we humans have been able to put together rivals the adaptive immune system. Perhaps that’s been one minor side effect of the pandemic: people who have taken the time to learn a bit about immunology can only come away with a sense of awe when they start to see the huge panorama of what’s been going on inside their bodies for every second of their lives.

29 comments on “Screening, Within and Without”

  1. JB says:

    If DELs could be subject to mutation, conversion to new slightly different structures, and subsequent rounds of enrichment you might get the efficiency of the immune system. But because the process of linking the DNA tag to the chemical structure is synthetic chemistry, not enzymatically encoded like the ribosome does, it can’t easily be done. The Schultz and Liu groups have taken some stabs at it but nothing that’s been a slam dunk improvement.

    1. Sanchopanza says:

      That is exactly what mRNA display, like the approach developed by Hiro Suga and commercialised by PeptiDream, gets you – a macrocyclic peptide full of unnatural amino acids, linked to their encoding mRNA. They are amenable to several rounds of enhancement and selection for better binders, and in the end you get picomolar affinity and compounds that disrupt even challenging protein–protein interactions.

      The disadvantage is that they are small but not really small molecules, so rarely cell permeable. But immune system like efficiency is definitely there.

    2. ScientistSailor says:

      Wasn’t DiCE trying something like that?

    3. K says:

      See Harbury lab and companies originated there.

  2. Jonathan says:

    The prospect of 10 to the 18 different distinct antibodies made me think. That would imply there are fewer than 20 each of each antibody molecule type in a human body. It would be an extraordinary screening feat even to reliably expose the right ones to a pathogen let alone to amplify them as necessary.

    I think there are fewer antibodies coded for than that (by a mix and match process of gene assortment) but those that pass the first “screen” are then subject to a mutational maturation process which generates multiple variants some of which have increased binding affinity.

    (Is that a recursive process that could be adapted for small molecule drug discovery?)

    1. sgcox says:

      Worse than that, it would imply there is at least one B cell encoding each antibody. To have them all in one human body would probably make a blob the size of the moon…

      1. Nesprin says:

        According to wolfram alpha 10^18x 5.5um diameter b-cells would have the volume of a blue whale, or 1.2 freight containers.

        1. Marko says:

          I’ve seen pictures of people like that. All this time I’ve been thinking they were just fat.

    2. ScientistSailor says:

      the *available* diversity is 10e18. That is the potential, obviously, nobody has that many different abs…

  3. Simon says:

    Wouldn’t identical twins, being natural clones have the same ab repertoire?

    1. sgcox says:

      No, they have identical starting material but then there is random shuffling as foetus grow.

      1. Simon says:

        Interesting! How does this shuffling occur?

        1. JB says:

          It’s called V(D)J recombination and it’s a pretty insane process- basically enzymes intentionally break your DNA at signal sequences in the immunoglobulin genes and randomly paste it back together. There’s a Wikipedia article summarizing it.

        2. Kaleberg says:

          That’s the whole magic at the heart of the combinatoric immune system. There’s an element of chance. The odds of two identical twins drawing the same sequence of cards aren’t all that good, even if they start with the same deck.

  4. Philip says:

    Derek, you left out my personal favorite, melanophores. I don’t know if anybody uses them anymore. I have been out of the HTS business for a long time.

    Arena used them for a long time, and may still. A very large pharma also used them, but I am not sure that is public knowledge. I had nothing to do with developing the experimental system, but I had a lot of fun getting it to work at a larger scale.

  5. Metacelsus says:

    “No living organism does anything like . . ., Förster resonance energy transfer.”

    Isn’t FRET involved in light-harvesting complexes for photosynthesis? See:

  6. Martin Bonner says:

    Actually, if there are 10**18 different repertoires of antibodies, then at about 10**9 humans, there is a 50-50 chance that there exists a pair of humans with identical repertoires. This is known as the birthday paradox (after the somewhat surprising fact that you only need a group of 26 people to get better than even odds of having a couple who share a birthday).

    Given that there are about 7 billion people alive today, and that the basic calculation assumes that each repertoire is equally likely (whereas in fact, I would expect similar genetics and similar environments to produce similar repertoires), the chances that there *aren’t* two humans with the same repertoire of antibodies is pretty small.

    1. Marko says:

      Now there’s an idea for a new business : ” Find your true soul-mate with Antibody Match! “

    2. Amigo, the combinatorial problem is 10^18 _choose_ a trillion or so. There aren’t 10^18 possible repertoires, there are (10^18 choose 10^12) repertoires, which is an astronomically large value. (Just look up any decent combinations calculator). The birthday paradox falls waaaay short here.

      1. Oudeis says:

        For once “astronomically” is selling things short.

        I’m as far from being a combinatorics expert as you can get while still knowing the word, but I believe “1000 choose 100” gives you an answer that’s about 10^40 times as great as the number of atoms in the universe. “10^18 choose 10^12” would require up-arrow notation or some other exotic way of writing huge numbers–scientific notation isn’t going to cut it.

      2. Martin Bonner says:

        Ah. Right. I hadn’t appreciated that. That does reduce the chances of collision a little! Thanks for the correction.

        1. Martin Bonner says:

          Actually, it’s not _quite_ as bad as I feared. If you expand nCk, apply Stirling’s formula (ignoring the O(log n) terms because those are simply too small here), and simplify by assuming that for n>>k, log n ~= log n-k, the log of nCk comes out at = k (log n – log k + 2). Plugging in 1E18 and 1E12 as n and k respectively (and using logs to the base 10), the log ends up as 8E12, so the actual value is:
          1 E 8000000000000 (or so).

          You wouldn’t want to write it out in base 10, but there’s no need for arrow notation.

          1. Oudeis says:

            You lost me, but thanks!

          2. Oudeis says:

            Okay, I thought through it, and I think I’m following a bit better. Again, thanks. How often do you get to say that you badly overestimated something and the right answer is only ten to the eight trillion?

      3. Marko says:

        So much for my new business idea. My investors are gonna be pissed! I’ve already spent their money.

  7. Konstantinos Spingos says:

    Immune system is like a second brain as it is a decision organ trading between specificity and sensitivity, as well as between inheritance and adaptiveness.

  8. Art Sciman says:

    Hello Medchem, my old friend, come to speak with us again. A few liberties duly taken with old friends S&G. Fans and non-fans alike please read on.

    (In memoriam SDM 1937-2021)

    Hello Medchem, my old friend
    Come to speak with you again
    Because a crystal softly creeping
    Left its seed while I was sleeping
    And the lead that was planted in my brain
    Still remains

    New Work, to that small molecule I come
    Flyin’ into the lab to get to the hood
    New Work, lookin’ down on the central ring
    Which the old lags say just ain’t the thing
    New Work, you’re real enough to me, there’s a heart
    A heart that lives and beats in New Work

    And I may be obliged to defend
    Every data point, every property
    For there’s never no obligations
    Never every reason to believe
    We all will be received
    In grace and favour land

    Cee CeeLogP, you’re breaking my heart
    You’re shaking my confidence daily
    Oh, Cee CeeLogP, I’m down on my knees
    I’m begging you please to come down
    Cee CeeLogP, you’re breaking my heart
    You’re shaking my confidence daily

    Years, or even decades, later…
    …If you’re one of a lucky few

    Sail on silver drug
    Sail on by
    Your time has come to shine
    All your dreams are on their way
    See how they shine
    I’m sailing right behind

    Medchem would not be convicted
    By a jury of mechem’s peers
    Still crazy after all these years
    Oh still crazy
    Still crazy
    Still crazy after all these years

    “D*****”, I said as we got in the Chevy in Nutley
    “Mountain View seems like a dream to me now”
    It took me four days on Eighty from Rahway
    I’d gone to look for America…


    …A long lost lyric, thought to have been penned by a long ago high school kid named Simon Paul, from Queens, New York, after a week’s work experience placement in the summer of ‘fifty seven, across the Hudson from New York, at Roche Laboratories, Nutley, New Jersey.

    Coincidentally the same summer Dr Leo Sternbach and co-workers did the legendary lab tidy up that unearthed the “few hundred milligrams of a white, nicely crystalline solid,” that, after two short hard years of animal experimentation and clinical trails, emerged from genesis in round bottom flask, primary rat locomotor test and chemical structure reassignment, as Librium, the first drug of a class that later became known as benzodiazepine receptor antagonists.

    No record exists of Simon Paul’s week with Roche, but a co-worker of the late Dr Sternbach, now aged ninety three, did recall in a recent email entitled, “My Ro 5-0690 of long ago,” that although the young Simon Paul didn’t fully get where all those flasks jamming up the lab had come from, or were headed to, from the moment he walked through the lab door the young man did seem to know instinctively that organic synthesis and medicinal chemistry were, back then, as much art as science.

    Genesis of Librium and Valium now two and more generations of synchemists and medchemists ago. World moved on. Internet tells Roche vacated campus at Kingsland St, Nutley, in 2012 (and Hillview Avenue CA94304 around 2008).

    Maybe synchem and medchem now become out and out sciences, just as immunology, virology and epidemiology generally portrayed, fallout of which has so dominated life, death, birdsong and headlines this last year.

    Maybe art left to dreamers, lyricists and dissenters. To maybe or not to maybe, may no longer even be the question?

  9. cynical1 says:

    As a former “small-molecule drug discovery person at heart” who played in the realm of immunology during my tenure, I enjoyed this post. Many thanks. And so refreshing to not have the armchair immunologists posting for a change.

Leave a Reply

Your email address will not be published. Required fields are marked *

Time limit is exhausted. Please reload CAPTCHA.

This site uses Akismet to reduce spam. Learn how your comment data is processed.