In my last essay, I spent a few thousand words cribbing about how most organoid work feels literally incapable of leading to anything useful. This was a lot of fun, but despite all of their issues, it’d be a lie to say that the sum impact of organoids has been zero. In truth, they’ve played a significant role in the stories of some scientific triumphs—at least according to papers—and it can be a fun exercise for us to dissect them.
What are these triumphs? And how accurate is the usual telling of them? All this, and more, is the subject of this essay. We’ll start with the most obvious settings, and slowly work our way to the stranger applications.
The success stories
Identifying which cancer patients will benefit from chemotherapy
This one is pretty straightforward.
How does chemotherapy work? Nearly all chemotherapies—paclitaxel, cisplatin, doxorubicin—work by futzing with the interior of individual cells. Paclitaxel jams the microtubules, cisplatin crosslinks the DNA, doxorubicin wedges itself between base pairs, all of which are meant to prevent the damn thing from dividing. Terrible news for a cancer cell, also terrible news for every other cell, but the net impact of it all is fine since most of your cells aren’t constantly dividing. Which is to say: chemotherapy acts via a cell-level knob, the level at which organoids are as good as any other in vitro system.
So it should not surprise you that this is precisely where organoids have posted their least ambiguous wins. The foundational result is a 2018 Science paper, ‘Patient-derived organoids model treatment response of metastatic gastrointestinal cancers’. The authors built a biobank of organoids from metastatic, heavily pretreated colorectal and gastroesophageal cancer patients. The authors then screened a library of 55 chemotherapy agents against the organoids and compared the results to what actually happened to the patients—who, importantly, had been enrolled in phase 1/2 trials, so there was a real clinical answer key against which to grade the organoids.
The concordance was startling: 100% sensitivity, 93% specificity, 88% positive predictive value, and 100% negative predictive value.
Holy shit, right? Yes, it wasn’t perfect-perfect across the board, but it did meet perfection on the last value: identifying non-responders. A 2019 Science paper found a similar result for irinotecan, another chemotherapy, in metastatic colorectal cancer. Chemotherapy isn’t traditionally considered precision medicine, but there is known heterogeneity in patient response, and being able to predict it should theoretically allow us to spare millions of cancer patients from a toxic drug that will functionally do nothing for them.
This is cool work, but it’s also so within the realm of ‘duh, you’d expect organoids to do this well’ that it probably isn’t worth discussing much further. Instead, let’s consider two likely follow-up questions you have.
First, was the organoid bit really necessary? For that matter, what does a tumor organoid even look like? Isn’t it just a blob of tumor cells? What organ-specific function would there be? So…isn’t a ‘tumor organoid’ an oxymoron? Cancer spheroids exist, sure, but surely nothing more.
It’s a good point. To answer the more basic question: a cancer organoid inherits its organoid-ness from the organ the cancer is a cancer of. Which is to say: a colorectal cancer organoid looks an awful lot like a typical colorectal organoid, just with messed up internals—a fact that the authors of the aforementioned irinotecan paper demonstrated. So yes, tumor organoids are a coherent concept.
But was the organoid part actually needed? Would a simpler spheroid not have done the same thing? In fact, it did.In a 2019 Scientific Reports paper, a spheroid made up of just dissociated ovarian tumor cells produced the following result across 44 patients: 89% accuracy, and all 32 ovarian cancer patients it identified as responders went on to respond to the tested chemotherapy.
Hmm. We’ll come back to this.
Moving on to the second question: if organoids nail chemotherapy response this cleanly, is this part of standard oncology care?
It is not. And somebody has tried: the authors behind the irinotecan result ran a prospective feasibility trial, concluding that ‘patients that received organoid-informed treatment did not experience clinical benefit.’
Strange! Why is this? Well, they do note that they could establish organoids from only 57% of patients, and it took about ten weeks from tissue to readout. You could imagine the problem here is the ten weeks bit; by the time your organoid is ready with a treatment recommendation, the cancer has mutated to an entirely new state. But the fun part about spheroids is that they are much easier to grow! The aforementioned spheroid paper spat out its verdict in seven days, not ten weeks, and spheroids are almost certainly easier to establish.
But…spheroid-guided chemotherapy selection is also not part of standard oncology care. Do we just need to run a trial here? Is the cancer cabal keeping this technology away from patients?
Maybe. Alternatively, the answer to this quandary may have nothing to do with organoids or spheroids, but rather a simple fact about medical care: being able to identify whether a patient will benefit from a therapy in advance is only useful if there’s something better on the table. Give chemotherapy to a patient, and maybe it’ll help. But if you don’t give chemotherapy to a patient who you don’t think will respond to it, it is not as if we have some secret chemotherapy that will work for them—at least not one that plops out of the organoid screening work. You’ll save them from needless suffering, yes, but that typically isn’t the goal in life-or-death situations like this.
But who knows what shape the future will take? Maybe we’ll start to extend outside of chemotherapy. A particularly interesting bit of research towards ‘arbitrary cancer drug modeling’ comes from Orakl Oncology, a French biotech.
In April 2026, they put out a preprint titled ‘SCOPE: Integrating Organoid Screening and Clinical Variables Through Machine Learning for Cancer Trial Outcome Prediction’. They do focus mostly on chemotherapy, yes, but also include in some things outside of that (ADCs and so on). And curiously, instead of asking ‘will this patient respond to this drug?’, they instead ask ‘will this trial arm hit its endpoint?’. To simulate a trials patient cohort, the trial’s eligibility criteria—cancer type and any stated molecular-subgroup criterion, like PDAC and KRAS-mut—is matched against a pre-existing biobank of organoids derived from 78 patients, and the matching ones had the associated trials drug screened against them. No fresh tumor tissue required!
What were the results of this? At first glance, remarkably good. Predicted progression-free survival (PFS) correlated with known outcomes at R² = 0.85—substantially better than the R² = 0.59 achieved by regressing over basic clinical data like median age/ECOG score/others—and the platform correctly picked the winner in 7 of 8 head-to-head arm comparison. Separately, it also predicted that the famous daraxonrasib would outperform chemotherapy before the famous phase III RASolute 302 result became public.
Still, you should exercise an abundance of caution when looking at work like this. A close reading turns up several details that are a bit strange. For instance, one of the eight included comparisons found no statistically significant difference between the treatments, but the paper still designates the arm with the numerically longer median PFS as the winner. There are other debatable bits like this, but it is difficult to begrudge the work too much. Forecasting trial outcomes is extraordinarily annoying to do entirely rigorously—I know personally!—and some tea-leaf reading is always needed. Excellent paper, and I hope more come out in the future!
Either way, the moral of the story here is that while organoids may turn out to be a poor instrument for deciding what to do with your cancer specifically, they may be a decent one for deciding which parts of a biotech’s portfolio are most worth focusing on. Neat!
Identifying which cystic fibrosis patients will benefit from treatment (?)
Let’s move on to something a bit weirder.
You may recognize cystic fibrosis as that nasty genetic disease that causes its sufferers to produce especially thick mucus, leading to a host of horrible downstream consequences, including recurrent lung infections, malnutrition, and interestingly, infertility. You may also recognize cystic fibrosis as one of those diseases that modern medicine has squarely triumphed over.
Imagine, though, that you had never been able to simply breathe. Imagine that mucus—thick, copious, dark—had been accumulating since the moment you were born, thwarting air and trapping microbes to fester inside your lungs. That you spent an hour each day physically pounding the mucus out of your airways, but even then, your lung function would spiral only downward, in what amounted to a long, slow asphyxiation. This was what it once meant to be born with cystic fibrosis.
Then, in the fall of 2019, a new triple combination of drugs began making its way into the hands of people with the genetic disease. Trikafta corrects the misshapen protein that causes cystic fibrosis; this molecular tweak thins mucus in the lungs so it can be coughed up easily. In a matter of hours, patients who took it began to cough—and cough and cough and cough in what they later started calling the Purge. They hacked up at work, at home, in their car, in bed at night. It’s not that they were sick; if anything, it was the opposite: They were becoming well. In the days that followed, their lungs were cleansed of a tarlike mucus, and the small tasks of daily life that had been so difficult became unthinkingly easy. They ran up the stairs. They ran after their kids. They ran 10Ks. They ran marathons.
Trikafta really is a miracle. It genuinely cured millions of people who would have otherwise suffered through a painfully short lifespan, returning them to an otherwise mostly healthy state.
Unfortunately, it does not work for everyone. The essay goes on to explain the heartbreaking reality that 10% of cystic fibrosis patients lack the necessary genetic mutations for the drug to work correctly. For these unlucky few, their community has moved on, living normal lives, but these 10% are still stuck with the disease.
Before we return to this 10%, we should first understand how Trikafta works. Cystic fibrosis is caused by a genetic mutation in CFTR, a chloride channel protein that, in aggregate, helps ensure plenty of water is added to newly produced mucus. The most common cause is a single-amino-acid deletion in CFTR—the deletion of a phenylalanine at position 508, a mutation also known as ‘F508del’—which destabilizes the folding of one of the protein’s domains and causes a patient’s cells to destroy the protein on sight. The small fraction of proteins that avoid destruction still cannot function properly.
Trikafta fixes both of these problems. The full details of how it works aren’t entirely relevant; the important bit is that the drug works when there is a semi-functioning CFTR that just needs a little help, which is true for roughly 90% of cystic fibrosis patients with the F508del mutation. It does not work when there is no CFTR protein at all. And, as you may have guessed by now, a large share of the non-responder 10% carry what are called ‘minimal-function mutations’ in the CFTR gene, making it so the cell essentially produces zero of the protein.
Seems like a closed case: no protein, no drug, end of story. It’s an awful shame for the 10%, and we should cross our fingers that the genetic therapy folks can someday make a dent here.
But…there is something off about the story here. There are more than 2,000 known CFTR mutations, many of which are exceedingly rare. How on earth did someone sit down and discover that only F508del leads to the particular CFTR that allows the drug to work?
They didn’t.
In the original trial behind Trikafta, the creator of the drug (Vertex Pharmaceuticals) did not test two thousand mutations; it tested F508del, because F508del is what ninety percent of patients carry and, not coincidentally, because that’s who you can actually fill a trial with. So the FDA did the only thing the evidence supported: patients with F508del can get it, patients without it cannot.
Nobody is being evil here. It’s genuinely a tricky situation. All parties involved here likely agree that, somewhere in the 2,000-plus mutations, are people whose CFTR is broken in precisely the semi-functional way that Trikafta is built to fix. But they’d also agree that running a trial for specifically these people is on the spectrum of cost-prohibitive to impossible.
Well, you may say, can’t we just try Trikafta on all patients with cystic fibrosis? It’s not a bad idea; over time, we’d naturally sort out who benefits and who doesn’t. But it is an economically iffy proposition. Trikafta’s list price hovers at around $311,741—not uncommon for a rare disease drug—and Vertex does deserve to be compensated for the twenty-year slog that it took to develop the three drugs that make up Trikafta. And why would insurance companies pay this extraordinary price when there’s no proof that it’ll actually help patients?
The FDA’s response to this, beginning around 2017, was an exceptionally radical regulatory decision: it would approve the drug for a mutation based on how that mutation behaved in a dish.
Not using organoids—we haven’t gotten to that part of the story yet—but rather using an assay called the Fischer rat thyroid (FRT) cell assay. Here, Vertex places a mutated CFTR into rat thyroid cells (which have no CFTR of their own), adds the drug, measures whether chloride transport rises ≥10%, and if it does, the FDA adds that mutation to the label. To be clear: this is a population-level, per-mutation tool. It asks “does mutation X in this construct respond to the drug?”, and the answer becomes a blanket rule that applies to everyone who carries mutation X.
By December 2022, the FDA had approved the drug for cystic fibrosis patients without an F508del variant but with at least one of 177 rare CFTR variants, all identified as responsive entirely through the FRT assay. And in 2024, a retrospective analysis found that the approval had clinically significant effects for the lucky group of patients who gained access to the drug.
All’s well that ends well, no?
Not entirely. No assay is without fault, and the FRT assay has plenty of faults. The full list is described in this 119-page document, but can be summarized by the following quote from it:
It is concluded that, a positive result in the FRT assay could be seen as indicative for providing a response upon clinical treatment for a tested CFTR variant. However, a negative result in the FRT is not regarded indicative for treatment failure in patient carrying that tested CFTR mutant.
Why is this? Not only are the cells we’re using in this assay thyroid cells, but they are from a rat, and do not natively express the protein we’re interested in studying. Honestly, it is shocking the assay tells us anything at all. If we hope to poke at this question of ‘what is the FRT assay missing?’, we’ll need to set up an assay that is a bit more realistic.
In 2013, Jeffrey Beekman’s group at UMC Utrecht published a paper in Nature Medicine describing exactly that assay. Notably, this method—called the forskolin-induced swelling assay (FIS)—uses organoids.
How does it work? First, you take a rectal biopsy from a cystic fibrosis patient. From that biopsy you grow intestinal organoids, which means they arrive with the patient’s own CFTR at the patient’s own expression level. Then you add a chemical called ‘forskolin’, which indirectly leads to the physiological signal that tells CFTR to open. If CFTR opens, chloride moves into the organoid’s lumen—an internal, fluid-filled cavity—and water follows it.
And because these organoids are closed, hollow spheres, the water has nowhere to go, so the organoid swells.
At least in healthy patients, that is. For patients with F508del mutations, the organoid did not swell, not until Trikafta was added. And, as a 2016 paper demonstrated, the organoid swelling responses correlated with lung function measures across treated patients.
In 2017, three Dutch health insurers responded by co-funded a €3 million validation trialto determine whether organoid swelling could be used as a payer-side test to determine who should get these drugs. Why insurers? Because it was in their best interest to figure out whether paying $300,000 for any particular patient is worthwhile, since giving the drug today to patients who’d benefit from it prevents costly medical events down the line.
This worked out surprisingly well. A 2019 paper followed patients who’d gotten the drug via this assay and found that in vitro swelling correlated with two in vivo endpoints. A 2022 case report discussed a patient with an uncharacterized rare allele who was initially rejected from a compassionate-use program for Trikafta, and FIS was the evidence used to argue her allele was a minimal-function mutation and get her in. Happily, the patient’s response was incredibly positive! This continues in the present day, with a 2025 paper finding six more examples of patients with rare alleles being allowed to access the drug on the basis of FIS.
It’s really quite extraordinary. I understand the tendency to get a little grumpy about how the cleverness of strange preclinical research seems to never quite connect with reality, but we really need to hand it to the organoid folks here. Unlike the chemotherapy case, the 3D-ness of organoids seemed necessary for the assay to work. And thanks to it, at least a few people have gotten the chance to access a drug that could dramatically improve their quality of life.
There is a caveat here, and it is the reason why this FIS business has not spread beyond one particular system: FIS is not standardized the way FRT is. Swelling is measured by imaging organoids over an hour and computing the change in cross-sectional area. Given everything I discussed in the last essay, you can only imagine that this number can vary with organoid size at baseline, passage number, and Matrigel lot. As a result, US regulators seem to have largely dialed in on simpler methods that rely entirely on cell lines. And these methods clearly have worked quite well! Yes, FIS alone could only cover 90% of CF patients, but circa April 2026, alternative cell-line methods have pushed that to 95%, representing more than 500 genetic variants.
It is undoubtedly the case that organoids were helpful in the saga of cystic fibrosis drug approvals. But it is also the case that simpler methods held their own quite well. So…were organoids needed here? I cannot find any head-to-head comparisons, but my instinct is that they were not. Complex!
Modeling how viruses impacts fetal development (??)
Zika is a virus that, until roughly eighteen months before the 2015 Brazilian outbreak, essentially nobody had thought about. It had been discovered in 1947 when it was isolated from a rhesus macaque in the Zika Forest in Uganda, but for the following seven decades, it retained a reputation as a mild illness. A rash, perhaps a fever.
Then, in February 2015, Brazil—which was experiencing a particularly aggressive outbreak of Zika—started to report a bizarrely high proportion of babies being born with severe microcephaly.
Let’s say you are a public health official who has just heard this news. What do you do here?
Obviously, you suspect some interplay between the two events. But can you be sure? Epidemiology on a novel exposure in an under-surveilled population isn’t exactly a pristine instrument upon which to declare a public health crisis. Perhaps you could test this in pregnant animals? As it turns out, mice don’t get infected by Zika. At least not wild-type mice; there are strain-specific susceptibilities with induced biological deficits, but that is again thin evidence. How about primates? Not bad, closer to human biology, but their gestation periods are on the order of four to eight months. Do you really have that time to spare?
In this dire moment, is it worth using a cortical organoid?
A cortical organoid is an interesting little object. Like many organoids, it is a gray blob the size of a boiled grain of rice, but unlike its kin, it has layers, with a wonderful spectrum of different types of cells as you move through it. There is a small lumen in the middle, and packed radially around it like petals is a dense ring of progenitor cells all pointing inward—the ventricular zone of a developing cortex. Move outward from it and the cell density drops and the identity changes, becoming a band of intermediate progenitors, then further out, neurons. The progenitors divide at the lumen, and their daughters crawl outward to settle in the outer band.
Exactly as a real cortex builds itself, over about a hundred microns instead of a few millimeters.
Of course, none of this means anything. The organoid is small, lacks immune cells entirely, is obviously fetal, and on and on and on. It is fun that it displays any organization at all, but it is fun, and nothing more. Obviously not useful for assessing the connection between Zika and damage to fetal brains.
Unless…Zika happens to infect progenitor neurons. Because if it does, it collapses the structure of cortical development, which means the virus ought to be uniquely harmful to the growth of brains that contain progenitor neurons. Such brains, as coincidence would have it, are mainly embryonic. If Zika does do this—and it’s a big if—an organoid composed of fetal neurons should model it astonishingly well.
…the virus infects human cortical progenitor cells, leading to an increase in cell death. We also report that the infection of human brain organoids results in a reduction of proliferative zones and disrupted cortical layers. These results indicate that ZIKVBR crosses the placenta and causes microcephaly by targeting cortical progenitor cells, inducing cell death by apoptosis and autophagy, and impairing neurodevelopment.
It all seems quite clean. And if you look up the success stories of organoids, the Zika story will pop up again and again. Even the LLMs have bought into it; Claude says it, Chatty says it, and Kimi says it. It isn’t a bad idea for them to do this, because the literature does keep repeating it. The review article ‘Can We Better Understand How Zika Leads to Microcephaly [via human brain organoids]?’ is emphatic about its answer: yes.
But let’s return to the public health official, trying to figure out whether Zika is actually causing this worrying blight of microcephaly. Do organoids help them all that much?
A 2016 NYT article compiled six reasons to believe that Zika causes microcephaly. The first is the coincidental rise of the condition alongside the viral infections. The second is that the same coincidence happened in another region. The third is a bit closer to biological truth: the virus was found in both amniotic fluid and in the brains of fetuses that died in the womb. And the fifth is that an n = 345 observational study found “29 percent of those who had been infected with Zika virus experienced ‘grave outcomes’ in their pregnancies, while none of the uninfected women did.”
Only the fourth reason gestures at organoids.
So, perhaps the public health official might not really have needed organoids to make a rapid decision to declare a public health emergency. Now, there is nothing wrong with an assay acting as confirmatory evidence, but I’d have expected something grander here. Perhaps a scientist trying to develop a cure would benefit more? Maybe. One of the other famous papers here is an organoid paper that conclusively rules out AXL, a tyrosine kinase inhibitor, as an otherwise promising target for Zika infection due to results from organoid studies.
Not bad. A negative screen is always useful.
But there is something troubling about these results in aggregate, because none of them—not the AXL one, not the cortical organoid one—seemed to depend on the actual structure of the organoid. Yes, there was certainly discussion of ‘layer thickness’ of the organoid pre- and post-infection in the aforementioned papers, but it does not seem to me that any of these bells and whistles were actually necessary to establish the core finding: Zika kills fetal progenitor neurons.
As of 2026, there is no approved treatment or vaccine regimen for Zika. So, were organoids useless for treatment development? There is a Nature Medicine paper that screened roughly 6,000 compounds—approved drugs, clinical candidates, tool compounds—through cortical organoids and came out with real hits! But by the time all these triumphant organoid papers came out in mid-2016, Zika had largely disappeared, mostly thanks to public health efforts.
Perhaps in the fullness of time, the organoid-specific findings—with some extra funding—might have proved useful in developing a therapy. But for the moment, it seems like the utility of organoids here has been in a very narrow, potentially substitutable, space.
Studying immune function (???)
This is our last and strangest one. Isn’t the hallmark trait of organoids that they cannot really model the immune system in a meaningful capacity? Didn’t I spend several thousand words on this in the last essay? Happily for us, biology is a lot more complicated than the silly blanket statements I am prone to making.
In 2021, the Mark Davis lab at Stanford found a fun little piece of ‘tissue arbitrage’ that they could take advantage of: tonsils. Every year, thousands of tonsillectomies are performed on children and adults alike, and the removed tissue is thrown into the trash. This is a shame, because tonsils are really quite special from a medical standpoint. Why? Because they are lymphoid organs, and lymphoid organs—which include bone marrow, spleens, and the thymus—are absolutely rich with dynamic immune activity, which is to say, filled to the brim with T and B cells.
Most of what we know about adaptive immunity has come from inbred mouse studies, using methods that are often difficult or impossible to confirm in humans. In addition, vaccine responses in mice are often poorly predictive of responses to those same vaccines in humans. Here we use human tonsils, readily available lymphoid organs, to develop a functional organotypic system that recapitulates key germinal center features in vitro, including the production of antigen-specific antibodies, somatic hypermutation and affinity maturation, plasmablast differentiation and class-switch recombination.
We use this system to define the essential cellular components necessary to produce an influenza vaccine response. We also show that it can be used to evaluate humoral immune responses to two priming antigens, rabies vaccine and an adenovirus-based severe acute respiratory syndrome coronavirus 2 vaccine, and to assess the effects of different adjuvants. This system should prove useful for studying critical mechanisms underlying adaptive immunity in much greater depth than previously possible and to rapidly test vaccine candidates and adjuvants in an entirely human system.
Before we get to how absolutely insane this is: how is this an organoid? At least under the definition we started this series with—stem-cell-derived—this is not one. Instead it is just…dissociated adult tissue. We could quibble about the terminology here, but nobody else really seems to and it is not as though we have a better name to describe this besides ‘immune organoids’, so let’s move on.
Returning to the paper: oh my god. You know? This was the whole dream behind organoids, that you get to peek behind the great veil of biology and look at processes your body typically keeps tucked away. That dream had failed for the last few decades, but in 2021, it finally worked. If you plop tonsil tissue onto a dish, dissociate the whole thing into a single-cell slurry, and then pack those cells back together, they reaggregate into something that really does let us see something that has never before been observed: adaptive, human immune responses.
The headline result here is that when a live-attenuated influenza vaccine was dropped into these tonsil organoids, the authors reported the canonical immune sequence of events unfolding—B cells modifying themselves, and so on—eventually producing influenza-specific antibodies that could neutralize the vaccine strains they were raised against. The same held true for a rabies vaccine, a SARS-CoV-2 vaccine, and even adjuvants.
There are a lot of follow-up papers here. There’s a 2023 paper finding that the live-attenuated vaccine induced a substantially stronger immune response in the immune organoids than the inactivated vaccine did, which matches up with reality. There is a 2025 paper finding that the baseline frequency of a particular cell type in an immune organoid correlated with immune response, and this phenomenon turned out to be true when the authors looked back at human results. There was also a move to use spleens instead of tonsils. How? You can sometimes get access to spleens from deceased transplant organ donors—since spleens are rarely transplanted. Why do this at all? Since spleens are physically large, a single one can support many different experiments, allowing you to escape the usual genetic donor noise that plagues tonsil organoid experiments.
As the above abstract states, this sets up a very interesting world, one in which adaptive immunity is fully and completely observable. What all could be done here? The mind boggles.
For one, we could find more adjuvants. There are five adjuvants in wide human use—alum, MF59, AS01/AS03, CpG 1018, Matrix-M—and this platform would likely allow us to find more. Do we need more? Maybe! And two, we could find general properties of immunodominance, which are useful across infectious-disease vaccines, cancer vaccines, and other vaccine contexts. Will having a perfect model of immunodominance allow us to solve vaccine design entirely? Probably not. The immunogens used in vaccines have two ideal properties: high immunogenicity and the ability to confer protection when injected. An immune organoid platform can test the former—does the immune system pay attention at all to this?—and kinda-sorta-not-really touch the latter. You may ask why not, and it’s a complicated answer, so let’s just say ‘neutralization is necessary, but not sufficient to confer protection against a pathogen’. This isn’t always true—neutralization is perfectly sufficient for a virus that uses an accessible entry protein to cause trouble—but it’s a decent enough rule of thumb. Protection is just a hard property to pin down!
This is all very interesting work. But surely you must feel the same way I do.
How is any of this possible?
Doesn’t the adaptive immune system run on large scales? Doesn’t it require this deep, intricate interplay between the T cells and the B cells and the dendritic cells and antigens and the lymphatic system? How is even an imperfect version of it being captured in such a tiny system? It is impossible.
One way to think about this is that what we want from any model of the adaptive immune system is actually quite limited, and consists mainly of a fascinating little temporary structure called a germinalcenter, which is an emergent phenomenon that arises whenever an antigen, a B cell, and a T cell are present. What goes on in this mess? A B cell touches an antigen held out on a dendritic cell, receives a survival signal from a helper T cell sitting directly beside it, divides, mutates its receptor, and repeats. That’s it. Here is what it looks like:
This thing is a few hundred microns across, involves on the order of ten thousand cells, and works entirely by local contact. Of course an organoid can model it; the tonsil contains everything you could need to create it—the cells, the immune history. It is not the fullness of the adaptive immune system, yes, but it does come awfully close.
How fun. And what an irony that the clearest triumph of organoids—the one case where the 3D structure of an organoid is actually necessary—comes from a setup that isn’t stem-cell-derived, and therefore isn’t an organoid, and whose entire contribution is an immune system, which organoids are famously incapable of having.
Where does this all stand today? Roughly where you’d expect a promising technology to be five years after it starts working: everywhere in the literature, and nowhere in a clinic. The recent output is quite fun. There is a late 2025 paper that puts these bad boys on ‘chips’, which is to say, there is some degree of medium flow over the immune organoids. There is also an April 2026 paper that used tonsil and spleen organoids to study facets of malaria vaccinology. More exist: the Davis lab’s 2021 paper has racked up 353 citations as of today.
But no vaccine exists because of an immune organoid, and it has, as far as I can tell, not yet been instrumental in the clinical-stage development of one. Still, five years is a short time for a new wet-lab method. We will see what the future holds.
And that’s it for this essay. There is one more left in this organoid series! Curious to know what that one will be about? Here is the teaser I included in the prior piece:
A case study into a particular disease area where we really have no choice but to use organoids. And they have been used here, continuously, for over a decade. Has it been fruitful? Not really, but I plan to make the argument that it is probably good to dial in on it further.
This essay is the second of three covering organoids. The full set is:
Introduction
In my last essay, I spent a few thousand words cribbing about how most organoid work feels literally incapable of leading to anything useful. This was a lot of fun, but despite all of their issues, it’d be a lie to say that the sum impact of organoids has been zero. In truth, they’ve played a significant role in the stories of some scientific triumphs—at least according to papers—and it can be a fun exercise for us to dissect them.
What are these triumphs? And how accurate is the usual telling of them? All this, and more, is the subject of this essay. We’ll start with the most obvious settings, and slowly work our way to the stranger applications.
The success stories
Identifying which cancer patients will benefit from chemotherapy
This one is pretty straightforward.
How does chemotherapy work? Nearly all chemotherapies—paclitaxel, cisplatin, doxorubicin—work by futzing with the interior of individual cells. Paclitaxel jams the microtubules, cisplatin crosslinks the DNA, doxorubicin wedges itself between base pairs, all of which are meant to prevent the damn thing from dividing. Terrible news for a cancer cell, also terrible news for every other cell, but the net impact of it all is fine since most of your cells aren’t constantly dividing. Which is to say: chemotherapy acts via a cell-level knob, the level at which organoids are as good as any other in vitro system.
So it should not surprise you that this is precisely where organoids have posted their least ambiguous wins. The foundational result is a 2018 Science paper, ‘Patient-derived organoids model treatment response of metastatic gastrointestinal cancers’. The authors built a biobank of organoids from metastatic, heavily pretreated colorectal and gastroesophageal cancer patients. The authors then screened a library of 55 chemotherapy agents against the organoids and compared the results to what actually happened to the patients—who, importantly, had been enrolled in phase 1/2 trials, so there was a real clinical answer key against which to grade the organoids.
The concordance was startling: 100% sensitivity, 93% specificity, 88% positive predictive value, and 100% negative predictive value.
Holy shit, right? Yes, it wasn’t perfect-perfect across the board, but it did meet perfection on the last value: identifying non-responders. A 2019 Science paper found a similar result for irinotecan, another chemotherapy, in metastatic colorectal cancer. Chemotherapy isn’t traditionally considered precision medicine, but there is known heterogeneity in patient response, and being able to predict it should theoretically allow us to spare millions of cancer patients from a toxic drug that will functionally do nothing for them.
This is cool work, but it’s also so within the realm of ‘duh, you’d expect organoids to do this well’ that it probably isn’t worth discussing much further. Instead, let’s consider two likely follow-up questions you have.
First, was the organoid bit really necessary? For that matter, what does a tumor organoid even look like? Isn’t it just a blob of tumor cells? What organ-specific function would there be? So…isn’t a ‘tumor organoid’ an oxymoron? Cancer spheroids exist, sure, but surely nothing more.
It’s a good point. To answer the more basic question: a cancer organoid inherits its organoid-ness from the organ the cancer is a cancer of. Which is to say: a colorectal cancer organoid looks an awful lot like a typical colorectal organoid, just with messed up internals—a fact that the authors of the aforementioned irinotecan paper demonstrated. So yes, tumor organoids are a coherent concept.
But was the organoid part actually needed? Would a simpler spheroid not have done the same thing? In fact, it did. In a 2019 Scientific Reports paper, a spheroid made up of just dissociated ovarian tumor cells produced the following result across 44 patients: 89% accuracy, and all 32 ovarian cancer patients it identified as responders went on to respond to the tested chemotherapy.
Hmm. We’ll come back to this.
Moving on to the second question: if organoids nail chemotherapy response this cleanly, is this part of standard oncology care?
It is not. And somebody has tried: the authors behind the irinotecan result ran a prospective feasibility trial, concluding that ‘patients that received organoid-informed treatment did not experience clinical benefit.’
Strange! Why is this? Well, they do note that they could establish organoids from only 57% of patients, and it took about ten weeks from tissue to readout. You could imagine the problem here is the ten weeks bit; by the time your organoid is ready with a treatment recommendation, the cancer has mutated to an entirely new state. But the fun part about spheroids is that they are much easier to grow! The aforementioned spheroid paper spat out its verdict in seven days, not ten weeks, and spheroids are almost certainly easier to establish.
But…spheroid-guided chemotherapy selection is also not part of standard oncology care. Do we just need to run a trial here? Is the cancer cabal keeping this technology away from patients?
Maybe. Alternatively, the answer to this quandary may have nothing to do with organoids or spheroids, but rather a simple fact about medical care: being able to identify whether a patient will benefit from a therapy in advance is only useful if there’s something better on the table. Give chemotherapy to a patient, and maybe it’ll help. But if you don’t give chemotherapy to a patient who you don’t think will respond to it, it is not as if we have some secret chemotherapy that will work for them—at least not one that plops out of the organoid screening work. You’ll save them from needless suffering, yes, but that typically isn’t the goal in life-or-death situations like this.
But who knows what shape the future will take? Maybe we’ll start to extend outside of chemotherapy. A particularly interesting bit of research towards ‘arbitrary cancer drug modeling’ comes from Orakl Oncology, a French biotech.
In April 2026, they put out a preprint titled ‘SCOPE: Integrating Organoid Screening and Clinical Variables Through Machine Learning for Cancer Trial Outcome Prediction’. They do focus mostly on chemotherapy, yes, but also include in some things outside of that (ADCs and so on). And curiously, instead of asking ‘will this patient respond to this drug?’, they instead ask ‘will this trial arm hit its endpoint?’. To simulate a trials patient cohort, the trial’s eligibility criteria—cancer type and any stated molecular-subgroup criterion, like PDAC and KRAS-mut—is matched against a pre-existing biobank of organoids derived from 78 patients, and the matching ones had the associated trials drug screened against them. No fresh tumor tissue required!
What were the results of this? At first glance, remarkably good. Predicted progression-free survival (PFS) correlated with known outcomes at R² = 0.85—substantially better than the R² = 0.59 achieved by regressing over basic clinical data like median age/ECOG score/others—and the platform correctly picked the winner in 7 of 8 head-to-head arm comparison. Separately, it also predicted that the famous daraxonrasib would outperform chemotherapy before the famous phase III RASolute 302 result became public.
Still, you should exercise an abundance of caution when looking at work like this. A close reading turns up several details that are a bit strange. For instance, one of the eight included comparisons found no statistically significant difference between the treatments, but the paper still designates the arm with the numerically longer median PFS as the winner. There are other debatable bits like this, but it is difficult to begrudge the work too much. Forecasting trial outcomes is extraordinarily annoying to do entirely rigorously—I know personally!—and some tea-leaf reading is always needed. Excellent paper, and I hope more come out in the future!
Either way, the moral of the story here is that while organoids may turn out to be a poor instrument for deciding what to do with your cancer specifically, they may be a decent one for deciding which parts of a biotech’s portfolio are most worth focusing on. Neat!
Identifying which cystic fibrosis patients will benefit from treatment (?)
Let’s move on to something a bit weirder.
You may recognize cystic fibrosis as that nasty genetic disease that causes its sufferers to produce especially thick mucus, leading to a host of horrible downstream consequences, including recurrent lung infections, malnutrition, and interestingly, infertility. You may also recognize cystic fibrosis as one of those diseases that modern medicine has squarely triumphed over.
In a great 2024 essay by Sarah Zhang in The Atlantic, she says the following:
Trikafta really is a miracle. It genuinely cured millions of people who would have otherwise suffered through a painfully short lifespan, returning them to an otherwise mostly healthy state.
Unfortunately, it does not work for everyone. The essay goes on to explain the heartbreaking reality that 10% of cystic fibrosis patients lack the necessary genetic mutations for the drug to work correctly. For these unlucky few, their community has moved on, living normal lives, but these 10% are still stuck with the disease.
Before we return to this 10%, we should first understand how Trikafta works. Cystic fibrosis is caused by a genetic mutation in CFTR, a chloride channel protein that, in aggregate, helps ensure plenty of water is added to newly produced mucus. The most common cause is a single-amino-acid deletion in CFTR—the deletion of a phenylalanine at position 508, a mutation also known as ‘F508del’—which destabilizes the folding of one of the protein’s domains and causes a patient’s cells to destroy the protein on sight. The small fraction of proteins that avoid destruction still cannot function properly.
Trikafta fixes both of these problems. The full details of how it works aren’t entirely relevant; the important bit is that the drug works when there is a semi-functioning CFTR that just needs a little help, which is true for roughly 90% of cystic fibrosis patients with the F508del mutation. It does not work when there is no CFTR protein at all. And, as you may have guessed by now, a large share of the non-responder 10% carry what are called ‘minimal-function mutations’ in the CFTR gene, making it so the cell essentially produces zero of the protein.
Seems like a closed case: no protein, no drug, end of story. It’s an awful shame for the 10%, and we should cross our fingers that the genetic therapy folks can someday make a dent here.
But…there is something off about the story here. There are more than 2,000 known CFTR mutations, many of which are exceedingly rare. How on earth did someone sit down and discover that only F508del leads to the particular CFTR that allows the drug to work?
They didn’t.
In the original trial behind Trikafta, the creator of the drug (Vertex Pharmaceuticals) did not test two thousand mutations; it tested F508del, because F508del is what ninety percent of patients carry and, not coincidentally, because that’s who you can actually fill a trial with. So the FDA did the only thing the evidence supported: patients with F508del can get it, patients without it cannot.
Nobody is being evil here. It’s genuinely a tricky situation. All parties involved here likely agree that, somewhere in the 2,000-plus mutations, are people whose CFTR is broken in precisely the semi-functional way that Trikafta is built to fix. But they’d also agree that running a trial for specifically these people is on the spectrum of cost-prohibitive to impossible.
Well, you may say, can’t we just try Trikafta on all patients with cystic fibrosis? It’s not a bad idea; over time, we’d naturally sort out who benefits and who doesn’t. But it is an economically iffy proposition. Trikafta’s list price hovers at around $311,741—not uncommon for a rare disease drug—and Vertex does deserve to be compensated for the twenty-year slog that it took to develop the three drugs that make up Trikafta. And why would insurance companies pay this extraordinary price when there’s no proof that it’ll actually help patients?
The FDA’s response to this, beginning around 2017, was an exceptionally radical regulatory decision: it would approve the drug for a mutation based on how that mutation behaved in a dish.
Not using organoids—we haven’t gotten to that part of the story yet—but rather using an assay called the Fischer rat thyroid (FRT) cell assay. Here, Vertex places a mutated CFTR into rat thyroid cells (which have no CFTR of their own), adds the drug, measures whether chloride transport rises ≥10%, and if it does, the FDA adds that mutation to the label. To be clear: this is a population-level, per-mutation tool. It asks “does mutation X in this construct respond to the drug?”, and the answer becomes a blanket rule that applies to everyone who carries mutation X.
By December 2022, the FDA had approved the drug for cystic fibrosis patients without an F508del variant but with at least one of 177 rare CFTR variants, all identified as responsive entirely through the FRT assay. And in 2024, a retrospective analysis found that the approval had clinically significant effects for the lucky group of patients who gained access to the drug.
All’s well that ends well, no?
Not entirely. No assay is without fault, and the FRT assay has plenty of faults. The full list is described in this 119-page document, but can be summarized by the following quote from it:
Why is this? Not only are the cells we’re using in this assay thyroid cells, but they are from a rat, and do not natively express the protein we’re interested in studying. Honestly, it is shocking the assay tells us anything at all. If we hope to poke at this question of ‘what is the FRT assay missing?’, we’ll need to set up an assay that is a bit more realistic.
In 2013, Jeffrey Beekman’s group at UMC Utrecht published a paper in Nature Medicine describing exactly that assay. Notably, this method—called the forskolin-induced swelling assay (FIS)—uses organoids.
How does it work? First, you take a rectal biopsy from a cystic fibrosis patient. From that biopsy you grow intestinal organoids, which means they arrive with the patient’s own CFTR at the patient’s own expression level. Then you add a chemical called ‘forskolin’, which indirectly leads to the physiological signal that tells CFTR to open. If CFTR opens, chloride moves into the organoid’s lumen—an internal, fluid-filled cavity—and water follows it.
And because these organoids are closed, hollow spheres, the water has nowhere to go, so the organoid swells.
At least in healthy patients, that is. For patients with F508del mutations, the organoid did not swell, not until Trikafta was added. And, as a 2016 paper demonstrated, the organoid swelling responses correlated with lung function measures across treated patients.
In 2017, three Dutch health insurers responded by co-funded a €3 million validation trial to determine whether organoid swelling could be used as a payer-side test to determine who should get these drugs. Why insurers? Because it was in their best interest to figure out whether paying $300,000 for any particular patient is worthwhile, since giving the drug today to patients who’d benefit from it prevents costly medical events down the line.
This worked out surprisingly well. A 2019 paper followed patients who’d gotten the drug via this assay and found that in vitro swelling correlated with two in vivo endpoints. A 2022 case report discussed a patient with an uncharacterized rare allele who was initially rejected from a compassionate-use program for Trikafta, and FIS was the evidence used to argue her allele was a minimal-function mutation and get her in. Happily, the patient’s response was incredibly positive! This continues in the present day, with a 2025 paper finding six more examples of patients with rare alleles being allowed to access the drug on the basis of FIS.
It’s really quite extraordinary. I understand the tendency to get a little grumpy about how the cleverness of strange preclinical research seems to never quite connect with reality, but we really need to hand it to the organoid folks here. Unlike the chemotherapy case, the 3D-ness of organoids seemed necessary for the assay to work. And thanks to it, at least a few people have gotten the chance to access a drug that could dramatically improve their quality of life.
There is a caveat here, and it is the reason why this FIS business has not spread beyond one particular system: FIS is not standardized the way FRT is. Swelling is measured by imaging organoids over an hour and computing the change in cross-sectional area. Given everything I discussed in the last essay, you can only imagine that this number can vary with organoid size at baseline, passage number, and Matrigel lot. As a result, US regulators seem to have largely dialed in on simpler methods that rely entirely on cell lines. And these methods clearly have worked quite well! Yes, FIS alone could only cover 90% of CF patients, but circa April 2026, alternative cell-line methods have pushed that to 95%, representing more than 500 genetic variants.
It is undoubtedly the case that organoids were helpful in the saga of cystic fibrosis drug approvals. But it is also the case that simpler methods held their own quite well. So…were organoids needed here? I cannot find any head-to-head comparisons, but my instinct is that they were not. Complex!
Modeling how viruses impacts fetal development (??)
Zika is a virus that, until roughly eighteen months before the 2015 Brazilian outbreak, essentially nobody had thought about. It had been discovered in 1947 when it was isolated from a rhesus macaque in the Zika Forest in Uganda, but for the following seven decades, it retained a reputation as a mild illness. A rash, perhaps a fever.
Then, in February 2015, Brazil—which was experiencing a particularly aggressive outbreak of Zika—started to report a bizarrely high proportion of babies being born with severe microcephaly.
Let’s say you are a public health official who has just heard this news. What do you do here?
Obviously, you suspect some interplay between the two events. But can you be sure? Epidemiology on a novel exposure in an under-surveilled population isn’t exactly a pristine instrument upon which to declare a public health crisis. Perhaps you could test this in pregnant animals? As it turns out, mice don’t get infected by Zika. At least not wild-type mice; there are strain-specific susceptibilities with induced biological deficits, but that is again thin evidence. How about primates? Not bad, closer to human biology, but their gestation periods are on the order of four to eight months. Do you really have that time to spare?
In this dire moment, is it worth using a cortical organoid?
A cortical organoid is an interesting little object. Like many organoids, it is a gray blob the size of a boiled grain of rice, but unlike its kin, it has layers, with a wonderful spectrum of different types of cells as you move through it. There is a small lumen in the middle, and packed radially around it like petals is a dense ring of progenitor cells all pointing inward—the ventricular zone of a developing cortex. Move outward from it and the cell density drops and the identity changes, becoming a band of intermediate progenitors, then further out, neurons. The progenitors divide at the lumen, and their daughters crawl outward to settle in the outer band.
Exactly as a real cortex builds itself, over about a hundred microns instead of a few millimeters.
Of course, none of this means anything. The organoid is small, lacks immune cells entirely, is obviously fetal, and on and on and on. It is fun that it displays any organization at all, but it is fun, and nothing more. Obviously not useful for assessing the connection between Zika and damage to fetal brains.
Unless…Zika happens to infect progenitor neurons. Because if it does, it collapses the structure of cortical development, which means the virus ought to be uniquely harmful to the growth of brains that contain progenitor neurons. Such brains, as coincidence would have it, are mainly embryonic. If Zika does do this—and it’s a big if—an organoid composed of fetal neurons should model it astonishingly well.
In March 2015, a paper titled ‘The Brazilian Zika virus strain causes birth defects in experimental models’ was submitted to Nature, with the following paragraph in the abstract:
This is not the only paper in this line of work. There is also ‘Zika virus impairs growth in human neurospheres and brain organoids’ and ‘Zika Virus Depletes Neural Progenitors in Human Cerebral Organoids through Activation of the Innate Immune Receptor TLR3’, both making roughly the same point: this virus destroys brains in this very particular manner that organoids happen to model well.
It all seems quite clean. And if you look up the success stories of organoids, the Zika story will pop up again and again. Even the LLMs have bought into it; Claude says it, Chatty says it, and Kimi says it. It isn’t a bad idea for them to do this, because the literature does keep repeating it. The review article ‘Can We Better Understand How Zika Leads to Microcephaly [via human brain organoids]?’ is emphatic about its answer: yes.
But let’s return to the public health official, trying to figure out whether Zika is actually causing this worrying blight of microcephaly. Do organoids help them all that much?
A 2016 NYT article compiled six reasons to believe that Zika causes microcephaly. The first is the coincidental rise of the condition alongside the viral infections. The second is that the same coincidence happened in another region. The third is a bit closer to biological truth: the virus was found in both amniotic fluid and in the brains of fetuses that died in the womb. And the fifth is that an n = 345 observational study found “29 percent of those who had been infected with Zika virus experienced ‘grave outcomes’ in their pregnancies, while none of the uninfected women did.”
Only the fourth reason gestures at organoids.
So, perhaps the public health official might not really have needed organoids to make a rapid decision to declare a public health emergency. Now, there is nothing wrong with an assay acting as confirmatory evidence, but I’d have expected something grander here. Perhaps a scientist trying to develop a cure would benefit more? Maybe. One of the other famous papers here is an organoid paper that conclusively rules out AXL, a tyrosine kinase inhibitor, as an otherwise promising target for Zika infection due to results from organoid studies.
Not bad. A negative screen is always useful.
But there is something troubling about these results in aggregate, because none of them—not the AXL one, not the cortical organoid one—seemed to depend on the actual structure of the organoid. Yes, there was certainly discussion of ‘layer thickness’ of the organoid pre- and post-infection in the aforementioned papers, but it does not seem to me that any of these bells and whistles were actually necessary to establish the core finding: Zika kills fetal progenitor neurons.
As of 2026, there is no approved treatment or vaccine regimen for Zika. So, were organoids useless for treatment development? There is a Nature Medicine paper that screened roughly 6,000 compounds—approved drugs, clinical candidates, tool compounds—through cortical organoids and came out with real hits! But by the time all these triumphant organoid papers came out in mid-2016, Zika had largely disappeared, mostly thanks to public health efforts.
Perhaps in the fullness of time, the organoid-specific findings—with some extra funding—might have proved useful in developing a therapy. But for the moment, it seems like the utility of organoids here has been in a very narrow, potentially substitutable, space.
Studying immune function (???)
This is our last and strangest one. Isn’t the hallmark trait of organoids that they cannot really model the immune system in a meaningful capacity? Didn’t I spend several thousand words on this in the last essay? Happily for us, biology is a lot more complicated than the silly blanket statements I am prone to making.
In 2021, the Mark Davis lab at Stanford found a fun little piece of ‘tissue arbitrage’ that they could take advantage of: tonsils. Every year, thousands of tonsillectomies are performed on children and adults alike, and the removed tissue is thrown into the trash. This is a shame, because tonsils are really quite special from a medical standpoint. Why? Because they are lymphoid organs, and lymphoid organs—which include bone marrow, spleens, and the thymus—are absolutely rich with dynamic immune activity, which is to say, filled to the brim with T and B cells.
We cannot easily extract marrow, spleens, or thymuses, but we can do so with tonsils! And in a 2021 paper, the Davis lab used them to do something quite extraordinary:
Before we get to how absolutely insane this is: how is this an organoid? At least under the definition we started this series with—stem-cell-derived—this is not one. Instead it is just…dissociated adult tissue. We could quibble about the terminology here, but nobody else really seems to and it is not as though we have a better name to describe this besides ‘immune organoids’, so let’s move on.
Returning to the paper: oh my god. You know? This was the whole dream behind organoids, that you get to peek behind the great veil of biology and look at processes your body typically keeps tucked away. That dream had failed for the last few decades, but in 2021, it finally worked. If you plop tonsil tissue onto a dish, dissociate the whole thing into a single-cell slurry, and then pack those cells back together, they reaggregate into something that really does let us see something that has never before been observed: adaptive, human immune responses.
The headline result here is that when a live-attenuated influenza vaccine was dropped into these tonsil organoids, the authors reported the canonical immune sequence of events unfolding—B cells modifying themselves, and so on—eventually producing influenza-specific antibodies that could neutralize the vaccine strains they were raised against. The same held true for a rabies vaccine, a SARS-CoV-2 vaccine, and even adjuvants.
There are a lot of follow-up papers here. There’s a 2023 paper finding that the live-attenuated vaccine induced a substantially stronger immune response in the immune organoids than the inactivated vaccine did, which matches up with reality. There is a 2025 paper finding that the baseline frequency of a particular cell type in an immune organoid correlated with immune response, and this phenomenon turned out to be true when the authors looked back at human results. There was also a move to use spleens instead of tonsils. How? You can sometimes get access to spleens from deceased transplant organ donors—since spleens are rarely transplanted. Why do this at all? Since spleens are physically large, a single one can support many different experiments, allowing you to escape the usual genetic donor noise that plagues tonsil organoid experiments.
As the above abstract states, this sets up a very interesting world, one in which adaptive immunity is fully and completely observable. What all could be done here? The mind boggles.
For one, we could find more adjuvants. There are five adjuvants in wide human use—alum, MF59, AS01/AS03, CpG 1018, Matrix-M—and this platform would likely allow us to find more. Do we need more? Maybe! And two, we could find general properties of immunodominance, which are useful across infectious-disease vaccines, cancer vaccines, and other vaccine contexts. Will having a perfect model of immunodominance allow us to solve vaccine design entirely? Probably not. The immunogens used in vaccines have two ideal properties: high immunogenicity and the ability to confer protection when injected. An immune organoid platform can test the former—does the immune system pay attention at all to this?—and kinda-sorta-not-really touch the latter. You may ask why not, and it’s a complicated answer, so let’s just say ‘neutralization is necessary, but not sufficient to confer protection against a pathogen’. This isn’t always true—neutralization is perfectly sufficient for a virus that uses an accessible entry protein to cause trouble—but it’s a decent enough rule of thumb. Protection is just a hard property to pin down!
This is all very interesting work. But surely you must feel the same way I do.
How is any of this possible?
Doesn’t the adaptive immune system run on large scales? Doesn’t it require this deep, intricate interplay between the T cells and the B cells and the dendritic cells and antigens and the lymphatic system? How is even an imperfect version of it being captured in such a tiny system? It is impossible.
One way to think about this is that what we want from any model of the adaptive immune system is actually quite limited, and consists mainly of a fascinating little temporary structure called a germinal center, which is an emergent phenomenon that arises whenever an antigen, a B cell, and a T cell are present. What goes on in this mess? A B cell touches an antigen held out on a dendritic cell, receives a survival signal from a helper T cell sitting directly beside it, divides, mutates its receptor, and repeats. That’s it. Here is what it looks like:
This thing is a few hundred microns across, involves on the order of ten thousand cells, and works entirely by local contact. Of course an organoid can model it; the tonsil contains everything you could need to create it—the cells, the immune history. It is not the fullness of the adaptive immune system, yes, but it does come awfully close.
How fun. And what an irony that the clearest triumph of organoids—the one case where the 3D structure of an organoid is actually necessary—comes from a setup that isn’t stem-cell-derived, and therefore isn’t an organoid, and whose entire contribution is an immune system, which organoids are famously incapable of having.
Where does this all stand today? Roughly where you’d expect a promising technology to be five years after it starts working: everywhere in the literature, and nowhere in a clinic. The recent output is quite fun. There is a late 2025 paper that puts these bad boys on ‘chips’, which is to say, there is some degree of medium flow over the immune organoids. There is also an April 2026 paper that used tonsil and spleen organoids to study facets of malaria vaccinology. More exist: the Davis lab’s 2021 paper has racked up 353 citations as of today.
But no vaccine exists because of an immune organoid, and it has, as far as I can tell, not yet been instrumental in the clinical-stage development of one. Still, five years is a short time for a new wet-lab method. We will see what the future holds.
And that’s it for this essay. There is one more left in this organoid series! Curious to know what that one will be about? Here is the teaser I included in the prior piece:
Stay tuned!