We have already written about what the regulatory shift away from animal models means. This is the bench-side companion: what organoids and other 3D cultures have actually been used to do, where they earned their place, and where they are still the wrong tool.
The short version is that 3D culture is worth its cost when your question depends on human tissue behaving like tissue. When it does not, a flat dish is cheaper, faster and easier to measure.
First, which kind of 3D
"3D culture" covers systems that differ by an order of magnitude in cost and effort, so it is worth being precise before talking about uses.
- Spheroids. Cells clumped into a ball, usually in low-attachment plates or hanging drops, often from a single cell line. Cheap, quick and easy to run in plates of 384. They get you three-dimensional geometry, cell-cell contact and gradients of oxygen and drug, but not organised tissue.
- Organoids from adult stem cells. Grown from a tissue sample, such as a gut, liver or tumour biopsy, embedded in a gel and fed niche factors. The method traces to a 2009 paper showing a single intestinal stem cell could build crypt and villus structures in a dish. They are epithelium only, they keep the genetics of the person they came from, and they can be expanded and frozen.
- Organoids from pluripotent stem cells. Grown from iPSCs or embryonic stem cells pushed through a developmental programme. This is how you get brain organoids and kidney organoids, tissues you cannot biopsy. They contain more cell types, take months, and tend to resemble fetal rather than adult tissue.
- Organs-on-chips. Cells in small channels with fluid flowing past them, sometimes with mechanical stretch. The lung-on-a-chip from 2010 put breathing motion on an air-blood barrier. Chips add the flow and force that a static gel lacks, at the cost of throughput.
- Assembloids and co-cultures. Two or more of the above combined: fused brain regions, organoids with added immune cells, tumour and stroma together. This is where most of the current development effort is going, because it addresses the biggest gap, which is that single organoids are missing most of the body around them.
1. Studying biology that only happens in human tissue
The clearest win for organoids is a question that has no animal or cell line answer at all.
Human norovirus is the leading cause of gastroenteritis worldwide, and for decades it could not be grown in a lab. In 2016 a group cultivated multiple strains in intestinal organoids grown as flat sheets, and found that bile was required for some strains to replicate, something no transformed cell line would have told them. When SARS-CoV-2 arrived, gut organoids showed within months that the virus productively infects human enterocytes, which helped explain the gastrointestinal symptoms.
The same logic applies to development. The original brain organoid paper modelled microcephaly using cells from a patient, because the mouse version of that mutation does not shrink the mouse brain the way it shrinks a human one.
Use it when: the species difference is the problem. A human-specific pathogen, a human-specific developmental feature, or a mutation whose mouse version does not reproduce the disease.
2. Testing a drug on one particular patient's cells
Because adult stem cell organoids carry the donor's genome, you can ask how that specific person's tissue responds to a drug.
Cystic fibrosis is the model case. Gut organoids from a healthy donor swell when treated with forskolin, because the CFTR channel moves fluid into the middle. Organoids from people with cystic fibrosis barely swell, and CFTR-restoring compounds bring the swelling back. That gives a simple, quantitative readout of whether a given modulator works on a given patient's mutations, which matters for people carrying rare variants that never appeared in a clinical trial.
Cancer is the larger and messier version. Organoids grown from metastatic gastrointestinal tumours of patients enrolled in early-phase trials tracked how those patients actually responded. A later prospective study in metastatic colorectal cancer is the more instructive one, because it shows where the method stops working. Organoids predicted response to irinotecan-based chemotherapy in more than 80% of patients, without wrongly flagging anyone who would have benefited. For 5-fluorouracil plus oxaliplatin, a standard regimen, they failed to predict outcome at all.
The catch: prediction works for some drugs and not others, and you cannot know which in advance without a clinical comparison. Turnaround also matters. If the organoids take longer to grow and test than the patient can wait for a treatment decision, the prediction arrives too late to use.
3. Catching toxicity before it reaches people
Drug-induced liver injury is one of the most common reasons drugs fail late or are withdrawn, and animal studies miss a share of it. This is where organs-on-chips have the most published evidence.
One study ran 870 liver-chips against a blinded set of 27 drugs with known human outcomes, chosen by an industry consortium as a benchmark. The chips caught 87% of the drugs that injure human livers and flagged none of the safe ones as toxic. For the same drug set, 3D spheroids of primary human hepatocytes, the model most widely used in industry, had previously caught 42% and wrongly flagged a third of the safe drugs.
The catch: the study was run by the company that makes the chip, and the spheroid figures come from earlier published work rather than a side-by-side run. The three toxic drugs the chip missed were also missed by spheroids, which suggests their toxicity involves cells or organs neither model contains. It is strong evidence, and it is still one study of 27 drugs.
4. Building a disease one mutation at a time
Organoids grown from normal tissue are genetically editable, which lets you rebuild a disease in steps instead of only studying the finished product.
In 2015, researchers used CRISPR to introduce the four most commonly mutated colorectal cancer genes (APC, TP53, KRAS and SMAD4) into normal human intestinal organoids. Most of the mutations freed the organoids from needing one of the growth factors in the medium, so withdrawing that factor selected the edited cells; TP53 loss was selected with a drug instead. Quadruple mutants needed no niche factors at all and grew as invasive tumours when implanted in mice. Losing APC and TP53 together was enough to produce widespread chromosomal instability.
Use it when: you want to know what a specific mutation does on a defined human background, and in what order mutations matter. A tumour sample only shows you the end state; an edited organoid shows you each step.
5. Putting the immune system back in
Standard tumour organoids are tumour epithelium only, which makes them useless for immunotherapy, because the immune cells that the drugs act on are gone. One answer is to grow tumour fragments at an air-liquid interface, which kept the patient's own immune cells embedded in organoids from more than 100 human and mouse tumours. The T cells retained the original tumour's receptor repertoire, and anti-PD-1 treatment expanded tumour-specific T cells and killed tumour cells in the dish.
The catch:the immune cells present are the ones that happened to be in the biopsy, and they are not replenished. That is a snapshot of the tumour's immune environment, not a working immune system with lymph nodes and circulation behind it. Building fuller immune compartments into organoids is an active field, and the honest position is that it is not solved.
6. Repairing tissue
The furthest-out use is putting organoid-derived cells back into a person. The most concrete evidence so far is in donor livers: cholangiocyte organoids, grown from bile duct cells, were transplanted into deceased donor livers kept alive by machine perfusion for up to 100 hours, where they engrafted and repaired damaged bile ducts.
Treat this as a direction rather than a service. If your work is in regenerative medicine you already know the manufacturing and regulatory distance between a working graft and a therapy. For everyone else, the practical uses are the five above.
Where 3D is the wrong choice
- A large primary screen. If you are testing thousands of compounds for a simple readout, start in 2D or in spheroids and save organoids for the hits. Organoid screens are possible, but every well is more expensive and more variable.
- Whole-body questions. How a drug is absorbed, distributed and cleared, anything involving circulation, and anything that needs a full adaptive immune response still need an animal or a person. Linking chips together narrows this gap; it does not close it.
- Mature adult tissue from pluripotent cells. Organoids made from iPSCs tend to look fetal. A careful comparison found that cortical organoids activate cellular stress pathways that impair cell identity, and contain broad cell classes without the specific subtypes of a real cortex. That does not rule them out for developmental questions, but it does mean an adult-onset disease may not show up.
- Anything that needs a blood supply. Most organoids lack vasculature, so nutrients and oxygen reach the middle only by diffusion. Large spheroids and organoids develop a dying core, which is useful if you are modelling a hypoxic tumour and a confound if you are not.
What it costs you in practice
The published successes hide a lot of unglamorous work. Before you commit, budget for these:
- Time. Adult stem cell organoids can be ready in weeks. Pluripotent-derived organoids take months, and the iPSC line has to be healthy before you start, which we cover in our guide to culturing iPSCs.
- The gel. Most organoids grow in Matrigel or similar basement membrane extracts, which come from a mouse tumour and vary from lot to lot. Record the lot number, and when results shift between experiments, check it before anything else. Defined synthetic gels exist and are improving, but do not yet support every organoid type.
- Variability between organoids. Organoids in one well differ in size and composition. Decide how you will normalise for size before you collect data, not after.
- What counts as a replicate. A hundred organoids from one donor are one biological replicate, not a hundred. If your claim is about people with a condition, you need several donors, and that, more than anything, drives the cost of a well-designed organoid study.
- Readouts. Thick 3D samples are harder to image than a monolayer and often need clearing and confocal or light-sheet microscopy. Dissociating them for flow cytometry or single-cell sequencing works, but you lose the structure you grew them for, so consider spatial methods if position matters.
Matching the model to the question
- Does a drug penetrate a solid mass and kill cells inside it? Spheroids.
- How does this patient's tissue respond? Adult stem cell organoids from that patient.
- What does this mutation do in human tissue?Edited organoids from normal tissue, or from a patient's iPSCs with an isogenic corrected control.
- How does a human brain or kidney develop, or fail to? Pluripotent-derived organoids, and assembloids if the question is about connections between regions.
- Is this compound toxic to a human organ? Organs-on-chips, where flow and repeat dosing matter.
- Does this immunotherapy activate T cells against the tumour? Tumour organoids that keep their immune cells, or co-cultures with the patient's own T cells.
If you are not going to build it yourself
Many labs will never need a permanent organoid set-up, and commissioning one study from a core facility or CRO that already runs the model is often faster than spending a year establishing it. If you go that way, ask the provider:
- Their success rate for establishing organoids from your tissue type, and what happens to the timeline and the price when a sample fails.
- How they characterise the organoids, and against what: matched patient tissue, published single-cell atlases, or nothing.
- How many donors the price covers, and whether the organoids or lines are yours to keep afterwards.
- Which gel they use, and whether they track lots.
- What raw data you get back, not just the summary figures.
The answers will tell you more about whether the study will stand up than the model name on the quote.