Brains in a Dish: Stem Cells and Organoids

In 2013 a laboratory in Vienna grew folded tissue in a dish from the skin cells of a person with severe microcephaly, a brain far smaller than normal. The tissue held zones of dividing cortical progenitors, and those zones were smaller than in tissue grown from control cells.

That experiment rests on older findings: a specialized cell keeps a complete set of instructions, a few transcription factors can return it to an embryonic state, and embryonic cells given room build tissue that resembles the early brain. The starting point is brain tissue culture: what can progenitors do once they are taken out of the embryo? And then the harder question: what does an organoid get right, and what does it get wrong?

A nucleus gets a second chance

As an embryo develops, its cells specialize. In the mid-twentieth century two explanations were open: cells lose the genes they no longer need, or they keep every gene and silence most. Nuclear transfer can tell the two apart: the nucleus of an egg is removed or destroyed, and the nucleus of another cell is put in its place. If the egg develops, the transplanted nucleus still had what development needs.

In 1952 Robert Briggs and Thomas King did this in the leopard frog Rana pipiens, using nuclei from blastula cells (the hollow ball of cells an early embryo forms). Some of the eggs developed into tadpoles. In their later experiments, nuclei taken from older embryos supported normal development less often, which looked like evidence that nuclei lose potential as cells specialize.

John Gurdon, working at Oxford with the clawed frog Xenopus laevis, pushed the test further. In his 1962 paper the donor nuclei came from the intestinal epithelium of feeding tadpoles, cells that had already specialized. Most transfers failed. A small fraction of the eggs developed into normal feeding tadpoles. One clean success is enough to answer the question: the gut-cell nucleus still held the instructions for a whole tadpole, nervous system included. The low success rate left a doubt, that the successes came from rare unspecialized gut cells, which later work addressed.

Nuclear transfer, later extended to mammals, showed that the genome is kept. It did not say what in the egg resets it.

Four factors

Shinya Yamanaka’s laboratory at Kyoto University asked that question from the other side. If an egg resets a nucleus with proteins, perhaps a small set of transcription factors (proteins that bind DNA and turn genes on or off) taken from embryonic stem cells could do the same in an ordinary cell. Embryonic stem cells come from an early mouse embryo, divide indefinitely in culture and are pluripotent: they can give rise to every cell type of the body.

Kazutoshi Takahashi and Yamanaka used mouse cells in which a stem-cell gene, Fbx15, had been joined to a drug-resistance gene, so that only cells switching it on survived the drug G418. They chose 24 candidate factors and delivered all of them into fibroblasts (connective tissue cells) with retroviruses, and resistant colonies grew. Removing one factor at a time showed which ones the colonies needed. The list fell to four: Oct3/4 (also called Oct4), Sox2, Klf4 and c-Myc. Their 2006 paper called the cells induced pluripotent stem cells, or iPS cells, and made them from both embryonic and adult mouse fibroblasts.

The first lines were imperfect: their gene expression differed from embryonic stem cells, and in the tests reported they gave no live chimeric mice (mice built partly from injected cells). In 2007 Keisuke Okita, Tomoko Ichisaka and Yamanaka selected cells with a stricter reporter, Nanog, and obtained lines that contributed to adult mice and their germ line. The laboratories of Rudolf Jaenisch and Konrad Hochedlinger reported similar results.

Human cells followed in November 2007, from two groups. Takahashi, Yamanaka and colleagues used the same four factors on adult human skin fibroblasts. Junying Yu, James Thomson and colleagues used OCT4, SOX2, NANOG and LIN28. A human iPS line carries the whole genome of its donor, so a laboratory could now make neural tissue carrying a patient’s mutation and the rest of that patient’s genome. Gurdon and Yamanaka shared the 2012 Nobel Prize in Physiology or Medicine.

Self-organizing tissue

Single progenitors from embryonic rat forebrain, grown in tiny wells in the late 1980s, made only neurons, only glia or both (see radial glia and progenitors). But a flat dish loses the shape of a tissue.

In 2008 Mototsugu Eiraku, Yoshiki Sasai and colleagues found that floating aggregates of mouse embryonic stem cells formed polarized cortical neuroepithelium: a sheet with its apical side facing a cavity, dividing progenitors near that side and young neurons further out. The cells arranged themselves. This is self-organization, and it is the idea behind every organoid.

Madeline Lancaster, working in Jürgen Knoblich’s laboratory in Vienna, applied it to human cells. In their 2013 method, human pluripotent stem cells were gathered into balls called embryoid bodies, steered toward neural tissue, and embedded in droplets of Matrigel. Matrigel is a gel from a mouse tumor that resembles basement membrane, rich in laminin and collagen IV. The droplets then moved to a spinning bioreactor, where moving medium improved the supply of oxygen and nutrients. The tissue grew to a few millimetres across, and they called it a cerebral organoid.

Cortex-like regions had a ventricular zone of radial glia around a fluid-filled cavity, with nuclei moving toward and away from the apical surface as the cells cycled (interkinetic nuclear migration), intermediate progenitors beyond them, and early neurons further out. Some dividing cells sat away from the apical surface with a basal process and no apical one, the shape of the outer radial glia that are abundant in human cortex (see human brain expansion).

Then came the disease test. Organoids from a person with severe microcephaly and mutations in CDK5RAP2, a gene for a centrosome protein, had smaller progenitor zones, and more of their cells became neurons early. The disorder belongs to microcephaly and malformations; here the question is what such a result can carry.

Design problem

Organoids grown from one person with microcephaly come out smaller than control organoids and have fewer progenitors. What do you need before you claim that the CDK5RAP2 mutation causes this, rather than the rest of that person’s genetic background, a quirk of one cell line, or a bad batch of culture?

Joining regions: assembloids

In an unguided organoid, regions form where they form, which makes traffic between two regions hard to study. Cortical inhibitory interneurons, for example, are born in the ventral forebrain and migrate tangentially into cortex (see tangential migration).

In 2017 Fikri Birey, Jimena Andersen, Christopher Makinson, Sergiu Pașca and colleagues at Stanford made the two regions separately. They patterned human pluripotent stem cells into cortical spheroids (dorsal forebrain) and subpallial spheroids (ventral forebrain), labelled the ventral ones, and placed one of each together until they fused. Labelled cells moved from the ventral partner into the cortical one far more often than in the reverse direction. They moved in jumps: a leading process extended, a swelling formed ahead of the nucleus, and the nucleus followed. Interneurons in fetal forebrain move in the same saltatory way. Once inside, the migrated cells received excitatory synaptic input.

The team then made spheroids from people with Timothy syndrome, caused by a gain-of-function mutation in CACNA1C, the gene for the L-type calcium channel Cav1.2. Patient interneurons jumped more often but covered less ground per jump, and blocking L-type channels improved their migration. Two other groups, Joshua Bagley with Knoblich and Yangfei Xiang with In-Hyun Park, reported fused regional organoids in the same months. Pașca’s group later named such fused preparations assembloids. Neural engineering and organoid assembly, combining parts that a single organoid cannot make in the right place, is where the field is heading. Cells in these preparations are labelled and altered with tools from hacking the cell.

The report card: why organoids are not brains

Resemblance has to be measured. Single-cell RNA sequencing counts the messenger RNAs in thousands of separate cells, so organoid cells can be compared with fetal cortex of known age (see tools for development). In 2015 J. Gray Camp, Barbara Treutlein and colleagues found that organoid cortex used gene programs similar to those of fetal neocortex, and in 2019 Silvia Velasco, Paola Arlotta and colleagues showed that a directed cortical protocol gave similar cell compositions from organoid to organoid.

In 2020 Aparna Bhaduri, Madeline Andrews, Arnold Kriegstein and colleagues graded more finely. They compared about 235,000 cells from 37 organoids with about 189,000 cells from developing human cortex. At the level of broad classes the organoids did well: radial glia, intermediate progenitors, neurons and interneurons were all there. At the level of subtypes they did worse. Organoid cells often resembled several primary subtypes about equally, and markers of particular subtypes, such as SATB2 in upper-layer neurons, were much rarer. Organoid cells also expressed genes of glycolysis and endoplasmic reticulum stress at higher levels. When organoid cells were transplanted into mouse cortex, those stress genes fell and some cells came to resemble primary subtypes more closely, which suggests that the culture environment contributes. Later protocols narrowed some gaps; the grade depends on protocol, age and cell type.

Other limits are built in. Organoids have no blood vessels, so the center of a large one dies. Most protocols lack microglia, which come from outside the neural tube. There are no sensory inputs and no distant targets, so long-range wiring cannot form normally, and cortical layering is at best partial. Organoids model early and mid-fetal stages, not a mature brain.

They are electrically active, though. In 2019 Cleber Trujillo, Alysson Muotri and colleagues recorded cortical organoids for months on electrode arrays and reported network oscillations that grew more complex with time; they reported that some features of this activity resembled features of EEG recordings from preterm infants. A New York Times article that year carried the headline “Organoids Are Not Brains. How Are They Making Brain Waves?” Coordinated firing in tissue with no inputs is a long way from a working brain, and the word “mini-brain” hides that distance. The ethical questions remain: consent from cell donors, the welfare of animals that receive organoid transplants, and how complex such tissue should become.