For successful disease modelling using iPSCs, it is essential to generate the cell type affected by disease in a quick and efficient manner with a high degree of purity. The tissues affected early in AD include the neocortex and hippocampus, primarily pyramidal, glutamatergic neurons [
6]. Additionally, there is evidence that pyramidal cell innervation by cholinergic neurons is lost early in AD, defining the basis of the cholinergic hypothesis of AD and suggesting that cholinergic cells initiate AD [
7]. From the neocortex and hippocampus, tau pathology is seen to spread in a defined manner that traces neuronal connectivity and tracks disease severity [
8]. Aβ accumulation focuses in the isocortex and its progression is less predictable than tau (reviewed in [
9]). For this reason, the major goal of stem cell differentiation in AD research is the generation of glutamatergic, cortical neurons.
Two-dimensional adherent neuronal cultures
Directed differentiation of stem cells towards disease-relevant neurons has gained momentum over the past decade and protocols now routinely generate cultures with high purity of the neuron of interest. Since the first reports of neural specification [
10,
11], information about extrinsic signalling conditions gleaned from developmental biology have been applied to stem cell differentiation, leading to defined fate choices and increasingly pure neuronal populations (reviewed by [
12]). With this in mind, stem cell differentiation closely mirrors developmental events, timing and signalling conditions—akin to applied developmental biology.
Stem cell neurogenesis progresses through three phases: neural induction, patterning, and terminal differentiation. In 2D cultures, dual SMAD inhibition has been widely adopted to enhance neural fate choice and closely mimic signalling conditions in the early embryo [
13,
14]. Following neural commitment, human neuroepithelium has a default differentiation potential to become cortical tissue. This is due to an anterior neural character with high levels of Wnt signalling, leading to dorsalisation of the cells [
15]. This provides a great efficiency to cortical differentiation protocols and can be further enhanced by inhibiting ventral fates via Sonic hedgehog (SHH) antagonists [
16]. At this stage, extrinsic signalling conditions can be manipulated to lead to differentiation to other cell types of interest [
15]. Committed neural progenitors are subsequently allowed to undergo protracted terminal differentiation to mature neurons, a process that follows in vivo corticogenesis. Protocols that are widely employed in the field follow these defined milestones [
14,
17,
18] and have been shown to be reliant on retinoid signalling [
18] and also are able to produce cells that functionally integrate with host circuitry after transplantation into rodent models [
17].
The cortex is made up of six defined neuronal layers with characteristic identity and marker expression. In vivo, these are generated in a time-dependent manner, with deeper layers generated first and upper layers later. Inhibitory interneurons migrate from ventral tissues into the cortex, modulating the activity and development of cortical excitatory neurons. These processes are all recapitulated by differentiating stem cells in vitro, whereby upper layer neurons are generated later in culture, necessitating differentiation periods of 100 days in vitro to generate a full complement of electrically active neurons [
17,
18].
Recent reports have attempted to accelerate stem cell differentiation, making the in vitro model more accessible. This has been achieved by either overexpression of the pro-neural transcription factor NGN2, which generates homogeneous cortical populations in 2–3 weeks [
19] or a combination of small molecule inhibitors targeting ERK, FGF, Notch and Wnt signalling pathways, which leads to a highly pure population of functionally active deep-layer neurons in just 13 days [
20].
In addition to cortical glutamatergic neurons, protocols for the directed differentiation of stem cells into other cell types relevant to AD have been developed. These include cortical interneurons [
21,
22] and cholinergic neurons [
23], two cell types that have been suggested to be involved in AD. Both cell types develop from the ventral forebrain and so are dependent on ventralising signals such as SHH. As in vitro models increase in complexity, such as investigating the interplay between neuronal subtypes, the employment of multiple models will become increasingly relevant.
After extended periods in culture, neurogenic cultures undergo a gliogenic switch, producing primarily astrocytes rather than neurons [
24]. Treatment with mitogens, such as FGF, and patterning molecules, such as BMP and LIF, leads to increasingly enriched astrocyte cultures [
25]. Astrocytes provide important support to neurons and are increasingly implicated in non-cell autonomous disease processes. These will be discussed in detail below.
The differentiation of stem cells in two dimensions benefits from consistent extracellular conditions throughout the culture, whereby all cells are exposed to similar growth factor concentrations throughout the adherent monolayer, giving great control to the user and high levels of purity of the resultant population. This is relevant not only for extrinsic growth factors but also for applications such as genetic manipulation or drug screening. The resultant cultures display electrophysiologically active neurons, vital for studying AD processes in the cell-relevant model. However, these cultures represent a reductionist approach and the small number of cell types may be seen as a drawback. Additionally, iPSC-derived neurons have a foetal developmental age [
26,
27], necessitating careful considerations when investigating ageing diseases such as AD.
Three-dimensional neuronal cultures
In an attempt to better model the developing brain, rigid scaffolds have been employed to produce 3D neuronal cultures. These are created from a variety of materials, including, but not limited to, polydimethylsiloxane with micrometric cavities [
28], super-porous/non-porous hydrogels, sintered titanium [
29] and hydrogels or matrigels [
30]. Mixtures of microfibre and nanofibre scaffolds made from polymers such as poly lactic acid and polyethylene terephthalate, respectively, enable structural integrity yet small pore size to increase cellular adhesion [
31].
Primary neurons and differentiated iPSCs have been used in a 3D soft matrix scaffold, conferring some advantages over 2D. Neuron bodies are more rounded and in proximity with cells on all sides and neurites can extend in all directions, closely modelling the axonal architecture in vivo [
28,
32].
The aforementioned support structures accomplish 3D tissue culture in different ways. Ridged and inorganic scaffolds serve more as support and guidance for cellular growth, while softer encapsulating scaffolds mimic the extracellular matrix, providing support as well as promoting growth within itself [
28,
33]. However, these 3D cultures are more akin to a cell suspension within a supportive matrix, rather than a fully 3D tissue.
Three-dimensional organoid cultures
Although 2D cultures and scaffolds offer the benefits outlined above, they lack the cellular diversity, structural complexity and physical architecture seen in vivo. Thus, efforts to create 3D models of neurons and glia are crucial for developing physiologically relevant models [
34].
Neural differentiation in three dimensions follows the same three processes as adherent monolayer protocols, namely neural induction, patterning and terminal differentiation. Upon exit from the pluripotent state, the small size of the 3D structures, termed embryoid bodies (EBs), enable the researcher to have close control over the neural fate choice. Similar to 2D cultures, SMAD inhibitors may be used at this stage to enhance neural fate choice [
14,
35,
36].
In contrast to differentiation in 2D, however, 3D strategies allow intrinsic cell behaviour to determine the development of the EB or organoid. Cell autonomous signals lead to events such as migration, polarisation of neuroepithelium and generation of a range of cell subtypes, culminating in self-organised heterogeneous neuronal tissue. Spatiotemporal signalling events and inherent gene cascades, with inherent genetic competencies and responses tightly controlled in time and space, strictly govern these complex cell behaviours. This self-organising phenotype gives 3D suspension organoids more in vivo relevance than the scaffolded 3D tissue cultures discussed above.
Organoids have been shown to exhibit defined radial glial cells, a cell type relevant to brain development and higher function. These show appropriate morphology and organisation, being representative of the developing human cortex [
37]. Supportive cells that develop alongside early neurons may well be vital in modelling disease initiation and progression [
38]. These events enable complex experimental questions to be asked, such as the cross-talk between cell types, e.g. neurons and glia. This is a further advantage for drug discovery when support cells may be crucial or the pathway of interest is not known [
34]. The fact that a 3D organoid more closely resembles tissue goes some way to reducing 2D cell culture artefacts and improving the robustness of findings in 2D cultures.
The development of organoids patterned towards the cortex has progressed in recent years. Endogenous patterning and terminal differentiation produce 3D structures with appropriate cortical cell layers and functional neurons [
39]. These structures can also be further patterned to other brain regions of interest by the addition of extrinsic growth factors, albeit the effects are seen at the outer portions of the 3D structure [
37,
40,
41].
In spite of these advances, organoids also have several limitations. The self-organising nature by which organoids are generated leads to increased heterogeneity between batches [
37]. This variability between organoids means analysis methods such as bulk-population transcriptomics or western blotting on whole tissue extracts may be difficult to interpret. However, single-cell transcriptomics has successfully shown that the diversity of cell types present in organoids closely resembles that present in developing foetal brain [
27]. This provides confidence in the physiological relevance of the system but once again raises the issue of developmental maturity and how relevant cerebral organoids are in representing the modelling of late-onset diseases. Finally, when organoids exceed 6–10 mm in diameter, the inability of nutrients to penetrate the organoids results in a necrotic centre, leading to the release of caspases and other detrimental cell signals, potentially increasing cell stress in larger organoids [
42]. Organoids have been successfully kept in continuous culture for up to one year, although some shrinkage can be observed from 6 months onward, when the necrotic core usually starts to develop [
43]. To halt proliferation of the organoid, reducing the chance of a necrotic core developing, inhibition of mitogenic signals such as Notch can increase terminal differentiation, which may enhance maturity [
44,
45].
On a practical level, culture of cerebral organoids within spinning flasks [
46] has been shown to offer enhanced diffusion of nutrients and oxygen, allowing for larger and more continuous organoid growth. However, these large custom-made organoid systems are expensive to purchase and require large amounts of costly media. The development of 3D-printed bioreactors provides a compact and economical method for organoid culture; 3D-printed bioreactors are plastic lids for standard six-well plates that incorporate mini-stirrers attached to an electric motor, allowing constant media movement around the organoids. This enables users to easily adopt them into existing cell culture facilities without the requirement of accommodating large orbital shakers or spinning bioreactors in tissue culture incubators [
37]. Further, mini-bioreactors reduce heterogeneity in organoids and reduces the amount of media required, reducing the costs associated with organoid generation [
47].
In spite of these drawbacks, organoids have been used very successfully to model neurodevelopmental processes and diseases, for example normal cortical folding [
48], microcephaly [
30] and lissencephaly [
49]. Importantly, organoids have been used to uncover species-specific differences between human and murine cortical development that lead to folding and expansion in human tissues specifically [
48]. Although reports of organoid models of neurodegenerative disease are limited at present, they have the potential to be an insightful new model in AD research, allowing researchers to experiment with more heterogeneous, naturally organised 3D cell models.