Gastruloid optimization

Gastruloids are often defined as three-dimensional aggregates of stem cells that faithfully recapitulate the spatial and genetic composition of the gastrulating embryo. These aggregates exhibit collective behaviors akin to those observed during early embryonic development, such as symmetry breaking and axis elongation [1–6]. However, there are no universal criteria for defining an ‘optimal' gastruloid. Moreover, optimality may depend on the scientific context, or utility one expects to gain from gastruloid experiments.

Given the emphasis on the presence of all germ layers and some form of spatial organization [5,7,8], we propose that optimization should primarily aim to achieve structures and cell compositions that faithfully represent the intricate characteristics of a gastrulating embryo, and to ensure the reproducibility and robustness of these results [8,9]. Even when employing consistent techniques and established protocols, the formation and progression of gastruloids can display significant variability [10,11]. This variability can be attributed to both intrinsic factors, stemming from the intricate dynamics and heterogeneity inherent in the stem cell population [5,6,10,12–14], and extrinsic factors, including variations in culture conditions and environmental cues [10,14]. Thus, understanding and mitigating this inherent variability is of great importance to propel the field forward and increase the usefulness of gastruloids as robust models for investigating the complexities of embryogenesis.

Below we define the problem of organoid variability, describing different measures for it. We then discuss different sources for gastruloid variability, and finally describe some approaches for reducing variability.

Variability between gastruloids (or organoids in general) can be defined and measured across different parameters. Various measurable parameters, spanning from morphology to intrinsic aspects like gene expression, can be used to characterize the state of a gastruloid. Imaging gauges size, shape, and structure, while cell viability, proliferation, and cycle progression can be assessed using different methods such as cell counting, BrdU labeling, and Ki-67 staining [15,16]. Developmental marker patterns quantify the differentiation progression and relations between different cell types [6,10,11,17]. In other organoid systems, domain-specific parameters such as membrane voltage (in brain organoids), or metabolic parameters like oxygen, glucose consumption, and lactate production (e.g. intestinal organoids) are often measured for specific functionality, displaying variability between and within organoids [18–21]. Cell type representation elucidates gastruloid complexity, function, and developmental state. It can be probed through single-cell RNA sequencing and spatial transcriptomics, revealing heterogeneity, differentiation trajectories, disease-associated changes, and rare cell types [11,22].

Gastruloid variability arises at multiple levels (Figure 1). At the experimental system level, multiple parameters define the gastruloid protocol. These include cell line choice, pre-growth conditions (affecting the starting cell epigenetic state), the cell aggregation method and parameters (e.g. number of cells per aggregate), and the precise differentiation protocol. Next, given a specific system (cell line and exact protocol), different repeats of the same protocol by the same lab could yield different results, due to multiple reasons. Finally, gastruloids within one experiment display a certain distribution of outcomes, whether in their overall morphology, or in their cell composition and spatial lineage arrangement. Being a complex dynamically evolving system, this gastruloid-to-gastruloid variability can change and often increase over time [11,23]. Similarly, brain organoids were shown to be more uniform in cell type composition during their early stages, but the variability increases in later developmental stages [24].

Depending on the experiment's purpose, there is some tradeoff between the need for quantity, the uniformity of the samples and their accessibility. There are different platforms to form and grow gastruloids, from 96-U-bottom and 384-well plates, to microwell arrays and shaking platforms. Each platform has its own balance between number of samples and variability between them; 96-U-bottom and 384-well plates allow stable monitoring of each gastruloid over time, and can be combined with liquid handling robots for high-throughput screening. They allow a medium number of samples, with some degree of initial variability between them, mainly in initial cell number. Using a shaking platform (for example with large well plates) allows many more samples, but obtaining uniform sizes is difficult, and live imaging of single gastruloids over time is not possible. Microwell plates may allow a more stable initial size of the sample, but monitoring and handling individual aggregates is more challenging. Moreover, when comparing dispersion of media in the well and movement of the gastruloid (especially between shaking and static platforms), the different growing platforms may affect the growth and differentiation direction. Different relevant growing platforms were recently reviewed in [25].

Besides the gastruloid growing platform, variation between experiments can also arise from different pre-growth conditions, medium batches, cell passage number and personal handling. Major problems with in vitro ESC culture stem from batch-to-batch differences of all media components, when the undefined media components (such as serum) have a potential to deeply affect the maintenance and differentiation of ESCs [6]. Different batches can affect cell viability, pluripotency state and differentiation propensity. For these reasons, many groups prefer to grow ESCs in more defined media, in the absence of serum and/or feeder cells. Different basal media affect the pluripotency state of the growing cells: feeder cells presence or absence and 2i/LIF vs. Serum/LIF can shift the pluripotency levels (ranging between Naive ICM-like and Epiblast-like pluripotency). Other component differences include different base media (DMEM, GMEM), percentage of serum in the media (10, 12.5, 15). The number of cell passages after thawing was also observed to affect gastruloid differentiation, for example, when trying to obtain somite-like structures in gastruloids. All these different pre-growth approaches can affect gastruloid outcome, and create bigger disparities in cell state and differentiation between different researches [6,10,11,26,27].

While there are stable gastruloid differentiation protocols, it is known that different cell lines and genetic backgrounds respond a bit differently to the same protocol. Different cell lines can have different propensities for different germ layers (or cell fates), as demonstrated for cardiac fates [14].

The differences discussed above are making the universal optimization of the gastruloid protocol harder. The researcher needs to make several experimental choices (pre-growth conditions, gastruloid growing platforms, cell-lines, etc.). Each of these choices could imply different timings or doses of different protocol steps for optimal results; For example, cell lines that tend to under-represent endoderm could be treated with Activin. Depending on cell line and pre-growth conditions, there may be a need to extend the aggregation under N2B27 only or shorten the Chiron pulse.

Once we have characterized the distribution of outcomes in a given gastruloid system (specific cell line and growth protocol), we can attempt to optimize it, in particular reducing its variability and/or steering it to desired outcomes. Steps towards reducing within-experiment gastruloid-to-gastruloid variability include:

• Improved control over seeding cell count. This may be obtained, for example, by aggregating cells in microwells, or in hanging drops.

• Increasing initial cell count. Assuming optimal cell mixing in suspension (which may compensate for local heterogeneity of ESCs in the 2D culture), a higher starting cell number may result in a less biased sample within each organoid (as the distribution of cell states will be closer to the overall distribution in the cell suspension). Aiming for higher cell counts can also decrease the sensitivity to technical variation in cell count per aggregate. This approach is of course limited by the biologically optimal cell count per aggregate, which varies between cell lines.

• Removal (or reduction) of non-defined medium components. Though the gastruloid protocol itself is based on defined medium alone, there are varying pre-growth (pluripotency) conditions in different protocols, some of which are based on growth in serum and feeders. These may result in batch-to-batch variability. The use of feeders could potentially increase heterogeneity in the 2D pre-culture, as the effect of feeders may not be spatially uniform.

• Short interventions during the protocol. These may serve to either buffer variability between organoids (by partially resetting them to the same state), or to generate a delay in one of the differentiation or morphogenetic processes, to improve its coordination with other processes.

• Personalized (gastruloid-specific) interventions. A more sophisticated approach for buffering the variability between gastruloids is matching the timing or magnitude (e.g. material concentration) of the next protocol step to the internal state of the gastruloid.

Definitive endoderm in the gastruloid model has shown large variability in its relative extent, the morphologies reported and their frequency. Endodermal gut-tube formation within the gastruloids does not lead to morphology changes such as elongation, but likely rather relies on it for its own progression. Therefore, for sufficient endodermal differentiation and gut-tube formation there is a need for a stable coordination between its progression and other layers, particularly the mesoderm, as it drives the A–P axis elongation. A shift in this fragile coordination can cause failure in endodermal progression, which can manifest in its developing morphology. This progression instability creates endodermal morphology variability in gastruloids [11,23].

To tackle this variability problem, we suggested a machine learning approach to better understand how early measurable parameters vary during the gastruloid development, and which of them are predictive of endodermal morphotype choice (Figure 2). For this goal, we used live imaging of developing gastruloids. Along the differentiation timeline we collected different morphological parameters, such as gastruloid size, length, width, aspect-ratio, as well as expression parameters based on fluorescent markers (we used a dual-marker Bra-GFP/Sox17-RFP cell line [28]).

At the end of each experiment, at 96 h, the different morphologies of the endoderm were classified into four main groups (tube, ball, ball-tube and dispersed). By analyzing learned decision trees that attempt to predict the morphotype from earlier parameters, we tried to infer the driving parameters — those that are most predictive of the future endodermal morphotype. Using these predictive parameters, we adapted interventions to the protocol in order to enhance the frequency of gut-tube morphologies and to reduce the variability of the endoderm morphology. We used two intervention approaches — a standard short intervention, where all gastruloids get the same treatment simultaneously, and a more ‘personalized' one, where each gastruloid gets its own protocol variant, based on selected parameters (in our case its shape). In both cases we obtained an increase in the frequency of tube morphologies, as well as a reduction in certain morphological parameters (e.g. endoderm domain diameter) across all gastruloids [23]. While these results improved our understanding of endoderm progression in gastruloids, they can be extended to other germ layers, other cell lines and also other protocols.

Robustness and reproducibility of the gastruloid, both within and between experiments, are key to its wide adoption for developmental modeling, as well as for applications such as drug screening. Developing approaches to control and curb the variability, will both increase this system's usefulness, and may teach us about the mechanisms employed by the embryo to generate developmental robustness.

Another critical aspect of optimizing gastruloids lies in harnessing their potential to develop towards specific lineages or tissue types. By directing gastruloid development along predetermined developmental trajectories, for example to generate cardiac or muscle tissue, researchers can unlock new avenues for studying organogenesis and for regenerative medicine. The ability to steer gastruloid development towards desired lineages requires a deep understanding of the signaling pathways, cellular interactions, and microenvironmental cues that drive tissue-specific differentiation during embryonic development. By carefully manipulating these factors, researchers can guide gastruloids to undergo targeted morphogenesis and lineage commitment, leading to the formation of tissue structures with functional properties resembling those found in the desired tissue type.

The approaches we survey here for gastruloid optimization apply to organoids in general. In organoid applications such as drug toxicity testing, compound screening, or treatment testing in personalized organoids, controlling and reducing the different levels of organoid variability may be crucial for reliable interpretation of these comparative assays [29–31].

• Gastruloids display variability at three different levels: between labs/protocols, within-lab between experimental repeats, gastruloid-to-gastruloid variability within experiment.

• Multiple sources affect each of the variability levels.

• Some generic considerations can reduce within experiment variability

• Statistical analysis/machine learning can guide the design of protocol interventions/modifications to reduce variability and enhance desired outcomes.

The authors declare that there are no competing interests associated with the manuscript.

This work was supported by the Israel Science Foundation (1491/22) and the European Union (Horizon-EIC-2021-PathfinderChallenges-01 101071203, SUMO).

Open access for this article was enabled by the participation of Tel Aviv University in an all-inclusive Read & Publish agreement with Portland Press and the Biochemical Society under a transformative agreement with MALMAD.