Decoding the mechanisms of embryo development with spatial biology

Spatial biology consists of several techniques that provide detailed cellular information within the spatial context of a 2D or 3D tissue and has been revolutionary across many areas of research due to its ability to map the spatial architectures of tissue and examine cellular interactions. Spatial techniques fall into two categories: image-based techniques, which saw major advances in the 1980s with the development of single-molecule fluorescence in situ hybridization (FISH) [1], and sequencing-based techniques, which, unlike FISH, capture RNA species in an unbiased manner.

The Visible Embryo Project & the advent of spatial biology

The origins of spatial genomics date back to the 1990s with the launch of the Visible Embryo Project (VEP). At the time, Mike Doyle was one of the investigators on this project (now the Vice President of Research at the New Mexico Institute of Mining and Technology, NM, USA) and explains that “it is only by visualizing the exact spatial distribution of function-correlated signals within the morphological context of the tissue where these functions occur that researchers can start to decode the intricate spatial interrelationships present in both normal and abnormal biology states.” Until the emergence of spatial biology, either the biological functions of cells or detailed morphological structures could be captured, but not both simultaneously.

The VEP started as a response to the Visible Human Project, an initiative Doyle was also part of, which set out to create a dataset of cross-sectional photographs of the human body. Whilst waiting for this data to become available, Doyle sought out alternative sources of high-resolution 3D images of human anatomy and came across a collection of 650 human embryos that were sectioned and preserved on slides housed in the Carnegie Collection of Human Embryology at the National Museum of Health and Medicine (Washington DC, USA). “I realized that if I could devise a way to digitize these sections and reconstruct them in 3D, I would effectively have an image dataset similar to 650 Visible Human Projects.”

To create this dataset, Doyle and his co-investigators developed a method for multidimensional morphological reconstruction of genome expression activity. This became the first spatial biology system and was named Spatial Analysis of Genomic Activity (SAGA), which was awarded a US patent in 2009 [2]. SAGA involves “generating a 3D morphological rendering of a tissue sample, subsampling the tissue along a regular grid pattern, bar-coding each subsection sample, analyzing each sample for gene expression, and then mapping the gene expression data back onto the morphological rendering of the tissue,” Doyle explains. This method laid the foundation for what became known as spatial transcriptomics.

The VEP had an ambitious goal of producing real-time 3D visualizations of high-resolution embryo datasets over the internet (Figure 1), “creating a comprehensive online digital archive of multidimensional data on normal and abnormal human development.” To create a digital resource from this massive dataset, the VEP also had to overcome limitations in computing technology as the “PCs of the early 90's – significantly less powerful than today's machines – stood no chance of managing the colossal embryo datasets that our site aimed to deliver.” So, they developed what is known today as ‘the cloud’, a system that remotely uses computational server capabilities located elsewhere to bypass limitations of local hardware – a technology that was far ahead of its time.

Despite these advances, it wasn't until 2016 when momentum and interest in spatial biology picked up following a paper first authored by Patrik Ståhl (Karolinska Institute, Stockholm, Sweden). This paper was published in Science and reported a method of measuring the spatial distribution of transcripts in brain and cancer tissue whilst maintaining the morphological context [3]. Since then, spatial biology has exploded, with spatial transcriptomics being awarded Nature‘s Method of the Year in 2020 [4], and the development of several -omics techniques, including spatial genomics, transcriptomics and proteomics.

Cellular interactions in embryo development

Single-cell RNA sequencing provides a high level of detail about cells in a tissue and has greatly advanced our ability to study cellular composition [5]. However, current methods require a single-cell suspension resulting in a loss of spatial data, limiting understanding of cellular interactions. The interaction between cells is a crucial aspect of many mechanisms; for example, “sometimes cancer cells are supported by their neighboring cells or sometimes they are chewed up,” explains Kyoung Jae Won, a researcher at Cedars-Sinai (CA, USA) developing computational software for spatial biology. “As single-cell RNA sequencing can lose spatial information, we cannot understand the cell's action directly.”

Cell–cell communication is essential in developing multicellular organisms as interactions with neighboring cells induce molecular signals that can determine lineage specification [6]. The correct spatiotemporal organization of cells is also crucial for proper embryo development and plays a role in pathophysiology [7]. “Embryogenesis is a really cool system,” comments Won. “Cell contact is really important to provide the cue for cell fate decisions for later time points … and the spatial information is important to understand embryogenesis.”

In a recent study, Won and his co-authors used computational analysis and physically interacting cell sequencing (PIC-seq) to identify genes that are observed when certain cells are in contact during embryogenesis [2]. PIC-seq combines the sorting of physically interacting cells with single-cell RNA sequencing to map cellular interactions. “Interestingly those genes that are influenced by cell contact are important for embryogenesis, and we believe that cell contact can provide a cue for later development.” They then used PIC-seq and developed computational algorithms to predict neighboring cell types based on the gene expression of a single cell.

The researchers are currently adapting this concept to use spatial transcriptomics data. “We found that without dissociation, we can still detect the cell contact-dependent genes using spatial transcriptomics,” says Won. Without generating PIC-seq data, this method, called cell neighbor-dependent gene expression (CellNeighborEX), identifies genes influenced by cells located nearby from image-based spatial sequential-FISH data [8]. Whilst CellNeighborEx is also applicable to NGS-based spatial transcriptomics data, identifying neighboring cells can be challenging as barcoded spots do not always match up with the cell boundary perfectly.

Looking under the hood during the ‘black box’ of human development

Weiyang Shi (Ocean University of China, Qingdao, China) also uses spatial omics technologies to study embryogenesis, focussing on early human embryogenesis, which remains a black box for developmental biologists [9]. Shi's team created a single-cell transcriptome atlas of early embryogenesis at 4–6 weeks after fertilization, a critical time when all the major organ types are laid out and tissue types start to differentiate. This timeframe is accompanied by a sharp increase in cellular diversity and occurs after gastrulation, when embryo cells are reorganized from a 1-dimensional ball called the blastula into a multilayered and multidimensional structure called the gastrula. This results in three germ layers – endoderm, mesoderm and ectoderm – which differentiate into the different organ systems (Figure 2) [10]. It is around this time that developmental defects start to arise, which can lead to miscarriage and birth defects.

“We wanted to fill a gap in existing human embryo data, which is either at an earlier stage, at around 2 weeks maximum, or much later,” explains Shi. This is due to the challenges of obtaining embryos in this 2-week timeframe.

This atlas of early human embryogenesis was made mostly using single-cell sequencing and utilized the Visium Spatial Gene Expression kit from 10x Genomics to map the cell, tissue and germ layer identities onto the spatial gene expression pattern. As Shi explains, in spatial biology “you still have to borrow cell type information from single cell analysis.”

They captured 185,000 cells from 22 dissected parts, identified 313 cell clusters across 18 developmental systems and uncovered 134 signaling interactions, based on a statistical method developed for the study. This included observing the beginning of pacemaker cell formation in the sinus node, a cluster of myocyte cells with pacemaker activity, which had previously not been obtained so early in human development. This was achieved using 234 biomarkers that the researchers identified in 157 previous papers.

Due to the low resolution of the Visium Spatial Gene Expression kit, mapping individual cell types precisely was a challenge. Shi hopes to carry out more precise spatial mapping in the future and expand their analysis to create a multi-omic reference atlas of early human embryogenesis. “We want to map these 300 plus cell types to an accurate 3D embryo. Within these cell types, about a hundred of them were not annotated before, so we don't know the exact identity and location of them in the embryo.” Mapping these cell types will give an idea of what the cells become in later development.

Spot the missing data

Despite its name, spatial biology is not an inherently visual method. The mesmerizing images that I associate with spatial biology (exemplified by Figure 3) often consist of only a handful of representative markers that are needed to tell a story about the data. “The data largely gets decomposed into plots where it quantifies different features about each cell type across different areas,” explains John Hickey, a researcher who will be starting a lab at Duke University (NC, USA) combining spatial biology tools and developing new computational methods. “What is funny, is we have all these very beautiful pictures, but very often we do not show them because these are more qualitative than quantitative.”

Spatial biology requires “a couple of months of optimization for your specific tissue and then a month or two for imaging and collecting data and then a year, or more, of data analysis,” Hickey describes. Additionally, “there is a lot of data that we are currently just not using because spatial techniques are far in front of computational techniques needed to interpret the data.” This means a lot of nuanced information is being left out due to current processing methods [11,12].

Computers have certainly come a long way since the VEP and SAGA but so has spatial biology. The computational power needed to handle these datasets results in high technological and financial barriers to utilizing these methods, something that future innovations will hopefully address.

Taking spatial biology from two dimensions to three dimensions

Hickey suggests that 3D spatial omics may be on the horizon and there is “a really big push to collect multiplexed data inside bodies,” something which the Human BioMolecular Atlas Program (HuBMAP) has been working towards. HuBMAP is bringing together more than 60 institutions and 400 researchers to generate a high-resolution spatial multi-omic reference map of the human body at a single-cell resolution [13].

Similarly, Shi suggests that spatial techniques will make a move towards 3D, enabling the study of the whole embryo rather than sectioned samples, but this would require better microscopes and a higher image acquisition speed to be feasible. Currently, imaging a 1 cm² slide can require 12 hours of imaging with techniques like multiplexed error-robust fluorescence in situ hybridization (MERFISH), Shi explained. However, spatial methods are much more efficient than in situ hybridization techniques that can only capture one or two genes at a time. “When I was doing my PhD and postdoc, I think half the time was spent doing in situ hybridization. However, spatial omics is like a multiplexed in situ technique… and can reduce the work of PhD and postdoc students!” remarks Shi.

Looking back at the 30 years since the advent of spatial biology for the VEP, Doyle concludes: “Spatial biology's journey from its origins in the VEP to the latest advancements has been a testament to the power of interdisciplinary collaboration between the biological sciences and information technologies. As we continue to push the boundaries of what is possible, I am convinced that we will uncover even more fascinating insights about the complexity and beauty of biological systems.”