SPARK-X and large spatial transcriptomic studies

Monika R, PSG College of Technology, Coimbatore

The fundamental unit of multicellular organisms is cells. The development of cellular and molecular biology has been boosted after the establishment of single-cell RNA sequencing (scRNA-seq), which provides information on the transcriptome of each individual cell to a group of cells with similar transcription patterns. Based on this, cells are then characterized into various types. To understand cell-type identity in multicellular organisms, one must integrate individual cells’ transcriptome profiles with their spatial position in certain tissue. But with solid tissues, to obtain cell suspension for the subsequent scRNA-seq analysis, a dissociation step should be performed. This process may lead to the loss of spatial information, which is essential for cellular fate and property. So, to preserve the spatial information, spatially resolved transcriptomic studies (techniques to resolve spatial information) are being introduced, which conducts gene-expression profiling with spatial localization information on cell and tissue cultures. These studies allow us to look at the spatial expression (SE) patterns of genes on the tissue, characterizing local structures, microenvironments and detecting cell-cell interactions across spatial locations.

Spatially resolved transcriptomic techniques

Single-molecule RNA fluorescence in-situ hybridization has been applied to quantitate RNA transcripts at single-cell resolution. But only a few genes can be measured. This led to the emerging of other imaging-based approaches – Multiplexed error-robust FISH (MERFISH), seqFISH, and sequencing-based methods – LCM-seq, Tomo-seq, spatial transcriptomics (ST), Slide-seq, Slide-seqV2, HDST, and 10X-Visium data, to obtain spatially resolved gene expression even at single-cell/subcellular resolution. 

• Slide-seq technology enables transcriptome-wide measurements by transferring RNA from tissue sections onto a surface covered in DNA-barcoded beads with known positions and inferring the locations of RNA employing a sequencing-by-ligation strategy.
• Slide-seqV2 technology emerges from Slide-seq with modifications to library generation, bead-synthesis and, array-indexing, improving mRNA capture sensitivity. 
• The high-definition spatial transcriptomics (HDST) method captures RNA from tissue sections on a dense, spatially barcoded bead array, for transcriptome-wide measurements.
• 10X-Visium technology emerges from original ST technology with improvements on resolution and experimental time.

These techniques have made it possible to study the spatial organization of transcriptomic landscape across tissue sections/within single cells, impacting many fields, including neuroscience, developmental biology and immunology.

Methods for Spatial Expression analysis

A key role in spatial transcriptomic studies is to identify genes that display any one of the three spatial expression patterns (hotspot, streak and gradient). So, the common methods for SE analysis include trendsceek, SpatialD, and SPARK. 

• SpatialD – Provides visualization of spatially resolved transcriptomic data and quick retrieval of SE expression in tissue of interest.

Figure 1. Overview of SpatialDB database.

Source: doi:10.1093/nar/gkz934 

• SPARK – Spatial pattern recognition via kernels (SPARK), builds upon a generalized linear spatial model with a variety of spatial kernels to accommodate count data generated from smFISH or sequencing-based spatial transcriptomics studies. With the penalized quasi-likelihood algorithm, SPARK is scalable and relies on χ2 distributions which further uses the Cauchy combination rule to combine information across multiple spatial kernels for calibrated P-value. SPARK controls type I error and is powerful for identifying SE genes.

But these methods are limited in the case of large-scale spatial transcriptomic (LSST) data because they would take days/months to analyze. Similarly, the memory requirement of both methods also scales cubically concerning the number of spatial locations, because they require large GB RAM. Finally, the LSST data are often in the form of sparse counts. Direct modelling of sparse data with binomial/Poisson/Gaussian distributions (SPARK-G) is not ideal, because they incur algorithm stability issues, and lead to failure of convergence in more than 90% of genes in LSST data. Such parametric approximation also leads to power loss and failure of type I error control at small P-values that are essential for detecting SE genes at the transcriptome-wide significance level. 

SPARK-X (SPARK-eXpedited), for SE analysis, is a scalable non-parametric test. SPARK-X builds upon a covariance test framework and extends it to incorporate a variety of spatial kernels for non-parametric spatial modelling of sparse count data from LSST studies. SPARK-X also reduces computational complexity and RAM requirement for SE analysis from cubic to linear concerning the number of spatial locations, resulting in computational speed improvements and RAM savings. Due to its non-parametric nature, SPARK-X is statistically robust and algorithmically stable, providing calibrated type I error control and improved power across a range of data types collected through a variety of technical platforms.

Simulation

Three LSST data collected by different technologies were analyzed for SE by various methods to find out the best SE analyzing method. 

1. Slide-seq mouse-cerebellum data contains gene expression measurements for 17,729 genes on 25,551 beads. SPARK-X identified 2336 SE genes, whereas SPARK-G identified 212 (180 overlapped with SPARK-X). SpatialDE was unable to detect any SE genes. SPARK-X took 3min to analyze the whole data, while SPARK-G took 56h and SpatialDE 47h.
2. Slide-seqV2 mouse-cerebellum data contains 20,117 genes on 11, 626 beads. SPARK-X identified 688 SE genes, while SPARK-G identified 112 (68 overlapped). SpatialD was unable to detect any SE genes. SPARK-X took 2min to analyze the whole data, while SPARK-G took 13h and SpatialDE 8h.
3. HDST mouse-olfactory bulb data contains 19,913 genes on 177,455 spots. This data is particularly challenging due to the large number of spots measured there. SPARK-G will take 114 days and 2100GB memory and SpatialDE takes 80days and 3500GB memory to analyze data, whereas SPARK-X requires 0.42GB memory and 3min of computing time and identified 125 SE genes. This is also the only SE method applicable for this data. 

In these analyses, many new SE genes including those that display spatial expression patterns within the same cell type were identified. These SE genes are involved in the synaptic organization and functional compartmentalization of the cerebellum also involved in lateral inhibition and odor discrimination in the olfactory system.

Visium human-ovarian cancer data through SPARK-X – The data contains 1198 genes on 3492 spots, SPARK-X identified 651 SE genes while SPARK identified 579 (474 overlapped). Importantly, many cancer-related KEGG and Reactome pathways can only be identified based on the SE genes detected by SPARK-X highlighting the benefits of SPARK-X analysis.

Also read: Doxorubicin and Epoxomicin loaded Nanoparticles for accelerating cancer cell apoptosis

Source:

• Zhu, J., Sun, S. & Zhou, X. (2021). SPARK-X: non-parametric modeling enables scalable and robust detection of spatial expression patterns for large spatial transcriptomic studies. Genome Biol 22, 184. doi:10.1186/s13059-021-02404-0
• Sun, S., Zhu, J., & Zhou, X. (2020). Statistical analysis of spatial expression patterns for spatially resolved transcriptomic studies. Nature methods, 17(2), 193–200.  doi:10.1038/s41592-019-0701-7
• Fan, Z., Chen, R., & Chen, X. (2019). SpatialDB: a database for spatially resolved transcriptomes. Nucleic Acids Research. doi:10.1093/nar/gkz934 

• The Corrosion Prediction from the Corrosion Product Performance
• Nitrogen Resilience in Waterlogged Soybean plants
• Cell Senescence in Type II Diabetes: Therapeutic Potential
• Transgene-Free Canker-Resistant Citrus sinensis with Cas12/RNP
• AI Literacy in Early Childhood Education: Challenges and Opportunities

About the author: Monika R is an enthusiastic Biotech student aspiring for an opportunity to develop skills and grow professionally in the research field. Extremely motivated and possess strong interpersonal skills and the ability to learn concepts quickly.