Analysis of RNA processing directly from spatial transcriptomics data reveals previously unknown regulation

  1. Julia Olivieri
  2. Julia Salzman

  • Curated by eLife

    eLife logo

    eLife assessment:

    This important study describes spatial RNA processing events by combining methods for single-cell transcriptomics data with spatial transcriptomics data. The evidence supporting the claims of the authors is solid, although the analysis could be further strengthened by including a broader range of samples as well as orthogonal validation either by experimental methods or simulated data. The work will be of general interest to researchers in the spatial transcriptomics field as well as researchers investigating alternative pre-mRNA processing across diverse tissues.

Reviewed by

Read the full article See related articles

Listed in

Log in to save this article

Abstract

Technical advances have led to an explosion in the amount of biological data available in recent years, especially in the field of RNA sequencing. Specifically, spatial transcriptomics (ST) datasets, which allow each RNA molecule to be mapped to the 2D location it originated from within a tissue, have become readily available. Due to computational challenges, ST data has rarely been used to study RNA processing such as splicing or differential UTR usage. We apply the ReadZS and the SpliZ, methods developed to analyze RNA process in scRNA-seq data, to analyze spatial localization of RNA processing directly from ST data for the first time. Using Moran’s I metric for spatial autocorrelation, we identify genes with spatially regulated RNA processing in the mouse brain and kidney, re-discovering known spatial regulation in Myl6 and identifying previously-unknown spatial regulation in genes such as Rps24, Gng13, Slc8a1, Gpm6a, Gpx3, ActB, Rps8 , and S100A9 . The rich set of discoveries made here from commonly used reference datasets provides a small taste of what can be learned by applying this technique more broadly to the large quantity of Visium data currently being created.

Article activity feed

  1. Version published to 10.7554/elife.87211.1 on eLife
  2. Version published to 10.7554/elife.87211 on eLife
  3. eLife

    Author Response

    Reviewer #1 (Public Review):

    In this study, the authors set out to investigate spatial RNA processing events, specifically alternative splicing and 3' UTR usage, in mouse brain and kidney tissues using ReadZS and SpliZ methodologies on spatial transcriptomics data. The research contributes to understanding tissue-specific gene expression regulation from a spatial perspective. The study introduces a novel approach for analyzing spatial transcriptomics data, allowing for the identification of RNA processing and regulation patterns directly from 10X Visium data. The authors present convincing evidence supporting the identification of novel RNA processing patterns using their methodology, which holds significant implications for researchers in the field of spatial transcriptomics and the study of alternative splicing and 3' UTR usage.

    Thank you for this thorough overview of our work.

    The conclusions of the study are mostly well-supported by the data; however, certain aspects could be improved to strengthen the findings.

    1. The conclusions of this study would be strengthened by conducting a more extensive tissue sample analysis and including biological replicates. Additionally, appropriate batch effect corrections should be applied when dealing with biological replicates.

    We agree that including biological replicates would strengthen our findings. We will include biological replicates of the mouse brain tissues in the revision.

    1. The 3' UTR usage and alternative splicing should be compared among clearly labeled clusters for a more comprehensive analysis.

    We understand that it can be difficult to see how the SpliZ quantiles map spatially onto the tissue images. For the splicing of Gng13, Myl6, and Rps24, we will include box plots broken down by spatial quadrant in the revision. However, this does result in an oversimplification of the spatial patterns found in the tissue slices, which make the plots less informative than the quantile plots to our view.

    1. The authors should clarify their rationale for choosing ReadZS and SpliZ approaches and provide comparisons with other methods to demonstrate the advantages and potential limitations of their chosen methodologies.

    Thank you for pointing out the lack of sufficient discussion of ReadZS and SpliZ in the manuscript. The ReadZS and SpliZ were chosen for this analysis because both of these methods provide an individual score for each cell-gene pair, which is easily adapted to providing a score for each spot-gene pair. Due to the sparsity and 3’ bias of Visium data, approaches designed to analyze RNA processing in full-length sequencing analysis are not applicable. The SpliZ and ReadZS are two of the limited number of tools available that are designed for the analysis of RNA processing in droplet-based data. Other available tools tend to rely on aggregating data across multiple cells using a method called pseudo-bulking (Li et al., 2021; Patrick et al., 2020). It is not clear how this could be used for spatial transcriptomics data without potentially obscuring subtle spatial patterns in the data. Others are based on PSI measurements, which are vulnerable to artifacts due to sparsity (Buen Abad Najar et al., 2020; Olivieri et al., 2022; Wen et al., 2022). The tradeoff between pseudo-bulking and a single score per spot-gene pair means that the ReadZS and SpliZ do not have the power to detect changes for genes with very low read counts. We will add text in the revision to clarify this point.

    Reviewer #2 (Public Review):

    The authors applied existing ReadZS and the SpliZ methods, previously developed to analyze RNA process in scRNA-seq data, to Visium data to study spatial splicing and RNA processing events in tissues by Moran's I. The authors showed several example genes in mouse brain and kidney, whose processing are spatially regulated, such as Rps24, Myl6, Gng13.

    Thank you for this thorough overview of our work.

    The paper touches on an important question in RNA biology about how RNA processing is regulated spatially. Both experimental and computational challenges remain to address it. Despite some potentially interesting findings, most claims remain to be validated by orthogonal methods such as RNA FISH and simulations.

    We appreciate that the reviewer finds the question important, and that the findings are potentially interesting. In the revision we will include biological replicates for our findings in the mouse brain. Unfortunately, experimental validation is outside of our budget for this project. It is unclear what further simulations could validate the biological discoveries in this manuscript: permutations were used to calculate the p value of each discovery, and the false positive and negative rates of the SpliZ have been assessed through simulation (Olivieri et al., 2022).

    In addition, the percentage of spatial processing events (splicing in 0.8-2.2% of detected genes, i.e. 8-17 genes and RNA processing in 1.1-5.5% of detected genomic windows, i.e. 57-161 windows) discovered is low. Does it suggest that most of RNA processing events were not spatially regulated across the tissue? Or does it question the assumption of treating spatial transcriptomics data similar to scRNA-seq data?

    We agree that the question of the prevalence of spatial RNA processing regulation is critical. Rather than the two options proposed here, we believe that the sparsity of the data limits our ability to call more of these events. In the revision, we will provide a supplemental figure showing the relationship between read depth and p value for each gene to quantify how the fraction of observed regulation changes with sequencing depth. It is worth noting that as these technologies improve, we expect the sequencing depth of spatial technologies to increase which would likely result in more discoveries.

    The unique features for ST data, such as mixture of neighboring cells, different capture biases and much smaller number of spots (pseudo cells here), may have significant effects on the power of scRNA-seq based methods, but it is not discussed in the manuscript. The lack of careful evaluation and low discovery rates could limit application of the approach to other tissues and subcellular data.

    We appreciate the concern that technical differences between scRNA-seq data and spatial transcriptomics data could affect our results. We agree that this point could be addressed more thoroughly in the text. None of the specificities of spatial transcriptomics data invalidate the assumptions of the SpliZ or ReadZS. The method we use to identify genes with significant spatial regulation of RNA processing was specifically created to be used for Visium data. It takes into account mixture of RNAs in neighboring cells by randomly sampling scores of neighboring cells, rather than randomization of the location of the spots themselves, which does indeed result in a high false positive rate (see “Permutations for Moran’s I” in the Methods). We do note that there is a limit to the power of this kind of analysis based on the number of spots and the read depth, which we will quantify in a plot in the revision.

  4. eLife

    eLife assessment:

    This important study describes spatial RNA processing events by combining methods for single-cell transcriptomics data with spatial transcriptomics data. The evidence supporting the claims of the authors is solid, although the analysis could be further strengthened by including a broader range of samples as well as orthogonal validation either by experimental methods or simulated data. The work will be of general interest to researchers in the spatial transcriptomics field as well as researchers investigating alternative pre-mRNA processing across diverse tissues.

  5. eLife

    Reviewer #1 (Public Review):

    In this study, the authors set out to investigate spatial RNA processing events, specifically alternative splicing and 3' UTR usage, in mouse brain and kidney tissues using ReadZS and SpliZ methodologies on spatial transcriptomics data. The research contributes to understanding tissue-specific gene expression regulation from a spatial perspective. The study introduces a novel approach for analyzing spatial transcriptomics data, allowing for the identification of RNA processing and regulation patterns directly from 10X Visium data. The authors present convincing evidence supporting the identification of novel RNA processing patterns using their methodology, which holds significant implications for researchers in the field of spatial transcriptomics and the study of alternative splicing and 3' UTR usage

    The conclusions of the study are mostly well-supported by the data; however, certain aspects could be improved to strengthen the findings.

    1. The conclusions of this study would be strengthened by conducting a more extensive tissue sample analysis and including biological replicates. Additionally, appropriate batch effect corrections should be applied when dealing with biological replicates.
    2. The 3' UTR usage and alternative splicing should be compared among clearly labeled clusters for a more comprehensive analysis.
    3. The authors should clarify their rationale for choosing ReadZS and SpliZ approaches and provide comparisons with other methods to demonstrate the advantages and potential limitations of their chosen methodologies.
  6. eLife

    Reviewer #2 (Public Review):

    The authors applied existing ReadZS and the SpliZ methods, previously developed to analyze RNA process in scRNA-seq data, to Visium data to study spatial splicing and RNA processing events in tissues by Moran's I. The authors showed several example genes in mouse brain and kidney, whose processing are spatially regulated, such as Rps24, Myl6, Gng13.

    The paper touches on an important question in RNA biology about how RNA processing is regulated spatially. Both experimental and computational challenges remain to address it. Despite some potentially interesting findings, most claims remain to be validated by orthogonal methods such as RNA FISH and simulations. In addition, the percentage of spatial processing events (splicing in 0.8-2.2% of detected genes, i.e. 8-17 genes and RNA processing in 1.1-5.5% of detected genomic windows, i.e. 57-161 windows) discovered is low. Does it suggest that most of RNA processing events were not spatially regulated across the tissue? Or does it question the assumption of treating spatial transcriptomics data similar to scRNA-seq data? The unique features for ST data, such as mixture of neighboring cells, different capture biases and much smaller number of spots (pseudo cells here), may have significant effects on the power of scRNA-seq based methods, but it is not discussed in the manuscript. The lack of careful evaluation and low discovery rates could limit application of the approach to other tissues and subcellular data.

  7. Version published to 10.1101/2023.03.13.532412v1 on bioRxiv