Data-driven Identification of Total RNA Expression Genes (TREGs) for Estimation of RNA Abundance in Heterogeneous Cell Types

Short talk at BioC22 Conference, July 2022

Next generation sequencing technologies have facilitated data-driven identification of gene sets with different features including housekeeping genes, cell-type specific expression, or spatially variable expression. Here, we sought to identify a new class of "control" genes called Total RNA Expression Genes (TREGs), which correlate with total RNA abundance in heterogeneous cell types of different sizes and transcriptional activity. We provide a data-driven method to identify TREGs from single nucleus RNA-seq data (snRNA-seq), available as an R/Bioconductor package at http://research.libd.org/TREG/.

Related to biorxiv.org/content/10.1101/2022.04.28.489923v1

Transcript

1. Data-driven Identification of Total RNA
   Expression Genes (TREGs) for
   Estimation of RNA Abundance in
   Heterogeneous Cell Types
   Louise Huuki-Myers
   Staff Scientist
   Lieber Institute for Brain Development
   @lahuuki
   1
   

   View Slide

2. Motivation
   ● Study different cell types in the human
   brain using smFISH with RNAscope
   ○ Cell size
   ○ Spatial organization in tissue
   ● Interested in total RNA expression of
   different cell types
   ● Can only measure four genes at a time
   with smFISH
   How do we measure total RNA content of a cell if we can only
   observe a few genes at a time? Use a TREG
   2
   

   View Slide

3. What is a TREG?
   ● Total RNA Expression Gene
   ● Expression is proportional to the
   overall RNA expression in a cell
   ● In smFISH the count of TREG
   “puncta” in a cell can estimate the
   RNA content Each point
   of light is a
   “puncta”
   3
   

   View Slide

4. Data Driven TREG Discovery
   Identify TREGs in Single Nucleus RNA-seq data
   1. Filter to top 50% expressed genes
   2. Filter out genes with high Proportion Zero
   expression
   3. Select genes with high Rank Invariance as
   candidate TREGs
   4
   

   View Slide

5. Evaluating for Rank Invariance
   ● High Rank Invariance = Expressed at a
   constant level in respect to other genes
   ○ Evaluate within and across cell types
   ○ Selecting for stable “Expression Rank:
   ■ By Cell: Higher Expression = Higher Rank
   5
   75
   1
   2
   3
   Cell 1 Cell 2 Cell 3 Cell 4 Cell 5
   Gene 1 75
   Gene 2 1
   ... 3
   Gene 100 50
   For each cell: rank genes
   = Expression Rank
   

   View Slide

6. ● High Rank Invariance = Expressed at a
   constant level in respect to other genes
   ○ Evaluate within and across cell types
   ○ Selecting for stable “Expression Rank:
   ■ By Cell: Higher Expression = Higher Rank
   Evaluating for Rank Invariance
   6
   75
   1
   2
   3
   Cell 1 Cell 2 Cell 3 Cell 4 Cell 5
   Gene 1 75 75 74 73 75
   Gene 2 1 90 55 75 20
   ...
   Gene 100
   Select for stable/invariant
   Expression Rank
   

   View Slide

7. Evaluating for Rank Invariance
   7
   

   View Slide

8. TREG R Package
   Functions
   8
   ● Gene filtering
   ○ get_prop_zero()
   ○ filter_prop_zero()
   ● Rank Invariance step-wise
   ○ rank_cells()
   ○ rank_group()
   ○ rank_invariance()
   ● rank_invariance_express()
   ○ Completes full RI calculations
   http://research.libd.org/TREG/
   

   View Slide

9. Experimental Design
   Step 2: Evaluate Candidate TREGs with smFISH
   smFISH with RNAscope in DLPFC tissue
   ● 9 tissue sections from one donor
   ● Evaluate three candidate TREGs vs. a
   Housekeeper Gene
   Step 1: Find Candidate TREGs in Single Nucleus
   data
   Human Brain postmortem 10x snRNA-seq
   ● Tran, Maynard et al. Neuron, 2021
   ● 70k nuclei
   ● Five Brain Regions, Nine broad cell types
   9
   

   View Slide

10. Selecting Candidate
    TREGs
    ● Selected AKT3, ARID1B, and
    MALAT1 as our candidate TREGs
    ● TREGs show high rank invariance
    within all cells and with in most cell
    types
    ● High correlation between total RNA
    expression and TREG expression
    across cell types
    10
    Total RNA per nucleus
    in snRNA-seq
    Expression of TREG per nucleus
    😈
    

    View Slide

11. Validation in smFISH + RNAscope
    ● Segmented nuclei and quantified TREG
    puncta with HALO software
    ● Segmentation of MALAT1 fails
    11
    
    

    View Slide

12. TREG quantification across DLPFC tissue
    12
    ● We observed more TREG expression in transcriptionally
    active grey matter
    ○ Aligns with location of neurons
    ● Follows with expected pattern of RNA expression
    

    View Slide

13. Patterns of Observed Puncta over Cell Types
    ● TREGs were expressed in most
    cells
    ● AKT3 tracks really well with
    pattern of expression across cell
    types seen in snRNA-seq
    ○ ARID1B also performs well
    RNAscope
    Gene
    Mean Prop.
    Cells with
    Expression
    Prop.
    non-zero in
    dlpfc snRNA
    Standard β
    (sn = -1.33)
    AKT3 0.88 0.92 -1.07
    ARID1B 0.86 0.94 -0.77
    MALAT1 0.98 1.00 -0.8
    POLR2A 0.78 0.30 -1.05
    13
    

    View Slide

14. Conclusion
    ● TREGs would allow the observation of
    total RNA expression via RNAscope
    ● Proportion Zero filtering + Rank Invariance
    allow selection of candidate TREGs in
    sn/scRNA-seq
    ● AKT3 appears to be a TREG compatible
    with RNAscope in the human brain
    ● Pre-print: doi.org/10.1101/2022.04.28.489923
    ● Bioconductor Release 3.15
    ○ bioconductor.org/packages/TREG
    ○ research.libd.org/TREG/
    ● Apply TREG methodology to find TREGs in
    other tissues or experimental settings
    14
    

    View Slide

15. Acknowledgements
    Leonardo
    Collado-Torres
    Kristen Maynard Stephanie C Hicks
    LIBD Johns Hopkins
    15
    Kelsey
    Montgomery
    Sang Ho Kwon Pre-print: doi.org/10.1101/2022.04.28.489923
    Stephanie Page
    Say hi on Twitter! @lahuuki
    

    View Slide