LipocyteProfiler

Discovering cellular programs of intrinsic and extrinsic drivers of metabolic traits in adipocytes using LipocyteProfiler

S. Laber and S. Strobel et al.

LipocyteProfiler is an unbiased high-throughput microscopy profiling assay that builds on the CellProfiler pipeline and uses a combination of cellular stains designed for large-scale profiling of lipid-accumulating cell types. We applied this approach to survey diverse cellular mechanisms by generating context-, process-, and allele-specific morphological and cellular profiles, that offer insights into metabolic disease risk mechanisms. This repository hosts analysis and visualisations of LipocyteProfiler and RNA-seq data as shown in S. Laber and S. Strobel et al.

Lipocyte Profiles

LipocyteProfiles (LP) are generated from 3,005 morphological and cellular features that map to three cellular compartments (Cell, Cytoplasm, Nucleus) across four channels differentiating the organelles, namely DNA (Hoechst), Mito (MitoTracker Red which stains mitochondria), AGP (actin, golgi, plasma membrane; stained with Phalloidin (F-actin cytoskeleton) and Wheat Germ Agglutinin (golgi and plasma membranes), and Lipid (BODIPY, which stains neutral lipids, multiplexed with SYTO14, which stains nucleoli and cytoplasmic RNA). We showed that LipocyteProfiles can be used to survey diverse cellular mechanisms of cell types, polygenic-risk of metabolic disease and allelic-risk of common complex diseases

• LipocyteProfiles of white adipocytes compared to brown adipocytes (generation of LP - different cell types.R; Figure 1g)

• LipocyteProfiles of subcutaneous vs. visceral adipocytes across differentiation (generation of UMAP LP.R; Figure 2b)

• LipocyteProfiles of AMSCs treated with isoproterenol (generation of LP - isproterenol treatment.R; Figure 4b)

• LipocyteProfiles of primary human hepatocytes with several drug treatments (generation of LP in hepatocytes with drug treatment.R; Figure; 4f+h)

• LipocyteProfiles of tail ends of polygenic-risk of insulin resistance (generation of LP - high vs low polygenic risk.R; Figure 5b)

• LipocyteProfiles of polygenic-risk for lipodystrophy (generation of LP - linear regression model.R; Figure 6b)

• LipocyteProfiles of allelic risk of 2p23.3 locus (generation of LP - risk vs non-risk haplotype single loci.R, Figure 7)

  • Visualisation tools:
    • Heatmap (visualisation of LP - heatmap.R)
    • Pie Chart (visualisation of LP - pie chart.R)
    • Barplot (visualisation of LP - barplot.R)

Extrinsic and Intrinsic Variance on LipocyteProfiles

Imaging based high-throughput profiling data was generated from samples derived from different donors and batches. We assessed the effect of both extrinsic and intrinsic variance on LipocyteProfiler features by performing:

• Variance component analysis across all data collected on LipocyteProfiles of 65 donor-derived differentiating AMSCs. We assessed intrinsic genetic variation compared to the contribution of other possible confounding factors such as batch, adipose depot, T2D status, age, sex, BMI, cell density, month/year of sampling, and passage numbers. (LP variance explained across data set.R; Figure S3)
• To obtain a measure of batch-to-batch variance associated with our experimental set-up, we differentiated hWAT, hBAT and SGBS preadipocytes in three independent experiments and tested batch effects by visualizing the data using a Principle component analysis and quantifying it using a Kolmogorov-Smirnov test implemented in the “BEclear” R package. (BEclear.R; PCA hWAT hBAT SGBS.R; Figure S2a)
• We performed a k-nearest neighbour (knn) supervised machine learning algorithm implemented in the “class” R package to investigate the accuracy of predicting biological and technical variation. For this analysis the data set, consisting of 3 different cell types (hWAT, hBAT, SGBS) distributed on the 96-well plate, imaged at 4 days of differentiation, was split into equally balanced testing (n=18) and training (n=56) sets. Accuracy of the classification model was predicted based on three different categories cell type, batch and column of the 96-well plate. (knn-analysis of hWAT hBAT SGBS.R; Figure S2)

Networks

Networks of LipocyteProfiles and RNA-seq data linked gene sets with morphological and cellular features, capturing a broad range of cell activity and identifying relevant cellular processes. We generated those networks using a linear regression model across 2,760 LipocyteProfiler features and expression of 52,170 genes across differentiation in both adipocytes depot. (LMM_gene_LP.R; Figure 3). Using this network we interrogated association of expression of a specific gene to LP features (LP gene profile - features extraction.R Figure 3c) or identified transcriptional pathways of a specific LP feature and correlated genes. (Enrichr pathway analysis.R; Figure 3b)

• Visualisation tools:
  • Network using “igraph” R package (visualisation of LMM network.R; Figure 3b)
  • Heatmap of LipocyteProfile of marker genes (visualisation LP of gene - heatmap; Figure 3c)