The complete cell atlas of C. elegans aging
Calico research

The complete cell atlas of C. elegans aging

Individual cell types in C. elegans age differently and activate distinct cell-protective responses
https://doi.org/10.1016/j.celrep.2023.112902
https://doi.org/10.1016/j.celrep.2023.112902
Antoine E. Roux
Calico
Han Yuan
Calico
Katie Podshivalova
Calico
David G. Hendrickson
Calico
Rex Kerr
Calico
Cynthia Kenyon
Calico
David R. Kelley
Calico

The complete cell atlas of C. elegans aging

This website is designed to share our aging C. elegans dataset and help users run their own analysis. All the details about the data collection are available in the publication. We measured gene expression in nearly every somatic cell type across the adult lifespan of C. elegans. Data were collected at six different timepoints ranging from young to old adults. It provides an original and near-complete resource for cell-type specific expression during the course of C. elegans adulthood. The user interface can be accessed at the next tab, or independently at User Inferface. If the page is stuck at 'loading user interface', please try again using Chrome as the browser.

C. elegans scRNA-seq uncovers nearly every known tissue type

single cell RNA-sequencing of adult C. elegans population reveals 47,423 cells quantifying 20,305 genes. Unsupervised clustering reveals 211 distinct cell type clusters. Cell type annotation is performed using WormBase anatomy gene sets.

single cell RNA sequencing of aging C. elegans

The C. elegans population was grown at 25˙C in killed bacteria and adults were harvested at 6 different timepoints from day1 to day15. W used gon-2(q388) mutants that significantly reduce gonad development. To further enrich for somatic cells, we stained dissociated cells and used FACS to sort out most germ cells with 4N and N ploidy. We filtered out the remaining germ cells in silico. Intestinal cells have a ploidy of 32N. Nonetheless, we recovered a small number of them. Intestinal and germline clusters might not represent the full expression pattern of this tissue and should be interpreted carefully. For more details, please refer to the publication.

The final dataset contained 47,423 cells quantifying 20,305 genes across the full time series. We recovered more cells and unique molecular identifiers (UMIs) per cell at early time points as compared to late time points. We performed Leiden clustering and identified a total of 211 distinct clusters. Here's the UMAP view of all 211 cell type clusters:

cell type annotation

We annotated the clusters by combining a systematic and manual analysis to find enrichment in anatomy marker genes available in Wormbase. The clusters cover a large portion of known cell types from every tissue, including the hypodermis, seam cells, various muscles, glands, glia, neurons, etc. Our annotation matched clusters to the large majority of known cell types. For instance, we found 133 neuronal clusters and 22 muscle clusters. 73% of the clusters are described by a single anatomy term, 18% by more than one anatomy term. Annotations for over 15 clusters were cross-verified by microscopy and confirmed that our method is reliable.

Today, annotation accuracy is limited due to the lack of known cell-type marker genes in some cases. Improving annotation will be an ongoing process as more single cell and imaging data are collected by the community. We wish to update our annotation with more recent observations, please contact us if you want to share your findings and help us improve our cluster annotations.

cell type specific gene expression and regulations

Our single cell data provide an unprecedented view into cell-type specific gene expression and transcription factor (TF) activity in the adult worm. You can explore the gene expression patterns and TF activity in our user interface.

cell type specific gene expression

We performed differential expression analysis for each cell type cluster to elucidate specific gene expression patterns. We identified an average of 14.4 marker genes per cell-type cluster. For example, we recapitulated dhp-1 as a marker for hypodermis, ceh-33 for head muscle and gcy-23 for the AFD neuron We also identified many novel markers. For example, the carbohydrate binding protein clec-166 emerged as a novel marker for M5 neurons, distinguishing them better than does the known marker vab-15. We validated M5 neuron-specific expression of clec-166 in vivo using a fluorescent reporter under the control of the clec-166 promoter and 3’-UTR:

cell type specific TF activity

Differential gene expression across cell types is driven by regulatory factor activity. We quantified TF target activity in each single cell by computing the AUCell statistic for TF targets (predicted by FIMO motif hits). We evaluated TF expression and activity across the 211 cell-type clusters and observed tissue-specific transcriptional regulation. We identified a total of 1,048 tissue-specific associations of the TF expression with its target genes activity across the atlas, including body-wall muscle specific HLH-1 activity, seam cell-specific ELT-1 activity, intestine-specific ELT-7 activity, and IL2 neuron-specific DAF-19 activity.

We also quantified the relationship between TF expression and activity in tissues where the TF of interest is expressed by computing the correlation between TF RNA expression and TF motif activity across single cells. We identified 48 TFs whose expression and activity significantly correlated in at least one cell type, as shown below:

C. elegans aging atlas

Our data revealed a high level of tissue-specificity in transcriptional changes during aging. We estimated the magnitude of aging-related change in each cell type using a single MMD statistic and found that this estimated rates of aging could be validated in vivo by examining age-dependent changes in mitochondrial morphology. Based on TF expression signatures, we identified candidates for new potential TFs regulating aging. You can explore the gene expression changes during aging in each cell type cluster using our user interface.

changes in gene expression and pathways during aging

Our data enabled the first comprehensive analysis of cell-type aging signatures across an entire animal. We first performed a coarse differential-expression analysis of young cells (day 1, 3 and 5) versus old cells (days 8, 11 and 15) to visualize global trends. The following plot displays a heatmap of log2 fold change for the 4,541 genes differentially expressed in at least one cluster:

We first looked for gene expression changes that were shared most broadly across cell types. We found that the most frequently up-regulated genes were the bidirectionally-transcribed HSP70 chaperone paralogs F44E5.4 and F44E5.5, orthologs of human HSPA6. Several additional heat-shock protein genes were frequently up-regulated. The most commonly down-regulated aging gene is hsp-4 in 45% of clusters. hsp-4 with the following top three down-regulated genes are involved in the ER-UPR (hsp-4, cup-2, pdi-6, xbp-1), suggesting that decreased ER resilience might contribute to aging.

We performed GSEA for each individual cluster based on the log2 fold change in old versus young cells comparison. We observed that pathways related to energy metabolism including mitochondrial respiration, ATP synthesis, glycolytic process and tricarboxylic acid cycle were the most frequently down-regulated:

global transcriptional characterization during aging

Gene expression drift is a common correlate of aging. We sought a full transcriptome quantification for how far gene expression changed with age in each cell type. We first quantified the average magnitude of transcriptome change between consecutive timepoints by maximum mean discrepancy (MMD). MMD is a statistical test measuring the distance between two distributions ; here, the cell embeddings in different timepoints. We assume that the degree to which MMD deviates with time is a proxy for how much each cell type is altered by aging, and we refer to it as the aging amplitude.

We observed significant changes in at least two consecutive timepoints for 158/165 cell types. Interestingly, MMD measurements exhibited substantial heterogeneity both between different cell types and within a single tissue type. In general, neuron clusters exhibited a significantly greater aging amplitude than did non-neuronal clusters.

In additional to MMD, we quantified global transcriptional changes in cell-cell variation and gene variability for each cell type cluster. For two selected cell types (uterine seam cells and coelomocytes), we predicted that uterine seam cells age faster, supported by greater MMD (0.10 vs. 0.07 and 0.04), cell-cell variation (3.04 vs. -1.54 and -0.47), and gene variability (3.23 vs. 2.66 and -3.12). Using fluorescent microscopy, we scored age-related change in mitochondria morphology concomitantly in the uterine seam and coelomocytes of multiple individuals. We observed that, the uterine seam cells have more fragmented mitochondria than the coelomocytes, which is associated with heath decline and increased mortality in previous studies.

changes in TF regulation during aging

Transcription factors play a crucial role in the regulation of longevity across many organisms. We summarized TF expression change across age for each cluster by its log2FC and target gene change by a t-statistic comparison of target gene AUCell scores in young versus old cells. Remarkably, of the eight top most globally age-regulated TFs, seven are well-known regulators of C. elegans aging in GenAge database. gei-3 is a universally up-regulated TF during aging that is not previously known to play a regulatory role. We found universal up-regulation of gei-3 and its predicted targets, suggesting that it could be an interesting new longevity regulator.

Based on the observation we investigated 55 TFs whose expression (or target activity) changed with age but were not linked previously to aging. We used RNA interference (RNAi) to turn down their expression starting from early adulthood and systematically recorded animal movement rates and survival over time. We found that RNAi of three such TF genes, gei-3, lsy-2, and mef-2 accelerated the rate of movement decline and mortality. gei-3 RNAi produced the largest effect (shown below), similar to that of RNAi inhibition of a positive control longevity regulator, the heat shock TF hsf-1.