@alecplot

PhD Candidate in Bioinformatics and Computational Biology at UNC-Chapel Hill. Optimal transport, cell fate, and CD8 T cells.

 Chapel Hill, NC
Joined August 2019
  • Tweets63
  • Following185
  • Followers105
Even after migration slows, differentiation continues within each tissue. Applying our OT + integration pipeline to gut-resident CD8s alone revealed distinct TRM paths, and highlighted CD52 as a potential late stem-like TRM marker. 7/
Left: Sankey diagram showing population flow over time for CD8 T cell trajectory clusters in the gut. Early effector and terminal effector clusters peak at day 4 and contract over time, the effector memory cluster maintains a consistent proportion over time, and the memory cluster gradually increases in proportion over time. Right: Line plot of Cd52 gene expression over time, with different colored lines for each trajectory cluster. All clusters have high Cd52 expression at day 7 that falls over time; however the memory cluster falls more slowly, and shows noticeably higher expression than the other clusters from days 14-90.

ALT Left: Sankey diagram showing population flow over time for CD8 T cell trajectory clusters in the gut. Early effector and terminal effector clusters peak at day 4 and contract over time, the effector memory cluster maintains a consistent proportion over time, and the memory cluster gradually increases in proportion over time. Right: Line plot of Cd52 gene expression over time, with different colored lines for each trajectory cluster. All clusters have high Cd52 expression at day 7 that falls over time; however the memory cluster falls more slowly, and shows noticeably higher expression than the other clusters from days 14-90.

1
44
Multi-compartment OT also let us compare early vs. late migration: 🌅 Early arrivals to the gut were enriched for resident memory (TRM)-associated genes 🌃 Late arrivals were more terminal effector-like In vivo antibody labeling confirmed these timing-dependent phenotypes! 6/
Volcano plot showing negative log10-adjusted p-values over log2-fold change expression for genes in early arrivals versus late arrivals to the gut. Early arrival genes include Cxcr6, Klrd1, Ifng, Ctla4, Cd69, Ccr9, Cd28, Hilpda, P2rx7, Klrc2, and Il7r. Late arrival genes include Lgals1, Hmgb1, Hsp90ab1, Mif, Anxa2, Hsp90aa1, Eno1, Pdia3, Calr, Peb1, Hnrnpu, P4hb, Ly6c2, and Lgals3.

ALT Volcano plot showing negative log10-adjusted p-values over log2-fold change expression for genes in early arrivals versus late arrivals to the gut. Early arrival genes include Cxcr6, Klrd1, Ifng, Ctla4, Cd69, Ccr9, Cd28, Hilpda, P2rx7, Klrc2, and Il7r. Late arrival genes include Lgals1, Hmgb1, Hsp90ab1, Mif, Anxa2, Hsp90aa1, Eno1, Pdia3, Calr, Peb1, Hnrnpu, P4hb, Ly6c2, and Lgals3.

1
46
We next asked: how is CD8 differentiation linked with migration into tissues? We modeled spleen (circulation) and gut (tissue) datasets jointly with OT and found that most migration happens around days 3-4 post infection, tapering by day 7, consistent with prior experiments. 5/
Left: UMAP dimensionality reduction shaded by CD8 compartment (gut/siIEL or Spleen). Right: Sankey diagram showing migration of CD8 T cells between compartments over time. All cells start in the spleen at day 3, with most gut cells arriving between days 3-4. A smaller proportion of cells is exchanged between compartments between days 4-7, while only minimal movement between compartments occurs after day 7.

ALT Left: UMAP dimensionality reduction shaded by CD8 compartment (gut/siIEL or Spleen). Right: Sankey diagram showing migration of CD8 T cells between compartments over time. All cells start in the spleen at day 3, with most gut cells arriving between days 3-4. A smaller proportion of cells is exchanged between compartments between days 4-7, while only minimal movement between compartments occurs after day 7.

1
1
51
Clustering in this trajectory space identified distinct differentiation patterns, visualized as “fate flows” using Sankey plots showing CD8 fate over time. This showed that OT recapitulates expansion/contraction of key CD8 subsets during the circulating infection response. 4/
Left: UMAP of spleen CD8 scRNA-seq data, shaded by trajectory cluster. Right: Sankey diagram showing population flows between trajectory clusters over time. Naive cells at day 0 differentiate into several clusters, including: effector early effector cells, which peak from days 3-4 and contract over time; terminal effector cells, which peak around day 10 before contracting; effector-memory cells, which maintain a stable proportion of the population over time; and memory cells, which gradually increase in proportion over time.

ALT Left: UMAP of spleen CD8 scRNA-seq data, shaded by trajectory cluster. Right: Sankey diagram showing population flows between trajectory clusters over time. Naive cells at day 0 differentiate into several clusters, including: effector early effector cells, which peak from days 3-4 and contract over time; terminal effector cells, which peak around day 10 before contracting; effector-memory cells, which maintain a stable proportion of the population over time; and memory cells, which gradually increase in proportion over time.

1
1
52
We next developed a method to integrate cells across time by embedding OT-derived trajectories into a shared space. This enabled us to separate stable features of lineages from transiently-expressed genes. 3/
Left: Schematic detailing trajectory kernel mean embedding (TKME) strategy. Cell probability densities are projected forward in time using OT. These distributions in cell state space are converted into kernel mean embedding vectors, which are then concatenated into the TKME. Right: UMAP dimensionality fit on spleen gene expression data, shaded by days post infection, compared to UMAP fit on TKME, showing how trajectory embedding integrates cells across time.

ALT Left: Schematic detailing trajectory kernel mean embedding (TKME) strategy. Cell probability densities are projected forward in time using OT. These distributions in cell state space are converted into kernel mean embedding vectors, which are then concatenated into the TKME. Right: UMAP dimensionality fit on spleen gene expression data, shaded by days post infection, compared to UMAP fit on TKME, showing how trajectory embedding integrates cells across time.

1
1
60
CD8 T cells protect us from infections but their rapid, non-stationary dynamics make them hard to model on a single-cell level. We used OT to fit temporal trajectories of circulating CD8 responses to acute infection, capturing clonal expansion and contraction. 2/
Left: schematic of simulated CD8 T cell kinetics during infection, showing naive, total CD8 and virus curves acros early activation, effector, and memory phases over days post infection. Right: line plot of OT-based CD8 popultion size estimates peaking then contracting over time.

ALT Left: schematic of simulated CD8 T cell kinetics during infection, showing naive, total CD8 and virus curves acros early activation, effector, and memory phases over days post infection. Right: line plot of OT-based CD8 popultion size estimates peaking then contracting over time.

1
100
me writing SLURM jobs vs. me debugging SLURM jobs
2
11
Load more