Optimal transport fate mapping resolves T cell differentiation dynamics across tissues

This study introduces an optimal transport-based framework that reconstructs continuous CD8 T cell differentiation and migration trajectories across tissues during viral infection, revealing distinct waves of intestinal entry and identifying AP4 as a key regulator distinguishing circulating from tissue-resident memory fates.

The Gist

Imagine the immune system as a massive, bustling city. When a virus invades, the city's defense force (the CD8 T cells) springs into action. But here's the problem: scientists usually only get to take a single, frozen photograph of the city at one specific moment. They see the soldiers, the factories, and the messengers, but they can't see how a rookie soldier became a general, or when a soldier decided to stay in a specific neighborhood to guard it forever.

This paper introduces a new way to watch the movie instead of just looking at the photos. The authors used a mathematical tool called Optimal Transport to reconstruct the entire journey of these immune cells over time and across different parts of the body.

Here is a breakdown of their discoveries using simple analogies:

1. The "Time-Traveling Map" (Optimal Transport)

Usually, scientists try to guess a cell's future by looking at its current DNA "snapshot." But immune cells are chaotic: they multiply like crazy, die off, and move around. It's like trying to predict the path of a single raindrop in a storm just by looking at a puddle.

The authors used Optimal Transport, which is like a sophisticated GPS system for cells. Instead of guessing, it calculates the most efficient "flow" of cells from one day to the next.

• The Analogy: Imagine you have a photo of a crowd of people at a train station at 8:00 AM and another at 10:00 AM. Standard methods might just say, "That person in the red hat is now over there." The Optimal Transport method says, "Okay, 50 people left the station, 20 new people arrived, and here is the specific probability that this specific person in the red hat boarded a train to the next city." It accounts for people getting on, off, and changing their minds.

2. The "Two-Way Street" of Migration

The researchers looked at two main locations: the Spleen (the main training camp) and the Small Intestine (a border town).

• The Discovery: They found that when a soldier arrives at the border town matters more than we thought.
• The Analogy: Think of the border town (the intestine) as a hotel.
  • Early Arrivals: Soldiers who arrived early (within the first few days) were like the first guests to check in. They settled in, got comfortable, and decided to stay forever. They became Tissue-Resident Memory (TRM) cells—the permanent security guards of the gut.
  • Late Arrivals: Soldiers who showed up later were like day-trippers. They were still in "fighting mode," didn't settle down, and eventually left or died off. They didn't become permanent guards.

3. The "ID Badge" (CD52)

Because the researchers could now track the "arrival time" of cells, they found a new way to tell them apart.

• The Discovery: They found a protein called CD52 that acts like a "Freshly Arrived" sticker.
• The Analogy: If you walk into a party, you might get a wristband.
  • No Wristband (CD52 Negative): You've been there a while; you know the staff; you're a regular (a long-term memory cell).
  • Wristband On (CD52 Positive): You just walked in 10 minutes ago; you're still looking around (a new arrival/effector cell).
  • This helps doctors distinguish between a soldier who is ready to stay and guard the gut forever versus one who is just passing through.

4. The "Switches" (Transcription Factors)

Finally, they looked at the "control panels" inside the cells (transcription factors) that decide what the cell becomes.

• The Discovery: They found that the "control panel" for becoming a permanent gut guard is different from the one for becoming a circulating soldier.
• The Analogy: Imagine two different types of drivers.
  • Driver A (Circulating): Drives a sports car. They are fast, aggressive, and keep moving. Their "engine" (regulated by a protein called AP4) is tuned for speed and short bursts of energy.
  • Driver B (Resident): Drives a heavy-duty truck. They are built to stay in one place and protect a specific neighborhood. Their engine is tuned for endurance.
  • The study found that if you force Driver A to use Driver B's engine, they don't become good guards. The "engine" (regulatory program) is specific to the job.

Why Does This Matter?

This paper changes how we understand immunity. Instead of thinking of immune cells as static categories (like "Good Guy" vs. "Bad Guy"), we now see them as a fluid flow.

• For Vaccines: If we want to create vaccines that protect the gut (like against foodborne illnesses), we need to make sure our "soldiers" arrive early and get the right "engine" to stay there permanently.
• For Cancer: If we can trick cancer-fighting T cells to arrive early and settle down in a tumor, they might be better at destroying cancer cells long-term.

In short: The authors built a time-machine map that shows us exactly how immune cells travel, when they decide to stay, and what makes them stick around. It turns a blurry, static photo of the immune system into a high-definition, slow-motion movie.

Technical Summary

1. Problem Statement

Immune responses, particularly CD8+ T cell differentiation during viral infection, are highly dynamic processes involving rapid proliferation, apoptosis, and migration across anatomical compartments (e.g., from lymphoid organs to tissues).

• Limitation of Current Methods: Standard single-cell RNA sequencing (scRNA-seq) provides static molecular snapshots. Traditional trajectory inference methods often fail in this context because they rely on assumptions of steady-state populations, mass conservation, and linear pseudotime ordering. They struggle to account for:
  • Massive clonal expansion and cell death (violating mass conservation).
  • Cell state cyclicity (e.g., dedifferentiation).
  • The entanglement of migration (transport between tissues) with differentiation (state changes within a tissue).
• The Gap: There is a need for a computational framework that can reconstruct continuous, probabilistic cell fate transitions across time and tissues while explicitly modeling population dynamics (birth/death) and migration.

2. Methodology

The authors developed a Optimal Transport (OT)-based fate mapping framework using the moscot package to model CD8+ T cell dynamics.

• Core Algorithm: They utilized Unbalanced Optimal Transport (UOT). Unlike standard OT which assumes mass conservation, UOT allows the probability mass to expand (proliferation) or contract (apoptosis) between timepoints.
  • Source Margins: To model growth rates, they calculated proliferation and apoptosis gene set scores for each cell. These scores were used to derive "source marginals" (prior growth rates), allowing the OT model to weight cells based on their likelihood of dividing or dying.
  • Transport Maps: The model computes a transport map ($P^*$) between adjacent timepoints, estimating the probability that a cell at time $t$ transitions to a specific state at time $t+1$.
• Multi-Compartment Modeling: The framework was extended to a two-compartment model (Spleen vs. Small Intestine Intraepithelial Lymphocytes - siIEL) to jointly model differentiation and migration. This allowed the inference of "fate flows" between circulation and tissue residency.
• Trajectory Integration (TKME): To overcome the limitations of static clustering, the authors developed Trajectory Kernel Mean Embeddings (TKME). Instead of clustering cells based on gene expression at a single timepoint, they embedded cells based on their entire inferred trajectory (future and past states). This allowed for the identification of "fluid" differentiation paths.
• Validation & Integration:
  • In Vivo Labeling: They used time-resolved intravascular antibody labeling (anti-CD45.2) to distinguish "early arrival" (tissue-resident >48h) vs. "late arrival" (circulating <48h) cells in the gut.
  • Perturbation: Genetic silencing (shRNA) and overexpression (retroviral vectors) of key transcription factors (Klf2, T-bet, TCF1, AP4) were performed to validate computational predictions.
  • Regulon Analysis: SCENIC was used to infer transcription factor (TF) regulon activities across the inferred trajectories.

3. Key Contributions

1. OT Framework for Immune Dynamics: Established OT as a principled method for reconstructing immune cell fate that explicitly handles non-equilibrium dynamics (proliferation/death) and migration, outperforming standard pseudotime methods in T cell contexts.
2. Fluid Differentiation Model: Demonstrated that CD8+ T cell differentiation is not a set of discrete, static states but a continuous "fate flow." They identified intermediate states (e.g., Effector Memory) that bridge terminal and memory fates, which are often obscured by static nomenclature.
3. Migration-Timing Hypothesis: Revealed that the time of tissue entry is a critical determinant of cell fate. Early-arriving cells are biased toward long-lived tissue-resident memory (TRM), while late-arriving cells tend toward terminal effector fates.
4. Novel Markers: Identified CD52 as a specific marker for recent tissue entrants and a discriminator between short-lived and long-lived TRM precursors.
5. Context-Dependent Regulation: Mapped the regulatory architecture of CD8+ T cells, showing that while effector programs are conserved across tissues, memory programs are highly context-dependent (spleen vs. gut).

4. Key Results

A. Reconstructing Circulating Dynamics (Spleen)

• Population Kinetics: The OT model accurately recapitulated the known expansion (peaking days 4-5) and contraction phases of the antiviral response, matching experimental population counts despite using snapshot data.
• Fate Flows: Sankey diagrams derived from OT maps showed coherent flows from stem-like precursors to terminal effectors and memory cells.
• Trajectory Clustering: TKME-based clustering revealed four distinct trajectories:
  • Traj-MEM: Long-lived memory (overlaps with TCM/TMP).
  • Traj-TERM: Terminal fated (overlaps with TTE).
  • Traj-EE: Early effector.
  • Traj-EM: Intermediate effector memory (high fate uncertainty, bridging the gap between terminal and memory).
• Validation: Silencing Klf2 impaired TTE formation and increased TMP-like cells, confirming the biological relevance of the inferred trajectories.

B. Multi-Compartment Migration Dynamics (Spleen $\to$ Gut)

• Migration Window: The model predicted a specific migration window (Days 3-7 post-infection) where cells egress from the spleen to the siIEL.
• Arrival Timing: Cells arriving early in the gut (Days 3-4) were distinct from those arriving later (Days 4-7).
  • Early Arrivals: Enriched for residency/memory genes (Cd69, Ccr9, Il7r) and CD52-negative.
  • Late Arrivals: Enriched for effector/circulation genes (Lgals1, Ly6c2) and CD52-positive.
• Experimental Validation: Intravascular labeling confirmed that >90% of siIEL cells at Day 7 were "early arrivals." These early arrivals exhibited a mature TRM phenotype (CD69+CD103+), while late arrivals were effector-like (KLRG1+, CD52+).

C. Differentiation of TRM Precursors

• Bifurcation: Within the gut, early-arriving cells diverged into two paths:
  1. Long-lived TRM: Characterized by upregulation of CD69, CD103, and CD52 (transiently), eventually becoming CD52-CD103+.
  2. Short-lived/Transient: Cells that failed to downregulate Ly6C and CD52, remaining in a transient state or re-entering circulation.
• Marker Discovery: The combination of CD52, CD69, and CD103 was identified as a robust axis to stratify TRM subsets in vivo.

D. Regulatory Architecture

• Shared vs. Context-Specific:
  • Effector Programs: Conserved across spleen and gut (e.g., T-bet activity).
  • Memory Programs: Divergent. Spleen memory relies heavily on Tcf7 (TCF1), whereas gut TRM programs are constrained and distinct.
• Key Regulators:
  • T-bet: Represses TRM differentiation in the gut.
  • TCF1: Promotes circulating memory but is attenuated in TRM; overexpression reduces TRM frequency.
  • AP4 (Tfap4): Identified as a context-specific regulator. AP4 overexpression enhanced circulating effector/memory accumulation in the spleen but had no effect on TRM accumulation in the gut, highlighting that tissue environment dictates the utility of specific regulators.

5. Significance

• Conceptual Shift: Moves the field from viewing T cell subsets as discrete, static endpoints to understanding them as continuous, probabilistic flows shaped by time and location.
• Methodological Advance: Provides a robust computational tool (scotty-tools and moscot integration) for analyzing dynamic biological systems where mass is not conserved (proliferation/death) and migration is prevalent.
• Therapeutic Implications:
  • Identifies CD52 as a potential target for distinguishing transient vs. durable tissue immunity.
  • Reveals that AP4 can boost circulating responses but not tissue-resident ones, suggesting that immunotherapies must be tailored to the specific tissue compartment.
  • Highlights that timing of intervention (early vs. late infection) is crucial for shaping the quality of memory (long-lived TRM vs. short-lived effectors).
• Broad Applicability: The "fate flow" framework is applicable to other dynamic systems, including chronic infection, autoimmunity, and cancer immunotherapy, where cell state transitions are complex and non-linear.