Regulatory network architecture constrains inflammatory responses in tissue-resident alveolar macrophages

By integrating multi-omics data with deep learning, this study reveals that tissue-resident alveolar macrophages exhibit more restrained inflammatory responses than recruited monocytes due to a stabilizing gene regulatory network architecture centered on PU.1 and CEBP{beta}.

The Gist

The Big Picture: The Lung's "Janitors" and the "Firefighters"

Imagine your lungs are a busy, high-tech city. Floating around in the airways are special cells called Alveolar Macrophages. Think of them as the city's Janitors.

Their main job is to keep the city clean and running smoothly (homeostasis). They eat up dust, clear out old surfactant (a slippery coating that helps you breathe), and generally maintain peace.

But sometimes, the city gets attacked—maybe by a virus or a chemical irritant (like LPS in the experiment). This is an inflammatory response, or a "fire." When this happens, the Janitors need to switch gears. They have to stop cleaning and start fighting the fire.

The big question this paper asks is: How do the Janitors know when to fight, and how do they know when to stop fighting and go back to cleaning?

The Two Types of Macrophages

The researchers discovered that there are actually two different groups of these cells in the lungs, and they handle the "fire" very differently:

1. The Native Janitors (Tissue-Resident): These are the original employees who have lived in the lung for a long time. They know the building inside and out.
2. The Hired Help (Recruited Monocytes): When the fire gets big, the body calls in reinforcements from the blood. These are new cells that rush into the lung to help.

The Discovery: The Native Janitors are much better at controlling the fire. They can fight the infection, but they don't overreact. They know how to calm down quickly once the danger is gone. The Hired Help, however, tend to go "all out." They fight harder, but they also cause more collateral damage and take longer to calm down.

The "Control Room" Analogy: Gene Regulatory Networks

Why is there a difference? It comes down to their internal "Control Room," which scientists call a Gene Regulatory Network (GRN).

Think of the Control Room as a massive switchboard with thousands of wires connecting different light switches (Transcription Factors). These switches turn genes on and off.

• The Hired Help (Recruited): Their control room is a bit chaotic. The switches are loosely connected. If you pull one wire (perturb a single factor), the whole system wobbles, and the lights flicker wildly. This means their response is unstable and can easily spiral out of control.
• The Native Janitors (Resident): Their control room is a fortress. The switches are tightly woven together in a dense, redundant web. If you pull one wire, the other wires hold it in place. The system is stable. It resists change. This stability allows them to fight the infection without accidentally destroying the lung tissue.

The Key Players: PU.1 and C/EBPβ

The researchers found two specific "Master Switches" (Transcription Factors) that act as the glue holding the Native Janitors' control room together: PU.1 and C/EBPβ.

• The Analogy: Imagine PU.1 and C/EBPβ are the Chief Engineers of the Native Janitors.
• They are everywhere in the control room, holding hands with other switches.
• When the researchers simulated "breaking" these engineers (a digital knockout), the Native Janitors' control room collapsed. Suddenly, the Janitors stopped being calm and started acting like the chaotic Hired Help, ramping up the inflammation way too much.

This proves that these two specific proteins are the secret sauce that keeps the Native Janitors stable and prevents them from causing too much damage to the lung while they fight.

The "Chromatin" Clue: The Blueprint

To understand how these engineers hold the room together, the researchers looked at the "blueprints" of the cells (the DNA and how open or closed it is, known as chromatin accessibility).

They found that in the Native Janitors, the blueprints for PU.1 and C/EBPβ are wide open and ready to work, even when the cell is trying to fight an infection. In the Hired Help, these blueprints are less accessible. This suggests that the Native Janitors are genetically "wired" to keep these stabilizing engineers on duty 24/7, ensuring they never lose their cool.

Why Does This Matter?

This is a big deal for human health.

• Chronic Disease: If our Native Janitors lose their stability (maybe due to aging or pollution), they might start overreacting to every little irritant. This leads to chronic inflammation, which is the root of diseases like asthma, COPD, and even fibrosis.
• Better Treatments: Instead of just trying to suppress the immune system (which stops the good fighting too), doctors might be able to design drugs that specifically boost the "stability" of these Native Janitors. We could teach the Hired Help to be more like the Native Janitors, or help the Native Janitors hold their ground better during a crisis.

Summary in One Sentence

This paper shows that the lung's native immune cells are better at fighting infections without causing damage because they have a super-stable internal "control network" held together by two key proteins (PU.1 and C/EBPβ), whereas the new cells recruited to help lack this stability and tend to overreact.

Technical Summary

1. Problem Statement

Tissue-resident alveolar macrophages (AMs) are critical for lung homeostasis, performing functions like surfactant catabolism and debris clearance. Upon inflammation, they must transition from a homeostatic state to a pro-inflammatory state and eventually resolve. However, the mechanisms governing the functional plasticity of these cells remain unclear. Specifically, it is not well understood how the gene regulatory networks (GRNs) of tissue-resident AMs differ from those of recruited monocyte-derived AMs, and how these networks constrain inflammatory responses to prevent chronic tissue damage. While individual transcription factors (TFs) are known, the higher-order regulatory interactions that coordinate identity and response restraint in vivo are poorly characterized.

2. Methodology

The authors employed an integrative multi-omics approach combining single-cell RNA sequencing (scRNA-seq), ATAC-seq, and deep learning-based chromatin modeling to reconstruct and analyze GRNs.

• Data Sources:
  • scRNA-seq: Utilized a published murine dataset (King et al.) of sorted AMs (CD88+CD64+ and intratracheal anti-CD45+) at steady state and 3, 6, and 15 days post-LPS challenge.
  • ATAC-seq: Generated new ATAC-seq profiles for tissue-resident AMs (CD45⁺CD64⁺CD11b⁻/loSiglecF⁺) from naïve mice and mice 3 days post-LPS.
• Computational Pipeline:
  1. Topic Modeling: Applied fastTopics (non-negative matrix factorization) to scRNA-seq data to identify distinct gene expression programs (topics). Top 1,000 genes per topic were annotated via KEGG pathway enrichment to define Homeostasis, Inflammation, and Resolution states.
  2. Cell Identity Clustering: Used k-means clustering and marker gene expression to distinguish Tissue-Resident AMs from Recruited (Monocyte-Derived) AMs.
  3. GRN Construction: Used CellOracle to build specific GRNs for resident and recruited populations. This involved scanning open chromatin regions for motifs (using gimmemotifs) and integrating scRNA-seq expression data.
  4. Network Analysis: Calculated Eigenvector Centrality to measure TF connectivity within the network.
  5. In Silico Perturbation: Systematically simulated TF knockouts (reducing expression to 0) within the GRNs. The impact was quantified by calculating dot products between perturbation trajectories and topic trajectories, generating Perturbation Scores for Homeostasis and Inflammation.
  6. Chromatin Accessibility Modeling: Used ChromBPNet (a deep learning model) on ATAC-seq data to predict sequence determinants of chromatin accessibility. TF-MoDISco was used for de novo motif discovery, and HOMER was used for motif enrichment analysis near transcription start sites (TSS).

3. Key Contributions

• Integration of Multi-Omics: Successfully combined scRNA-seq, ATAC-seq, and deep learning (ChromBPNet) to infer GRN architectures specifically for tissue-resident vs. recruited macrophages in an in vivo context.
• Quantification of Network Stability: Introduced a metric-based approach (vector magnitude of perturbation scores) to demonstrate that tissue-resident AMs possess a more stable, redundant regulatory architecture compared to recruited cells.
• Identification of Stabilizing TFs: Pinpointed PU.1 (Spi1) and CEBP/β (Cebpb) as central nodes that not only maintain homeostatic identity but actively constrain inflammatory responses in tissue-resident AMs.

4. Key Results

• Distinct Functional States: Topic modeling confirmed that tissue-resident AMs maintain a strong association with the "Homeostasis" topic (enriched for PPAR signaling and lipid processing) even during LPS challenge, whereas recruited AMs shift predominantly toward the "Inflammation" topic (enriched for cytokine signaling and NF-κB pathways).
• Network Architecture Differences:
  • Centrality: TFs in tissue-resident AM GRNs exhibited significantly higher eigenvector centrality than those in recruited AMs, suggesting a more interconnected, redundant network.
  • Perturbation Stability: Global in silico TF knockouts in recruited AMs resulted in large vector magnitudes (large functional shifts), indicating a fragile network. In contrast, tissue-resident AMs showed smaller shifts, indicating a stabilizing regulatory architecture that buffers against perturbation.
• Role of PU.1 and CEBP/β:
  • Motif Analysis: ATAC-seq and ChromBPNet revealed that ETS (including Spi1/PU.1) and CEBP family motifs were the most abundant in accessible chromatin.
  • Specific Perturbation: Unlike global TFs, in silico knockout of PU.1 and CEBP/β in tissue-resident AMs caused large functional shifts (large vector magnitudes), driving cells away from homeostasis and toward inflammation.
  • Contrast: This effect was significantly stronger in tissue-resident AMs than in recruited AMs, suggesting these TFs are the "keystone" stabilizers of the tissue-resident network.
• Chromatin Landscape: The study highlights that the distinct chromatin accessibility landscape of tissue-resident AMs, driven by PU.1 and CEBP/β, underlies their ability to restrain inflammation compared to recruited cells.

5. Significance

• Mechanistic Insight: The paper provides a mechanistic explanation for why tissue-resident macrophages are less prone to hyper-inflammatory responses than recruited monocytes. It is not just a difference in gene expression, but a fundamental difference in network topology stabilized by specific pioneer factors (PU.1 and CEBP/β).
• Therapeutic Implications: Understanding that PU.1 and CEBP/β act as stabilizers suggests that targeting these nodes could modulate macrophage plasticity. Enhancing their activity might help restore homeostasis in chronic inflammatory diseases, while their loss could explain the transition to dysregulated states.
• Methodological Framework: The study establishes a robust framework for analyzing tissue-specific GRNs using integrative single-cell and chromatin accessibility data, which can be applied to other tissue-resident immune cell populations to understand their functional plasticity and disease susceptibility.