3D-iseqPLA: Spatiotemporal Immune Inflammation modulates 3D NFκB signaling interactomics of multiprotein supercomplexes

Nicholas Zhang^1,2,3, Collin Leese-Thompson^4,5, Sriya Sirigireddy^1,3, Dhruv Nambiar^1,3, Lakshana Ramanan^1,3, Rabindra Tirouvanziam^4,5, and Ahmet F. Coskun^1,2,3,*

¹ Wallace H. Coulter Department of Biomedical Engineering, Georgia Institute of Technology and Emory University, Atlanta, GA, USA ² Interdisciplinary Bioengineering Graduate Program, Georgia Institute of Technology, Atlanta, GA, USA ³ Parker H. Petit Institute for Bioengineering and Bioscience, Georgia Institute of Technology, Atlanta, GA, USA ⁴ Department of Pediatrics, Emory University, Atlanta, GA, USA ⁵ Center for CF & Airways Disease Research, Children's Healthcare of Atlanta, Atlanta, GA, USA ^* Corresponding author: ahmet.coskun@bme.gatech.edu

---

Overview

This repository contains code, data, and analysis pipelines for the first volumetric, in situ profiling of endogenous NFκB protein-protein interactions (PPIs) using iterative sequential proximity ligation assay (iseqPLA) combined with spinning disk confocal microscopy and 3D reconstruction.

Key Features

• 3D spatial interactomics of NFκB signaling supercomplexes at single-cell resolution
• iseqPLA workflow for multiplexed PPI detection across sequential imaging cycles
• PRISMS-based 3D reconstruction pipeline for volumetric quantification
• ~50,000 cells imaged across multiple experimental conditions
• scGPT foundation model validation of NFκB gene panel relevance
• Analysis of cystic fibrosis patient-derived macrophages in coculture systems

---

Abstract

The NFκB signaling pathway orchestrates inflammatory responses through the dynamic assembly and dissociation of membrane-proximal multiprotein supercomplexes, yet their spatiotemporal organization within the three-dimensional (3D) intracellular space has remained unresolved at single-cell resolution. Here, we present the first volumetric, in situ profiling of endogenous NFκB protein-protein interactions (PPIs) using iterative sequential proximity ligation assay (iseqPLA) combined with spinning disk confocal microscopy and 3D reconstruction.

Across 01-3T3 mouse fibroblasts, IMR-90 human fibroblasts, and cystic fibrosis patient-derived macrophage cocultures, we characterize supercomplex dissociation kinetics, p65 nuclear translocation dynamics, and negative feedback engagement over a 105-minute cytokine time course. We demonstrate that:

• 3D volumetric quantification resolves PPI distributions obscured by conventional 2D maximum intensity projections
• Extracellular matrix coating critically determines the fraction of NFκB-responsive cells
• CF airway-conditioned macrophages amplify paracrine NFκB signaling in adjacent fibroblasts

A transfer learning-based scGPT foundation model, trained on curated in vitro and in vivo transcriptomic datasets, confirms statistically significant enrichment of our selected NFκB gene panel within inflammation-relevant transcriptional feature space.

---

Repository Structure

3D-iseqPLA/
├── code/
│   ├── image_processing/          # 3D confocal image processing scripts
│   ├── iseqPLA_analysis/          # PPI quantification and analysis
│   ├── foundation_model/          # scGPT training and evaluation
│   ├── visualization/             # Figure generation scripts
│   └── utils/                     # Helper functions
├── data/
│   ├── raw/                       # Raw microscopy images (not included - see Data Availability)
│   ├── processed/                 # Processed single-cell measurements
│   ├── transcriptomics/           # scRNA-seq datasets for foundation model
│   └── metadata/                  # Experimental metadata
├── figures/                       # Publication-quality figures
├── notebooks/                     # Jupyter notebooks for analysis
├── requirements.txt               # Python dependencies
└── README.md                      # This file

---

Installation

Prerequisites

• Python 3.8 or higher
• CUDA-compatible GPU (recommended for foundation model training)
• Minimum 32GB RAM (64GB recommended for large image processing)

Setup

# Clone the repository
git clone https://github.com/coskun-lab/3D-iseqPLA.git
cd 3D-iseqPLA

# Create a virtual environment
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

Key Dependencies

• numpy>=1.21.0
• pandas>=1.3.0
• scipy>=1.7.0
• scikit-learn>=1.0.0
• matplotlib>=3.4.0
• seaborn>=0.11.0
• opencv-python>=4.5.0
• scikit-image>=0.18.0
• napari>=0.4.0 (for 3D visualization)
• torch>=1.10.0 (for scGPT)
• scanpy>=1.8.0 (for scRNA-seq analysis)
• scgpt (foundation model)

---

Experimental Design

Cell Models

1. 01-3T3 mouse fibroblasts (n=2,780 cells)

   • TNFα (10 ng/mL), IL-1β (1 ng/mL), DMSO control
   • Time course: 0–105 min in 15-min intervals

2. IMR-90 human fibroblasts (n=1,425 cells)

   • Upstream supercomplex panel: TRAF-5_TRADD, TRAF-5_TRAF-2
   • Same cytokine conditions

3. CCL2 macrophage + IMR-90 cocultures (n=16,961 cells)

   • Control condition (CCL2 chemokine attractant)
   • LPS (10 ng/mL), TNFα, IL-1β, DMSO
   • Time course: 0, 30, 60, 120, 240, 480 min

4. CFASN macrophage + IMR-90 cocultures (n=15,617 cells)

   • CF airway supernatant-conditioned macrophages
   • Same stimulation conditions

5. IMR-90 monocultures (n=13,221 cells)

   • Baseline comparison without macrophages

iseqPLA Panel

Cycle 1: Reporter proteins (H2B, p65 protein) Cycle 2: p105/p50 & p65 heterodimer Cycle 3: A20 & IKKβ (negative feedback) Cycle 4: A20 & IKKγ (negative feedback)

Upstream panel: TRAF-5_TRADD, TRAF-5_TRAF-2

Imaging Parameters

• Microscope: Cephla Squid spinning disk confocal
• Objective: Nikon 60× water lens
• Z-spacing: 0.5 μm
• Z-planes: 40 per field of view
• Channels: DAPI, A488/GFP, ds-RED/TRITC, Cy5/647 nm
• Stitching: 3×3 grid per FOV

---

Usage

1. Image Processing and 3D Reconstruction

Process raw confocal z-stacks into 3D volumetric renderings:

python code/image_processing/reconstruct_3d.py \
    --input data/raw/experiment_01/ \
    --output data/processed/3d_renderings/ \
    --z-spacing 0.5 \
    --num-planes 40

2. PPI Quantification

Quantify PPI puncta from iseqPLA images:

python code/iseqPLA_analysis/quantify_ppis.py \
    --input data/processed/3d_renderings/ \
    --output data/processed/ppi_measurements/ \
    --panel upstream  # Options: upstream, feedback, reporters

3. Single-Cell Analysis

Extract single-cell features and generate quantitative metrics:

python code/iseqPLA_analysis/single_cell_analysis.py \
    --input data/processed/ppi_measurements/ \
    --output data/processed/single_cell_features.csv \
    --compute-nc-ratio  # Nuclear-to-cytoplasmic p65 ratio

4. Foundation Model Training

Train scGPT model on curated transcriptomic datasets:

# In vitro training
python code/foundation_model/train_invitro.py \
    --data data/transcriptomics/invitro_datasets.h5ad \
    --output models/scgpt_invitro/ \
    --epochs 20 \
    --batch-size 32

# In vivo fine-tuning
python code/foundation_model/train_invivo.py \
    --pretrained models/scgpt_invitro/best_model.pt \
    --data data/transcriptomics/invivo_datasets.h5ad \
    --output models/scgpt_invivo/ \
    --epochs 10

5. Generate Figures

Reproduce publication figures:

python code/visualization/generate_all_figures.py \
    --data data/processed/ \
    --output figures/ \
    --format pdf

---

Key Results

1. 3D vs 2D Quantification

3D volumetric analysis provides:

• Reduced variance in nuclear-to-cytoplasmic p65 ratios
• More accurate discrimination of nuclear vs. cytoplasmic PPIs
• Elimination of artifacts from z-plane signal overlap

Example: 2D analysis yielded N/C ratios ~3 AU at peak activation vs. ~2.5 AU in 3D, with systematically higher variance.

2. Supercomplex Dissociation Kinetics

• TNFα drives the most rapid and complete dissociation of TRAF-5_TRADD and TRAF-5_TRAF-2 supercomplexes
• Dissociation begins at 45 min and approaches baseline by 90–105 min
• IL-1β produces attenuated dissociation, consistent with distinct TNFR1 vs. IL-1R signaling architectures
• Peak p65 activation (30–45 min) coincides with maximal supercomplex dissociation

3. ECM Coating Effects

Substrate coating profoundly affects NFκB activation:

Coating	TNFα-activated cells	IL-1β-activated cells
Collagen I	87.5% (*** p<0.001)	87.6% (ns)
Poly-L-lysine	67.2%	81.8%
Matrigel	18.2% (*** p<0.001)	29.8%

Implication: Matrigel dramatically suppresses cytokine-induced NFκB activation, likely through cytokine sequestration.

4. CF Macrophage Paracrine Amplification

CFASN-exposed macrophages show:

• Elevated IL-1β-induced p65 activation (0.47 at 120 min vs. 0.25 for CCL2)
• Sustained LPS response with secondary elevation at 240–480 min
• Hyperinflammatory phenotype transmitted to adjacent fibroblasts via paracrine signaling

5. Foundation Model Validation

scGPT model confirms NFκB gene panel relevance:

• In vitro model: Mann-Whitney U test p = 0.0295* (significant enrichment)
• In vivo model: Non-significant (p = 0.264) due to greater transcriptomic heterogeneity
• Model performance: Accuracy, precision, recall, F1, AUC all >0.95

---

Data Availability

Due to the extremely large size of raw microscopy datasets (>2 TB), raw images are available upon request from the corresponding author (ahmet.coskun@bme.gatech.edu).

Processed data included in this repository:

• Single-cell PPI measurements (CSV format)
• 3D volumetric features
• Foundation model training datasets (GEO accessions listed in Supplementary Tables 1–2)

External datasets used:

• In vitro: GSE94383, GSE199404, GSE189062, GSE132791, GSE197031, GSE120000, GSE226488 (n=243,268 cells)
• In vivo: 16 lung disease cohorts (COVID-19, IPF, CF, COPD, asthma, tuberculosis; n=10,000 cells)

---

Citation

If you use this code or data, please cite:

@article{zhang2025spatiotemporal,
  title={Spatiotemporal Immune Inflammation modulates 3D NFκB signaling interactomics of multiprotein supercomplexes},
  author={Zhang, Nicholas and Leese-Thompson, Collin and Sirigireddy, Sriya and Nambiar, Dhruv and Ramanan, Lakshana and Tirouvanziam, Rabindra and Coskun, Ahmet F.},
  journal={bioRxiv},
  year={2025},
  doi={10.1101/XXXX.XX.XX.XXXXXX}
}

---

Funding

This work was supported by:

• Lung Spore and the National Cancer Institute (P50CA217691)
• National Institutes of Health (R35GM151028, 1R33CA291197)
• Winship Cancer Institute of Emory University (P30CA138292)

---

Contact

Ahmet F. Coskun, Ph.D. Associate Professor Wallace H. Coulter Department of Biomedical Engineering Georgia Institute of Technology and Emory University Email: ahmet.coskun@bme.gatech.edu Lab Website: coskun.gatech.edu

Nicholas Zhang PhD Candidate Interdisciplinary Bioengineering Graduate Program Georgia Institute of Technology Email: nzhang326@gatech.edu

---

License

This project is licensed under the MIT License - see the LICENSE file for details.

---

Acknowledgments

We thank:

• Dr. Rabindra Tirouvanziam and the Center for CF & Airways Disease Research for providing CF patient-derived macrophages
• The Winship Cancer Institute Shared Resources for imaging support
• The scGPT development team for the foundation model framework

---

Keywords

spatial interactomics, NFκB signaling, supercomplexes, 3D confocal, spatiotemporal dynamics, proximity ligation assay, cystic fibrosis, inflammation, single-cell analysis, foundation models