LANTERN Collaboratory — Lymphedema & Lipedema GWAS Pipeline

Genome-wide association study pipeline for primary lymphedema, secondary lymphedema, and lipedema using the All of Us Research Program Controlled Tier Dataset v8.

Citation

Bowler SA, Hampilos NH, Ndhlovu LC, and on behalf of the LANTERN Collaboratory. Divergent Genetic Architectures Underlying Primary and Secondary Lymphedema and Lipedema: A Genome-wide Association Study. (In preparation)

---

Study Overview

Three parallel GWAS conducted on All of Us v8 WGS data (GRCh38, Hail v0.2):

Condition	Cases	Controls	GW Significant	Suggestive
Primary lymphedema	93	930	1 (chr15:36,333,087 — MEIS2 proximal)	30
Secondary lymphedema	4,336	43,360	0	49
Lipedema	110	1,100	2 (SLC39A3, chr19p13.3)	325

Total participants: 49,929

---

Repository Structure

lantern_gwas/
├── phenotyping/
│   ├── lymphedema_phenotyping.py      # Primary/secondary case ID + 1:10 matching
│   ├── lipedema_phenotyping_v4.py     # Lipedema case ID, tiering, anthropometric QC
│   └── lipedema_controls.py          # Hierarchical control matching
├── gwas/
│   ├── gwas_lymphedema.py            # Primary + secondary GWAS (Hail, Wald)
│   └── gwas_lipedema_v2.py           # Lipedema GWAS (Hail, Wald, female-only)
├── qc/
│   └── merge_and_artifact_detection.py  # Per-chr merge + segmental dup screening
├── figures/
│   ├── manhattan_qq.py               # Figure 2 + Supplementary Figure 1
│   └── slc39a3_gene_plot.py          # Figure 3 — SLC39A3 locus plot
├── utils/
│   ├── annotation.py                 # Ensembl VEP variant annotation
│   ├── pathway_enrichment.py         # Enrichr pathway analysis
│   └── ld_calculation.py             # LD r² between lead variants (Hail)
├── .gitignore
└── README.md

---

Requirements

Data Access

All analyses were performed on the All of Us Researcher Workbench using the Controlled Tier Dataset v8. Access requires:

1. Registration at researchallofus.org
2. Completion of required ethics and data use training
3. Approval for Controlled Tier access
4. A registered workspace on the Researcher Workbench

These scripts cannot be run outside the All of Us environment — individual-level genomic data is not publicly available and must remain within the Workbench.

Software

All analyses run within the All of Us Jupyter notebook environment. The following are available pre-installed or installable within the Workbench:

Package	Version	Purpose
Hail	0.2	GWAS, VDS operations, variant QC
Python	3.10	All scripting
pandas	>=1.5	Data manipulation
numpy	>=1.23	Numerical operations
matplotlib	>=3.6	Figure generation
scipy	>=1.9	Statistical tests, QQ plots
requests	>=2.28	Ensembl VEP API calls

Installation within AoU Workbench

No installation is required for Hail — it is configured as part of the Dataproc cluster when you create a Hail-enabled notebook environment.

For any missing Python packages, install within a notebook cell:

import subprocess
subprocess.run(["pip", "install", "requests", "--quiet"])

Cluster Configuration

Lymphedema pipeline (larger cohort, N ~48,000):

• 10 worker nodes, 32 CPUs, 208 GB RAM each
• Driver: 32 CPUs, 416 GB RAM

Lipedema pipeline (smaller cohort, N ~1,200):

• 8 worker nodes, 16 CPUs, 104 GB RAM each
• Driver: 16 CPUs, 104 GB RAM

---

Usage

Scripts should be run in the following order within the All of Us Workbench:

1. Phenotyping

# Identify lymphedema cases and match controls
phenotyping/lymphedema_phenotyping.py

# Identify lipedema cases (tiered phenotyping)
phenotyping/lipedema_phenotyping_v4.py

# Match lipedema controls (hierarchical)
phenotyping/lipedema_controls.py

2. GWAS

# Run lymphedema GWAS (primary + secondary in parallel)
gwas/gwas_lymphedema.py

# Run lipedema GWAS
gwas/gwas_lipedema_v2.py

3. QC

# Merge per-chromosome TSVs and screen for artifacts
qc/merge_and_artifact_detection.py

4. Annotation and Enrichment

# Annotate suggestive variants via Ensembl VEP
utils/annotation.py

# Pathway enrichment via Enrichr API
utils/pathway_enrichment.py

# LD between lead variants
utils/ld_calculation.py

5. Figures

# Manhattan plots (Figure 2) + QQ plots (Supplementary Figure 1)
figures/manhattan_qq.py

# SLC39A3 locus plot (Figure 3)
figures/slc39a3_gene_plot.py

---

Key Implementation Notes

gwas_lipedema_v2.py contains three important fixes relative to earlier versions:

1. chrX source mismatch — PAR region filter_rows must occur before dosage expressions are defined. Moving it after annotate_cols creates a new MT object; covariate expressions remain bound to the pre-filter MT, causing source mismatch in logistic_regression_rows.

2. Firth non-convergence — Hail's Firth solver fails wholesale at N ~110 cases. Switched to Wald regression, which is standard GWAS practice at this sample size.

3. Null model explosion — sex_at_birth_num is a constant in the female-only lipedema cohort, causing Newton iteration to explode. Removed from covariates. Only age and PC1-PC4 are used.

---

Data

Input data paths are available to registered All of Us researchers within the Controlled Tier Workbench environment. Key inputs:

• All of Us v8 WGS VDS (GRCh38, Hail format)
• All of Us ancestry PCs (ancestry_preds.tsv, controlled tier)
• Cohort CSV files generated by the phenotyping scripts above

Output files are stored in your personal workspace bucket on Google Cloud Storage, accessible only within your registered Workbench workspace. Do not share workspace bucket IDs or paths publicly.

---

License

Code in this repository is released under the MIT License.

Individual-level data from the All of Us Research Program is not included and is not publicly available. Access to the underlying data requires independent registration and approval through the All of Us Researcher Workbench.

---

Support

For questions about the analysis pipeline, please open a GitHub Issue or contact Scott A Bowler (scb4005@med.cornell.edu):

• GitHub: Sabowler
• Institution: Division of Infectious Disease, Department of Medicine, Weill Cornell Medicine, New York, NY

For questions about All of Us data access:

• All of Us Support
• Researcher Workbench Documentation