Multi-omic profiling of human and mouse dorsal root ganglia enables targeted gene delivery to nociceptors

By integrating cross-species multi-omic profiling with in vivo screening and machine learning, this study identifies conserved regulatory elements to engineer targeted AAV vectors that specifically deliver therapeutic genes to nociceptors, offering a promising toolkit for treating refractory chronic pain.

The Gist

Imagine your body's pain system as a massive, bustling city. In this city, there are millions of messengers (neurons) running around. Most of them are just doing regular jobs, like delivering mail or checking the temperature. But a specific group of messengers, called nociceptors, are the "alarm systems." Their only job is to scream "DANGER!" when you touch something hot or sharp.

The problem with chronic pain is that these alarm systems get stuck in the "ON" position. They scream even when there's no fire. Currently, the medicine we use to stop the screaming (like opioids) is like a city-wide blackout. It shuts down the alarms, but it also shuts down the mail carriers, the temperature checkers, and the traffic lights. This leads to dangerous side effects like addiction or confusion.

Scientists wanted a way to turn off only the alarms without touching the rest of the city. But here's the catch: the alarms look almost identical to the other messengers. It's like trying to find a specific red car in a sea of identical red cars.

This paper describes how a team of scientists built a universal remote control to fix this, using a mix of detective work, computer modeling, and viral delivery trucks.

1. The Detective Work: Mapping the City

First, the scientists needed a map. They couldn't just look at the cars; they had to look under the hood to see the engine codes.

• The Multi-Omic Atlas: They took samples from the "nerve ganglia" (the nerve hubs) of both mice and humans. They didn't just read the genes (the instruction manual); they also looked at the "chromatin accessibility" (which parts of the manual are currently open and ready to be read).
• The Analogy: Imagine every cell has a library. In a pain cell, the "Pain" books are open on the tables, while the "Touch" books are locked in the basement. In a touch cell, it's the opposite. The scientists mapped exactly which books were open in which cell type for both mice and humans.
• The Discovery: They found that the "open books" (regulatory switches) for pain cells are surprisingly similar in mice and humans. This meant they could do the heavy lifting in mice and trust the results would work in people.

2. The Viral Delivery Trucks (AAVs)

To deliver a new instruction to these cells, the scientists used AAVs (Adeno-Associated Viruses). Think of these as tiny, harmless delivery trucks that can drive into cells and drop off a package.

• The Problem: Usually, these trucks drop their packages everywhere. If you want to fix the pain alarms, you don't want the truck to drop the package in the mail carrier's house.
• The Solution: They needed a specific "address label" (an enhancer) that would only stick to the pain cells.
• The Screening: They took hundreds of potential address labels (DNA sequences) and put them on the trucks. They released these trucks into mice and watched where they went.
  • Some trucks went everywhere (bad).
  • Some trucks went only to the "C-PEP" pain cells (good!).
  • Some trucks went only to the "C-NP" pain cells (also good!).
  • They found a few "Golden Tickets" (specifically CRE1 and CRE8) that acted like perfect GPS coordinates, guiding the truck only to the pain alarms.

3. The Computer Brain (PAIN-net)

Screening hundreds of labels one by one is slow and expensive. So, the scientists built an AI model called PAIN-net.

• The Analogy: Imagine you are trying to guess the password to a safe. You could try every combination, or you could learn the pattern. PAIN-net learned the "grammar" of the pain cells. It looked at the DNA sequences and learned: "If you see this specific pattern of letters, it belongs to a pain cell."
• The Result: Now, instead of testing every single label in a mouse, they can feed a DNA sequence into the computer, and the AI predicts: "Yes, this label will only work on pain cells." This allows them to design brand-new, custom-made labels that are even better than the natural ones.

4. The Test Drive: Turning Down the Volume

Finally, they tested if their new remote control actually worked.

• The Payload: They loaded their "Golden Ticket" trucks with a gene for Kir2.1. Think of Kir2.1 as a "volume knob" that turns the electrical signal down. It makes the cell harder to excite.
• The Result: When they used the CRE1 truck, the pain cells received the volume knob. The cells became much harder to trigger. They stopped screaming "DANGER!" at the slightest touch.
• Human Test: They also tested this on human cells grown in a dish (derived from stem cells). The same "Golden Ticket" label worked on human pain cells, ignoring the human heart cells. This proved the technology works across species.

Why This Matters

This paper is a huge leap forward because it gives us a universal toolkit.

1. Precision: We can now target pain cells specifically, leaving the rest of the nervous system alone.
2. Speed: The AI model means we can design new tools much faster.
3. Hope: This opens the door for new pain treatments that don't cause addiction or sedation, potentially changing how we treat chronic pain for millions of people.

In short, the scientists built a map of the pain city, found the perfect address labels, taught a computer to read the map, and built a delivery truck that can silence the pain alarms without shutting down the whole city.

Technical Summary

1. Problem Statement

Chronic pain is frequently driven by the hyperexcitability of nociceptors (peripheral sensory neurons detecting noxious stimuli). Current analgesic strategies, such as opioids, lack cell-type specificity, leading to severe side effects. While single-cell transcriptomics has mapped the molecular identity of distinct dorsal root ganglion (DRG) subtypes, there is a critical lack of tools to selectively target specific nociceptor subtypes across species (mouse and human). Existing transgenic mouse models are not easily translated to humans, and viral vectors (AAVs) currently lack the cis-regulatory elements (CREs) necessary to drive transgene expression exclusively in nociceptors while sparing other DRG cell types (e.g., proprioceptors, mechanoreceptors, glia).

2. Methodology

The authors employed a multi-pronged approach combining high-throughput genomics, functional screening, and machine learning:

• Cross-Species Multi-Omic Atlases:

  • Generated matched single-nucleus RNA sequencing (snRNA-seq) and ATAC-seq (snATAC-seq) datasets for DRGs from 71 adult mice and 66 human donors.
  • Optimized neuronal enrichment protocols (sorting GFP+ nuclei in mice; NeuN+ nuclei in humans) to overcome the low abundance of neurons (~1% of total DRG cells).
  • Integrated data to identify conserved cell types and differential chromatin accessibility peaks.

• Regulatory Network Analysis:

  • Used SCENIC+ to infer gene regulatory networks (GRNs), linking transcription factors (TFs), CREs, and target genes to define cell identity.
  • Identified evolutionarily conserved TFs (e.g., ISL1, RUNX3, BCL11A) and their associated CREs active in specific neuronal subtypes.

• In Vivo AAV Enhancer Screening:

  • Individual Screening: Constructed AAVs with candidate CREs driving an EGFP reporter. Delivered via intracerebroventricular (ICV) injection in neonatal mice (using AAV-PHP.S capsid for broad DRG transduction).
  • Multiplexed Screening: Created a library of 106 prioritized CREs, each linked to a unique DNA barcode. Delivered as a pooled library to mice and quantified expression via scRNA-seq to assess specificity across dozens of subtypes simultaneously.
  • Validated specificity using RNAscope in situ hybridization and scRNA-seq.

• Machine Learning (PAIN-net):

  • Developed PAIN-net, a 1D Residual Convolutional Neural Network (CNN).
  • Trained on mouse DRG snATAC-seq data to predict cell-type-specific chromatin accessibility and differential accessibility (Log2FC) directly from DNA sequence.
  • Used in silico saturation mutagenesis to identify nucleotides critical for enhancer activity.

• Functional and Translational Validation:

  • Electrophysiology: Tested AAVs driving Kir2.1 (an inward-rectifying potassium channel that hyperpolarizes neurons) in dissociated mouse DRG cultures to measure changes in excitability (rheobase).
  • Human iPSC Assay: Transduced human iPSC-derived nociceptors and cardiomyocytes with candidate CRE-AAVs to test cross-species portability.

3. Key Contributions

• First Cross-Species Multi-Omic DRG Atlas: A comprehensive resource linking chromatin accessibility and gene expression for mouse and human DRG, revealing conserved regulatory programs.
• Discovery of Nociceptor-Specific Enhancers: Identification of specific CREs (e.g., CRE1, CRE6 for peptidergic C-nociceptors; CRE8 for non-peptidergic C-nociceptors) that drive biased expression in vivo.
• PAIN-net Model: A deep learning framework that decodes the "grammar" of nociceptor cis-regulatory logic, enabling the prioritization of native CREs and the design of synthetic enhancers.
• Functional Viral Toolkit: Demonstration that these CREs can drive therapeutic effectors (Kir2.1) to silence nociceptors without affecting other cell types, validated in both mouse and human iPSC models.

4. Key Results

• Conserved Regulatory Logic:

  • Identified ~35–44% overlap in differentially accessible peaks between mouse and human DRG cell types, suggesting conserved regulatory mechanisms.
  • SCENIC+ analysis confirmed that TFs like ISL1 (neuronal identity) and BCL11A (C-PEP nociceptors) drive conserved eRegulons in both species.

• Enhancer Screening Outcomes:

  • CRE1 and CRE6 drove reporter expression specifically in C-PEP (peptidergic, Tac1+) nociceptors, showing ~3–5 fold higher specificity than ubiquitous controls (CAG).
  • CRE8 drove specific expression in C-NP (non-peptidergic, Mrgprd+) nociceptors (~4–6 fold specificity).
  • The multiplexed screen successfully identified CREs for other subtypes (e.g., A-PEP, proprioceptors) and revealed that some highly accessible CREs act as silencers or require specific chromatin context.

• PAIN-net Performance:

  • Achieved a median Pearson correlation of 0.72 between predicted and measured chromatin accessibility in held-out data.
  • Successfully predicted the in vivo specificity of CRE1, CRE6, and CRE8.
  • In silico mutagenesis confirmed that disrupting specific TF motifs (e.g., BCL6, RUNT, ISL1) significantly reduced predicted accessibility in a cell-type-specific manner.

• Functional Silencing:

  • AAVs expressing Kir2.1 under CRE1 significantly increased the rheobase (current required to fire action potentials) and hyperpolarized the resting membrane potential in mouse DRG neurons, effectively reducing excitability.
  • In human iPSC-derived models, CRE1 and CRE8 drove robust reporter expression in nociceptors but not in cardiomyocytes, demonstrating cross-species functionality and safety.

5. Significance

This study provides a foundational viral toolkit for pain research that bridges the gap between mouse models and human therapeutics.

• Translational Potential: By identifying CREs that function in both mouse and human, the authors offer a pathway to develop cell-type-specific gene therapies for refractory chronic pain that avoid the side effects of current systemic analgesics.
• Methodological Advance: The integration of multi-omics with machine learning (PAIN-net) establishes a generalizable strategy for engineering synthetic enhancers, moving beyond trial-and-error screening to rational design.
• Resource Availability: The multi-omic atlases and code (available at painseq.com) serve as a critical reference for the neuroscience community to study peripheral sensory neuron biology and develop targeted neuromodulation strategies.

In summary, the paper successfully translates basic epigenomic insights into a functional, cross-species viral platform capable of selectively silencing pain-sensing neurons, representing a major step toward precision medicine for chronic pain.