Atlas of HIV cis-regulatory elements reveals extensive transcriptional variation across clades, isolates, and within individuals

By integrating massively parallel reporter assays with comparative sequence analysis across diverse HIV isolates and cell types, this study constructs a comprehensive regulatory atlas that reveals extensive, non-consistently selected transcriptional variation driven by distinct transcription factor configurations and conserved intragenic elements, ultimately enabling accurate sequence-based prediction of viral transcriptional activity.

The Gist

Imagine HIV not just as a tiny virus, but as a hacker that has broken into your computer (your body's cells) and installed a piece of malicious software. To cause damage, this software needs to "run." In the world of biology, "running" means the virus starts reading its own instructions to make more copies of itself.

The part of the virus that tells it when to start running is called the LTR (Long Terminal Repeat). Think of the LTR as the ignition switch and the gas pedal of the virus.

For decades, scientists thought all HIV viruses were roughly the same, so they studied just a few "standard models" of this ignition switch. But this new study is like a massive investigation that says: "Wait a minute! There are thousands of different versions of this switch, and they all work differently!"

Here is a simple breakdown of what the researchers discovered, using some everyday analogies:

1. The "Universal Remote" vs. The "Custom Remote"

Scientists used a high-tech tool called an MPRA (Massively Parallel Reporter Assay). You can think of this as a massive testing lab where they took thousands of different HIV ignition switches (from different people and different parts of the world) and plugged them into a test cell to see how loud the engine revs.

• The Finding: They found that the "gas pedal" sensitivity varies wildly. Some HIV strains are like a sports car that revs instantly with a light tap. Others are like a heavy truck that needs a massive push to get moving.
• The Surprise: Even two viruses living inside the same person can have completely different ignition switches. One might be quiet and hiding (latent), while its neighbor is loud and active.

2. The "Recipe Book" Problem

The virus has a recipe book (its DNA) that tells it how to build itself. The researchers realized that the "ingredients" for the ignition switch (called Transcription Factors) aren't just about what ingredients you have, but how you arrange them on the page.

• The Analogy: Imagine a recipe for a cake. If you have flour, eggs, and sugar, you can make a cake. But if you put the sugar before the flour, or if you have too many eggs, the cake might taste terrible or not rise at all.
• The Discovery: The study showed that HIV viruses rearrange their "ingredients" (binding sites for proteins) in millions of different ways. Some arrangements make the virus very sensitive to the body's immune signals (like inflammation), while others make it ignore those signals. This explains why some people's bodies can't clear the virus as easily as others.

3. The "Hidden Backup Generators"

For a long time, scientists thought the virus only had one way to start: the main ignition switch (the LTR).

• The Discovery: The researchers found hidden backup generators inside the virus's own code. These are called intragenic CREs.
• The Analogy: Imagine your house has a main light switch. If that switch breaks, the house goes dark. But this study found that the virus also has emergency lights hidden inside the walls (inside the virus's coding regions). Even if the main switch is broken or turned off, these hidden lights can still turn on and keep the virus "alive" or active in a sneaky way. This is a big deal because it means the virus might be able to keep a low-level hum going even when we think we've turned it off.

4. The "Crystal Ball" (AI Prediction)

Testing thousands of viruses in a lab is slow and expensive. So, the team built two AI models (named CREST and LARM).

• The Analogy: Think of these models as a Crystal Ball. You can feed the AI the virus's "DNA code" (the text of the recipe), and it instantly predicts:
  1. How loud the engine will rev on its own (Baseline activity).
  2. How much it will rev when you hit the gas (Response to immune signals).
• Why it matters: Now, instead of waiting weeks to test a virus in a lab, doctors or researchers can just type the virus's code into a website (HIVRegDB) and get an instant report on how dangerous or active that specific strain might be.

5. The "Transmission Mystery"

When HIV passes from one person to another, it usually goes through a "bottleneck"—only one or two viruses make the jump. Scientists wondered: Does the virus pick the "loudest" engine to start the infection?

• The Finding: Surprisingly, no. The virus doesn't seem to care if the ignition switch is super sensitive or weak when it jumps to a new host. It's like a gambler rolling dice; the virus doesn't seem to be "choosing" the best engine to start the infection. It just takes what it gets. This suggests that the virus's ability to spread isn't about how loud its engine is, but about other tricks (like how well it hides from the immune system).

Why Should You Care?

This study changes how we think about curing HIV.

• The "Shock and Kill" Problem: Current cure strategies try to "shock" the virus out of hiding (latency) so the immune system can "kill" it. But this study shows that because every virus has a different ignition switch, a "one-size-fits-all" shock might not work. Some viruses might need a harder shock, while others might need a different kind of trigger.
• The Future: By understanding the specific "grammar" of each virus's ignition switch, we can design smarter, more personalized treatments that target the specific way a patient's virus is hiding.

In short: HIV is a master of disguise with thousands of different "on" switches. This study mapped them all, found hidden backup switches, and built an AI to predict how they work, giving us a much better roadmap for finally turning off the virus for good.

Technical Summary

1. Problem Statement

Human Immunodeficiency Virus (HIV) establishes persistent infection by integrating into the host genome, forming a latent reservoir that evades immune clearance and antiretroviral therapy (ART). The reactivation of these proviruses is driven by the Long Terminal Repeat (LTR), a cis-regulatory element responsive to host transcription factors (TFs).

• Knowledge Gap: While the mechanisms of LTR activation (e.g., via NF-κB) are well-studied in a few laboratory strains, the functional impact of HIV's immense genetic diversity (across clades, recombinant forms, and within individual patients) on transcriptional regulation remains poorly understood.
• Clinical Implication: "Shock-and-kill" cure strategies rely on the assumption that diverse proviruses respond similarly to latency-reversing agents. However, variations in TF binding site configurations and intragenic regulatory elements may lead to heterogeneous reactivation, contributing to the failure of these strategies.
• Unexplored Territory: The extent and conservation of intragenic cis-regulatory elements (CREs) outside the LTR, and how they compensate for LTR variability, are largely unknown.

2. Methodology

The authors employed a multi-faceted approach combining high-throughput experimental assays, comparative genomics, and machine learning.

• Massively Parallel Reporter Assays (MPRAs):
  • Tiling: The genomes of HIV-1 (clades A, B, C, 01_AE, etc.) and HIV-2 were tiled with overlapping 200 bp sequences.
  • Saturation Mutagenesis: Selected active regions underwent saturation mutagenesis (substituting every nucleotide with the other three bases) to map nucleotide-level contributions to activity.
  • Isolate Screening: Thousands of natural HIV isolates (6,676 HIV-1 and 41 HIV-2) were synthesized and tested.
  • Cellular Context: Experiments were performed in Jurkat cells (unstimulated and stimulated with αCD3+PMA, TNFα, or IFNγ) and human primary CD4+ T cells from four donors to assess cell-type and stimulus-specific variability.
• Validation in Integrated Contexts:
  • Chimeric Proviruses: Full-length HIV-1 constructs with swapped U3 regions were generated to infect Jurkat cells. Reactivation was measured via GFP expression to validate MPRA findings in an integrated proviral context.
  • CASCADE (Protein-Binding Microarrays): Used to identify specific transcription factors and chromatin-modifying cofactors (e.g., p300, NCOA3, RBBP5) recruited by viral DNA elements.
• Computational Modeling:
  • CREST (Cis-Regulatory Element Sequence Tester): A deep learning model (transfer learning from Malinois and Sei) trained to predict baseline MPRA activity from DNA sequence.
  • LARM (LTR Activation Response Model): A gradient-boosting model trained to predict stimulus-induced fold-activation (specifically for TNFα) based on TF occupancy features.
• Longitudinal & Transmission Analysis:
  • Applied predictive models to longitudinal plasma samples from 7 treatment-naive patients and transmission pairs (mother-infant and heterosexual) to track evolutionary trajectories of regulatory activity.

3. Key Contributions

1. First Functional Atlas: Created the first comprehensive, high-resolution functional map of HIV-1 and HIV-2 cis-regulatory elements across thousands of natural isolates.
2. Quantification of Variability: Demonstrated that transcriptional activity varies drastically (up to 18-fold) across isolates, even within the same clade, driven by specific TF motif configurations rather than phylogenetic grouping alone.
3. Discovery of Intragenic CREs: Systematically identified conserved intragenic enhancers/promoters in gag, env, and pol that function independently of the 5' LTR, particularly in defective proviruses.
4. Predictive Framework: Developed and validated CREST and LARM, enabling the in silico prediction of viral transcriptional behavior from sequence alone, bypassing the need for costly wet-lab assays for every new isolate.
5. Regulatory Compensation: Revealed a "regulatory grammar" where activity in one LTR region (e.g., U3) can be compensated by another (e.g., U5) or by intragenic elements, maintaining balanced transcriptional output.

4. Key Results

A. Extensive Transcriptional Variation

• Clade Differences: Clade C isolates generally showed higher baseline activity in the U3 region (attributed to an extra NF-κB site), while Clade 01_AE showed weaker TNFα response but stronger IFNγ response.
• Within-Clade Heterogeneity: Activity varied significantly among isolates of the same clade. This was driven by the number, strength, and spacing of TF binding sites (NF-κB, SP/KLF, IRF, ETS).
• Cooperativity: Structural modeling (AlphaFold3) and mutagenesis confirmed that NF-κB and SP1/SP/KLF sites must be in close proximity to cooperate effectively. The absence of SP/KLF sites drastically reduced inducibility.
• Recombinants: Recombinant viruses did not necessarily inherit the regulatory properties of their parental clades; their LTR activity was highly variable and often distinct from parents.

B. Donor-Specific Variability

• In primary CD4+ T cells, isolates lacking specific TF configurations (e.g., SP/KLF sites in U5) showed high variability in activity across different human donors.
• Specific regulatory architectures (like the presence of SP/KLF sites) acted as "buffers," reducing donor-to-donor variability and ensuring robust transcription across diverse cellular environments.

C. Within-Host Evolution

• Transmission: Analysis of transmission pairs showed no consistent selection for high or low LTR activity during transmission. The bottleneck appears driven by other factors (e.g., envelope tropism), not intrinsic LTR strength.
• Longitudinal Evolution: Within individuals, baseline LTR activity showed heterogeneous trajectories (some increasing, some decreasing, some stable) over 5–8 years. However, stimulus-induced responsiveness (reactivation potential) remained relatively stable, suggesting that while constitutive activity evolves, the capacity to reactivate is constrained once infection is established.

D. Intragenic CREs and HIV-2

• Intragenic Elements: Identified active CREs in gag (regulated by ETS/bZIP) and env (regulated by ETS/CREB/CEBP). These elements are under evolutionary constraint and can drive transcription in defective proviruses lacking a functional 5' LTR.
• HIV-2 vs. HIV-1: HIV-2 LTRs are generally less active due to a single NF-κB site and lack of IRF sites. However, HIV-2 compensates via intragenic CREs (e.g., in gag) that recruit similar cofactors (RBBP5) via IRF motifs, suggesting a convergent regulatory strategy despite different LTR architectures.

5. Significance and Implications

• Cure Strategies: The findings challenge the "shock-and-kill" paradigm by showing that genetically diverse reservoirs have vastly different reactivation potentials. A single latency-reversing agent may fail to reactivate specific proviral subpopulations due to distinct TF configurations.
• Pathogenesis: Variations in LTR responsiveness to cytokines (e.g., IFNγ vs. TNFα) may explain differences in disease progression between clades (e.g., the rapid decline in CD4+ T cells in 01_AE infections).
• Defective Proviruses: The discovery of functional intragenic CREs suggests that "defective" proviruses (which make up the majority of the reservoir) may still produce viral RNA and proteins, contributing to chronic inflammation and immune activation despite ART.
• Resource: The authors released HIVRegDB, a web resource containing all experimental data, predictive models (CREST/LARM), and tools for users to submit sequences for functional annotation.

In summary, this study moves HIV research from a "one-strain-fits-all" model to a nuanced understanding of viral regulatory diversity, providing the tools and data necessary to predict how specific viral sequences will behave in different hosts and therapeutic contexts.