Promoter mutagenesis and a massively parallel reporter screen of the MAPT locus identifies cis-regulatory elements and genetic variation effects

This study employs massively parallel reporter assays, CRISPR interference, and saturation mutagenesis to map cis-regulatory elements and identify the functional impact of genetic variants on the MAPT locus, revealing novel neuron-specific regulatory mechanisms relevant to tauopathies.

The Gist

The Big Picture: The "Tau" Trouble

Imagine your brain is a bustling city. Inside the city, there are roads (microtubules) that transport important packages. Tau is the "glue" or the "tape" that holds these roads together, keeping traffic flowing smoothly.

However, in diseases like Alzheimer's and Frontotemporal Dementia, this glue gets sticky and clumps together, forming "traffic jams" (neurofibrillary tangles). These jams stop the city from functioning, leading to memory loss and cell death.

The gene that makes this glue is called MAPT. Scientists want to stop the glue from clumping. One way to do this is to turn down the volume on the MAPT gene so less glue is made in the first place. But to do that, we first need to understand exactly how the "volume knob" for this gene works.

The Mission: Mapping the "Volume Knobs"

Think of the MAPT gene as a radio station. The gene itself is the music, but it needs Cis-Regulatory Elements (CREs) to act as the volume knobs, switches, and dials that tell the gene when to play loud, when to play soft, and when to stay silent.

For a long time, scientists only knew about a few of these knobs. This study went on a massive treasure hunt to find all the hidden knobs controlling the MAPT gene, and to see how tiny changes (mutations) in the instructions could break them.

How They Did It: The "Massive Parallel" Experiment

The researchers used a clever technique called MPRA (Massively Parallel Reporter Assay).

The Analogy: The "Taste Test" Factory
Imagine you have a giant factory (the cell) and you want to test 10,000 different recipes to see which one makes the best cake.

1. The Ingredients: Instead of flour and sugar, they took tiny snippets of DNA from the MAPT region (the "recipes").
2. The Test: They put these snippets into a special "reporter" machine inside the factory. If a snippet is a real "volume knob," the machine lights up and glows.
3. The Scale: They didn't test them one by one. They tested thousands of snippets all at once, like a massive taste test where every recipe gets a try simultaneously.
4. The Locations: They tested these in two types of "factories":
   • HEK cells: A generic, standard factory (like a test kitchen).
   • Neurons: The actual brain cells (the real restaurant where the cake is served).

The Discovery: They found that most of the "volume knobs" only work in the Neuron factory. If you tested them in the generic factory, they looked like broken switches. This proves that to understand brain diseases, you have to test in brain cells, not just generic cells.

The "Saturation" Screen: Breaking Every Letter

Once they found the most important area (the MAPT promoter, which is the main switch), they did something even more extreme called Saturation Mutagenesis.

The Analogy: The "Typo" Game
Imagine the main switch is a sentence written in a book: "Turn on the lights."
The researchers took this sentence and created a library of books where they changed every single letter to every other possible letter.

• They changed the "T" to an "A", a "C", and a "G".
• They even deleted small chunks of the sentence.
• They tested every single possible "typo" to see which ones broke the sentence and which ones made it work better.

The Result: They found specific spots where changing even one letter completely broke the switch. They also found that some "typos" only broke the switch in the brain factory, not the generic one.

The "AI" Check-Up

The researchers also asked two super-smart AI computers (AlphaGenome and PromoterAI) to predict what would happen if they made these changes.

The Analogy: The Weatherman vs. The Rain
The AI computers are like weather forecasters. They looked at the DNA and said, "If you change this letter, it will rain (gene expression goes down)."

• The Surprise: In some cases, the AI was right.
• The Glitch: In other cases, the AI got it wrong. For example, the AI predicted a change would stop the gene, but the experiment showed the gene actually got louder.
• Why? The AI didn't account for a hidden "splice" (a way the gene gets cut and pasted) that changes the final product. This teaches us that while AI is great, it can't replace real experiments yet, especially in complex cells like neurons.

The "Villains" and "Heroes" Found

1. New Switches: They discovered new "volume knobs" (CREs) that no one knew existed, including one for a nearby gene called KANSL1 (which is linked to developmental disorders).
2. The "Hot Spot": They found a tiny, 26-letter stretch of DNA that is critical. If you change any letter there, the gene stops working. This area is a binding site for three specific "managers" (transcription factors named EGR2, ZBTB14, and TCLF5) that hold the switch together.
3. Real-World Impact: They tested DNA variations found in real Alzheimer's patients. They found that some of these patient variations actually broke the switches, explaining why those patients developed the disease.

The Takeaway

This study is like a detailed map of the control panel for the brain's "glue" gene.

• We found new buttons we didn't know existed.
• We learned that the buttons only work in the brain, not in test tubes.
• We tested every possible typo to see which ones cause the system to crash.
• We learned that AI is helpful but not perfect yet.

By understanding exactly how these switches work, scientists can now design better drugs (like "antisense oligonucleotides") to specifically turn down the volume on the MAPT gene, potentially slowing down or stopping the progression of Alzheimer's and other dementias.

Technical Summary

1. Problem Statement

Tauopathies, including Alzheimer's Disease (AD), Frontotemporal Dementia (FTD), and Progressive Supranuclear Palsy (PSP), are characterized by the accumulation of Tau protein aggregates. While therapeutic strategies (immunotherapy, antisense oligonucleotides) aim to reduce Tau levels, the precise mechanisms regulating the MAPT gene (which encodes Tau) remain incompletely understood.

• Knowledge Gap: The MAPT locus is located in a complex chromosomal inversion region (17q21.31) with two primary haplotypes (H1 and H2). H1 is associated with increased disease risk, while H2 is protective. However, the specific cis-regulatory elements (CREs) controlling MAPT expression, particularly in a neuron-specific context, and the functional impact of genetic variations (SNVs/InDels) within these regions are largely unannotated.
• Objective: To systematically map the regulatory landscape of the MAPT locus, identify novel CREs, and comprehensively evaluate how genetic variation alters regulatory activity in neuronal and non-neuronal cell types.

2. Methodology

The authors employed a multi-pronged approach combining large-scale screening, targeted validation, and saturation mutagenesis in human cell models.

A. Cell Models

• Neurons: KOLF2.1J-hNGN2 induced pluripotent stem cells (iPSCs) differentiated into excitatory neurons (14 days post-differentiation).
• Control: HEK293FT cells (used for lentivirus production and comparative regulatory assays).
• Mixed Cultures: Differentiated neural progenitor cells (NPCs) containing excitatory neurons, inhibitory neurons, and astrocytes (used for CRISPRi validation).

B. Massively Parallel Reporter Assays (MPRAs)

Three distinct MPRA strategies were utilized:

1. BAC MPRA: 24 Bacterial Artificial Chromosomes (BACs) covering a ~3 Mb region around MAPT were sheared into ~250 bp fragments and cloned into a lentiMPRA vector.
2. Oligo MPRA: A synthesized oligo pool (270 bp) tiled every 90 bp across regions of low BAC coverage, including the MAPT gene itself and upstream regions.
3. Variant/ Mutagenesis MPRA:
   • Known Variants: Tested SNVs and InDels (<10 bp) from the Alzheimer's Disease Sequencing Project (ADSP) overlapping top CREs.
   • Saturation Mutagenesis: A 2,000 bp region encompassing the MAPT promoter was saturated with oligos representing every possible single nucleotide variant (SNV) and 5 bp centered deletions.

C. CRISPR Interference (CRISPRi)

To validate candidate CREs identified by MPRA:

• 14 regions were nominated based on MPRA activity and 3D-omics data.
• dCas9-KRAB guides were transduced into mixed neuronal cultures.
• RNA-seq was performed, and expression changes were analyzed using DESeq2, incorporating Seurat-derived cell-type scores to distinguish direct gene repression from changes in cell population composition.

D. Computational Integration

• AI Comparison: Results from the saturation mutagenesis were compared against two AI predictive tools: AlphaGenome and PromoterAI.
• Motif Analysis: MotifbreakR was used to identify disrupted transcription factor (TF) binding sites.
• Splicing Prediction: SpliceAI was used to assess potential splicing alterations.

3. Key Contributions & Results

A. Discovery of Cell-Type Specific CREs

• The BAC and Oligo MPRAs identified thousands of regulatory bins.
• Specificity: The majority of CREs were cell-type specific.
  • Neurons: 1,335 CREs (BAC) and 864 CREs (Oligo).
  • HEKs: 1,402 CREs (BAC) and 254 CREs (Oligo).
• Novelty: Most CREs identified via BAC MPRA were previously unannotated in ENCODE v4. Oligo MPRA CREs largely overlapped with distal enhancer-like signatures (dELS).

B. Validation of Novel CREs via CRISPRi

• MAPT: Region r5 (~113 kb upstream) was confirmed as a novel, high-confidence CRE for MAPT (adj. p < 0.05).
• KANSL1: Regions r12 and r13 were identified as CREs for the neighboring gene KANSL1 (associated with Koolen-de Vries Syndrome and AD risk).
• Differentiation: Region r8 (~72 kb upstream) showed significant MAPT repression in unadjusted models but lost significance after correcting for cell type scores, suggesting it is critical for neuronal differentiation rather than direct MAPT regulation.

C. Genetic Variation Effects

• ADSP Variants: Testing known AD variants at top CREs revealed strong cell-type specificity.
  • Neurons: 55 SNVs increased activity; 61 decreased activity.
  • Shared vs. Specific: Most effects were cell-type specific. Only 9 SNVs and 2 InDels showed effects in both cell types.
  • Notable Variants: Variants in KANSL1 introns showed significant regulatory shifts. Specific variants at the previously identified CRE (+77,758 bp) showed loss or gain of activity in neurons.

D. Saturation Mutagenesis of the MAPT Promoter

• Scope: Comprehensive assessment of a 2,000 bp promoter region (chr17:45,893,536–45,895,535).
• Findings:
  • Identified 199 variants with increased activity and 192 with decreased activity in neurons.
  • HOT Site: A high-occupancy target (HOT) region (chr17:45,893,810–45,894,854) showed dense regulatory activity and enrichment for variant effects.
  • Common Variant: The common SNV rs11575896 (G>A) showed a specific increase in activity in neurons.
• AI Tool Correlation:
  • Agreement: AI tools (PromoterAI, AlphaGenome) correlated moderately with MPRA data overall.
  • Discrepancy: A region at the Exon 1-Intron 1 boundary showed increased MPRA activity but predicted loss of expression by AI tools. SpliceAI predicted this was due to alternate splicing, highlighting the limitation of AI tools that predict expression without accounting for splicing context.
  • Neuron-Specific Loss: A ~26 bp region (chr17:45,894,271–45,894,297) showed a consistent loss of activity upon mutation. This region overlaps high conservation and TF binding sites for EGR2, ZBTB14, and TCFL5.

4. Significance

• Regulatory Map: This study provides the most comprehensive map of MAPT regulatory elements to date, identifying previously unannotated CREs that are highly specific to neuronal cell types.
• Therapeutic Targets: The identification of specific TF binding sites (EGR2, ZBTB14, TCFL5) and critical CREs (like r5) offers new targets for therapeutic intervention to modulate Tau levels.
• Genetic Risk Interpretation: The study demonstrates that the functional impact of genetic variants (including those from the ADSP) is highly context-dependent (cell-type specific). Variants that appear neutral in non-neuronal cells may have profound effects in neurons.
• AI vs. Experimental Data: The work highlights the current limitations of AI predictive tools in capturing complex genomic contexts (e.g., splicing vs. expression) and underscores the necessity of experimental validation (MPRA) for interpreting non-coding genetic variation in neurodegenerative diseases.
• Resource: The data serves as a foundational resource for understanding the H1/H2 haplotype effects and the regulatory architecture of the MAPT locus in tauopathies.

5. Limitations

• Haplotype Bias: All experiments were conducted in cells with the H1 haplotype; results may differ in H2-containing cells.
• Cell Type Scope: Experiments were limited to excitatory neurons (KOLF2.1J-hNGN2) and HEK293FT cells; other neuronal subtypes (e.g., inhibitory neurons) or glial cells were not directly tested in the MPRA.
• Direct Expression: While MPRA measures regulatory activity, the direct effect of specific variants on endogenous MAPT protein levels was not measured (requires prime editing, suggested for future work).