Treating hearing loss with sound. In Russian we say klin klinom: driving a wedge out with a wedge. The problem itself becomes the cure.
Every gene therapy vector needs a promoter: a switch that tells the cell when to make the protein. Standard approach is always-on. Ours: put the gene behind a switch that responds to sound. Hair cells already have this machinery. Sound bends stereocilia, calcium floods in, a signaling cascade fires. We hijack it. Sound in = gene on. Silence = gene off. The child wears the hearing aid he already has. No special frequencies. Just everyday sounds.
Does the protein vanish at night? No. Stereocilin has a half-life of ~30 days. It takes ~13 hours of hearing aid use to hit 50% of normal levels. Sleep doesn't reset anything.
The ODE model is mathematically correct: equations are sound, parameters are from peer-reviewed literature, and the system dynamics are thoroughly analyzed. However, several biological assumptions have not been validated in cochlear outer hair cells. A literature review (April 2026) identified the following critical gaps.
A more biologically plausible activity-dependent pathway exists. Adenylyl cyclase 1 (AC1/ADCY1) is a calcium-calmodulin-activated enzyme located directly in stereocilia, where MET-channel calcium is abundant. It converts Ca²⁺ signals to cAMP, a small molecule that diffuses freely to the soma, bypassing the calcium compartmentalization barrier entirely.
| NFAT pathway | AC1/CREB pathway | |
|---|---|---|
| Ca²⁺ sensor location | Soma (Ca²⁺ doesn't reach) | Stereocilia (where Ca²⁺ is) |
| Second messenger | Ca²⁺ itself (heavily buffered) | cAMP (freely diffusing) |
| Pathway in OHC | Only activated by damage | AC1 mutations cause deafness |
| Safety | Triggers apoptosis (TNF, Caspases) | Neuroplasticity pathway |
| Promoter element | 6xNFAT-RE (never tested in ear) | CRE (TGACGTCA, well characterized) |
| Tested in hair cells | No | No |
In Strc-/- mice, stereocilia lose horizontal top connectors by P15, bundle stiffness drops 60-74%, and stereocilia progressively disconnect. It is possible that this structural disorganization disrupts calcium compartmentalization, allowing MET-channel Ca²⁺ to leak into the soma, making the NFAT pathway viable only in DFNB16 cells.
If true, this would create an elegant self-dosing mechanism: the disease activates the promoter, the cure silences it. But no one has measured somatic Ca²⁺ in Strc-/- OHCs. This remains pure speculation. It is testable by calcium imaging in cochlear explants from stereocilin-null mice.
This is a computational hypothesis. Each individual component is proven (NFAT promoters, MET channel biophysics, mini-STRC packaging). Nobody has combined them for inner ear gene therapy yet.
Words are cheap. We built a mathematical model to check whether this cascade actually produces enough protein. Five differential equations, every parameter from published literature, four scenarios tested. The question: does a hearing-aid-wearing child produce therapeutic levels of stereocilin through sound alone?
With a realistic hearing aid schedule (16 hours ON at 70 dB, 8 hours sleep), the model predicts 29,571 stereocilin molecules per OHC after 72 hours (target: 15,000). In silence, only 1,023 molecules accumulate (6.8%). This gives a 29-fold dynamic range between sound-activated and silent states. The 50% therapeutic threshold is reached in just 13 hours of hearing aid use. The system self-regulates: protein saturates at the available binding sites on stereocilia, preventing overexpression.
The system saturates at ~20 dB. Between 22 dB (quiet bedroom) and 70 dB (hearing aid), protein output is identical: ~29,750 molecules. The 29x dynamic range exists only between 0 and 14 dB, a range humans rarely experience. Self-dosing via hearing aid gain modulation does not work with the current 6xNFAT promoter.
All 5 eigenvalues at 70 dB have negative real parts. System is asymptotically stable. No oscillations, no chaos. Five timescales: Ca²⁺ (seconds), NFAT cycling (minutes), mRNA (hours), protein (days).
Path-dependence detected due to slow protein degradation (t½ = 30 days). Kinetic memory, not bistability. One steady state per parameter set. System is monostable.
Parameter optimization across 245 combinations found that increasing the NFAT activation threshold (0.05→0.30) and cooperativity (Hill 4→8) moves the switch to 40-50 dB. Biologically: use 9 NFAT binding sites with reduced individual affinity (A→G at position 4 of each site). One nucleotide change, repeated 9 times, transforms the therapy from "always ON" to "hearing-aid controlled."
| Sound | 6xNFAT (current) | 9xNFAT-weak |
|---|---|---|
| 0 dB | 11.6% | 0.1% |
| 20 dB | 179.7% | 0.1% |
| 40 dB | 198.3% | 5.7% |
| 45 dB | ~198% | 51.2% |
| 50 dB | 198.3% | 163.3% |
| 70 dB | 198.3% | 198.1% |
Full bifurcation analysis: Jacobian eigenvalues at 0/70/90 dB, two-parameter sweep (Kd_CaN × NFAT_threshold, 90 points), hysteresis check (ramp up vs down over 72h), and promoter fold-induction sensitivity (10×-200×). All passed. Code: bifurcation_analysis.py
Previous ODE models used fixed sound profiles. This model closes the feedback loop: protein accumulates → hearing improves → audiologist reduces hearing aid gain → less MET activation → less STRC production → equilibrium. The hearing aid becomes a true dosing device.
Day 3: Hearing improves enough that HA adds zero gain. System reaches equilibrium at 27,617 molecules (184% of target), 3.1 dB residual hearing loss, maintained by ambient sound alone.
The promoters behave identically under normal conditions. The difference appears when sound is removed. This is the test of real self-dosing.
9x responds to silence. 6x doesn't.
One half-life. Full recovery in 2 weeks.
9x needs real sound. 6x is constitutive.
| Parameter | Value | Source |
|---|---|---|
| MET channel conductance | 150 pS | Beurg et al. 2006 |
| Channels per bundle | 134 | Fettiplace 2017 |
| Endolymphatic potential | +80 mV | Standard |
| Calcineurin Kd (Ca²⁺) | 500 nM | Stemmer & Klee 1994 |
| NFAT nuclear import t½ | ~2 min | Tomida et al. 2003 |
| Promoter fold induction | up to 62x | Wu et al. 2023 |
| Promoter leakage | Zero (3 weeks) | Wu et al. 2023 |
| Apical compartment volume | 0.05 pL | Lumpkin & Bhatt 2001 |
How fragile is the result? We varied each parameter by ±50% to find what matters most. If the model breaks when one number changes, that parameter needs the most careful experimental measurement.
Transcription and translation rates dominate (sensitivity 1.50). This means the most important experiment is characterizing the 6xNFAT promoter strength specifically in hair cells. Channel count and calcineurin affinity matter less: the system is robust to biological variation in these parameters.
The complete ODE model is available as a Python script. Dependencies: numpy, scipy. Run it yourself to reproduce these results or modify parameters.
View on GitHub: ode_model.py