Hidden Hearing Loss, Gene Therapy and Regeneration

In male CBA/CaJ mice, noise that left hearing thresholds back to normal within two weeks still cut the ribbon count at one cochlear place by about half. A threshold is a statement about the quietest sound that gets through, and it turns out to say little about how many synapses carry the rest. This lesson asks what the ribbon synapse needs in order to fire, what noise does to it, how a gene too large for its vector was delivered anyway, and why the reasons birds regrow hair cells do not simply transfer to us.

Hearing and balance covered transduction, the amplifier and the ribbon synapse in one paragraph, and said that mature mammalian hair cells are not replaced. We start at the synapse’s molecules and at what noise does to it. For quantal release and calcium sensors in general, see synaptic transmission and vesicle traffic.

The synapse that must fire for a lifetime

An inner hair cell releases transmitter continuously, graded with its receptor potential, onto auditory nerve fibres. The release site is a ribbon, an electron-dense structure at the presynaptic active zone. What the ribbon contributes was tested by Becker and colleagues (2018) in mouse by deleting its main protein, RIBEYE. The ribbon supported transmission by keeping a larger resident pool of vesicles, and removing RIBEYE did not stop inner hair cell synapses from forming. So the ribbon tunes the supply of vesicles, and it is not what assembles the synapse.

Which molecule senses calcium and triggers fusion? Roux and colleagues (2006) recorded from inner hair cells of Otof-null mice. Calcium current and ribbon morphogenesis were preserved, but exocytosis was almost abolished. They proposed that otoferlin, a protein with several C2 domains (modules that often bind calcium), may be the major calcium sensor for fusion. A knockout shows that a protein is needed, not that it is the sensor, because a protein could be needed to dock or prime vesicles without sensing anything.

A subtler perturbation addresses that. Michalski and colleagues (2017) made a knock-in mouse in which two alanines (Ala515 and Ala517) replaced residues in the C2C domain, reducing calcium binding. In inner hair cells recorded ex vivo at postnatal days 14 to 18, with patch clamp and calcium uncaging, the calcium sensitivity of both fast and sustained exocytosis dropped. The protein was present, so its absence could not explain the change. That fits a calcium-sensing role, and it does not show that otoferlin is the only sensor.

A normal audiogram over a damaged cochlea

An audiogram plots behavioural threshold against frequency. Neither it nor a physiological threshold counts synapses. Kujawa and Liberman (2009) asked what happens to them after noise that does not leave a threshold shift. In male CBA/CaJ mice exposed in vivo at 16 weeks to 8 to 16 kHz octave-band noise at 100 dB SPL for two hours, response thresholds returned to pre-exposure values by two weeks. Hair cells remained intact. Afferent terminals were lost acutely, ribbon counts in the 32 kHz region were about 50 percent below control at eight weeks, and the cochlear nerve degenerated later. This shows that recovered sensitivity can conceal injury in this strain, sex, age and exposure. It does not give a prevalence or a perceptual consequence in people.

Two human studies then tried to find the same lesion without counting synapses, since no living human synapse count exists. Liberman and colleagues (2016) studied 34 young adults, mean age about 24 to 25, grouped by self-reported noise exposure and use of hearing protection. The higher-risk group had a higher SP/AP ratio on electrocochleography (the ratio of the summating potential, a hair cell response, to the compound action potential of the auditory nerve), which the authors said was consistent with selective neural loss. The same group also had raised thresholds at 10 to 16 kHz despite normal conventional audiograms. Guest and colleagues (2018) studied adults aged 18 to 40 with normal audiograms from 0.25 to 8 kHz: 16 with verified difficulty understanding speech in noise and 16 closely matched controls, matched on thresholds up to 14 kHz. They found no association between the impairment and ABR (auditory brainstem response) or EFR (envelope-following response, a measure of neural phase locking to a sound’s slow amplitude changes) markers of synaptopathy. The authors allowed that these measures might not be sensitive enough.

Read the pair without choosing a side. The first study could not separate synapse loss from the high-frequency threshold difference, since a raised 10 to 16 kHz threshold is itself a cochlear difference. The second matched on that, which is a stronger comparison, but it used proxies whose sensitivity in people is unknown. A complaint, a proposed lesion and a validated test are three different things.

Design problem

Design a study that could show whether noise-induced synapse loss explains trouble understanding speech in noise in people with normal audiograms. Then say what the published work can and cannot support.

One way to do it

No living human synapse count exists, so the design must name its physiological proxy in advance (for example ABR wave I amplitude or an envelope-following measure) and set the sample size from that proxy’s sensitivity, not from hope. Match groups on extended high-frequency thresholds through 14 kHz, as Guest and colleagues did, so a raw cochlear threshold difference cannot masquerade as synapse loss. Measure noise exposure prospectively, because recalled history is biased. Fix the speech-in-noise test and its endpoint before looking at the proxy.

What the published work supports: in male CBA/CaJ mice a defined exposure left thresholds recovered and ribbons halved at one place. It supplies the lesion and its timing. It does not supply the prevalence in people, a diagnostic criterion, or proof that the lesion causes a listening complaint.

Replacing a gene too big for its vector

Loss of otoferlin function is a recessive deafness, and the gene is a good target for replacement because the missing activity sits in cells that survive. There is a size problem. A conventional adeno-associated virus (AAV) vector carries about 4.7 to 5 kb, and the otoferlin coding sequence is about 6 kb, before a promoter is added.

Al-Moyed and colleagues (2019) split the cDNA between two AAV2/6 vectors, so that a cell receiving both could reconstitute the full-length cassette. They tested two designs that rebuild it by different routes, trans-splicing and a hybrid strategy. Otof-null mice were injected in vivo through the round window (a membrane at the base of the cochlea) at postnatal days 6 and 7, and cells were recorded at days 14 to 18. Full-length otoferlin appeared in up to 50 percent of inner hair cells. Fast exocytosis was restored in transduced cells, measured as a rise in membrane capacitance, but vesicle replenishment reached only 35 to 50 percent of wild-type values. Click-evoked ABR thresholds in treated mice were 40 to 60 dB SPL.

A simple consideration explains the ceiling. Suppose, in a made-up case, each vector independently reaches 70 percent of inner hair cells. A cell needs both, so the expected fraction is 0.7 times 0.7, or 0.49, close to the reported maximum. Real transduction is not independent, so this is a way to see why coverage is capped, and it is not a prediction. The result is partial rescue in neonatal mice. It does not show normal speech perception in anyone.

Cells reached by vector 1 70% Cells reached by vector 2 70% Resample

Expected with both (p1 × p2)

49.0 of 100

Observed in this drawn field

0 of 100

Reported maximum

up to 50 percent

In this sketch: independent delivery, every cell with both halves counted as reconstituted, made-up reach values. The reference line is from Otof-null mice injected at P6 to P7 with dual AAV2/6 (Al-Moyed and colleagues, 2019).

This sketch does not say: that the product is a prediction (real transduction is not independent); that a cell with both vectors works normally (vesicle replenishment reached only 35 to 50 percent of wild type in transduced cells); that coverage equals hearing (click ABR thresholds in treated mice were 40 to 60 dB); or anything about children or speech.

A cell can rebuild full-length otoferlin only if it received both halves, so, in this sketch, the fraction with full-length otoferlin is limited by how many cells receive both, not by either vector alone.

Reading an early-phase trial without inflating it

The human trials use OTOF replacement, which regenerates no hair cells and treats no other form of deafness. Lv and colleagues (Lancet, 2024) reported a single-centre, single-arm study of six children with biallelic OTOF variants, treated in one ear with an AAV1-hOTOF vector and followed for 26 weeks. Five showed hearing recovery, with average ABR threshold reductions of 40 to 57 dB among the five. No dose-limiting toxicity or serious adverse event was reported in that six-child, 26-week report. Wang and colleagues (Nature Medicine, 2024) reported an interim analysis of five children, aged 1.2 to 11.0 years, treated in both ears. All five improved, and all five localized sound better, with 13 or 26 weeks of observation and 36 adverse events, none dose-limiting or serious. Qi and colleagues (Advanced Science, 2024) reported two children, one treated in one ear and one in both, and normal-range hearing in the treated ear of the five-year-old one month after injection. (A separate Qi trial of ten participants appeared in Nature Medicine in 2025, and a 2026 Nature report of 42 participants followed some for up to 2.5 years. That is evidence of persistence over those years, and it is not proof of lifelong durability.)

The DB-OTO trial (New England Journal of Medicine) treated twelve children with OTOF-related profound deafness in an open-label single-group design. Nine of twelve met the week-24 primary endpoint, an average of 70 dB HL or better, and three of twelve reached average normal sensitivity. It recorded 67 adverse events. The United States Food and Drug Administration is reported to have granted accelerated approval on 23 April 2026, for patients with confirmed biallelic OTOF variants, preserved outer hair cell function and no prior implant in the treated ear, with a larger analysis of 16 of 20 efficacy-evaluable participants. Regulatory status changes, and none of this supports a claim about lifelong benefit.

Worked problem: reading a censored number

In the Wang trial, patient 1 had baseline average ABR thresholds above 95 dB in each ear and 58 dB at week 26. A student writes: “the treatment improved hearing by exactly 37 dB, so it restored 39 percent of hearing.”

Step 1, the bound. Let the true baseline be B, with B above 95. The improvement is B minus 58, which exceeds 95 minus 58, or 37 dB. So 37 dB is a strict lower bound, because the baseline was above the test ceiling and was not measured.

Step 2, the percentage. The student’s 39 percent is 37 divided by 95. It divides a bound by the ceiling and calls it a fraction of hearing. These censored values do not identify a percentage of hearing restored.

Step 3, a different kind of fraction. In the DB-OTO report in the New England Journal of Medicine, 9 of 12 is 75 percent. That is the share of children meeting a stated endpoint (70 dB HL or better), not 75 percent of normal hearing. The later label analysis of 16 of 20 (80 percent) is a different analysis set and must not be pooled with it.

Birds and fish regrow hair cells, and adult mice barely do

Birds and fish show that hair-cell loss need not be permanent. Corwin and Cotanche (Science, 1988) reported replacement of damaged hair cells in the chicken cochlea after acoustic trauma. In a separate Science paper that year, Ryals and Rubel studied adult Coturnix quail: hair-cell counts recovered to within 5 percent of the original count, and thymidine incorporation appeared over hair-cell and supporting-cell nuclei within ten days of the trauma. Those are different species and different experiments, and neither is a mammalian result. In the zebrafish lateral line, a surface mechanosensory organ built of neuromasts, Pinto-Teixeira and colleagues (2015) found regeneration after repeated severe injury and in adult animals, with hair cells restored within 72 hours in animals aged three months, one year and two years. That is not a cochlea.

In mice the question became which barriers stop supporting cells from converting. Iyer and colleagues (eLife, 2022) used Cre-inducible mice to express ATOH1 alone and with GFI1 and POU4F3 in cochlear nonsensory cells, inducing at postnatal day 1 or day 8. At day 1 every ATOH1-containing combination tested gave cells with hair-cell morphology and rudimentary mechanotransduction, measured partly by rapid uptake of the dye FM1-43. At day 8 the three-factor combination gave about 50 to 60 new hair cells per 200 micrometres at day 15, against fewer than 5 for ATOH1 alone or with GFI1. Joint single-nucleus RNA and ATAC sequencing (ATAC reads which stretches of chromatin are open) of untreated mice found that hair-cell gene loci were less accessible at day 8, and hair-cell-to-supporting-cell signalling, including Notch (see Notch signaling), could act as a further barrier.

McGovern and colleagues (PNAS Nexus, 2024) pushed the age further. Using Lfng-CreERT2 lineage tracing in mature mouse organ of Corti, they induced at three weeks and collected at four to six weeks, and found that three factors could reprogram some supporting cells into hair-cell-like cells with rudimentary stereociliary bundles, more of them when the native hair cells had been killed around the time of induction. The new cells were reported to be innervated and to carry ribbon-synapse proteins. A cell with a rudimentary bundle is not a restored hearing organ. The unsettled step is whether such cells form working synapses with spiral ganglion neurons, which would be tested by recording from them and by measuring auditory responses in deafened mature animals. For the biology of regeneration in general, see injury and regeneration and stem cells; for the devices that substitute for what is lost, see sensory prostheses.