2019 Zhu — Single-cell proteomics reveals changes in expression during hair-cell development

Single-cell proteomics of E15 chick utricle: TMSB4X actin-buffer is 10× lower in hair cells than supporting cells (riBAQ 0.006 vs 0.056), releasing G-actin for stereocilia.

Citation

Zhu Y, Scheibinger M, Ellwanger DC, Krey JF, Choi D, Kelly RT, Heller S, Barr-Gillespie PG. “Single-cell proteomics reveals changes in expression during hair-cell development.” eLife 8:e50777, 2019. DOI 10.7554/eLife.50777. Published 2019-11-04. CC0 public domain.

Filename note: PDF is krey2019.pdf because Krey is corresponding-author lab. First author is Zhu.

Earlier-note correction (2026-04-25): previous version of this note attributed the paper to “Krey JF, Chatterjee P, Halford J, Cunningham CL, Perrin BJ, Bharat TAM…” — that author list was hallucinated. Verified-correct list above is taken verbatim from the locally-parsed MinerU output of the eLife PDF (~/BookLibrary/mineru-output/krey2019/krey2019/auto/krey2019.md, line 19).

TL;DR

NanoPOTS single-cell proteomics on FACS-sorted E15 chick utricle cells (1 cell ≈ 1 picoliter) revealed that the actin-monomer buffer TMSB4X is at near-equimolar concentration with actin in supporting cells (progenitors) but drops ~10-fold in hair cells. The authors propose this drop releases sequestered G-actin for stereocilia bundle assembly, explaining why hair cells build elaborate F-actin without upregulating actin gene transcription. CellTrails developmental-trajectory inference reproduces the supporting-cell → hair-cell transition from protein expression alone (30 single cells) and corroborates with scRNA-seq from 254 cells.

Numbers that matter

Cell biophysical parameters

ParameterValueMethodSource (page)
FM1-43 high (hair cell) volume1.01 ± 0.20 pL (mean ± SD, N=9)post-FACS volumetricp.3
FM1-43 low (supporting cell) volume0.67 ± 0.15 pL (N=8)post-FACS volumetricp.3
HeLa cell volume (comparator)4.9 ± 1.1 pL (N=5)same methodp.3
Stereocilia per E15 utricle hair cell72 ± 8 (N=26 striolar + extrastriolar)wholemount countsp.5
Total actin per E15 hair cell~15,000,000 moleculesACTG1 riBAQ × volumep.5
Actin per E15 stereocilium~200,000 moleculesderived from 15M / 72p.5
Actin per E20 stereocilium (Shin 2013 ref)~400,000 moleculesexternal priorp.5
Single FM1-43 high cell, unique peptides~200 (50% MS/MS, 50% Match)nanoPOTS LC-MS/MSp.4
20-cell pool FM1-43 high, unique peptides~2,500 (~70% MS/MS)samep.4
Proteins detected in single FM1-43 high cell~60identification thresholdp.4
Proteins detected in 20-cell FM1-43 high pool~600samep.4

TMSB4X vs actin stoichiometry (load-bearing for h09)

Verbatim from Figure 2C, 20-cell pool samples (mean ± SEM):

ProteinFM1-43 high (hair cell) riBAQFM1-43 low (supporting cell) riBAQFold (low/high)
ACTG1 (total actin)0.043 ± 0.0010.060 ± 0.0051.4×
TMSB4X0.006 ± 0.0020.056 ± 0.0129.3×

Stoichiometry consequence: TMSB4X binds actin 1:1 (Goldschmidt-Clermont 1992).

  • Supporting cells: TMSB4X ≈ ACTG1 → most actin monomers sequestered
  • Hair cells: TMSB4X / ACTG1 ≈ 0.14 → ~86% of actin not bound by TMSB4X

Statistical confidence (verbatim Table 1, p.6)

ComparisonEstimateStd. errordft-value95% CI lower95% CI upperp-value
ACTG1 FM-high − ACTG1 FM-low−0.0170.00922.3−1.830−0.0360.0020.081
ACTG1 FM-high − TMSB4X FM-high0.0360.00812.04.6070.0190.0540.001
ACTG1 FM-low − TMSB4X FM-low0.0040.00812.00.451−0.0140.0210.66
TMSB4X FM-high − TMSB4X FM-low−0.0500.00922.3−5.356−0.069−0.031<0.001

Mixed-effects model with random intercept for samples, lmerTest 3.1-0 (Kuznetsova 2017).

Single-cell trajectory analysis parameters

ParameterValueSource (page)
Single cells in proteomics trajectory30p.7
Proteins/groups passing detection ≥5 cells75p.7
High-variance proteins for embedding37 (log2 niBAQ > 1.0 + above variance fit)p.7
Manifold dimensions4Methods
scRNA-seq cells (post-QC)254 (S1–S7), of 384 startingp.10
scRNA-seq genes after filtering186 assay genes (182 prior + AGR3, AK1, GPX2, TMSB4X)p.10 + Table 3
Smoothing spline DOF4Methods
Sequencing depth2 × 150 bp paired-end on NextSeq 500/550 High OutputMethods

Derived [F-actin] (calculation, NOT in paper)

Re-derived from this paper’s Numbers + standard stereocilium geometry. Flagged as derived.

ParameterValueNotes
Mature OHC stereocilium diameter~200 nmexternal (Tilney)
Mature OHC stereocilium length~3 μmexternal (Tilney)
Stereocilium volume~9.4 × 10⁻¹⁷ Lπ·r²·h
Total actin per stereocilium200,000 molecules (this paper)p.5
Local [actin] in stereocilium~3,500 μM (3.5 mM) totalAvogadro
[F-actin] assuming 90% filamentous~3.15 mMderived
Practical working estimate1–5 mM [F-actin] in mature corebracketed

Method essentials

  • Tissue: E15 Gallus gallus utricle, peeled epithelium → cells dissociated (Accutase + thermolysin).
  • Sorting: FM1-43 dye loading distinguishes hair cells (high) from supporting cells (low). FACS into nanowells (single + pools of 3, 5, 20).
  • Sample prep: nanoPOTS (Zhu 2018b, Nat Commun) — all protein extraction, reduction, alkylation, proteolysis (trypsin + Lys-C) inside one nanowell, robotic dispensing.
  • LC-MS: 30-μm-i.d. column → Orbitrap Fusion Lumos Tribrid, data-dependent acquisition.
  • ID/quant: MaxQuant + Andromeda, Match Between Runs, iBAQ (intensity-based absolute quantification, Schwanhäusser 2011); per-protein iBAQ normalized as riBAQ (relative molar fraction) and niBAQ (log2 batch-corrected for trajectory).
  • Trajectory inference: CellTrails R package (Bioconductor 10.18129/B9.bioc.CellTrails), spectral embedding into 4D manifold, pseudotime via diffusion-pseudotime style approach.
  • scRNA-seq: Smart-seq2 (Picelli 2014) on 384 FACS-sorted cells, paired-end 2×150 bp NextSeq, STAR + RSEM, SCnorm normalization.
  • Statistics: mixed-effects model (lmerTest 3.1-0); Benjamini-Hochberg FDR for volcano plot.
  • Antibody validation: anti-TMSB4X validated by knockdown in two prior papers (Zhou 2013, Li 2018); chicken TMSB4X = mouse/human within 2 amino acids of 44 (95.5% identity).

Data deposit: ProteomeXchange PXD014256 (PRIDE).

Limitations

  • Only 30 single cells in proteomics trajectory — insufficient to recover the TrS / TrES / TrES* triple-branching seen in the 254-cell scRNA-seq.
  • Membrane proteins systematically underrepresented (general shotgun-MS limitation; Kar 2017). Only ATP1A1, SLC17A8, ATP2B2 detected. STRC absent — neither in proteomics nor in the 186-gene scRNA-seq panel (Table 3, verified). Chick utricle ≠ mammalian cochlear OHC; STRC’s HTC role is mammalian-OHC-specific.
  • ACTB vs ACTG1 cannot be discriminated by mass spec here (only one isoform-distinguishing peptide detected); only scRNA-seq separates them.
  • Total iBAQ scaled non-linearly with cell number (Figure 1F) → significant protein loss to nanowell-surface adsorption. Absolute counts are lower bounds.
  • TMSB4X peptides covered 75% of the ~5 kDa protein (5 peptides). One of the 5 peptides is shared with paralog TMSB15B (Figure 1—supplement 2). Functional argument hinges on chick Tmsb4x vs Tmsb10/Tmsb15 dominance — supported by gEAR mouse data, not directly measured here.

Relevance to STRC

h09 (Hydrogel) — load-bearing biology

This paper supplies the first quantitative cell-type-resolved measurement of TMSB4X:actin stoichiometry in hair cells. h09’s WH2-peptide bundling mechanism is functionally analogous to TMSB4X’s WH2-class actin-monomer-binding role.

Direct implications for h09 modeling:

  1. Hair-cell free G-actin pool exists — TMSB4X consumes ~14% of actin, leaving ~86% as either free monomer or in F-actin. Working competition target for synthetic WH2 peptide is the unsequestered pool.
  2. Supporting cells are NOT the therapeutic target compartment in h09 (the bundle is in hair cells), but the supporting-cell TMSB4X excess illustrates that endogenous β-thymosin-class binders saturate ~100% of actin when at equimolar abundance — consistent with peptide binding equilibria assumed in Recipe — Profilin Thymosin-β4 Monomer Pool Partitioning.
  3. STRC is NOT measured here. The placeholder STRC_NORMAL_OHC_M = 1 µM remains unsourced (audit confirmed below).

STRC absence — confirmed (carried over from 2026-04-25 audit)

  • Main article body, Figure 1—source data 1 (full MaxQuant output), Figure 5—source data 1 (CellTrails scRNA-seq), Table 3 (186 trajectory genes): STRC / stereocilin / Q7RTU9 not present.
  • Reason: chick utricle E15 ≠ mammalian cochlear OHC. STRC’s horizontal-top-connector role is mammalian-OHC-specific.
  • Verdict: STRC_NORMAL_OHC_M = 1 µM stays an in-house placeholder. Propagate # WARNING: no primary measurement to scripts.

h26 (Engineered Homodimer Avidity) — minor

  • Establishes baseline stereocilia-protein abundance in immature hair cells. Useful for calibrating effective avidity windows, but not load-bearing.

Connections