When a sheet of skin cells closes a wound, it crawls across a soft, protein-rich meshwork called the extracellular matrix (ECM) — the fibrous material, made of collagens, laminins, and attached signaling molecules, that surrounds every cell in a tissue. Almost everything we know about which genes control this crawling comes from a standard laboratory test, the scratch assay, in which a cell monolayer is scraped and filmed as it grows back across the dish. The problem is that the dish is glass-hard plastic, roughly a million times stiffer than skin, and biochemically inert. Meanwhile, biomaterials groups have built increasingly faithful tissue-matrix mimics — decellularized tissue hydrogels, aligned nanofibers, chiral self-assembling gels — but test them only with genetically untouched cells, inferring receptor mechanisms rather than testing them. And geneticists routinely delete genes in specific cell types using Cre-LoxP conditional knockout (a system where an enzyme called Cre snips out a chosen gene only in cells where a selected promoter is active), but they run their migration readouts on bare plastic.
No study in the literature crosses all three: the genes of the migrating cells themselves, the physical and biochemical properties of the matrix they migrate on, and a standardized wound-closure readout. This proposal builds that cross — a factorial map of wound closure speed as a function of conditional genotype and matrix properties — and delivers it with a machine-vision measurement pipeline, plus an annotated benchmark for tracking cells on optically messy biomimetic matrices.
Core Claim
The rules of collective migration inferred from plastic and inert elastomers do not transfer to ligand-rich, physiologically compliant matrices, and the transfer failure is gene-specific. Concretely, we predict two things. First, the monotonic stiffness–speed relationship reported on inert PDMS rubber by The effect of stiff substrates on the collective migration of A549 cells — average speeds of 7.7, 7.1, and 10.0 µm/hr at 1.4, 3.4, and 18.3 MPa, rising to 10.5 µm/hr on 7.7 GPa glass — will flatten, plateau, or become biphasic when stiffness is varied on a matrix that presents native basement-membrane ligands, consistent with the biphasic single-cell prediction from Recent advance in surface modification for regulating cell adhesion and behaviors. Second, deleting a single matrix-binding receptor in the migrating cells (integrin α3, the main keratinocyte receptor for laminin) will selectively abolish the response to laminin-rich matrices while leaving migration on collagen-I matrices intact — a dissociation no existing experiment could detect, because none has varied genotype and matrix composition independently.
The key quantity is the genotype-by-matrix interaction. Let $v(g, m)$ be the wound-edge closure speed for genotype $g$ on matrix $m$. The claim rests on measuring the interaction contrast
$$ I(g; m_A, m_B) = \big[v(g, m_A) - v(\text{WT}, m_A)\big] - \big[v(g, m_B) - v(\text{WT}, m_B)\big], $$
which is zero if a gene's effect on closure is the same on every matrix, and large only if the gene's necessity depends on what the cells are crawling on. Everything in this proposal is an instrument for making $I$ measurable, honestly, and at scale.
Research Program
The matrix panel. We produce two families of biomimetic matrices with independently tunable mechanics and composition. The first follows the detergent-free, weakly alkaline isolation protocol of A simple method to isolate structurally and chemically intact brain vascular basement membrane for neural regeneration following traumatic brain injury, which preserves native architecture while removing over 99% of native DNA, and whose genipin crosslinking raised the elastic modulus from 0.54 MPa to 1.12 MPa — our lever for varying stiffness at constant composition. The second follows the SDS-decellularized hydrogel route of Decellularized Human Kidney Cortex Hydrogels Enhance Kidney Microvascular Endothelial Cell Maturation and Quiescence, whose proteomic profile (~37% collagen-IV, ~20% collagen-I, ~8.6% laminins) we adopt as a quantitative reference composition, verified by LC-MS/MS on every batch to control the batch-to-batch heterogeneity that the field itself flags as its biggest reproducibility problem. By dilution and crosslinking we span 1 kPa to ~1 MPa — covering the entire soft-tissue range (1–150 kPa) that even the most careful existing stiffness study never reached, since its softest substrate was 1.4 MPa. Purified collagen-I gels at modulus-matched points, plus bare PDMS and glass, anchor the panel to the old regime.
The genotype panel. Primary murine keratinocytes carrying floxed alleles of Itga3 (the laminin-binding integrin α3β1), Ptk2 (FAK, the intracellular hub that turns adhesion into signaling), and RhoA (the cytoskeletal controller), each crossed to an inducible K14-CreERT2 driver so deletion happens in adult cells on demand — deliberately avoiding both embryonic lethality and the compensatory rewiring that the conditional-knockout literature identifies as a masking artifact. Littermate floxed-without-Cre cells are the matched controls for every experiment. This trio is chosen so a phenotype can be localized: if α3 deletion breaks adhesion to laminin but FAK deletion breaks speed everywhere, the mechanism separates grip from signaling.
The assay and measurement layer. Scratch wounds are made by stencil release rather than scraping, eliminating the ragged-edge variability the field names as its main artifact, with mitomycin C suppression and a primary 12-hour analysis window to keep proliferation from masquerading as migration. Imaging feeds an automated tracking pipeline evaluated with the Cell Tracking Challenge metrics SEG (segmentation overlap accuracy) and TRA (trajectory graph-matching accuracy) from The Cell Tracking Challenge: 10 years of objective benchmarking. Because fibrillar matrices scatter light and confound even nnU-Net-class models — the "Generalizability Wall" that benchmark's own authors describe — we will release an annotated benchmark of monolayer wound closure on light-scattering matrices, with silver-standard dense labels built by fusing top algorithmic predictions, the same trick that lifted annotation coverage from a sparse 17.8% to 99.1% in the CTC corpus.
A secondary experiment addresses a live contradiction in the literature: whether wound-edge migration is dominated by soluble inflammatory signals or by the cells' own matrix-engagement machinery. MED1 Deficiency in Macrophages Accelerates Intimal Hyperplasia via ROS Generation and Inflammation showed that macrophage-specific MED1 knockout produces a secretome that accelerates wild-type vascular smooth muscle cell scratch closure — but on plastic, with the responder cells genetically wild-type. We will apply that pro-migratory conditioned medium to α3-knockout and wild-type sheets on basement-membrane matrices. If the secretome rescues the knockout, soluble cues dominate; if it cannot, cell-autonomous matrix engagement is the bottleneck that paracrine stimulation cannot bypass.
Baselines
The do-nothing reference is the field default: wild-type cells in a manually scratched well of uncoated polystyrene, gap width measured by hand at endpoints. It is not a competitor for accuracy but a calibration of the artifact we claim exists — by Sasaki et al.'s own measurements, near-glass rigidity pushes collective speed to 10.5 µm/hr, far above anything soft tissue sees, so this baseline tells us how big the plastic artifact is in our own hands.
The strongest existing competitor for our central question is Sasaki's stiffness-series design, because it is the only study that has systematically varied mechanics under a scratch assay with careful measurement. We replicate its logic exactly — same stiffness ladder concept, same 12-hour window, same speed and directionality metrics — but on our ligand-matched dECM series and with genotype as a second factor, using the same cells, imaging rig, tracker, and analysis code for every condition so only matrix and genotype vary. Beating it means one of two results: a statistically resolved change in the shape of the speed–stiffness curve on ligand-rich matrices relative to its monotonic PDMS behavior, or a nonzero interaction contrast $I$ for at least one genotype, which its wild-type-only design is structurally incapable of detecting.
The ablation that isolates our key genetic ingredient is transient siRNA knockdown of the same three targets in the same cells on the same matrices, following the approach Nitric oxide promotes epidermal stem cell migration via cGMP-Rho GTPase signalling used for RhoA and Rac1. If siRNA and true knockout produce indistinguishable phenotypes, the expensive genetics adds nothing and we will say so; if they diverge, the divergence itself quantifies how much the field's knockdown-based conclusions understate gene necessity.
For the tracking layer, the strongest baseline is MALA, the affinity-prediction-plus-agglomeration tracker that achieved a TRA of 0.997 and a SEG of 0.907 on dense Drosophila epithelial sheets in A Benchmark for Epithelial Cell Tracking, with the off-the-shelf Tissue Analyzer tool (0.758 SEG, 0.865 TRA on proper disc movies) as the easy lower bar and the human baseline of TRA ~0.95 from CTMC: Cell Tracking with Mitosis Detection Dataset Challenge as the ceiling. All trackers get the same frames, the same annotation budget, and the same hyperparameter search budget; beating the baseline means an adapted model closing a measured share of the gap between MALA's out-of-the-box performance on our fibrillar data and its 0.997 reference, and we will report the drop itself as a benchmark result either way.
Why not just…?
Why not just use siRNA instead of breeding Cre-LoxP mice? This is the most serious alternative, and for a single 12-hour assay it might genuinely suffice — Zhan et al. got clean mechanistic answers from siRNA. The counter-evidence is that knockdown is incomplete and decaying: Membranous type matrix metalloproteinase 16 induces human prostate cancer metastasis found siRNA suppressed but did not abolish migration, and silencing fades over 48–96 hours as cells divide. Rather than argue, the siRNA-versus-knockout ablation above settles it empirically; if siRNA matches knockout everywhere, the simple method wins and our genetics is reduced to a validation role.
Why not just add chemical inhibitors to the medium? Pharmacology is cheaper and faster, but it bathes every cell at once, and the dose-response is treacherously biphasic in exactly this literature: the chiral DMOG hydrogel study found 0.1 mM DMOG promotes endothelial proliferation while 0.5–1 mM inhibits it, and Zhan et al. found 100 µM SNAP accelerates closure (82.1% vs. 48.8% at 12 h) while 300–500 µM suppresses it. More decisively, Zhang et al.'s MED1 work showed the NF-κB inhibitor BAY11-7082 could not rescue the PGC1α-driven antioxidant network that genetic manipulation rescued — a case where chemistry failed and genetics succeeded.
Why not just coat plates with collagen-I? Single-protein coatings are reproducible and optically clean, but Nagao et al. showed purified collagen-I actively pushes renal microvascular endothelium into a proliferative, fibrogenic-like state — the wrong phenotype for studying physiological repair — while only the complex native-composition matrix restored quiescence. A collagen-I-only panel would answer a question about fibrosis, not about normal wound closure, and it cannot detect composition-dependent gene effects because it has no composition axis.
Why not just run Transwell assays instead of scratching? Transwell chambers measure single cells squeezing vertically through pores toward a chemoattractant; wound re-epithelialization is a horizontal, collective movement of a cohesive sheet governed by cell-cell junctions and leader-follower polarity. Transwells also physically cannot present an aligned nanofibrous sheet or an intact basement membrane as a migration substrate, so the contact-guidance half of our question — central to results like the aligned SIS periosteum scaffold — is untestable in that format. We retain a Transwell arm only as a sanity check that individual chemotaxis and collective closure can dissociate.
Proposed Experiments and Decisive Checks
Experiment one builds the matrix panel and verifies, per batch, modulus by AFM and composition by LC-MS/MS against the Nagao reference profile; a batch outside tolerance is discarded, which is itself a contribution given that the field lists batch heterogeneity as its top open problem. Experiment two establishes the wild-type speed–stiffness curves on every matrix family, delivering the first direct test of whether Sasaki's monotonic scaling survives on ligand-rich compliant substrates; the decisive check is a pre-registered model comparison between monotonic and biphasic fits, with the fibroblast mechanosensing threshold above ~2,000 Pa reported by Cai et al. as the theoretical anchor for where the curve should bend. Experiment three runs the full genotype-by-matrix factorial and tests $I \neq 0$, with the headline prediction that α3 knockout cripples closure on laminin-rich matrices only; failure of this prediction would itself falsify the receptor-specificity story and is treated as a real outcome, not a null result to bury. Experiment four runs the siRNA-versus-knockout comparison described above. Experiment five applies MED1-knockout macrophage conditioned medium, made exactly as in Zhang et al., to knockout and wild-type sheets on basement-membrane matrices, settling the secretome-versus-machinery contradiction. Experiment six builds and releases the fibrillar-matrix tracking benchmark, running MALA, CELLECT (which processes a frame in 2 seconds, 56 times faster than linajea, with 46% long-trajectory accuracy on C. elegans), and a detection-driven tracker out of the box, then measuring how large the Generalizability Wall actually is and how much of it fine-tuning on our silver-standard corpus recovers.
The program closes with the two-sentence deliverable we want a reader to walk away with: on matrices that look like real tissue, which genes are necessary for wound closure depends on which matrix the cells are standing on, and the field's plastic-dish conclusions about gene necessity are, at minimum, unproven until this map exists.
References
- The effect of stiff substrates on the collective migration of A549 cells
- Recent advance in surface modification for regulating cell adhesion and behaviors
- A simple method to isolate structurally and chemically intact brain vascular basement membrane for neural regeneration following traumatic brain injury
- Decellularized Human Kidney Cortex Hydrogels Enhance Kidney Microvascular Endothelial Cell Maturation and Quiescence
- The Cell Tracking Challenge: 10 years of objective benchmarking
- MED1 Deficiency in Macrophages Accelerates Intimal Hyperplasia via ROS Generation and Inflammation
- Nitric oxide promotes epidermal stem cell migration via cGMP-Rho GTPase signalling
- A Benchmark for Epithelial Cell Tracking
- CTMC: Cell Tracking with Mitosis Detection Dataset Challenge
- Membranous type matrix metalloproteinase 16 induces human prostate cancer metastasis
- Chiral Supramolecular Hydrogel Loaded with Dimethyloxalyglycine to Accelerate Chronic Diabetic Wound Healing by Promoting Cell Proliferation and Angiogenesis
- Small Intestinal Submucosa Biomimetic Periosteum Promotes Bone Regeneration
- CELLECT: contrastive embedding learning for large-scale efficient cell tracking