Spatially resolved in vitro tissue models: a curation pass and a schema gap (2026-09-02)

Date: 2026-09-02 Source: Brenden CK, Williams JT, Rivas D, Forro C, Choi S. Spatiotemporal instrumentation of in vitro tissues: Toward understanding human disease. Cell Biomaterials (2026). DOI:10.1016/j.celbio.2026.100576 Scope: What this perspective's argument and citation list yield for dismech's experimental_models section. Purpose: Record one curation tranche against Myocardial_Infarction, and the schema limitation the exercise exposed.

The source is a lead-generator, not a citable reference

The perspective is in press and not indexed in PubMed (title search returns zero hits), so it has no PMID. just fetch-reference doi:10.1016/j.celbio.2026.100576 succeeds but returns content_type: unavailable — a header-only cache entry with no body. Nothing in it can be quoted as a validated snippet, and no evidence item in this pass cites it.

That is the correct outcome rather than a problem to work around. It is a perspective: its value here is its citation list and its argument, both of which point at primary experimental papers that are indexed and do have abstracts. Those are what got curated. Where the perspective's framing is used in the KB it appears in a notes: field as provenance, never as evidence.

What the perspective argues

Two claims are load-bearing for dismech.

A spatially uniform disease stimulus is a different experiment from a spatially graded one. The paper's recurring example is myocardial infarction: uniform hypoxia shifts a whole construct into a disease state, but the actual lesion is an interface between injured and viable tissue. Models that engineer a gradient reproduce phenotypes that uniform models do not. This is an empirical claim, and the primary papers below test it directly.

Endpoint assays lose the phenomenon. Cytokine dynamics, action-potential morphology and contractile mechanics change on timescales that fixed-timepoint immunoassays cannot resolve, which motivates the instrumented-tissue programme the paper is arguing for.

Curation output

Five models added as an experimental_models block on kb/disorders/Myocardial_Infarction.yaml, which previously had none. All were selected because pathology is spatially graded within the construct, which is the property the perspective is about.

PMID	Model	Type	Links → node	Fidelity
32284552	Cardiac infarct organoid with an internal oxygen-diffusion gradient	ORGANOID	Myocardial Ischemia and Cardiomyocyte Death (RECAPITULATES); Cardiac Fibrosis and Ventricular Remodeling (PARTIALLY_RECAPITULATES)	MODERATE
36475790	Myocardial infarct border-zone-on-a-chip	ORGAN_ON_CHIP	Myocardial Ischemia and Cardiomyocyte Death (PARTIALLY_RECAPITULATES); Post-Infarction Inflammation and Cardiac Repair (PARTIALLY_RECAPITULATES)	MODERATE / LOW
33055246	Human heart-on-a-chip IRI assay with endothelial EV rescue	ORGAN_ON_CHIP	Ischemia-Reperfusion Injury (RECAPITULATES)	MODERATE
32092276	Bioelectronically instrumented heart-on-a-chip under acute hypoxia	ORGAN_ON_CHIP	Myocardial Ischemia and Cardiomyocyte Death (PARTIALLY_RECAPITULATES)	LOW
38683053	Epicardial-myocardial Biowire II heart-on-a-chip	ORGAN_ON_CHIP	Ischemia-Reperfusion Injury (PARTIALLY_RECAPITULATES); Cardiac Fibrosis and Ventricular Remodeling (PARTIALLY_RECAPITULATES)	MODERATE

Eight modeled_mechanisms links against four of the entry's five existing pathophysiology nodes, with 11 evidence-backed readouts. Every snippet was checked as an exact substring of its cached reference before being written, not after. All eight links carry model_scale and typed divergences (see below).

Two gradings are worth flagging because they are the ones a reviewer would challenge:

PMID:32092276 is graded LOW fidelity and kept anyway. Its cells are HL-1, an immortalized murine atrial line — wrong species and wrong chamber for human ventricular myocardium. It earns its place because the instrumentation is the contribution: multiplexed extra- and intracellular electrodes give a continuous time course through the ischemic insult, showing tachycardia resolving into bradycardia and then arrhythmia. No other model in the entry produces that. The limitations field says so explicitly, and a notes: field records why the record was retained despite the cell source.

PMID:36475790's inflammation link is graded LOW, not MODERATE. The chip raises inflammatory-cascade expression under a gradient with no leukocytes present, which is a real and interesting result — but Post-Infarction Inflammation and Cardiac Repair is a node about neutrophil and monocyte recruitment, efferocytosis, and the inflammatory-to-reparative transition. The model reaches the cardiac-tissue half of that and none of the cellular half.

Not curated, and why

Candidate	Disposition
Basara et al. 2024, 3D bioprinted aged post-infarct myocardium (perspective ref 29)	Plausible sixth MI model. Identified from the reference list, not assessed this pass — not fetched, not read.
Yamasaki et al. 2021, bioengineered cardiac tissue in hypoxia/re-oxygenation (ref 20)	Same: identified, not assessed.
Lin et al. 2023, stretchable nanoelectronics in cardiac microtissues (ref 26)	Not a disease model. It studies electrical maturation of healthy microtissue, so it has no pathophysiology node to link to.
Kim et al. 2023, iPSC-CM arrhythmogenic cardiomyopathy (ref 39)	Relevant to Arrhythmogenic_Right_Ventricular_Cardiomyopathy, not to MI. Out of scope for this entry.
Kim et al. 2016, gut-on-a-chip inflammation (ref 22)	Belongs to the IBD/barrier work already tracked in projects/NAMO_RD_MODELS.md.
Yang et al. 2024, Kirigami electronics for neural organoids (ref 84)	Instrumentation method, no disease claim.

The unassessed rows are recorded as unassessed deliberately. Reading a title in a reference list is not evidence about a paper, and the distinction matters more here than usual because the whole point of the exercise was evidence discipline.

What the schema does and does not carry

Correction to an earlier draft of this report. It claimed that an exhaustive search of the schema found no slot for the measurement context these models need. That was wrong by the time it was written. model_scale and divergences had already landed on ModelMechanismLink, between this branch's original base and the main it was rebased onto, and the survey behind the claim was run against the older base and never re-run. Both slots are now populated on all eight links in this pass. What follows is the corrected position.

What the new slots do cover

model_scale (BiologicalScaleEnum) records the scale a model actually observes, so a scale gap against the target node's biological_scale is computable rather than buried in prose. divergences types the caveat that limitations writes as prose, each entry naming a kind, explaining why it applies here, and grading whether it bears on this link's claim.

Three of the four problems the first draft listed are answered by divergences:

Claim	Now carried as
HL-1 is a murine atrial line, not human ventricular myocardium	SPECIES_MISMATCH, INVALIDATING
The border-zone chip's inflammation link has no leukocytes at all	BOUNDARY_OMISSION, INVALIDATING
Its inflammatory readout is transcript abundance standing in for a cellular response	PROXY_QUANTITY, QUALIFYING
Ischemia is imposed by diffusion limit or gas control, not by coronary occlusion	CAUSE_UNREPRESENTED, QUALIFYING

That last one is worth flagging: every model in this entry is CAUSE_UNREPRESENTED against a disease whose defining lesion is atherothrombotic, and the taxonomy makes that queryable across the KB rather than leaving it as a sentence eight times over. All eight links now carry the tag, each with its own description of how the ischemia is imposed in that model — a diffusion limit in the organoid, device gas control on the two gradient chips, a medium-switch protocol on the two ischemia-reperfusion constructs. An earlier revision made this claim in prose while tagging only two links, so querying the entry returned two. That is the same defect as a stale notes: sentence: prose asserting a queryability the data did not provide.

What still has nowhere to go

Both new slots describe how a model falls short. Neither records positive metadata about how a measurement was made, which is the axis the perspective argues about. Three things from this pass still survive only as prose:

1. An internal spatial comparator. The Richards calcium readout compares interior to edge cardiomyocytes within the same organoid. model_scale says the model observes cellular state; it cannot say the control is a different region of the same construct, which is the methodological point. STRUCTURAL_IDEALIZATION is the nearest divergence type and is the wrong shape — the spatial structure here is the model's strength, not its idealization.
2. Temporal resolution. The two-photon light-sheet calcium imaging runs at 20 ms, which is what makes the arrhythmia visible. It sits in free-text culture_system next to the culture format. TEMPORAL_SCOPE again types a mismatch, not a sampling rate.
3. An unnamed comparator. ModelReadoutDirectionEnum is defined "relative to the model's control or comparator arm", yet no slot names that arm. The Bannerman cell-death readout is DECREASED against epicardium-free tissue, not against untreated tissue; read without the prose, DECREASED is close to meaningless. This is the smallest and most tractable of the three, is not specific to spatial models, and every one of the KB's readouts inherits it.

The fourth item in the first draft — a non-monotonic beat-rate time course recorded as ALTERED — is a real loss of information but is adequately handled by ALTERED plus interpretation, and is withdrawn as a schema concern.

Two observations, one of them corrected

• assays is populated on 284 readouts and ontology-bound on none of them. An earlier revision of this report said the slot was "populated on 0 of 289 readouts" and "universally unused". That was wrong. 284 of the KB's 3,091 readouts carry assays — 514 entries across 82 files. What is at zero is the binding: every one of those 514 entries is a bare preferred_term with no term:. OBI is absent from conf/oak_config.yaml and has no cache/enums/ membership cache, so a curator who tried to bind one could not validate it. The tooling-gap conclusion survives, and is in fact stronger than the original framing: curators did reach for this slot and were stopped at the binding, which is better evidence for closing the gap than disuse would have been.
• The expressive slots that already exist are under-populated, which is worth weighing before adding more: just model-scale-audit reports 2,122 of 2,186 model-to-mechanism links as scale-UNDETERMINED, and many links carry no relationship at all. These are snapshot figures in a dated report and will drift — re-run the recipe rather than quoting them. The decision register deliberately carries no numbers for the same reason.

A follow-on this pass deliberately did not take

The eight links here set model_scale, but all eight stay UNDETERMINED in just model-scale-audit because no Myocardial_Infarction pathophysiology node carries biological_scale. Populating those five nodes would make the comparison computable and is probably right, but it edits nodes this PR did not author and that main revised independently, so it belongs in its own change.

Notes for future scans

• Re-run a schema survey against the base you actually ship on. The gap section of this report was written from a survey run against the branch's original base, and by the time it was committed main had added model_scale and divergences to the very class it claimed had no such slot. The rebase brought the new schema in and silently invalidated the prose. A survey is a claim about repository state and rots exactly like a notes: sentence does.
• Fetch the reference before believing a perspective's characterization of it. The perspective describes PMID:32092276 among cardiac instrumentation advances without foregrounding that its cells are murine; that only surfaced on reading the abstract, and it changed the fidelity grade from MODERATE to LOW.
• just fetch-reference on a DOI succeeds with an empty body. Check content_type: in the cache header before planning to quote a source — unavailable means no snippet from it will ever validate.
• The richest existing pattern to copy for an instrumented in vitro model is kb/disorders/High_Altitude_Pulmonary_Edema.yaml (lung-on-chip with fidelity, limitations, and directional readouts). It also independently hit the gap above, carrying "spatially uneven … overperfusion" in free-text limitations.