TDAR AOP Network

Scope

The T-cell dependent antibody response (TDAR) is a functional immunotoxicology endpoint. Three AOP-Wiki pathways terminate at it — AOP 154 (calcineurin inhibition), AOP 277 (impaired IL-1R1 signaling) and AOP 315 (JAK3 inhibition) — and a fourth, AOP 14 (glucocorticoid receptor activation), terminates at increased disease susceptibility while sharing intermediate Key Events with them.

This project uses that cluster as a comparator against dismech's own modeling of humoral immunity. It is a two-way comparison, and the two directions are kept separate because they have different audiences and different standards of proof.

This page records findings. It does not decide adoption: whether dismech gains a module, an edge, or a schema construct as a result belongs in its own issue.

The AOP-side observations are not being tracked as dismech work. They are findings about the framework rather than about this repository, and no issue, worklist, or follow-up here depends on them. They are recorded on this page because the comparison produced them, and that is all.

Relationship to AOP_EMOD_ALIGNMENT. That project asks how dismech's constructs map onto the AOP/EMOD data model, grounded on Lead_Poisoning. This one asks what a specific, well-populated AOP cluster reveals when set against a specific, well-populated dismech module. The first is a schema comparison; this is a content comparison. They are expected to inform each other.

Lines of questioning

A. What elements of dismech are exposed through comparison of TDAR AOPs?

Open.

B. How does dismech expose knowledge gaps in TDAR AOPs?

Open.

Further questions under each line are expected to follow once the first are answered.

Provenance

The starting material is an OpenScientist autonomous deep-research run, commissioned to assemble a TDAR-centred AOP network with assay-anchored Key Events and to aggregate evidence for a causal linkage between TDAR and increased susceptibility to infection.

Two companion documents sit beside the report in TDAR_AOP/. Neither renders on the website — only top-level projects/*.md does — so they are read in the repository:

• TDAR_AOP/failed-claims.md — the report's claims that do not survive checking against their own sources, each with the claim in the report's own words, what was checked, and what the source says.
• TDAR_AOP/findings-to-actions.md — a checklist over all twelve finding sections, separating the ones implying work in this repository from the ones that are observations about the AOP framework with no dismech action.

The report is a lead, not a source. Its citations have not been put through just validate-research-references, and the verification below was done by hand against AOP-Wiki and PubMed. The provider bundle also contained iteration transcripts, container logs and a PDF rendering of the same report; those are recoverable from the job and are not committed.

AOP-Wiki snapshot. AOP-Wiki figures below are from the 09-03-2026 export unless a figure names another snapshot: the F010 ratings check and the infection-outcome census were both run against 09-15-2026. Export counts move between dates — 1,602 events in 09-15-2026 against 1,598 in 08-06-2026 — so which snapshot produced a figure is load-bearing for anyone re-running it. Both were read through aop-wiki-cli. AOP-Wiki is not a citable reference in dismech's validation stack — there is no AOP: prefix and no fetcher — so anything curated from it must cite the primary literature instead.

Verification record

What follows was checked directly, and in several places it does not agree with the report.

The TDAR → infection edge is not curated anywhere

KE984 (Impairment, T-cell dependent antibody response, AOPs 154 and 277) and KE1719 (Impairment of T-cell dependent antibody response, AOP 315) are both terminal Adverse Outcomes with zero downstream Key Event Relationships. The link the report was commissioned to establish is not a KER, so it carries no weight-of-evidence or empirical-support table.

This is a consequence of the OECD review of AOP 277. That AOP was originally titled "Impaired IL-1R1 signaling leading to increased susceptibility to infection"; reviewers found "susceptibility to infection" hard to define as a measurable Adverse Outcome, and it was replaced by the TDAR endpoint. Nothing re-attached the infection outcome downstream, so the claim was orphaned rather than strengthened.

What survives in AOP-Wiki is one field, and it is about a different assay

The only quantitative bridge to infection in the cluster is the quantitative_understanding field of KER2928 (Suppression of T cell activation → Impairment, TDAR), 207 characters:

"Luster et al (1993) demonstrated that Concanavalin A response of splenocytes showed the linear dose-response relationship with the host resistnace to Listeria monocytogenes or Streptococcus pneumoniae."

Concanavalin A splenocyte proliferation, not TDAR — one Key Event upstream. The report inherited that substitution rather than introducing it.

The High/High weight-of-evidence ratings are not in the export

The report's F010 presents a table of terminal-KER ratings — Evidence High, Quantitative Understanding High for all three AOPs — introduced as "Extraction of the AOP-Wiki Key Event Relationship (KER) weight-of-evidence tables". No such tables are present.

In the 09-15-2026 export, each KER's four evidence blocks carry free_text, tables and headers. For all three terminal KERs, tables is empty, headers is empty, has_any_tables is false, and no High / Moderate / Low token appears anywhere in the record:

Character counts are measured after HTML stripping; AOP-Wiki free-text fields hold raw HTML, and a direct cache read has to strip it. The three KER records were read from all_kers_09-15-2026.json, materialized by the command in Reproducing it above and keyed by the AOP-Wiki KER IDs shown.

Nor are the ratings on the parent AOP records, whose woe_evidence is narrative prose rather than a ratings table. AOP 277 and AOP 315 contain no rating token at all; AOP 154 contains only Moderate, never High.

So the claim that the terminal KER is rated High/High in all three pathways, and the "exceptional confidence" the report draws from it, cannot be confirmed from the export, and the one rating word that does appear anywhere in the three AOP records is weaker than the claim. KER1510 is the sharpest case: AOP 154 is the OECD-endorsed pathway of the three, and its terminal KER carries no evidence text whatsoever. AOP 315, by contrast, is still Under Development.

This does not prove the ratings do not exist — they may be held in a structured field the XML export does not carry, or be visible on the rendered wiki page. It does mean the report's stated method, extraction from KER weight-of-evidence tables, cannot have produced them from this data.

The primary source does not support the reading placed on it

PMID:8365588 (Luster et al. 1993, Risk assessment in immunotoxicology. II) is the origin of the quantitative claim. Three statements in its own abstract qualify it:

• "No single immune test could be identified which was fully predictive for altered host resistance, although most assays were relatively good indicators (i.e., > 70%)."
• "A good correlation exists between changes in the immune tests and altered host resistance in that there were no instances where host resistance was altered without affecting an immune test(s). However, in some instances immune changes occurred without corresponding changes in host resistance." — this establishes the immune tests as sensitive detectors of immunotoxicity, not as predictors that a suppressed response means more infection. The AOP needs the converse direction.
• The linear model was confirmed for cyclophosphamide only: "For most of the relationships this could not be confirmed using a large chemical data set and, thus, a more mechanistically based approach for modeling will need to be developed."

The frequently quoted "enumeration of lymphocyte populations and quantitation of the T-dependent antibody response were particularly beneficial" describes predicting immunotoxicants, and cites the 1992 companion paper rather than this one. The 1993 abstract names it outright — "(Luster et al., Fundam. Appl. Toxicol., 18, 200-210, 1992)" — immediately before that sentence, so the identification rests on the paper's own citation and not on inference.

One statement in the 1993 abstract cuts the other way, toward the AOP's reading, and is recorded here so the page carries both sides — conclusion (3):

"The ability to resist infectious agent challenge is dependent upon the degrees of immunosuppression and the quantity of infectious agent administered."

That is a dose-dependence claim about immunosuppression and host resistance, and it is the paper's clearest support for a graded relationship. What it does not supply is the part the pathway needs: it names neither TDAR nor any single immune test as the measure of "degrees of immunosuppression", which is what conclusion (2) explicitly declines to do.

The 1992 companion paper does not say TDAR is best, or that it predicts infection

Following that citation to its source — PMID:1534777 (Luster et al. 1992, Risk assessment in immunotoxicology. I. Sensitivity and predictability of immune tests) — settles two things the "particularly beneficial" phrase is routinely used to imply.

TDAR ranked second, not first. The abstract reports:

"The tests that showed the highest association with immunotoxicity were the splenic antibody plaque forming cell response (78%) and cell surface marker analysis (83%)."

The splenic antibody plaque-forming cell response is the TDAR assay. At 78% it is outscored by cell surface marker analysis at 83%. "Among the most predictive" is fair; anything stronger is not.

And the 78% measures a different endpoint. It is the association with immunotoxicity — the ability to detect an immunotoxic compound — not with susceptibility to infection. The same abstract says the host-resistance question was still open at the time:

"Efforts are currently underway using this database to determine the relationships between these immune tests and susceptibility to challenge with infectious agents or transplantable tumor cells."

That work became the 1993 paper above, whose conclusion (2) is that no single immune test was fully predictive of altered host resistance.

So the chain of attribution closes without ever establishing the claim it is cited for. The phrase is real and sits in the 1993 abstract; it points to 1992; and 1992 measures compound detection, ranks TDAR second, and explicitly defers the infection question to the paper that then declines to answer it in TDAR's favour.

Reframed from TDAR to impaired antibody response, the evidence is strong

PMID:20675197 (Orange et al. 2010), a meta-analysis of 17 studies, 676 patients and 2,127 patient-years in primary immunodeficiency with hypogammaglobulinemia:

"Pneumonia incidence declined by 27% with each 100mg/dL increment in trough IgG (incidence rate ratio, 0.726; 95% confidence interval, 0.658-0.801). Pneumonia incidence with maintenance of 500 mg/dL IgG trough levels (0.113 cases per patient-year) was 5-fold that with 1000 mg/dL (0.023 cases per patient-year)."

Human, quantitative, with confidence intervals, and interventional rather than correlational — immunoglobulin replacement moves the exposure, and restoring the antibody level restores resistance to infection in a graded way.

Compare what the TDAR framing rests on: a single rodent database — assembled by the NTP tier-approach screening battery reported in 1988, and analysed for host-resistance relationships in 1993 (PMID:8365588) — whose own abstract states that no single immune test was fully predictive of altered host resistance. The reframed claim is not a stronger version of the TDAR claim. It is a different claim — about a measured serum protein in patients rather than an assay endpoint in mice — and it is the one the evidence actually supports.

The infection outcome is barely represented, as nodes or as edges

This is a claim about the whole of AOP-Wiki, so it is stated with the query that produces it.

Reproducing it

# 1. Pin the CLI. c24ce8c is published on origin/main.
#    Point the data directory anywhere that is NOT a dismech checkout.
export AOP_WIKI_CLI_DATA_DIR=~/aop-wiki-data

# 2. Materialize a dated snapshot. Any KER command parses the full XML and
#    writes all_events_<date>.json and all_kers_<date>.json under
#    $AOP_WIKI_CLI_DATA_DIR/outputs/cache/<date>/. This is the slow step.
uvx --from git+https://github.com/gingin77/aop_wiki_cli@c24ce8c aop-wiki-cli \
  find-kers-for-events --ke-terms "antibody" --date 09-15-2026

3. The filter. Over all_events_<date>.json, select every event whose title matches, case-insensitively, infect|host resist|susceptib|pathogen load|viral (load|titer). For each, record level_of_biological_organization and the character counts of description and measurement_method after HTML stripping. Then over all_kers_<date>.json, select every KER whose downstream_ke.id is one of those events, and record the character counts of the four evidence blocks.

That last step is an ad-hoc read of the CLI's cache, which is what the aop-wiki-cli skill permits while working in that directory. It is deliberately not committed as a dismech script: the CLI owns the parsers and the entity model, and the durable home for this query is a command there rather than parsing code in this repository.

The census

Snapshot 09-15-2026: 1,602 events, 2,373 KERs.

Every Key Event in the entire wiki denoting an infection or host-resistance outcome:

Four events. Two incoming relationships. The evidence on those two:

Separately, KER3702 (AOP 618) is the only downstream edge from AOP-Wiki's one Reduced antibody production Key Event (KE2398). It runs to Diminished vaccine response rather than to infection, and its four evidence blocks are also empty.

What it means

Across the whole immunosuppression space the wiki carries no evidence-bearing edge into an infection outcome, and the outcome nodes themselves are mostly empty: three of the four carry no definition and no statement of how they would be measured. The framework names the outcome and then leaves it unspecified, which is consistent with what the AOP 277 reviewers said when they removed it — that susceptibility to infection was hard to define as a measurable Adverse Outcome. It also explains KE323 having no incoming relationship at all: an undefined node is hard to draw an evidenced edge into.

The relationship is well established in medicine — it is why immunoglobulin replacement is standard care in primary antibody deficiency — and absent from the framework.

KER2498 is the one exception, and it is not a counterexample. It is the only evidence-bearing edge into any infection node in the wiki, and it is richly evidenced. But it runs from SARS-CoV-2 production to onward transmission: its description is about respiratory droplets, aerosols, fecal-oral spread, masks and surface disinfection, and its assay detects virus or antibody in a person. It measures a virus establishing itself and spreading between hosts, not a host losing the capacity to resist one. Note that level_of_biological_organization does not carry this distinction — KE576 and KE1939 are both tagged Individual — so the difference has to be read from the records rather than inferred from that field.

Module classification

The modules in kb/modules/ that touch antibody response, with their terminal nodes and conformer counts. Tiering matters: a module in which antibodies appear as pathogenic effectors says nothing about whether a patient can mount a protective response, and must not be counted as coverage.

Tier 1 — models the antibody response itself

Tier 2 — sibling arms of host defense

Each is declared complementary in germinal_center_reaction's own description.

Tier 3 — antibodies as pathogenic effectors, not as response capacity

Conformance state of the antibody-deficiency entries

At the time of this survey, germinal_center_reaction's description named five entries as intended conformers and three of the six antibody-deficiency entries declared conforms_to. That gap is what issue #11897 was opened against, and it has since been closed — so both states are recorded here, because the survey finding is what motivated the work and the current state is what a reader needs.

All anchors are on germinal_center_reaction; the module stem is elided in the table for width.

So five of six now declare, not three. Selective_IgA_Deficiency remains unconformed on purpose: #11897 recorded it as genuinely uncertain — the defect is isotype-restricted, the entry records intact heavy-chain alpha loci, its genetics are associative rather than a lesion in the chain, and mucosal IgA sits partly in the T-independent territory the module explicitly scopes out. A recorded decision not to conform was named there as an acceptable outcome, and that is the outcome.

The module description was updated in the same work and now names seven entries in the decreased-or-absent direction rather than five.

Also conforming: Activated_PI3K-delta_Syndrome, Kabuki_Syndrome (both in the decreased direction) and Follicular_Lymphoma (dysregulated direction).

The structural observation

phagocyte_oxidative_burst_failure terminates at Recurrent Bacterial and Fungal Infection. germinal_center_reaction terminates one step earlier, at Durable Protective Humoral Immunity, with nothing downstream — the same place AOP-Wiki's KE984 stops.

dismech already carries "increased susceptibility to infection" in its module vocabulary, on the innate arm. It is absent from the humoral one. Both frameworks stop at the same node, independently.

Where the report's chemical branches would go

germinal_center_reaction's notes list, as explicitly out of scope and "to be argued as its own module", iatrogenic B-cell depletion (rituximab, transplant immunosuppression). The report's calcineurin, JAK3, AhR and glucocorticoid branches fall within that reservation. Note that dismech currently holds no drug- or chemical-induced immunosuppression disease entries: cyclosporine, tacrolimus and JAK inhibitors appear throughout kb/disorders/ as treatments, not as toxicity entries. Any such module would begin with no conformers.

Not yet explored

• The remainder of the OpenScientist report. The verification above covers its central causal claim and the network structure. Six of its twelve finding sections have not been checked: the assay mapping (F004), the 2026 B6C3F1/N in vivo dataset (F006, F013), the convergent-literature section citing Burleson, Descotes and White (F007), the 2024 Front. Toxicol. mapping said to confirm the four-AOP network independently (F008), the regulatory-framework and NAM 3R section (F011), and the AhR/PAC branch (F012, F014).
• One checked section was only partly checked. F009's sources (PMID:30068597, Bohrer 2018; PMID:26430088, Cippà 2015) were assessed for what they bear on — both skip the TDAR node — but not verified against their abstracts.
• The report's own five limitations and six proposed follow-up experiments have not been assessed at all.
• Whether the F010 High/High ratings exist anywhere outside the XML export — in a structured field the export does not carry, or on the rendered AOP-Wiki KER pages. The check above establishes only that they are not extractable from the export by the method the report states.
• Only the abstract of the 1992 companion paper (PMID:1534777) was consulted, not its full text.
• AOP 14's bridge into Increased, Disease susceptibility (KE323). Note this is not the same gap as KER570, whose empty evidence blocks are recorded above: KE323 has no incoming relationship at all, so there is no KER there to evidence.