2026-10-20 · Abridged edition · r1
Background. Multi-component botanicals occupy a durable but structurally awkward position in modern therapeutics and nutrition. The U.S. Food and Drug Administration's Botanical Drug Development guidance (2016 revision) explicitly permits multi-component botanicals as drug substances, yet as reflected in the FDA Center for Drug Evaluation and Research list of botanical drug products (accessed 2026-10-20) only four such products have cleared the pathway — sinecatechins (Veregen, 2006, NDA), crofelemer (Fulyzaq/Mytesi, 2012, NDA), birch triterpenes (Filsuvez, 2023, NDA), and anacaulase-bcdb (NexoBrid, 2022, BLA) — with Chemistry, Manufacturing and Controls (CMC) reproducibility of the drug substance across lots documented as a persistent challenge by FDA CDER authors in a peer-reviewed review of the botanical-drug regulatory experience [7], consistent with the extended CMC and botanical raw material characterization sections of the FDA 2016 Guidance [1]. Finished dietary supplements under the Dietary Supplement Health and Education Act (DSHEA, 1994) face a parallel problem on the structure-function side: mechanism anchoring for a complex botanical mixture is non-trivial, and the New Dietary Ingredient (NDI) notification lane requires defensible identity and safety documentation.
Approach (architectural posture only). We describe an architectural platform that pairs food-grade enzyme and characterized botanical substrates under pinned fermentation conditions. Every output carries a persistent pending-wet-lab invariant and an explicit validation-state badge; novel-composition territory and the engine's internal ranker remain NDA-gated. This abridged edition presents the regulatory framing and the terminology discipline that constrain the architecture.
Scope of this abridged edition. The methods, machine-readable dossier schema, illustrative Hypothesis-tier records, provenance-check tooling, bundle manifest, and the auditability-boundary section are deliberately out of scope here. The complete technical preprint is available on request to qualified reviewers under mutual NDA.
Conclusions. Directed fermentation with pinned conditions, mechanism anchoring, and a persistent validation-state badge offers a tractable path to reproducible botanical drug substance and defensible structure-function framing. All computational outputs are explicitly research intelligence, pending wet-lab co-validation; no disease, efficacy, or treatment claim is advanced here.
Multi-component botanicals sit at an uncomfortable regulatory and scientific intersection. On the drug side, the U.S. Food and Drug Administration's Botanical Drug Development: Guidance for Industry [1] formally recognizes multi-component botanical mixtures as acceptable drug substances under an Investigational New Drug (IND) application, provided Chemistry, Manufacturing and Controls (CMC) data demonstrate acceptable lot-to-lot consistency. The pathway has been open since 2004 (guidance revised 2016). As reflected in the FDA Center for Drug Evaluation and Research list of botanical drug products (accessed 2026-10-20) [2], four botanical drug products have cleared the pathway: sinecatechins (Veregen, approved 2006) for external genital and perianal warts, under an NDA; crofelemer (Fulyzaq, later Mytesi, approved 2012) for HIV-associated non-infectious diarrhea, under an NDA; birch triterpenes (Filsuvez, approved 2023) for the treatment of partial-thickness wounds in epidermolysis bullosa, under an NDA; and anacaulase-bcdb (NexoBrid, approved 2022) for eschar removal in deep-partial-thickness and full-thickness burns, under a BLA. One of several factors that have been discussed in the context of botanical drug development — including by FDA CDER authors in a peer-reviewed review of the botanical-drug regulatory experience [7] — is Chemistry, Manufacturing and Controls reproducibility of the drug substance across lots, which the FDA 2016 Botanical Drug Development Guidance [1] addresses explicitly (CMC and botanical raw material characterization sections); biology is not asserted here to be solved, only that the CMC-reproducibility axis is one of several axes a sponsor must address.
On the dietary-supplement side, DSHEA (Public Law 103-417, 1994) [3] permits structure-function claims for finished supplements and requires an NDI notification for any dietary ingredient not marketed in the United States before 15 October 1994. The DSHEA framework imposes several distinct requirements that are easy to collapse and that this manuscript keeps separate: (i) structure-function claims require the manufacturer to possess substantiation that the claim is truthful and not misleading before the claim is made; (ii) FDA does not pre-approve dietary supplement structure-function claims (the manufacturer notifies FDA within 30 days of first marketing a claim, and FDA may object post-hoc); (iii) the label must carry the mandatory FDA disclaimer “This statement has not been evaluated by the Food and Drug Administration. This product is not intended to diagnose, treat, cure, or prevent any disease.”; and (iv) a New Dietary Ingredient notification must be submitted to FDA at least 75 days before the ingredient is introduced into interstate commerce. Mechanism anchoring at the molecular level is non-trivial for mixtures where dozens of bioactive constituents coexist, each with its own ADME profile and its own enzyme-substrate chemistry during fermentation or digestion.
This preprint describes the architectural posture of a platform that attempts to close the reproducibility-and-mechanism-anchoring gap at the deliverable level. We make no therapeutic claim and no efficacy claim; the artifact is a mechanism-hypothesis indication-discovery dossier, labeled hypothesis-tier where appropriate and persistently flagged as pending wet-lab. The engine's internals, scoring weights, and novel-composition structures are confidential; what is disclosed is the shape of the pipeline, the shape of the dossier, and the regulatory posture that disciplines both. This revision is deliberately scoped as an architectural methodology paper: it does not advance an engine-benchmark claim and does not describe the engine's ranker beyond the public behavioral surface disclosed in this preprint.
Earlier passes of this manuscript risked treating five distinct terms (GRAS, QPS, food-grade, biosafety, and safety of the fermentation output) as near-synonymous proxies for “safe.” They are not. Each term has a distinct issuer, scope, and legal/analytical meaning, and conflating them produces claims the underlying evidence does not support. This section fixes the vocabulary used throughout the paper and specifies the replacement language.
| Term | Issuer / framework | Attaches to | What it actually says | What it does not say |
|---|---|---|---|---|
| GRAS | US FDA (21 CFR 170.30; 21 U.S.C. 321(s)) | A specific substance for a specified food use | Qualified experts generally recognize the substance as safe under the intended conditions of use, based on scientific procedures or (pre-1958) common use in food. | Not a global “safe” label. Not a claim about any new substance the organism may produce. Not an EU authorisation. Not a toxicological pass for an unrelated fermentation output. |
| QPS | EFSA (Qualified Presumption of Safety list) | A taxonomic unit used in the food/feed chain, with explicit qualifications | EFSA presumes the organism poses no safety concern for the specified use, provided named qualifications are met (absence of acquired AMR determinants, defined strain identity). | Not a safety determination for a product the organism synthesises or releases. Not transferable across use categories. Not equivalent to US GRAS. Does not substitute for strain-level AMR / virulence screening. |
| Food-grade | Industry / supply-chain convention (not a regulatory term) | An ingredient, enzyme preparation, or process input | The material is produced, graded, and supplied for use in food manufacturing, typically under food-contact and purity specifications. | Not a safety determination of any end-product mixture. “Food-grade enzyme” does not mean the enzyme’s catalytic products are themselves safe for consumption. |
| Biosafety (BSL-1 / Risk Group 1) | WHO LBM / NIH / national biosafety authorities | Handling of the live organism in laboratory or production settings | The organism is not known to cause disease in healthy adult humans under standard laboratory handling. | Not a consumer-exposure safety claim. Not an output-safety claim. Not a dose/route determination for ingestion. |
| Safety of the fermentation output | Established per product via analytical + toxicological evidence | The end-product mixture (post-fermentation broth, isolate, or formulated ingredient) | Determined de novo by: analytical identity (LC-MS, NMR), contamination screens (mycotoxins, heavy metals, residual solvents, endotoxin, viable-cell count), and toxicological profiling (in vitro + in vivo as scope requires). | Cannot be inherited from the organism’s GRAS or QPS status, nor from the food-grade status of inputs. Must be established for each biotic x substrate x condition tuple. |
A scan of earlier drafts identified the following recurrent misuses, each of which is corrected in this revision: (i) “GRAS organism, therefore the output is safe” is replaced with “the organism holds a GRAS determination for its specified food use; this is a precondition for consideration and does not establish safety of the fermentation output”; (ii) “QPS-listed, so no further safety assessment is required” is replaced with “the species is QPS-listed by EFSA under the stated qualifications; QPS attaches to the organism, the fermentation output requires its own analytical and toxicological basis”; (iii) “food-grade enzyme” used as a safety claim is replaced with a supply-chain-grade description that does not imply safety of catalytic products; (iv) “BSL-1 biosafety” used as a consumer-safety proxy is replaced with an occupational-handling-only framing; and (v) interchangeable use of “safe,” “generally safe,” “regulatorily cleared,” and “non-toxic” for predicted products is uniformly downgraded to “pending wet-lab analytical identity and toxicological profiling,” consistent with the manuscript’s computational-only posture.
Standing statement adopted throughout this manuscript. Organism-level GRAS or QPS status is a precondition for consideration but is not an output-level safety claim. Safety of the fermentation output requires analytical identity (LC-MS / NMR), contamination screens (mycotoxin, heavy metals, residual solvents, endotoxin), and toxicological profiling, all of which are pending wet-lab. “Food-grade” describes an input’s supply-chain grade, not a safety determination of the end-product mixture. “Biosafety” (BSL-1 / Risk Group 1) describes laboratory handling of the live organism, not consumer exposure to its output. Each of the five terms — GRAS, QPS, food-grade, biosafety, and output safety — is used in its narrow technical sense only; they are not substituted for one another.
This paper describes an architectural alignment between the directed-fermentation surface and the control vocabulary used by several regulatory frameworks (DSHEA for dietary supplements, the FDA Botanical Drug Guidance, GRAS self-affirmation, and the NDI notification pathway). Architectural alignment is not qualification: each pathway imposes its own identity, safety, and manufacturing-control basis — botanical raw material characterization, lot-to-lot stability, CMC, and (for drug and NDI tracks) non-clinical safety and human-safety evidence — none of which this manuscript asserts or substitutes for. Claims here are limited to design-time structure (pinned-condition pairing, bioprocess method, and the dossier-shape posture) and are explicitly pre-qualification. International frameworks (EFSA novel food, Health Canada NHP, PMDA foods-with-function-claims, TGA listed/complementary medicines) impose analogous but non-identical burdens and are out of scope here.
Current FDA botanical drug approvals (FDA Center for Drug Evaluation and Research list of botanical drug products, accessed 2026-10-20): four products [2]. The three NDAs are sinecatechins (Veregen, 2006), crofelemer (Fulyzaq/Mytesi, 2012), and birch triterpenes (Filsuvez, 2023). The one BLA is anacaulase-bcdb (NexoBrid, 2022). Over-the-counter botanical actives covered by monographs (e.g. psyllium, witch hazel) are a separate regulatory lane and are not included in the four-product count.
This paper makes no disease claim, no efficacy claim, and no treatment claim. All outputs are research intelligence, computationally predicted or in-silico hypothesis, pending wet-lab co-validation. Nothing in this preprint should be read as medical, legal, regulatory, or freedom-to-operate advice.
The architectural posture described here changes several things for three distinct partner profiles.
For botanical drug co-development partners, reproducibility of the drug substance across lots — discussed in the botanical-drug regulatory literature as one of several factors in the development pathway — is proposed to be addressed at the deliverable level, not merely at the policy level. Pinned fermentation conditions, publicly sourceable strains with repository catalog numbers, and recreation protocols at competent-lab granularity are the preconditions for the CMC consistency package [1] that reviewers have historically found lacking. The mechanism-hypothesis dossier does not replace an IND submission; it supplies the structural ingredients that an IND submission needs.
For formulators and ingredient-portfolio partners, mechanism anchoring under DSHEA [3] structure-function framing is the gap most commonly cited by partners evaluating novel bioactive fermentation outputs. A mechanism-hypothesis dossier that pairs a reproducible biotic with a known-pathway mechanism rationale may contribute to a DSHEA structure-function claim substantiation file, subject to manufacturer evaluation and independent corroboration, and subject to the standard limitation that the structure-function narrative is not an efficacy claim and must carry the FDA disclaimer.
For food-grade enzyme × substrate ingredient tracks, this B2B track supplies pinned-condition pairings with a shared control set, allowing downstream partners to assess ingredient fit without bespoke per-candidate negotiation.
Platform direction. The immediate next step is wet-lab co-validation of selected hypothesis-tier rows under a research-partnership arrangement with a competent fermentation and analytical-chemistry wet-lab partner; the deliverable would be the validation-tier promotion from Hypothesis to In-vitro, carrying the badge and the raw-data appendix. Further out, orthogonal mechanism readouts (analytical identity under LC-MS and NMR, enzyme-kinetic parameters, biotic-level reproducibility replicates) are the natural extensions of the architecture.
Pending wet-lab is universal. Every row in the current public catalog carries the pending_wet_lab invariant. Computational prediction, however well grounded in known enzyme-substrate chemistry, is a hypothesis, not a validated effect. The validation-tier badge on each row is the honest record of that state.
Current catalog is Hypothesis-tier. No row in the current public catalog carries an In-vitro or Lead-tier badge. Promotion requires documented wet-lab co-validation under a documented protocol.
Mechanism anchoring is necessary but not sufficient. A credible known-pathway rationale narrows the hypothesis space, but it does not substitute for empirical demonstration of the predicted chemistry under the pinned conditions. Abstention is also a first-class output of the engine: pairs outside the engine's coverage are refused rather than scored.
Architectural scope only. This revision is deliberately scoped as a methodology paper describing the architecture of mechanism-anchored biotic indication-discovery dossier generation. It does not report an engine-benchmark result, does not advance an engine performance claim, and does not describe the engine's ranker beyond the public behavioral surface disclosed in this preprint. Benchmark reporting against an engine-appropriate held-out task is a separate deliverable scheduled as follow-up and will appear in a subsequent revision only when the task, the metric, and the engine surface are matched in design.
No disease, efficacy, or treatment claim. Nothing in this preprint, in the public catalog, or in any dossier output is a claim that any product diagnoses, treats, cures, prevents, or mitigates any disease. The posture is DSHEA structure-function plus mechanism anchoring for the supplement lane, and reproducible drug-substance preparation for the Botanical Drug Guidance lane, with no therapeutic claim beyond what each regulatory lane itself permits.
Scope of disclosure. The engine's internals, ranker weights, novel composition-of-matter structures, and per-cluster details are confidential and are not disclosed in this preprint. Qualified reviewers under a non-disclosure agreement may access the NDA-gated layer.
Honesty about missing work is part of the posture. The following items are named deliberately as not-yet-present in this revision; each is scheduled as follow-up and will be reported in a later revision only when it has actually run:
Each item above is a known gap, not a surprise. The paper's standing posture — architectural methodology for compositional claims — is unchanged by their absence; the posture is precisely what makes the absence legible.