ENIGMA Carbon Census 1

1. The census is mostly dark: 74 of 83 compounds have no isolate-level utilizer call

The governing result is a gap, not a coverage claim. Of 83 compounds (59 SSO-groundwater + 24 necromass), all 83 resolved to structures (InChIKey via PubChem), 54 linked to KEGG, but only 9 are "callable" (review I1: callable = an ENIGMA-isolate utilizer call or a Tier-1 measured RB-TnSeq carbon-source fitness experiment). Eight are callable on an enigma_isolate_call basis (≥1 ENIGMA-isolate prediction with a catabolic, compound-degrading reaction in a genome); a ninth — lauric acid — is callable on a measured_fitness basis: it has a Tier-1 measured carbon-source growth experiment in the Fitness Browser, confirmed structurally by an InChIKey re-match (NB02c). Lauric acid's Fitness Browser organism is a reference bacterium, not an ENIGMA isolate, so the ENIGMA-isolate deliverable (a) is unchanged. The remaining 74/83 (89%) are organism-dark: the genetic determinants of their utilization are not linkable via the queried BERDL and curated resources. This is resource-darkness, not proof of absence from science — class-level catabolic literature exists for several of these compounds (e.g. monoterpenes, nicotine), and the project's own literature-rescue channel (NB02b) returned zero only because it used a shallow PubMed-title screen, which is a method floor rather than evidence of absence (see Limitations). This dark list, stratified by reason and chemical class, is the actionable enrichment-design output — it separates not-linkable-here compounds (candidate discovery targets) from the small characterize-known set, and a literature/PaperBLAST pass would reclassify some of the dark set.

The funnel: 83 compounds → 83 structure-resolved → 54 KEGG-linked → 9 callable; 74 organism-dark.

Organism-dark breakdown by bucket (post-I1, 74 compounds; NB09 Part 4): KEGG-linked/no-reaction-in-genomes 33, fully-orphan (no KEGG) 29, biosynthesis-known/catabolism-unknown 6, only-generic-reactions 6. By class, the dark set is dominated by exactly the predicted classes: Alkaloids, Shikimates/Phenylpropanoids, Terpenoids, and Fatty acids. The dark fraction is nearly identical across sources (groundwater 53/59 = 90%, necromass 21/24 = 88%), so darkness is a chemical-class property, not a sampling-source artifact.

(Notebooks: 01_identity_resolution, 02_pathway_linkage, 02b_linkage_deepening, 02c_fb_inchikey_rematch, 03_organism_mapping, 08_synthesis, 09_deepening)

2. H1 (coverage gradient by chemical class) — not formally supported (underpowered); directional only

The 9 callable compounds are overwhelmingly pollutant-adjacent aromatics: salicylic acid, 3-hydroxybenzoic acid, 4-hydroxybenzaldehyde, phthalic acid, terephthalic acid (all Shikimates/Phenylpropanoids), plus Phenylethylamine and xanthine (Alkaloids), Abscisic acid (Terpenoid, a single 2-strain call), and lauric acid (Fatty acid, on a measured_fitness basis only — review I1). One caveat on this set: xanthine is mis-scored as carbon-catabolic — its allowlisted reaction R02107 (xanthine→urate, xanthine oxidase) is the purine pathway, which is nitrogen acquisition, not carbon catabolism, so the carbon-callable set is effectively 8, not 9 (see Limitations). The aromatic/phthalate concentration is unaffected. The apparent direction matches the prediction — callables cluster in the aromatic/phthalate subset that curated biodegradation knowledge and KEGG cover. But the pre-registered null (uniform coverage across classes) is not rejected: a χ² test on the 8 enigma_isolate_call callables by class (Shikimates 5/25, Alkaloids 2/26, Terpenoids 1/17, Fatty acids 0/10, Polyketides 0/2, AA/Peptides 0/2, mixed 0/1) gives χ²=5.07, p=0.53 (df=6); adding the single measured-fitness lauric acid (Fatty acids 1/10) does not change the verdict. With only ~9 callables across 83 compounds the design is underpowered to detect a class gradient, and the directional pattern is further confounded with annotation coverage (which classes KEGG/ModelSEED/genome_depot annotate) rather than a clean biological degradability gradient. Verdict: H1 not formally supported. The per-class Tier-0 list remains a useful descriptive enrichment-design deliverable, which does not require H1 to hold.

(Notebook: 08_synthesis)

3. H2 (cross-module modularity) — not supported beyond a shared aromatic funnel

Among the 8 enigma_isolate_call capacities across genomes carrying ≥1 capacity, catabolic versatility is right-skewed (mean ~1.19, median 1, max 4 capacities per genome). NB05 first reported co-occurrence with the φ coefficient, but φ compresses toward zero when a capacity is rare (most are present in tens-to-hundreds of 3109 genomes), which can hide a real multiplicative enrichment. Review I4 (NB05b) recomputes two interpretable effect sizes per pair from the saved 2×2 counts — a Haldane-corrected odds ratio + 95% CI and the Jaccard index — with no Spark re-run.

The effect sizes sharpen but do not overturn the verdict. By pair class: within-aromatic median OR 1.12 (3/10 pairs with CI entirely > 1, median Jaccard 0.026); within-other median OR 7.43 (1/3, but driven by Haldane correction on near-zero counts); and the genuine modularity test — cross-block pairs (distinct lower pathways) — median OR 2.28 with 4/15 CI > 1 but median Jaccard ≈ 0. The one biologically real cross-block signal the OR surfaces (and φ hid) is phenylethylamine co-occurring with aromatic-funnel acids: 4-hydroxybenzaldehyde + phenylethylamine OR 2.84 (95% CI 1.49–5.40, q=0.035, n11=11); phthalic acid + phenylethylamine OR 3.37 (95% CI 1.46–7.75, q=0.071, n11=6). This is expected and mechanistically coherent, because phenylethylamine is itself an aromatic-derived substrate (deaminated to phenylacetate, then the paa/phenylacetyl-CoA route), so versatile aromatic degraders tend to carry both. Crucially the Jaccard stays ≈ 0.04 even for these enriched pairs — only ~4% of genomes with either capacity carry both — and only 25 of 3109 genomes (0.8%) carry both an aromatic and a non-aromatic capacity. Verdict: no broad cross-module modularity; the one positive signal is itself aromatic-derived, i.e. co-occurrence within the broad aromatic-catabolism phenotype, not modular co-assembly of mechanistically-distinct modules. What is supported is phylogenetic concentration — Burkholderiales dominate utilizer counts, and high-versatility genera (Polaromonas, Alcaligenes, Burkholderia, Paraburkholderia, Comamonas, Hydrogenophaga) cluster in that order.

(Notebooks: 05_cooccurrence, 05b_cooccurrence_effects)

4. H3 (source-tracking: groundwater vs necromass) — untestable / confounded, reported as such

Of the 8 ENIGMA-isolate-callable compounds, only 2 are necromass-sourced (terephthalic + phthalic acid), and both are phthalate-class aromatics with Actinomycetota-heavy utilizers. (The 9th callable, lauric acid, is also necromass but has no ENIGMA-isolate utilizers and no field data, so it adds nothing to the source contrast.) Source is therefore inseparable from chemical class at n=2 — there is no honest statistical contrast to run. Rather than fabricate a confounded result, NB07 was reframed as a straight SSO field-occurrence atlas (Zhou Lab 16S, enigma_coral bricks): 62 of the implicated utilizer genera are observed in the SSO field at genus resolution, with top field prevalences ~0.7–0.9 for the aromatic utilizers (e.g. 3-hydroxybenzoic-acid utilizers at 0.90 field prevalence).

(Notebook: 07_environmental_atlas)

5. Deliverables (a) + (b): tier-stratified isolate predictions, phylogenetically concentrated

Deliverable (a) — 569 ENIGMA-isolate utilizer prediction rows across the 8 ENIGMA-isolate-callable compounds (lauric acid, callable only on measured fitness, has no ENIGMA-isolate rows). The aromatics carry the bulk: salicylic acid 129 strains, 3-hydroxybenzoic acid 127, 4-hydroxybenzaldehyde 125; terephthalic acid 34; phthalic acid 36; Phenylethylamine 28; xanthine 13; Abscisic acid 2. (The "34 all high-certainty" for terephthalic acid is a single-reaction-signature artifact, not a strength — see below and Limitations.)

Deliverable (b) — 494 strain placements on the GTDB taxonomy (359 distinct strains), each carrying a tier × certainty score: 64 high-certainty (gene-complete pathway, Tier 2), 387 medium (Tier 2 pathway), 43 medium (Tier 3 signature enzyme). Utilizers concentrate in Pseudomonadota / Burkholderiales, with Pseudomonadales and Sphingomonadales contributing to specific compounds (e.g. Sphingomonadales 47 strains for 4-hydroxybenzaldehyde).

H4 verdict (tier-stratified, phylogenetically concentrated isolate predictions per compound): partially supported — strongly for the 8 ENIGMA-isolate-callable compounds, null for the other 75. The 8 callable-with-isolate compounds each yield a tier-stratified prediction set that concentrates phylogenetically (Pseudomonadota/Burkholderiales, with compound-specific Sphingomonadales/Pseudomonadales contributions), exactly as H4 predicted. The remaining 75 are Tier 0 with no isolate-level placement, so H4 is null there. (Lauric acid, the 9th callable, qualifies only on measured RB-TnSeq fitness in a reference bacterium and contributes no ENIGMA-isolate phylogenetic prediction, so it sits outside both arms.)

(Notebooks: 04_enigma_utilizers, 06_phylo_maps)

6. Deliverable (c), global arm: a cross-environment abundance atlas for the 86 implicated genera

Beyond the local SSO field atlas, the implicated utilizer genera were placed across global environments — terrestrial/freshwater (NMDC, kbase.nmdc_arkin) and marine (Planet Microbe, as a contrast). This is a biome-abundance proxy, not catabolic activity: no environmental dataset measures the census compounds, so genus abundance indicates where the organisms live, not where the compounds are degraded.

NMDC scale: 3825 taxonomy-bearing metagenomes (denominator = covstats files with taxonomy), 83/86 genera detected in 1719 metagenomes, 99% sample-labeled via two independent ontologies (ENVO MIxS triad + GOLD ecosystem path). Macro-environment distribution: soil 2260, freshwater 723, periphyton 472, plant 119, sediment 42.

Soil-vs-freshwater abundance contrast (exploratory — see Limitations) recovers textbook biogeography: soil-enriched genera are classic soil taxa (Mycobacterium log2 +5.09, Terriglobus +5.29, Nitrobacter +4.79, Afipia, Streptomyces, Nocardioides, Mesorhizobium), and freshwater-enriched genera are aquatic Betaproteobacteria (Cellvibrio log2 −2.19, Curvibacter, Rhodoferax, Comamonas, Acidovorax). The newly surfaced periphyton class (freshwater biofilms: epilithon/epipsammon/epiphyton) is a strong reservoir — Burkholderiales/Comamonadaceae (Rhizobacter, Variovorax, Polaromonas, Methylibium, Sphingomonas, Hydrogenophaga) at ~96–97% prevalence and mean relative abundance ~0.005–0.009.

Label-free outlier discovery (the most defensible signal, robust to label noise): top genus×sample relative-abundance spikes occur in periphyton (Nocardioides 0.43 in epipsammon, Hydrogenophaga 0.28 in epiphyton) and soil (Mycobacterium 0.22). Outliers peak in periphyton (34 genera) and soil (32).

Marine (Planet Microbe, 302 runs): 68/68 listed genera show positive abundance, but marine is mostly a negative biome contrast — terrestrial/freshwater genera sit at ~1e-3 to 1e-4 in open ocean. Genuine marine members are the exception (Alteromonas 0.048 in 240/302 runs, prevalence 0.79; cosmopolitan Pseudomonas 0.011, Sphingomonas 0.0063).

(Notebook: 07b_environmental_atlas_global)

7. Bioavailability/physicochemistry is a ceiling on the atlas: callable compounds are smaller and simpler (directional, underpowered)

NB09 Part 1 resolves PubChem physicochemical descriptors for all 83 compounds and contrasts callable vs dark. With only n=9 callable this is directional, not inferential (Mann–Whitney, reported as direction + uncorrected p). Callable compounds are consistently smaller and structurally simpler than dark ones: median Complexity 133 vs 207 (p=0.034), MolecularWeight 152 vs 179 (p=0.066), HeavyAtomCount 11 vs 13 (p=0.057), with directionally higher polarity (TPSA 57 vs 41, H-bond donors 2 vs 1). The honest reading is that this is at least as much an annotation-coverage ceiling as a biological-bioavailability one — simple, common, pollutant-adjacent aromatics are exactly the molecules KEGG/ModelSEED/genome_depot annotate, so "callable" tracks structural simplicity partly because the knowledge base does. Notably there is no physicochemical separation between groundwater and necromass compounds (all contrasts p>0.14), reinforcing that source does not predict chemistry here.

(Notebook: 09_deepening)

8. The callable phenotype is a specialist clade trait, not a portable generalist module set

NB09 Part 2 transposes the deliverable to genomes. Of genomes carrying ≥1 of the 8 aromatic/alkaloid capacities, 675 are specialists (exactly 1 capacity) and only 18 are generalists (≥3; max 4). Generalist chassis concentrate in Paraburkholderia (5), Burkholderia (4), Hydrogenophaga (2) and scattered Burkholderiaceae — the same clade that dominates the versatility ranking, confirming versatility is a clade trait, not a widely portable module. All callable evidence is catabolic-direction by construction of the NB03 filter (846 Tier-2 pathway + 102 Tier-3 signature genome-compound calls; biosynthetic-only signatures are excluded and land in the dark set). Part 3 (per-clade conservation) shows compound-specific clade specialization rather than a shared generalist toolkit. These conservation fractions are denominated over depot genomes that already carry some catabolic call — not a census of all genomes of each genus — so they describe within-called-set specialization, not absolute prevalence: e.g. Castellaniella and Hylemonella carry 3-hydroxybenzoic-acid capacity in ~100% of such genomes, Sphingomonads (Novosphingobium/Sphingobium) carry 4-hydroxybenzaldehyde capacity at ~92–100%, Paenarthrobacter carries phthalic-acid capacity at 100%, and Comamonas spans both 3-hydroxybenzoate (92%) and 4-hydroxybenzaldehyde (85%). Capacity breadth across the chassis is led by 3-hydroxybenzoic acid (296 genomes, 53 genera) and salicylic acid (197 genomes, 28 genera); Abscisic acid is a 2-genome edge case.

(Notebook: 09_deepening)

9. The dark set has structure: 6 biosynthesis-known compounds are MIBiG-consult targets, not true orphans

NB09 Part 4 taxonomizes the 74 organism-dark compounds (post-I1) into four buckets: 33 KEGG-linked but no reaction in any queried genome, 29 fully orphan (no KEGG link at all), 6 biosynthesis-known/catabolism-unknown, and 6 only-generic-reactions. The 6 biosynthesis-known compounds — Tyramine, guanidineacetic acid, cinnamic acid, caffeic acid, palmitic acid, farnesol — have annotated biosynthetic-direction signatures but no catabolic call, which is exactly the profile to triage against MIBiG/biosynthetic literature before committing to discovery enrichment (flagged as an external consult; MIBiG was not queried in BERDL). The 29 fully-orphan compounds (no KEGG link) are the hardest discovery targets and skew necromass-heavy (13/24 necromass vs 16/59 groundwater are orphan). This per-bucket structure is the operational refinement of the headline gap: it ranks the 74 dark compounds by why they are dark and therefore by how to attack them.

(Notebook: 09_deepening)