Translating Phage Host-Range Evidence into Patient-Specific Therapy

Metagenome-Prioritized Phage Cocktails for Crohn’s Disease and IBD develops a translation framework in which phage host-range evidence is combined with patient-specific target presence, ecological state, strain coverage, and non-phage alternatives rather than treated as a sufficient proxy for clinical efficacy. ₁

Translation principle

Patient-specific phage therapy requires at least four linked decisions: whether a target is present in the patient, whether a lytic phage can infect the relevant strain, whether the target is mechanistically and ecologically worth removing, and whether the proposed intervention has evidence beyond in-vitro susceptibility. ₁

The report therefore distinguishes target prioritization from therapeutic tractability: the scoring returned 6 actionable candidates of 71 scored, and those six species reached the actionable Tier-A threshold, but their phage evidence ranged from clinical-trial-stage evidence for Escherichia coli (the EcoActive precedent) to lytic-literature evidence for Eggerthella lenta (PMBT5) and Enterocloster bolteae (PMBT24), temperate-only evidence for Mediterraneibacter gnavus (6 phages), and coverage gaps for Hungatella hathewayi and Flavonifractor plautii. ₁

This distinction supports a hybrid-cocktail model in which direct phage targeting is used where lytic host-range evidence exists, monitored or engineered approaches are considered where evidence is limited, and non-phage interventions are retained for targets without a usable phage. ₁

The targets are also not independent: 5 of the 6 actionable Tier-A species form a single pathobiont module per subnet, which the report reads as making multi-target cocktails appropriate rather than single-species targeting, and which connects this page to species-cooccurrence-modules-as-intervention-units. ₁

The report’s biological reading of that module is interpretive rather than measured: Tier-A pathobionts are said to form a single ecologically-linked co-occurrence module within CD ecotypes, to co-favour similar conditions (the report proposes bile-acid dysregulation plus low-oxygen inflammation), and the ecological coupling is read as suggesting that a cocktail hitting 3+ of {M. gnavus, E. lenta, E. bolteae, H. hathewayi, E. coli} will have compounding effects. That is a hypothesis about an untested intervention, not a measured treatment outcome. ₁

The report’s closing thesis states the framework as a whole: Crohn’s disease at the gut-microbiome level is a single principal-direction phenomenon (CC1 r=0.96) within which 6 actionable Tier-A pathobionts and 2 cross-corroborated mechanism narratives — iron acquisition and bile-acid 7α-dehydroxylation — define a state-dependent, hybrid-cocktail design framework with concrete per-patient drafts for 14 of 23 UC Davis Crohn’s disease patients. That is an interpretive synthesis across the project’s pillars rather than a single measurement, and its multi-omics component is argued in multi-omics-integration. ₁

The framework is handed off as three deliverables: a per-patient master table for the 23 UC Davis Crohn’s disease patients combining all per-pillar attributes, design category, cocktail strategy and longitudinal status; a target decision matrix of the 6 actionable Tier-A species × 5 attributes resolving to a final priority class; and a 4-phase clinical-translation roadmap (Immediate / Near / Mid / Long term). ₁

How the Tier-A target list was scored

The candidate list this framework translates was produced by an explicit four-criterion score rather than by expert choice. Four criteria (A3–A6 from the project’s research plan) were applied to the 71 unique rigor-controlled candidates — 51 E1 plus 40 E3 provisional plus 5 cross-ecotype engraftment, deduplicated. ₁

The E1 arm of that pool is the most strongly supported: E1 Tier-A comprised 51 candidates, all 100 % sign-concordant across sub-studies, at FDR (false-discovery rate, the expected share of false positives among calls) < 0.10 and pooled CLR-Δ (centred-log-ratio abundance difference) > 0.5. ₁

The top ten E1 Tier-A species, with pooled CLR-Δ and FDR as the report gives them: ₁

• Ranks 1–5: Mediterraneibacter gnavus +4.85, 2e-12; Streptococcus salivarius +3.26, 2e-9; Streptococcus thermophilus +2.69, 4e-6; Erysipelatoclostridium innocuum +2.65, 4e-7; Streptococcus parasanguinis +2.44, 2e-6. ₁
• Ranks 6–10: Enterocloster asparagiformis +2.41, 2e-9; Intestinibacter bartlettii +2.36, 3e-5; Hungatella symbiosa +2.23, 1e-5; Gordonibacter pamelaeae +2.18, 2e-5; Erysipelatoclostridium ramosum +2.16, 4e-6. ₁

Ranks 2, 3 and 5 — S. salivarius, S. thermophilus, S. parasanguinis — are canonical oral-cavity species, and the report reads their enrichment in Crohn’s-disease stool as consistent with the oral-gut axis ectopic-colonization literature for IBD (Xiang 2024, PMID 39188957; Guo 2024, PMID 38545880; Tanwar 2023, PMID 37645044) and with the specific finding that S. salivarius is a salivary biomarker for orofacial granulomatosis co-occurring with Crohn’s disease (Goel 2019, PMID 30796823); this is cited external literature, not measurement in this corpus. ₁

Rank 4 raises the therapeutic stakes of a species the phage layers do not cover: Erysipelatoclostridium innocuum is specifically documented as a vancomycin-resistant pathobiome in IBD with clinically consequential phenotypes — creeping-fat formation and intestinal strictures in Crohn’s disease, reduced ulcerative-colitis remission rates — and FMT (faecal microbiota transplantation) is under evaluation as an intervention (Le 2025, PMID 40074633). ₁

Rank 9 cuts the other way and weakens any blanket reading of enrichment as pathobiont activity: Gordonibacter pamelaeae produces urolithins from dietary ellagitannins (Selma 2014, PMID 24744017) and increases during microbiome recovery from a dysbiosis-inducing insult (Tierney 2023, PMID 36840551), which the report says makes its CD-up signal in E1 difficult to interpret as pathobiont activity. ₁

A3 — literature plus cohort CD-association (0–5) sums five independent signals per candidate: the confound-free meta-analysis, HMP2 external replication concordance, ref_cd_vs_hc_differential (Kumbhari reference; log₂FC > 0.5 plus FDR < 0.10), ref_species_ibd_associations (UHGG-indexed dxIBD mixed-effects) and ref_phage_biology curated top-tier targets. ₁

Most candidates carry only partial literature support: the A3 distribution was 13 candidates scoring 0, 22 scoring 1, 25 scoring 2, 5 scoring 3, 5 scoring 4 and 1 scoring 5. ₁

A4 — protective-analog exclusion (0 / 1) fails a candidate whose within-IBD-substudy CD-versus-nonIBD effect is negative (protective-analog risk) or that appears on the curated-protective-species list (F. prausnitzii, A. muciniphila, R. intestinalis, R. hominis, L. eligens, A. rectalis, C. scindens, C. eutactus, B. adolescentis, B. longum). ₁

Three candidates fail A4 — Anaerostipes hadrus (−0.32 confound-free effect), Clostridium scindens (curated protective list) and Roseburia faecis (−2.74 effect) — which supports the translation principle above by removing from consideration species whose depletion could be harmful before any phage question is asked. ₁

A5 — strain and engraftment evidence awards 1.0 to the 5 donor-2708-engraftment pathobionts (M. gnavus, E. lenta, E. coli, E. bolteae, H. hathewayi) and 0.5 for a Kumbhari strain-competition disease-dominance or IBD-adapted-strain gene signal; 3 candidates hit the 0.5 tier (A. hadrus, B. cellulosilyticus, F. plautii). ₁

A6 — BGC inflammatory mediator (0 / 0.5 / 1) takes a biosynthetic gene cluster (BGC, a genomic locus encoding a secondary-metabolite pathway) count per candidate from ref_bgc_catalog, using synonymy-inverted matching so that pre-GTDB (Genome Taxonomy Database) names such as “Ruminococcus gnavus” match the canonical “Mediterraneibacter gnavus”; a score of 1.0 requires ≥ 1 BGC containing CD-enriched CB-ORFs — the report’s per-ORF enrichment units from ref_cborf_enrichment — at effect > 0.5 and FDR < 0.05. ₁

Fourteen candidates score 1.0 on A6, including M. gnavus (39 BGCs, 26 CD-enriched CB-ORFs — the largest count), E. coli (93 BGCs with MIBiG (Minimum Information about a Biosynthetic Gene cluster) matches to Colibactin / Yersiniabactin / Enterobactin), S. salivarius (98 BGCs with Salivaricin 9 / A / Cochonodin I), Streptococcus parasanguinis (51 BGCs, 11 CD-enriched) and Hungatella hathewayi (6 BGCs, 4 CD-enriched). These A6 counts come from the project’s candidate-scoring query against ref_bgc_catalog with synonymy-inverted name matching, whereas the per-species counts in the mechanism section below come from the project’s later genomic-content enrichment test over the same catalog (Elmassry 2025; 10,060 BGCs across 6,221 species-annotated entries), and the two analyses do not agree: E. coli 93 BGCs here against 146 there, M. gnavus 39 against 58, H. hathewayi 6 against 13, and E. bolteae recorded here as having no BGC hit in the catalog against 18 BGCs there. The source does not reconcile any of these pairs, so both members of each are preserved as analysis-specific counts and none should be quoted as the species’ BGC total. ₁

Total Tier-A score = A3/5 + A4 + A5 + A6, range 0–4, with the actionable threshold at 2.5; on that scale the scoring returned 6 actionable candidates, the top 6 of 71 scored, listed below with their evidence provenance. ₁

• 1 — Hungatella hathewayi (E1 | engraftment), 4.0: all 5 A3 signals pass; 6 BGCs with 4 CD-enriched CB-ORFs. ₁
• 2 — Mediterraneibacter gnavus (E1 | E3_prov | engraftment), 3.8: 4/5 A3; 39 BGCs with 26 CD-enriched CB-ORFs (inflammatory glucorhamnan mechanism). ₁
• 3 — Escherichia coli (E1 | engraftment), 3.6: engraftment plus MIBiG match to Colibactin + Yersiniabactin + Enterobactin. ₁
• 4 — Eggerthella lenta (E1 | engraftment), 3.3: engraftment plus 4/5 A3 signals plus a Kumbhari IBD-adapted-strain gene signal. ₁
• 5 — Flavonifractor plautii (E1 | E3_prov), 3.3: Kumbhari strain-competition plus a BGC with CD-enriched CB-ORFs. ₁
• 6 — Enterocloster bolteae (E1 | engraftment), 2.8: engraftment, but no BGC hit in the catalog under this scoring query, which the report flags as a potential blind spot. ₁

The engraftment-derived evidence is mostly concordant with the cohort differential evidence: 5 of 6 engraftment pathobionts pass the confound-free CD-up test. ₁

Host-range evidence and cocktail coverage

The PhageFoundry evidence layer phagefoundry_strain_modelling, derived from Gaborieau 2025, contained 96 phages, 188 Escherichia coli strains, and 17,672 experimentally tested susceptibility pairs, of which 3,929 were susceptible and 13,743 resistant, corresponding to a 22 % susceptibility rate. ₁

That layer is Escherichia coli-rich and sparse for everything else. The report states that 96 phages × 188 Escherichia coli strains × 17,672 susceptibility pairs is sufficient to design a 5-phage cocktail at 95 % strain coverage for Escherichia coli AIEC, while the other 5 actionable Tier-A species are not present in phagefoundry_strain_modelling at all, so their phage availability rests on ref_phage_biology literature curation alone. ₁

The central digest describes the same collection as 188 Escherichia coli strains, 96 phages and 17,672 binary infection outcomes, together with a machine-learning model at AUC (area under the receiver-operating-characteristic curve) = 0.883 and SHAP (SHapley Additive exPlanations, a per-feature attribution method) importances for 1,582 gene cluster features; no source states whether every strain × phage combination was tested, so the reported outcome count should be read as the number of experimentally tested pairs rather than as a complete strain-by-phage matrix, and the matrix layout remains undocumented. Within that matrix, Lambdavirus 411_P1 (phage lambda) infects only 1/188 strains (0.5 %) against 43.4 % for Myoviridae, which the digest reads as widespread ManYZ receptor loss making lambda-mediated infection rare — an interpretation of the host-range pattern rather than a receptor measurement. ₂

The project report cites AUC=0.88 for the same Gaborieau phage-prediction experiment and states the susceptibility rate as 3,929 / 17,672 = 22 % across the matrix, so the two sources give slightly different model figures for one dataset. _1,2

A third source reports the same experiment with one metric the other two omit: the machine-learning model reached AUC = 0.883, Accuracy = 84.3%, trained on 1,582 gene cluster presence/absence features with SHAP importance scores, which supports the digest’s AUC figure rather than the rounded value the project report gives for the same experiment and that is recorded above. ₃

The same source gives the overall infection rate as 22% (3,929 positive / 13,743 negative) and describes the underlying layer as PhageFoundry Strain Modelling (phagefoundry_strain_modelling): 188 E. coli strains × 96 phages, binary infection outcomes plus an ML model, attributed to Gaborieau et al. 2024, Nature Microbiology. That attribution year differs from the Gaborieau 2025 the project report gives for the same layer; both are recorded as each source states them. _3,1

Within that panel the susceptibility rate is strongly family-structured, which refines the single headline rate into a per-family one: ₃

• Myoviridae — 43.4% (2,531/5,828), described as broad host range (tail fibers, not ManYZ-dependent). ₃
• Podoviridae — 13.0% (978/7,520), moderate host range. ₃
• Siphoviridae — 9.7% (420/4,324), narrow host range. ₃
• Lambdavirus (single phage) — 0.5% (1/188), read as near-universal resistance via ManYZ receptor loss/modification. ₃

The lowest row is a single phage: Lambdavirus 411_P1 (genus Lambdavirus) infects only 1 of 188 strains — NILS06 — a 0.5% infection rate and the lowest of any phage in the dataset, which supports the same observation recorded from the central digest above and adds the identity of the one susceptible strain. ₃

The receptor explanation stays an inference: the source states the pattern is consistent with widespread ManYZ variation/loss conferring near-universal lambda resistance across diverse E. coli strains, which interprets a host-range pattern rather than measuring receptor presence. ₃

The same source also carries the originating paper’s comparative conclusion — that “adsorption factors” (surface receptors) are the primary predictors of phage–host interaction, not antiphage defense systems — and reads it as supporting a niche for the SNIPE defence system, which it describes as cleaving DNA during ManYZ transit and so operating at exactly the interface where phage success is determined. Applying a panel-level predictor ranking to SNIPE is an extrapolation, because SNIPE was not among the systems tested inside this interaction matrix, and it refines the host-defence caveat recorded below. ₃

A greedy minimum-set-cover analysis selected five phages—DIJ07_P2, LF73_P1, AL505_Ev3, 55989_P2, and LF110_P2—that covered 94.7 % of the 188 tested strains; the report’s own summary describes the same five-phage design as covering 95 % of the 188 PhageFoundry-tested strains. The report’s synthesis of this analysis pairs the 94.7 % strain-coverage figure with a separately stated AIEC phylogroup B2/D coverage of 65/94 phages, which is a count of the collection’s isolation hosts rather than a coverage property of the selected cocktail. ₁

The greedy ladder reports 63.8 % coverage for DIJ07_P2 (Phapecoctavirus) alone, 81.4 % after adding LF73_P1 (Tequatrovirus), 92.6 % at four phages with AL505_Ev3 and 55989_P2, 94.7 % at five with LF110_P2, and 98.4 % at eight with NIC06_P2, LF73_P4 and BCH953_P4. ₁

The ladder rows are labelled with coverage targets the selected cocktails do not reach: the row labelled ≥95 % delivers 94.7 % and the row labelled ≥99 % delivers 98.4 %, so neither nominal threshold was met and neither should be quoted as achieved coverage. ₁

The project’s own synthesis row restates the dataset as 96 phages × 188 strains × 17,672 pairs and reports a “5-phage cocktail covers 95 % of E. coli strains” together with 69 % of phages in the AIEC-relevant phylogroup. Both of those summary figures rest on looser denominators and rounding than the detailed tables: the 69 % is taken over the 94 phages that carry host-phylogroup information rather than over the full 96-phage collection, and 95 % is a label for an achieved 94.7 %, so the synthesis row is a hand-off summary and the table values are the measured ones. The same synthesis row pairs that cocktail figure with the project’s other headline phage result — a gut-anaerobe Tier-A coverage gap confirmed across all 3 layers — so the two are reported together at summary grade. ₁

An extended eight-phage design reached 98.4 % coverage of the tested Escherichia coli strains. ₁

Among 94 phages with host-phylogroup information, 65 (69 %) were isolated against B2/D hosts, the phylogroups emphasized as AIEC-relevant in the report. ₁

Grouped by isolation-host phylogroup, 50 phages were isolated against B2 hosts (median host range 21.3 %, maximum 63.3 %), 15 against D hosts (13.8 %, 63.8 %) and 10 against A hosts (16.5 %, 39.4 %); these medians and maxima are host ranges measured across the tested 188-strain panel, not coverage within the respective phylogroup, so they cannot be read as phylogroup-specific efficacy. ₁

The AIEC phylogroup distribution motivating that emphasis — predominantly B2 (~80 %) and D (~20 %) — is attributed to Dogan 2014 and Dubinsky 2022, i.e. external literature rather than measurement in this corpus. ₁

The report reads the 65 of 94 phages (69 %) isolated against B2/D phylogroup hosts as AIEC-relevant partly through the identity of those isolation hosts: the broadest-host-range B2/D phages include LF82_P8 (60.1 %, Mosigvirus), where LF82 is identified as the AIEC reference strain (Darfeuille-Michaud 2004), LF73_P1 (62.8 %), where LF73 is identified as a CD-associated AIEC strain, and 536_P7 / 536_P9, where Escherichia coli 536 is identified as a UPEC/B2 archetype. ₁

PhageFoundry itself carries no explicit AIEC-versus-commensal strain annotation. The 188 strains include LF82, LF73, 536, NIC06 and BCH953, named as phage isolation hosts and listed by the report as canonical AIEC strains — though the same report elsewhere characterizes 536 as a UPEC/B2 archetype, so both labels it applies to that strain are recorded here rather than resolved in favour of either — and the report states that accurate AIEC coverage estimation still requires cross-reference to literature isolate annotations, so no AIEC-specific coverage figure follows from the panel as distributed. ₁

Twenty-six strains (14 %) were phage-resistant at ≤5 % susceptibility across the tested phage panel, and the report flags them as escape candidates whose clinical significance it says needs cross-reference to AIEC pks-island annotation. Low susceptibility across the panel as a whole does not establish resistance to the specific phages chosen by the set-cover analysis, so this group should be read as a panel-wide low-susceptibility set rather than as measured escape from the selected cocktail. ₁

These coverage values describe the PhageFoundry strain panel rather than the UC Davis patient isolates, because the dataset did not provide patient-isolate matching, explicit AIEC-versus-commensal labels, burst-size measurements, titer data, or in-vivo delivery validation. The 95 % strain-coverage figure likewise belongs to the 188 PhageFoundry-tested Escherichia coli strains, and per-patient applicability requires an AIEC strain-resolution diagnostic on UC Davis patient stool isolates, which the report queues in its near-term roadmap. ₁

Accordingly, the minimum-set-cover result supports prioritizing a candidate cocktail for laboratory testing, but it does not establish patient-level coverage or in-vivo efficacy. ₁

The susceptibility matrix is binary, carrying no titer or burst-size measurement, while real-world cocktail dosing depends on phage burst size, receptor-binding kinetics and gastric-passage survival — all of which the report places out of project scope. ₁

A strain-level diagnostic requirement attaches to the cocktail itself: it covers broad Escherichia coli strain diversity but does not by itself distinguish AIEC from commensal Escherichia coli, so the report requires per-patient pks-island, Yersiniabactin and Enterobactin gene-presence detection, via PCR or amplicon panel, before cocktail selection. ₁

The two phage layers do not intersect at all: HMP2 viromics × PhageFoundry overlap = 0, because the 7 unique Escherichia coli phages observed in HMP2 fact_viromics — D108, EC6, ECML-117, Murica, slur16, vB_EcoM-VpaE1 and vB_EcoM_AYO145A — do not name-overlap with any PhageFoundry phage. That 7-phage count stands beside the 16 unique Escherichia coli phages the same report counts in fact_viromics elsewhere on this page; the source does not reconcile the two, so both are preserved and neither should be used as the in-vivo Escherichia coli phage total. ₁

The report treats the two datasets as complementary rather than overlapping — PhageFoundry as research and clinical isolates with experimental susceptibility, HMP2 viromics as natural phages observed in patient stool — and draws the consequence that in-vivo gut delivery of the cocktail phages is not validated by the project’s data and would require in-vivo testing. ₁

Phage-availability tiering and the gut-anaerobe gap

The targetability matrix rests on ref_phage_biology, a 12-organism literature-curated synthesis assembled from an indicator-taxa literature review, so the phage-availability layer is a curation of published reports rather than new measurement. ₁

Phage availability is scored on a 0–3 ordinal scale: 0 = no known phages, 1 = temperate or limited, 2 = lytic literature, 3 = clinical trial or commercial cocktail. ₁

• Hungatella hathewayi — Tier-A score 4.0, phage score 0, no known phages: the highest-scored species is a coverage gap, flagged for external database query (INPHARED + IMG/VR). ₁
• Mediterraneibacter gnavus — Tier-A 3.8, phage score 1: 6 known phages, all temperate, leaving lytic-locked engineering or a biochemical glucorhamnan-synthesis target. ₁
• Escherichia coli (AIEC) — Tier-A 3.6, phage score 3: the EcoActive cocktail (7 lytic phages, clinical trials) plus HER259 (FimH-targeting) make it the most advanced target. ₁
• Eggerthella lenta — Tier-A 3.3, phage score 2: PMBT5 siphovirus characterized. ₁
• Flavonifractor plautii — Tier-A 3.3, phage score 0, phage status unknown: not present in ref_phage_biology at all, and carrying the highest bile-acid-coupling cost. ₁
• Enterocloster bolteae — Tier-A 2.8, phage score 2: PMBT24, a virulent 99,962 bp Kielviridae phage, the best-characterized lytic phage among the gut-anaerobe Tier-A set. ₁

Across the six actionable Tier-A species this yields four phage-availability classes: one clinical-trial-stage species (Escherichia coli, the EcoActive cocktail), two with lytic literature (Eggerthella lenta with PMBT5, Enterocloster bolteae with PMBT24), one temperate-only (Mediterraneibacter gnavus, 6 temperate phages) and two gaps (Hungatella hathewayi, Flavonifractor plautii). Read the other way, three of six actionable Tier-A core species have no lytic phage in the ref_phage_biology curation — Hungatella hathewayi and Flavonifractor plautii as coverage gaps, Mediterraneibacter gnavus temperate-only — which the report treats as a structural property of gut-anaerobe phage availability rather than a project limitation, with INPHARED and IMG/VR external database queries named the highest-priority extension but ruled out of scope under plan v1.9, which restricted the work to platform-internal data. ₁

Phage availability runs inverse to target priority here, which refines the translation principle above from a caveat into a structural feature: the two highest-scored species, Hungatella hathewayi at 4.0 and Mediterraneibacter gnavus at 3.8, have the weakest phage availability, and the two phage-gap species are scored 4.0 and 3.3, so INPHARED and IMG/VR external database queries remain the highest-priority follow-up rather than a completed evidence layer. Flavonifractor plautii additionally carries the highest bile-acid-coupling cost, so the report notes that phage targeting of it may be deprioritized in favor of bile-acid-binding co-therapy. ₁

Direct Tier-1 phage targeting is proposed only for the Escherichia coli AIEC subset, using either the EcoActive clinical-trial precedent or the project’s own five-phage design (DIJ07_P2 + LF73_P1 + AL505_Ev3 + 55989_P2 + LF110_P2), and is conditional on a strain-resolution AIEC diagnostic — pks-island plus Yersiniabactin plus Enterobactin gene-presence — that the patient data did not supply; Tier-2 phage targeting with monitoring, explicitly including bile-acid pool monitoring, is proposed for Eggerthella lenta (PMBT5) and Enterocloster bolteae (PMBT24), and for Mediterraneibacter gnavus only if lytic-locked phage engineering succeeds. ₁

For the gaps, Hungatella hathewayi is the highest priority for INPHARED and IMG/VR external-database query, with GAG-degrading enzyme inhibitors drawn from the ref_phage_biology therapeutic-targets layer as the fallback if no phages are found, while Flavonifractor plautii receives the lowest priority because of its bile-acid cost, the report suggesting either deprioritization or co-administration of UDCA or of a bile-acid-binding agent, which it lists as two separate options rather than describing UDCA as a binding agent. ₁

The actionable set is also not closed: 9 Tier-B candidates in the 2.2–2.4 range, including S. salivarius with Salivaricin MIBiG (Minimum Information about a Biosynthetic Gene cluster) matches, E. asparagiformis, V. parvula and S. parasanguinis, remain candidates subject to phage-availability promotion, so the target list could widen on evidence not yet gathered. The top E1 Tier-A candidates S. salivarius, S. thermophilus and S. parasanguinis are all oral streptococci, and the report reads their within-substudy CD enrichment as consistent with that framework. ₁

Named in full, the nine sub-threshold Tier-B candidates at score 2.2–2.4 are Enterocloster asparagiformis, Streptococcus salivarius, E. citroniae, E. clostridioformis, Blautia coccoides, Veillonella atypica, S. parasanguinis, Actinomyces oris and V. parvula; each has a BGC hit, an A4 pass and A3 = 1–2 signals but lacks direct engraftment or strain-adaptation evidence, so phage-targetability scoring could promote any of them on B-tier phage-availability evidence. ₁

Direct observation supports the curated gap from an independent data type: across 3,039 HMP2 fact_viromics sample-rows, 16 unique Escherichia coli phages and 13 unique Klebsiella phages were observed in patient stool, with no observations of phages targeting Hungatella hathewayi, Mediterraneibacter gnavus, Flavonifractor plautii, Eggerthella lenta or Enterocloster bolteae. ₁

The report’s combined availability verdict therefore stacks three layers — curated literature (ref_phage_biology), experimental susceptibility (PhageFoundry phagefoundry_strain_modelling, 96 phages × 188 Escherichia coli strains × 17,672 pairs) and in-vivo phageome (HMP2 fact_viromics, 630 samples) — which it reports as converging on the same per-target classification: for Escherichia coli the clinical cocktail, the experimental susceptibility matrix and the in-vivo phage correlation all align, which it reads as high feasibility, whereas for Mediterraneibacter gnavus, Hungatella hathewayi and Flavonifractor plautii the gap is confirmed across all three layers — no clinical phages, no PhageFoundry coverage and no in-vivo Podoviridae/Myoviridae signal in HMP2. That convergence is agreement among three evidence types rather than a proof of absence: a species being absent from the PhageFoundry collection and showing no in-vivo phage-family correlation does not establish that no phage for it exists. ₁

Per species, the project tabulates the three layers side by side, with the in-vivo column reporting endogenous phage-family correlations rather than direct observation of a phage for that host: ₁

• E. coli AIEC — clinical-trial (EcoActive); 5-phage cocktail at 95 % strain coverage (the report’s summary label for the measured 94.7 %), 65/94 AIEC-relevant phylogroup; +0.18 Podoviridae, +0.13 Myoviridae endogenous correlation; classed Tier-1, phage-feasible at high confidence with all 3 layers aligned. ₁
• E. lenta — PMBT5 siphovirus literature; not in the PhageFoundry collection; no detectable in-vivo correlation; classed Tier-2, phage-feasible with monitoring. ₁
• E. bolteae — PMBT24 (Kielviridae, virulent); not in the PhageFoundry collection; +0.12 Myoviridae correlation; classed Tier-2, phage-feasible with monitoring. ₁
• M. gnavus — 6 temperate siphoviruses; not in the PhageFoundry collection; weak in-vivo signal; classed Limited, temperate-only, alternatives needed. ₁
• H. hathewayi — NO known phages; not in the PhageFoundry collection; no detectable in-vivo phage correlation; classed GAP, external-database query priority. ₁
• F. plautii — not in the literature curation; not in the PhageFoundry collection; no detectable in-vivo phage correlation; classed GAP with the highest bile-acid-coupling cost, alternatives preferred. ₁

The same six resolve into a target decision matrix that adds ecotype association, bile-acid-coupling cost and UC Davis prevalence to a final priority class: ₁

• H. hathewayi — 4.0, E1+E3, no bile-acid coupling listed, GAP, 83 %: Tier-1 phage GAP, INPHARED/IMG-VR query priority. ₁
• M. gnavus — 3.8, E1+E3, none, temperate-only, 91 %: Tier-1 limited — lytic-locked engineering OR biochemical glucorhamnan target. ₁
• E. coli — 3.6, E3-specific, none, clinical-trial, 35 %: Tier-1 phage with strain-resolution — the five-phage cocktail, AIEC diagnostic required. ₁
• E. lenta — 3.3, universal Tier-1, moderate, lytic (PMBT5), 70 %: Tier-2 phage — PMBT5 with bile-acid monitoring. ₁
• F. plautii — 3.3, E1-specific, HIGHEST bile-acid cost, GAP, 78 %: Tier-2 deprioritize — triple penalty; UDCA/BA-binding co-therapy. ₁
• E. bolteae — 2.8, E1+E3, moderate, lytic (PMBT24), 83 %: Tier-2 phage — PMBT24 with bile-acid monitoring. ₁

The report reads that matrix as exposing the structural shape of the clinical-translation problem: the 2 highest-scored species (H. hathewayi, F. plautii) are both phage GAPs; the species with the most complete phage precedent (E. coli) is the rarest in UC Davis at 35 % carriage; F. plautii carries a triple penalty and should be deprioritized despite its actionable status; and M. gnavus is near-universal but temperate-only. Its naming of H. hathewayi and F. plautii as the two highest-scored species sits awkwardly with the same table ranking M. gnavus at 3.8 above F. plautii at 3.3; the wording is recorded as the source states it rather than reconciled. ₁

HMP2 viromics also carried no CRISPR-spacer-derived phage–host predictions, so discovering phages for the Tier-A hosts out of that dataset would itself require an IMG/VR cross-reference. ₁

Host-side defence content is a further, largely unmeasured modifier of translation: the PhageFoundry Klebsiella genome browser carries 1 DUF4041 annotation (3 proteins at 558 aa, the same length as the Escherichia coli SNIPE protein) and 4,619 PTS_EIIC (PF02378/ManY family) annotations, making Klebsiella the only PhageFoundry species with both SNIPE and the ManX-family PTS domain PF00358; the suggestion that this affects phage-therapy efficacy in that pathogen is speculative, and it links this page to phage-defense-syndromes-and-arms-race and snipe-defense-system. ₂

A four-species genome-browser search puts that single Klebsiella annotation in comparative context, counting defence-related annotation terms per species: ₃

• Klebsiella — DUF4041 1, GIY-YIG 1, Nuclease 98, Restriction 42, Abortive infection 2. ₃
• Acinetobacter — DUF4041 0, GIY-YIG 3, Nuclease 91, Restriction 33, Abortive infection 2. ₃
• P. aeruginosa — DUF4041 0, GIY-YIG 2, Nuclease 125, Restriction 54, Abortive infection 2. ₃
• P. viridiflava — DUF4041 0, GIY-YIG 0, Nuclease 90, Restriction 41, Abortive infection 1. ₃

The source reads the detection of DUF4041 in Klebsiella as clinically relevant on the grounds that SNIPE could affect phage therapy efficacy in this pathogen; that is a proposed consequence of an annotation hit, not a measured change in lysis, and it supports the same speculative link already recorded from the central digest rather than supplying independent evidence for it. ₃

Mechanism and ecological cost of candidate targets

Mechanism evidence is concentrated in one target, and the project’s per-target cocktail-design profile is built on three axes: iron/AIEC-mediated versus other CD-specialization mechanism, bile-acid coupling cost, and species-abundance-mediated versus strain-content-mediated mechanism. A Fisher’s exact test of biosynthetic gene cluster (BGC) themes in Tier-A core versus background BGCs, with Benjamini–Hochberg false-discovery-rate (FDR) correction across themes, found iron_siderophore enriched at 54 / 286 versus 51 / 9,774 (Fisher OR 44.4, FDR 6.5e-56), genotoxin_microcin at 25 / 286 versus 4 / 9,774 (OR 234.0, FDR 3.3e-35) and NRPS_PKS_hybrid at 32 / 286 versus 739 / 9,774 (OR 1.54, FDR 0.042), while bacteriocin_RiPP at 154 / 286 versus 5,217 / 9,774 (OR 1.02, FDR 0.90) sat at background rate. ₁

Within the actionable Tier-A core, Escherichia coli holds 146 BGCs with 54 iron and 25 genotoxin MIBiG matches, whereas Eggerthella lenta (41 BGCs), Mediterraneibacter gnavus (58), Enterocloster bolteae (18), Hungatella hathewayi (13) and Flavonifractor plautii (10) each have zero in both columns. ₁

Escherichia coli’s 54 iron BGCs comprise 19 Yersiniabactin, 16 Enterobactin and 19 of class siderophore (some redundant per-strain assemblies), and its 25 genotoxin BGCs comprise 8 Colibactin, 15 Microcin B17 and 2 Microcin J25; this refines the iron-and-genotoxin rationale in iron acquisition and pathobiont expansion to a single species rather than a property of the pathobiont module. ₁

Two therapeutic readings of that iron signal are proposals rather than demonstrated interventions: that per-patient iron-supplementation decisions should account for ecotype-specific pathobiont iron-system expression, and that phage targeting could specifically deplete the iron-system-enriched AIEC subset without affecting iron-pathway-deficient commensals. ₁

The other five Tier-A species are BGC-mechanism dark, so their inclusion rests on Tier-A scoring, co-occurrence module membership and a within-carrier metabolic signature rather than on BGC presence; relatedly, ebf/ecf RPKM (reads per kilobase of transcript per million mapped reads) functions as a sample-level CD biomarker with no per-species attribution, which makes it usable for treatment-response monitoring but not for cocktail target selection. ₁

Polyamine and PUFA (polyunsaturated fatty acid) elevations share that limitation — sample-level CD biomarkers without per-species attribution, hence clinical follow-up measures rather than per-species cocktail-design targets. ₁

Strain resolution, which the Escherichia coli component needs in order to target Yersiniabactin/Enterobactin/Colibactin-positive AIEC specifically, was dropped from cMD raw reads under plan v1.9; the proposed substitute is a kbase_genomes pks plus iron-BGC genome-content query, that is, a genome-content proxy standing in for patient-strain typing. ₁

Removing a target also carries a bile-acid cost. Flavonifractor plautii carries the highest: a strong negative correlation with primary cholate (−0.26) together with a positive correlation with secondary lithocholate (+0.15), read as 7α-dehydroxylation activity whose loss would shift the bile-acid pool toward primary inflammatory forms — a prediction, not a measured depletion. Enterocloster bolteae is moderate (deoxycholate +0.17, lithocholate +0.18), and Eggerthella lenta moderate with a partial pattern (cholate −0.13, ketodeoxycholate −0.14) consistent with Eggerthellaceae bile-acid metabolism (Devlin & Fischbach 2015 PMID 26412091). ₁

The project’s per-species mechanism matrix states the resulting targeting decisions explicitly. Escherichia coli, at Tier-A 3.6, is labelled a Tier-1 phage target with a strain-resolution requirement: its dominant mechanism is read as strain-content-mediated, combining a within-carrier CD-down signal with 54 iron MIBiG matches counted across the species, while the AIEC-subset focus — targeting pks/Yersiniabactin/Enterobactin-positive strains, at low bile-acid cost — is the report’s targeting rationale rather than a strain-resolved assignment of those matches to AIEC genomes. ₁

The three bile-acid-coupled species are labelled Tier-2 phage targets with graded coupling cost: Flavonifractor plautii (Tier-A 3.3) as an active 7α-dehydroxylator carrying the highest bile-acid-coupling cost (cholate −0.26, lithocholate +0.15), where depletion is predicted to shift the bile-acid pool toward inflammatory primary tauro-conjugated forms and the report suggests co-administering UDCA or a bile-acid-binding agent; Enterocloster bolteae (Tier-A 2.8) as an active 7α-dehydroxylator at moderate cost (deoxycholate +0.17, lithocholate +0.18), with no BGC-based mechanism and a mixed mechanism assignment; and Eggerthella lenta (Tier-A 3.3) as a partial 7α-dehydroxylator at moderate cost (ketodeoxycholate −0.14) whose primary CD association is a non-BGC drug-metabolism mechanism (Koppel 2018 Cgr2; CB-ORF CD-DOWN 54%) that the report says may not require strain resolution. ₁

Hungatella hathewayi, Mediterraneibacter gnavus and Escherichia coli are assigned low bile-acid-coupling cost because they are not in the 7α-dehydroxylation network in this dataset; that is an annotation of low cost within one dataset’s correlation network, not evidence that removing them would have no bile-acid consequence. ₁

Because the Flavonifractor plautii CD association is species-abundance-mediated, the report infers that phage targeting would deplete bile-acid 7α-dehydroxylation activity proportionally to the abundance reduction, with no strain-level escape route via gene-content selection within the species; this is an inferred consequence of an abundance model, not an observed phage intervention. ₁

A paired-sample analysis weakens one earlier cost argument, and the two parts of it differ in evidential status. Measured: cross-feeding between the pathobionts and Anaerostipes caccae is not supported by sample-level paired evidence, the substrate–product relationship is undetectable at sample level, and shared environment is the more parsimonious explanation of the species-level coupling — a null association result. Inferred: from that null the report concludes phage targeting of pathobionts is unlikely to substantially deplete A. caccae through metabolic coupling, which is a prediction about an untested intervention rather than a measured absence of depletion. Together they refine community-metabolic-interdependence. ₁

Patient-level targetability

Among 23 UC Davis Crohn’s disease patients, Mediterraneibacter gnavus was present in 21 (91 %), Hungatella hathewayi in 19 (83 %), Enterocloster bolteae in 19 (83 %), Flavonifractor plautii in 18 (78 %), Eggerthella lenta in 16 (70 %), and Escherichia coli in 8 (35 %); the report records these as per-target prescribing rates — M. gnavus 91 %, H. hathewayi 83 %, E. bolteae 83 %, F. plautii 78 %, E. lenta 70 %, E. coli only 35 %. ₁

Escherichia coli is present in only 35 % of UC Davis patients — much lower carriage than expected for an AIEC-canonical species — so the five-phage cocktail is directly applicable to ~8/23 patients only, which weakens any reading of the set-cover design as a general-purpose Crohn’s intervention. ₁ Detection of Escherichia coli does not by itself establish AIEC strain status, so the report’s ~35 % applicability estimate extrapolates from species presence rather than measuring the AIEC-targetable fraction; conversely the near-universal carriage of Hungatella hathewayi and Mediterraneibacter gnavus makes their phage gap the dominant unmet need. The report states this as a literature-versus-cohort mismatch: Escherichia coli AIEC is over-emphasized in the literature relative to UC Davis cohort prevalence, with only 8 of 23 patients carrying detectable Escherichia coli (35 %) against the canonical framing of AIEC as the prototypical Crohn’s-disease pathobiont (Darfeuille-Michaud 2004), so the five-phage AIEC cocktail is directly applicable to ~35 % of patients while the universal Hungatella hathewayi (83 %) and Mediterraneibacter gnavus (91 %) phage gaps represent the unmet need. ₁

The report’s own wording of that unmet need labels both species gaps — “while H. hathewayi (universal 83 %) and M. gnavus (universal 91 %) — both Pillar-4 GAP — represent the unmet need” — whereas the availability table above classifies Mediterraneibacter gnavus as temperate-only at phage score 1 with 6 temperate phages, not as a gap. The inconsistent label is recorded as the source’s wording rather than resolved in favour of either classification. ₁

The patient-level target profile therefore cannot be inferred from population-level target prioritization alone, because target prevalence and phage tractability differed across the six actionable species. ₁

The report assigned direct phage targeting to Escherichia coli, monitored lytic-phage targeting to Eggerthella lenta and Enterocloster bolteae, limited or engineered targeting to Mediterraneibacter gnavus, and non-phage alternatives or external database searches to Hungatella hathewayi and Flavonifractor plautii. ₁

Concrete per-patient cocktail drafts were produced for 14 of 23 patients (61 %), and all 9 E1 patients carried the full five-species E1 pathobiont module, but Hungatella hathewayi and Flavonifractor plautii were phage gaps and Mediterraneibacter gnavus was temperate-only. ₁ Broken out by ecotype, concrete phage components reached 0 of the 7 E0 patients (mean 1.71 Tier-A targets per patient), all 9 E1 patients (mean 4.89 targets; PMBT24 plus PMBT5 plus the five-phage Escherichia coli design where that species was present), 4 of the 6 E3 patients (mean 2.33 targets) and the 1 mixed-ecotype patient 6967 (3 targets); the E0 patients lack concrete components precisely because their priority targets sit in the availability gap or the temperate-only class. ₁

The report also records an operational stratification of the same cohort by cocktail strategy — 12 reserve for flare, with no active cocktail but calprotectin monitoring plus Mediterraneibacter gnavus qPCR; 4 on the E1 hybrid 3-strategy without Escherichia coli, receiving PMBT24 plus PMBT5 plus alternatives for H. hathewayi, F. plautii and M. gnavus plus UDCA/BA-binding; 4 E0 limited because their priority targets are GAP, receiving a limited cocktail with non-phage strategies considered; 1 on the full E1 hybrid 3-strategy with Escherichia coli AIEC, receiving the five-phage Escherichia coli design plus PMBT24, PMBT5, alternatives and UDCA; 1 E3 focused with Escherichia coli, receiving the five-phage design plus PMBT5 and alternatives; and 1 on state-dependent dosing, patient 6967, whose cocktail rebalances on E1↔E3 ecotype shifts — while calling the per-patient recommendations exemplars or templates rather than statistically robust per-patient validation, with generalization requiring multi-center validation and the stratification itself to be revisited on a larger Crohn’s disease cohort. The report does not state how these strategy counts map onto the ecotype assignment of the same 23 patients, so both groupings are recorded here as given. ₁

The 12 quiescent patients in that stratification are defined by calprotectin < 250 or unmeasured, and the report gives them the same Tier-A panel but as a “reserve for flare” rather than an active cocktail, with calprotectin monitoring plus Mediterraneibacter gnavus qPCR. ₁

The E1 pattern makes a pure phage cocktail structurally infeasible for those patients and supports a three-strategy hybrid consisting of direct phage targeting where lytic options exist, alternatives such as GAG-degrading enzyme inhibitors or bile-acid co-therapy, and limited or engineered approaches for Mediterraneibacter gnavus. ₁ The named components of that E1 draft are direct phage for Escherichia coli (the five-phage design, if present), Enterocloster bolteae (PMBT24) and Eggerthella lenta (PMBT5); alternatives for Hungatella hathewayi (GAG-degrading enzyme inhibitors) and Flavonifractor plautii (bile-acid-binding co-therapy, or deprioritization); and a limited option for the temperate-only Mediterraneibacter gnavus via lytic-locked engineering or a biochemical glucorhamnan-synthesis target — with the report stating plainly that a pure phage cocktail is not feasible for any E1 patient, only a hybrid. The report pairs that principle with its cohort-level counts — concrete phage cocktails for 14 of 23 patients (61 %), and Escherichia coli present in only 35 % of the cohort — with the Flavonifractor plautii bile-acid cost as the dominant E1 design constraint, and with a stratification of the patients into 4 cocktail-design categories (active disease with many targets, active disease with few targets, quiescent disease, and mixed ecotype). Its own statement of the principle is internally loose: it says all 9 E1 patients carry the full 5-species Tier-A pathobiont module while qualifying the direct-lytic subset as Escherichia coli, Enterocloster bolteae and Eggerthella lenta “when present”, so the stated module composition is ambiguous and is recorded here as given. ₁

The drafted regimen strata are: ~3 patients in E1 with active calprotectin and Escherichia coli present, receiving the full Tier-1 combination of the five-phage Escherichia coli cocktail plus PMBT24, PMBT5 and alternatives for Flavonifractor plautii, Hungatella hathewayi and Mediterraneibacter gnavus; ~5 E1 patients with active calprotectin but no Escherichia coli, receiving Tier-2 PMBT24 plus PMBT5 plus alternatives; 1-2 quiescent E1 patients, for whom a cocktail is reserved for flares under calprotectin monitoring; the 7 E0 patients, receiving a limited strategy with deprioritization considered because their priority targets are gap or temperate-only; ~3 E3 patients with Escherichia coli, receiving the five-phage cocktail plus PMBT5 and alternatives — a conditional recommendation, not patient-specific AIEC validation; and ~3 E3 patients without Escherichia coli, receiving a limited Eggerthella lenta plus Hungatella hathewayi strategy. ₁

Patients are also grouped into clinical-priority classes. Class A, active disease with multiple targets, lists patients 3701 (calprotectin 8000), 1835 (calprotectin 3340), 5843 (calprotectin 2150) and 6434 (calprotectin 251 plus 6 targets) as the highest priority for a full hybrid cocktail of phages plus alternatives for the gap species; class C, quiescent disease, lists 1112 (41), 2708 (3), 1492 (9) and 1460 (7), for whom the cocktail is reserved for flares with calprotectin monitoring; class D is the single mixed-ecotype patient 6967 (E1 ↔ E3), assigned state-dependent dosing and treated as the central per-patient stability test. ₁

The active-versus-quiescent split behind those classes uses a fecal-calprotectin threshold of 250 μg/g, which the report describes as a clinical convention not validated against per-patient response data: some patients with calprotectin 41–248 μg/g, currently placed in the quiescent class, may still benefit from an active cocktail, so the threshold should be re-evaluated against clinical response in a pilot. ₁

Flavonifractor plautii was deprioritized despite its Tier-A score because it combined the highest bile-acid coupling cost with a phage gap; it is present in 78 % of patients (18/23) including all 9 E1 patients, which makes that double penalty the dominant E1 design constraint. The report itself calls this a triple penalty — presence in 78 % of UC Davis patients, the highest bile-acid coupling cost as an active 7α-dehydroxylator, and phage-gap status — and labels Flavonifractor plautii a “leave alone” target despite its score of 3.3, recommending co-therapy to maintain the secondary bile-acid pool that its depletion would otherwise erode, a predicted rather than measured erosion. The report writes that co-therapy as “BA-binding co-therapy (UDCA / cholestyramine)”, grouping UDCA with cholestyramine under a binding agent, whereas elsewhere it lists UDCA and a bile-acid-binding agent as two separate options; the inconsistent grouping is recorded here as the source’s wording rather than adopted as a pharmacological classification. ₁

These per-patient outputs are template-grade by the report’s own statement: the 23-patient cohort is small, so the profiles are exemplars or templates rather than statistically robust per-patient validation, and there are no per-patient bile-acid measurements at all — the bile-acid cost driving several of the decisions above is an ecotype-level annotation, not a patient measurement. ₁

Two specific diagnostics are missing. No patient-specific AIEC strain-resolution diagnostic exists for the 8 Escherichia coli-positive patients — the 8 / 23 carry detectable Escherichia coli by Kaiju, a taxonomic read classifier, but no strain typing — so the cocktail recommendation for them assumes AIEC subset prevalence per Dogan 2014 rather than measuring it, and per-patient AIEC genotyping (pks-island plus Yersiniabactin plus Enterobactin) is listed as a near-term clinical-translation prerequisite; and the Flavonifractor plautii deprioritization is precautionary, resting on the correlation-based mechanism, with clinical validation requiring per-patient bile-acid panels. ₁

A pooled-cohort design would mismatch this patient structure. The report projects that a cocktail built against the pooled top-pathobiont list from the preliminary donor 2708 engraftment analysis (R. gnavus, E. bolteae, E. coli, E. lenta, C. difficile, K. oxytoca) would mismatch the ~27 % of UC Davis patients in E0, who may have near-healthy microbiomes and not need the cocktail at all, and may be over-aggressive for the 42 % in E1, whose pathobiont burden differs from the severe E3 cluster — a design inference from ecotype composition, not a measured treatment response. Those shares are quoted as the report states them; the same report elsewhere assigns the 23 patients as 7 E0, 9 E1, 6 E3 and 1 mixed-ecotype, and it does not state how the percentages relate to those counts, so neither figure is derived from the other here. ₁

The E0 stratum is additionally unserved by the evidence base: with no E0-specific Crohn’s-disease contrast available, the report proposes that Tier-A targets for the 27 % of the cohort it places in E0 — the same unreconciled share noted above — be drawn either from cross-ecotype cohort-level evidence (the 5 engraftment-confirmed pathobionts) or from an ecotype-agnostic within-substudy CD-versus-nonIBD comparison, a conditional fallback rather than an E0 target list. ₁

Ecological state and treatment updating

The four-ecotype framework assigned the 23 UC Davis patients to 7 E0, 9 E1, 6 E3, and 1 mixed-ecotype patient, with no UC Davis patient assigned to E2. ₁

The proposed cocktail-design workflow starts from that assignment: a patient’s stool MetaPhlAn3 taxonomic profile is projected onto the ecotypes to give E0, E1 or E3 — E2 being absent in active Crohn’s disease — and a per-ecotype Tier-A target list is then read off the co-occurrence modules, with E1_CD module 0 containing Hungatella hathewayi, Flavonifractor plautii, Enterocloster bolteae, Eggerthella lenta and Mediterraneibacter gnavus and E3_CD module 1 containing Eggerthella lenta, Hungatella hathewayi, Escherichia coli and Mediterraneibacter gnavus, which makes Flavonifractor plautii E1-specific and Escherichia coli E3-specific. ₁

That assignment step is itself a metagenomics requirement rather than a clinical-data shortcut: the clinical-data-only H1c classifier’s nominal AUC (area under the receiver-operating-characteristic curve) passes the project’s bar, but in the UC Davis translation it could not separate E1 from E3 without metagenomics. The report therefore recommends stool metagenomics as a trial screening step, listing a rapid qPCR panel on ecotype-defining species and better within-IBD training data as unproven alternatives. ₁

The module-anchored design logic may also not transfer across ecotypes. The two oral Actinomyces anchors of E3_CD module 1 couple only weakly with module pathobionts (~ρ=0.17–0.19), which the report reads as E3 being characterized by inflammation-driven oral-gut ectopic colonization rather than metabolic structure; the cocktail-design implication is therefore E1-specific, and the 31 % of the UC Davis cohort the report places in E3 may need a different module-anchor framework. That 31 % is the report’s own figure and is not reconciled in the source with its assignment of 6 of the 23 patients to E3, so both are kept as stated. ₁

The taxonomic ecotype framework at NB01b consensus K=4 remains the project’s primary ecotype basis for the cocktail-design pillars, so every state-dependent recommendation below inherits that clustering choice and the sensitivities discussed in ecotype-clustering-validity. ₁

The report recommends retaining a universal Tier-1 trio of Mediterraneibacter gnavus, Hungatella hathewayi, and Eggerthella lenta, dropping Flavonifractor plautii after an E1-to-E3 transition, and considering Escherichia coli targeting in E3 when an AIEC strain diagnostic is positive. The trio spans both E1 and E3, so the report treats it as the cocktail backbone for any active-disease Crohn’s patient that does not need re-evaluation on an ecotype shift, whereas Flavonifractor plautii is an E1_CD module 0 species and not an E3_CD module 1 species — dropping it on an E1→E3 transition also reduces the bile-acid-coupling-cost concern — and Escherichia coli sits on the E3_CD module 1 priority list rather than the E1_CD list, so an E3→E1 transition deprioritizes it, with inclusion in either direction subject to AIEC strain detection. ₁

This state-dependent strategy connects condition-specific-fitness to therapy selection by treating ecological state as a variable that can alter target priority, intervention composition, and the expected cost of removing a species. ₁

The per-patient layer was assembled for all 23 UC Davis patients by combining ecotype, Tier-A profile, bile-acid-coupling-cost annotations and medication harmonization into concrete cocktail drafts with per-patient components and caveats, which the report describes as a multi-attribute decision framework rather than a single-axis Tier-A score, and within which patient 6967’s longitudinal E1 ↔ E3 shift serves as the central per-patient stability test. ₁

Patient 6967, the only multi-timepoint UC Davis patient with biological-replicate samples, shifted from E1 to E3 across two visits, while Mediterraneibacter gnavus increased from 0.53 to 7.45 reads, a 14.0× expansion; across those two visits the cocktail Jaccard similarity was 0.60, with Hungatella hathewayi, Mediterraneibacter gnavus, and Eggerthella lenta shared between the two visits’ target lists and Enterocloster bolteae plus Flavonifractor plautii appearing as priority targets at visit 1 only — a change in the ecotype-derived target list rather than a disappearance of those species, which were still detected at visit 2. The ecotype calls themselves are only moderately confident — visit 1 confidence 0.64, visit 2 confidence 0.41 — and the report states that the ecotype-call shift could partly reflect classifier uncertainty, while noting that the 14× Mediterraneibacter gnavus and 3× Eggerthella lenta expansions are quantitative biological observations independent of the ecotype label. ₁

The report calls that 0.60 Jaccard moderate overlap, implying non-trivial cocktail re-design rather than a complete rewrite, but its generalizability is unknown: whether 3-of-5 component overlap implies clinically meaningful re-design depends on patient-response heterogeneity, which the report places out of project scope. ₁

In the same patient, Eggerthella lenta increased 3.1×, Flavonifractor plautii 1.9×, Enterocloster bolteae 2.1×, and Hungatella hathewayi 1.3×, whereas Escherichia coli remained absent at 0.00 reads on both visits. ₁ The report reads the Mediterraneibacter gnavus fold-change — 14×, from 0.53 to 7.45 reads — as the dominant signature of the E1→E3 transition, with the other Tier-A pathobionts expanding 1.3–3.1× as general worsening of dysbiosis; because this rests on one patient’s two visits, it is a single-case observation rather than an established transition signature. ₁

Technical replication bounds how much of such a trajectory could be measurement noise. In patient 1112’s technical replicates the six Tier-A taxa gave 0.445 versus 0.436 (2 %) for Hungatella hathewayi, 7.916 versus 7.792 (1.6 %) for Mediterraneibacter gnavus, 0.000 versus 0.000 for Escherichia coli, 0.712 versus 0.642 (10 %) for Eggerthella lenta, 2.019 versus 1.994 (1.2 %) for Flavonifractor plautii and 0.864 versus 0.862 (0.2 %) for Enterocloster bolteae, with Spearman ρ = 1.000 (p < 0.001) on the 6 Tier-A taxa — perfect rank concordance across resequencing replicates. That fixes a technical-noise floor for the longitudinal contrast but is not itself a biological-replicate longitudinal trajectory, so the 5 state-dependent dosing rules of the project’s longitudinal verdict, and the 0.60 cocktail Jaccard alongside them, still rest on patient 6967 alone. ₁

The drafts tracked that transition: at visit 1 (E1) the module gave 5 priority targets, all 5 present, so the cocktail was Hungatella hathewayi, Mediterraneibacter gnavus, Enterocloster bolteae, Eggerthella lenta and Flavonifractor plautii; at visit 2 (E3) the module gave 4 priority targets of which 3 were present — Escherichia coli absent — so the cocktail narrowed to Eggerthella lenta, Hungatella hathewayi and Mediterraneibacter gnavus, the three species shared across both visits that the report labels the universal Tier-1 trio. The report reads Escherichia coli’s absence at both visits as meaning patient 6967 is not an AIEC carrier, which is an inference from species-level non-detection rather than a strain-typed result. ₁

These observations support reassessing the target and cocktail profile over time rather than assuming that one baseline cocktail remains appropriate after ecological state changes. ₁

The proposed workflow recommends ecotype reassessment every 3–6 months and a five-fold change in Mediterraneibacter gnavus abundance, measured by qPCR as a cheap clinical proxy — motivated by the 14× Mediterraneibacter gnavus expansion in patient 6967’s E3 transition and proposed so that full metagenomics is not needed at every visit — as the trigger for full ecotype retesting, keeps the universal Tier-1 trio of Mediterraneibacter gnavus, Hungatella hathewayi and Eggerthella lenta as the cocktail backbone for any active-disease Crohn’s disease patient, and conditions the remaining targets on ecotype (Flavonifractor plautii dropped on an E1→E3 transition, Escherichia coli included in E3 subject to AIEC strain detection); the report explicitly labels the qPCR proxy, trigger, and dosing rule as hypotheses requiring prospective validation. ₁

The hand-off packages those rules as a complete clinical workflow in three parts — an initial visit, follow-up calprotectin and Mediterraneibacter gnavus qPCR monitoring, and an ecotype-shift decision tree — none of which the project ran prospectively. ₁

Three limits qualify that workflow. No timing information is available between patient 6967’s two visits, so the duration of the E1→E3 drift is unknown and the 3–6 month re-test interval is a clinical convention default rather than a figure derived from the data; the state-dependent dosing rule is theoretical, not yet validated in clinical practice, and rests on n=1, which the report states limits statistical generalization and makes the rule a hypothesis for prospective validation; and the Mediterraneibacter gnavus qPCR trigger is a hypothesis rather than an established assay — the 14× expansion in patient 6967’s E3 transition suggests a 5-fold change might be the threshold for triggering a full ecotype re-test, but it comes from a single trajectory and validation requires a prospective cohort with paired qPCR and metagenomics across timepoints, which would also test whether ecotype drift is common at all and whether the 0.60 cocktail Jaccard generalizes beyond this one trajectory. ₁

External phage-database queries (INPHARED + IMG/VR) and multi-cohort prospective validation are flagged in the project’s own roadmap as out-of-scope clinical-translation prerequisites, which places both load-bearing evidence gaps of this framework outside what the present corpus can close. The roadmap presents these items as out-of-scope follow-ups required to bridge from computational deliverables to a clinical pilot, and sizes the two external resources as INPHARED ~25K phages and IMG/VR ~3M UViGs covering the gut-anaerobe phage-discovery space. It names the 3 gut-anaerobe coverage gaps those queries would address as H. hathewayi, M. gnavus lytic alternatives and F. plautii, and flags two further near-term-tier prerequisites: multi-cohort prospective validation of the state-dependent dosing rule, whose basis is the n=1 patient 6967 trajectory, and a per-patient AIEC strain-resolution diagnostic. ₁

Evidence limits and safety of translation

HMP2 endogenous phageome analysis covered 3,039 rows across 630 samples, with 97 % ecotype-call coverage, and identified Gokushovirus WZ-2015a as CD-down cross-ecotype (a Microviridae depletion), the E1 contrast giving cliff δ = −0.358, FDR = 5e-7, n_CD = 231, and n_HC = 125, which the report reads as consistent with Norman 2015 and Clooney 2019. ₁

Escherichia coli correlated with Podoviridae at ρ = +0.183 and Myoviridae at ρ = +0.125, while the maximum absolute endogenous-phage correlation was ≤ 0.18. ₁ These two statements as reported cannot both hold — the Podoviridae coefficient falls outside the stated maximum — and the source does not reconcile them, so both are preserved here as conflicting source statements and neither should be used as a bound. ₁

The report’s pillar summaries state the same two correlations in rounded form as Escherichia coli × Podoviridae +0.18 and × Myoviridae +0.13, Spearman coefficients it reads as confirming endogenous Escherichia coli phage carriage in vivo among T7-like and T4-like Caudovirales families; both the rounded and the finer figure pairs are recorded here as the report gives them, and only the rounded pair is consistent with the report’s |ρ| ≤ 0.18 characterization. ₁

These associations are compatible with ecological coupling but do not demonstrate that endogenous phages control target abundance or that adding exogenous phages will reproduce the observed relationship. ₁

External literature supplies a precedent for the Escherichia coli arm: Galtier 2017 (PMID 28130329) is cited as demonstrating that AIEC-specific bacteriophages targeting CEACAM6-bound E. coli can reduce AIEC-associated intestinal mucosal signal in Crohn’s disease models — cited literature precedent, not a clinical outcome measured in this project. ₁

The citation list frames that precedent as promising rather than as established clinical efficacy — bacteriophages targeting adherent-invasive Escherichia coli strains as a promising new treatment for Crohn’s disease — and adds a second, community-level precedent: Sheikh 2024 (PMID 38532703) reports that transplanting microbiota from Crohn’s disease patients into germ-free mice results in colitis, a mouse result that bears on community involvement without identifying which species’ removal would help. ₁

The phageome is a noisier signal channel than species-level taxonomy. The report describes the phageome × ecotype × diagnosis effect sizes as modest, giving a maximum of |ρ| ≤ 0.18 — the same figure recorded above as inconsistent with the +0.183 Podoviridae coefficient, so it should be read as the report’s characterization rather than a bound — against species-level taxonomy effect sizes routinely |CLR-Δ| (centred-log-ratio abundance difference) > 1.0, and states that the three-layer phage-evidence convergence therefore rests on directional concordance, not on effect-size matching across layers. ₁

Cross-cohort replication is weak at the level of the target list itself: single-study E3 evidence combined with marginal ecotype stability means the Tier-A list, particularly its E3 component, should be treated as hypotheses for further experimental validation rather than established targets, and the UC Davis cocktail drafts inherit that limitation, with per-candidate cross-cohort support to be annotated explicitly. ₁

Against that, the report records that the rigor-controlled Tier-A list replicates at 88.2 % sign concordance on HMP2, which it elsewhere compresses to “HMP2 88.2 % E1 Tier-A external replication”; that endpoint is a sign-concordance rate for the Tier-A candidate list, not a replication rate for the six actionable species, which the report names separately as H. hathewayi, M. gnavus, E. coli, E. lenta, F. plautii and E. bolteae. The replication endpoint therefore refines rather than removes the caveat above, because it is reported for the E1 list and no equivalent replication figure is given for the E3 component. ₁

The report also notes that HMP2 viromics had an 80 % family-classification Unknown fraction — the VirMAP family-classification gap, covering 80 % of observations — leaving substantial uncertainty in phage identity and host assignment; Hungatella hathewayi and Mediterraneibacter gnavus show negative correlation with that dominant Unknown family, and all in-vivo correlations are |ρ| ≤ 0.18, modest relative to the species-level Crohn’s axis, so what non-detection of phages for those hosts can establish is bounded by how little of the phageome is classified. ₁

The unresolved gaps motivate phage-therapy-evidence-translation as a distinct translational problem rather than collapsing host-range measurement, ecological association, and clinical efficacy into one evidence category. ₁

The framework should also be interpreted alongside pangenome-integration, because susceptibility was measured across strains and the proposed AIEC-relevant selection depends on strain-level diagnostics that were not available in the current patient data. ₁

It also intersects environmental-resistome, because patient-specific treatment requires distinguishing a target’s abundance association from evidence that phage-mediated removal will alter disease-relevant ecology. ₁

Finally, the integration of PhageFoundry, HMP2 viromics, and UC Davis profiles illustrates the cross-dataset provenance and comparability issues captured by cross-tenant-data-bridging. ₁

Figures

• NB05_tier_a_scored.png — top-30 A3–A6 scoring matrix heatmap plus total-score bar chart with the 2.5 actionable threshold. ₁
• NB12_phage_targetability_matrix.png — per Tier-A core × phage-availability score heatmap with 4-class stratification (clinical-trial / lytic-literature / temperate-only / GAP) and a therapeutic-targets cross-reference for alternative strategies. ₁
• NB13_phage_cocktail.png — three panels: cumulative strain-coverage curve (5-phage = 94.7 %), per-phage isolate-host phylogroup distribution (AIEC B2/D dominant), and cocktail-composition strain-coverage heatmap. ₁
• NB14_phageome_ecotype.png — three panels: per-phage-family CD-vs-HC by ecotype (Gokushovirus E1 cliff=-0.358 FDR=5e-7 highlighted), Tier-A pathobiont × phage-family endogenous correlation heatmap (E. coli × Podo/Myo +0.18/+0.13), and the family-classification gap (80 % “Unknown”). ₁
• NB15_patient_cocktail_draft.png — three panels: 23 UC Davis patient × Tier-A presence heatmap sorted by ecotype and calprotectin, per-ecotype prescribing-rate bars (M. gnavus 91 %, E. coli 35 %), and calprotectin × n_targets scatter with the 4 patient-design categories. ₁
• NB16_longitudinal_dosing.png — three panels: patient 6967 per-visit Tier-A abundance (E1 → E3 transition, M. gnavus 14× expansion highlighted), patient 1112 technical-replicate scatter (Spearman ρ=1.000), and cocktail-composition shift Venn (visit 1 ∩ visit 2 = 3 universal Tier-1; Jaccard 0.60). ₁

Tensions

The project’s mechanism profile and its phage-availability matrix disagree about two species. On the mechanism side, Hungatella hathewayi is profiled as a Tier-1 phage target — highest Tier-A score 4.0, not in the bile-acid network, mechanism read from species abundance plus a within-carrier metabolic shift (biosynthesis-up, sugar-down) — with low bile-acid cost and broad targeting judged acceptable. ₁

Mediterraneibacter gnavus is likewise profiled as a Tier-1 phage target on the mechanism side: Tier-A 3.8, not in the bile-acid network, no Kumbhari strain-adaptation signal, and a well-characterized mucin-glucorhamnan-producing mechanism (Henke 2019, cited as reporting that the Crohn’s-disease-associated Ruminococcus gnavus produces an inflammatory polysaccharide — external literature, not a measurement in this corpus). ₁

On the availability side the same two species carry phage score 0 (no known phages) and 1 (6 temperate phages), so the “Tier-1 phage target” label records mechanistic desirability and low bile-acid cost, not the existence of a usable phage. The disagreement is between layers of one project rather than between projects, and it is recorded rather than averaged: a reader taking the mechanism table alone would over-read tractability for precisely the two species the host-range layer cannot currently serve. ₁

A second internal disagreement sits inside the patient stratification. The operational strategy table assigns 1 patient to the full “E1 hybrid 3-strategy (full, with E. coli AIEC)” — the five-phage Escherichia coli design plus PMBT24, PMBT5, alternatives and UDCA — while the project’s own state-dependent rule makes Escherichia coli inclusion E3-specific and drops it on an E3→E1 transition. The report does not reconcile the two; both are recorded here, the strategy table as a per-patient assignment conditioned on detected Escherichia coli and the dosing rule as a proposed ecotype-level default. ₁

A third cost argument is recorded as a within-project revision rather than a live disagreement. The earlier cocktail-design concern was that a phage cocktail depleting M. gnavus / F. plautii / H. hathewayi / E. bolteae may incidentally reduce Anaerostipes caccae abundance through loss of substrate — A. caccae being the only genuine butyrate-producer among the module-anchor commensals, and butyrate anti-inflammatory, so its incidental depletion could partially offset the cocktail’s therapeutic benefit — and that each actionable Tier-A target therefore needed a “metabolic-coupling cost” annotation before cocktail finalization; that was a hypothesis about an untested intervention, not a measured effect. The later paired-sample metabolite-level test reframed it: the substrate–product relationship is not detectable at sample level, the implication is less load-bearing than originally framed, phage targeting of pathobionts is unlikely to substantially deplete A. caccae through metabolic coupling, and bile-acid coupling cost replaces metabolic-coupling cost as the primary per-target ecological-cost annotation for the actionable Tier-A set. The earlier framing is preserved here as the project’s historical position rather than an equally supported side, and community-metabolic-interdependence carries the underlying coupling evidence. ₁

Open Directions

• Match the five-phage and eight-phage designs against raw UC Davis isolates using adsorption, killing-curve, resistance-frequency, burst-size, and titer assays to determine patient-isolate coverage rather than extrapolating from the 188-strain PhageFoundry panel. ₁
• Query INPHARED and IMG/VR for Hungatella hathewayi, Flavonifractor plautii, and Mediterraneibacter gnavus, then validate candidate host assignments and lytic activity experimentally to resolve the three gut-anaerobe coverage gaps. ₁
• Add AIEC strain diagnostics and pks, Yersiniabactin, Enterobactin, and other strain-level markers to patient-matched susceptibility models to test whether phylogroup and virulence-gene status improve Escherichia coli cocktail selection. ₁
• Prospectively follow patients every 3–6 months with metagenomics, qPCR, ecotype reassignment, bile-acid measurements, and phage-resistance assays to test whether the proposed reassessment interval and five-fold Mediterraneibacter gnavus trigger predict clinically relevant state transitions. ₁
• Perform controlled longitudinal phage or hybrid-intervention studies with target abundance, metabolomics, resistance, and clinical outcomes to distinguish phage-mediated causality from the modest endogenous-phage associations observed in HMP2. ₁
• Compare direct phage targeting, GAG-degrading enzyme inhibitors, bile-acid co-therapy, and engineered-phage approaches in patient-derived communities to determine whether hybrid treatment can address phage gaps without worsening bile-acid coupling costs. ₁
• Document the PhageFoundry matrix layout — whether the reported tested pairs cover every strain × phage combination implied by the reported strain and phage counts, and which pairs went untested — and re-derive the coverage percentages from the documented matrix before any of them is used as a design figure. _2,1
• Test whether Klebsiella strains carrying both the DUF4041/SNIPE annotation and ManX-family PTS domains are measurably harder to lyse, which would convert an annotation co-occurrence into a therapy-relevant defence effect. ₂
• Carry the bile-acid coupling cost annotation — the per-target ecological-cost measure the project adopted in place of metabolic-coupling cost — onto every actionable Tier-A target with per-patient bile-acid panels, and test the Anaerostipes caccae substrate case directly in co-culture or a defined community, to settle whether cocktail-driven pathobiont depletion erodes butyrate supply. ₁