Method reference: intrinsic versus acquired AMR, and what BacFlux can actually say¶

A regulator asking about a resistance gene wants to know one thing: did this strain always have it, or did it acquire it? This page covers what BacFlux can answer, and what it deliberately refuses to answer. For the short version, see The mobility ladder.

Terms used on this page


AMR	antimicrobial resistance
intrinsic	resistance shared by essentially all strains of a species — a property of the species, not of this isolate
acquired	resistance gained horizontally, from a plasmid or mobile element
EFSA	European Food Safety Authority, whose guidance drives the intrinsic/acquired question
FEEDAP	the EFSA panel that issues the guidance for feed additives
QPS	Qualified Presumption of Safety, EFSA's list of species accepted as safe
MIC	minimum inhibitory concentration — the lab measurement of resistance
ANI	average nucleotide identity, a genome-to-genome similarity measure
IS	insertion sequence, the smallest kind of mobile element

Three questions arise when an 08.mobilome table is read next to a regulatory checklist:

AMRFinderPlus reports point mutations. How is that different from a BLAST search like ABRicate's, and why does it need --organism?
EFSA defines an "intrinsic" AMR gene as one shared by the vast majority of strains of a species. Does that mean BacFlux would have to compare against every genome of that taxon in NCBI?
EFSA publishes a catalogue of AMR genes in Bacillus. Is that a new AMR database? Should BacFlux use it?

Short answers: it is a fundamentally different question, not a stricter search; yes, and that is why BacFlux does not attempt it; and no, it is not a database and BacFlux does not use it.

1. Two kinds of resistance, and why one is invisible to BLAST¶

A bacterium becomes resistant in one of two ways.

It acquires a gene it did not have. A blaCTX-M or a tet(M) arrives by horizontal transfer. Before, the gene was absent; after, it is present.

It mutates a gene it already had. A single amino-acid substitution in gyrA (fluoroquinolones), rpoB (rifampicin) or rpsL (streptomycin). These are essential genes, present in every strain of the species whether resistant or susceptible. One residue differs.

ABRicate answers "is this gene present?" by BLAST, and reports identity and coverage. That question is exactly right for the first mechanism and structurally blind to the second: query a susceptible genome for gyrA and you get a hit at ~99.9% identity; query a resistant one and you get a hit at ~99.9% identity. The substitution that decides the phenotype is averaged away inside the percentage. No threshold separates them, because the difference is not one of degree — it is one residue at one coordinate. Lowering the cutoff only adds noise.

AMRFinderPlus with --organism asks a different question. It loads a curated set of known resistance-conferring substitutions for that species, aligns the query to the reference allele, and inspects the specific coordinate: not "is gyrA here" but "is residue 83 of GyrA a serine or a leucine". Hits are reported with Method: POINTX / POINTP, naming the substitution.

Two further differences matter even where no mutation is involved:

Per-gene curated cutoffs rather than one blanket threshold. Gene families differ in natural diversity, so a single cutoff is either too permissive for conserved families or too strict for variable ones. This is why BacFlux applies the EFSA 80%/70% thresholds to the ABRicate leg only — imposing them on AMRFinderPlus would replace curation with a blunter rule.
Structured output: element type (AMR / STRESS / VIRULENCE), class and subclass, and a partial-at-contig-end flag. The mobilome module consumes these.

Why this matters for the intrinsic/acquired question¶

Point mutations are the chromosomal, non-transferable mechanism. So this is the one place a single genome can speak to the intrinsic side. Everything else the module does — replicon, mobile-element context, mobility tier — addresses the acquired and mobile side.

The caveat¶

--organism only engages for organisms AMRFinderPlus curates mutations for: roughly three dozen, overwhelmingly clinical — Escherichia, Salmonella, Klebsiella, Staphylococcus aureus, Campylobacter, Pseudomonas aeruginosa, Enterococcus, Vibrio, Neisseria and similar.

On a batch of plant- and environment-associated strains — the Bacillaceae and soil Pseudomonadota that make up most environmental collections — --organism was applied to none of them. Every {sample}_amrfinder_organism_audit.tsv recorded the same reason: no curated organism for this taxon. Every {sample}_amrfinderplus_mutations.tsv was therefore header-only.

Read those empty files as "not assessed", never as "no mutations found." BacFlux writes the refusal and its reason to the audit file so the distinction survives.

2. Why BacFlux does not classify genes as intrinsic or acquired¶

EFSA's definitions are population-level, and the wording is explicit:

Intrinsic AMR gene — "Gene inherent to strains of a bacterial species … An AMR gene is considered 'intrinsic' when it is shared by the vast majority of wild type strains of the same species (or subspecies) and is restricted to those located on the chromosome."

Acquired AMR gene — "A resistance gene novel for the strain under assessment, acquired through horizontal transfer … Acquired AMR genes could be integrated in the bacterial chromosome or harboured on a separate genetic element."

— EFSA Scientific Committee (2025), Glossary

Two consequences follow, and both constrain what this workflow may claim.

You cannot establish "intrinsic" from one genome. The claim is about the species, not the isolate. Determining it means comparing gene presence across many genomes of that taxon — which is why EFSA publishes a separate method (EFSA BIOHAZ Panel, 2023a) and a separate tool for it, described as a "Pipeline for the automated analysis of gene distribution in microbial species". BacFlux implements no population comparison of any kind: every analysis is per-genome.

Chromosomal does not mean intrinsic. An acquired gene may sit on the chromosome. This is why mobility tier 1 is labelled intrinsic_candidate and not intrinsic. The qualifier is doing real work and should not be dropped: the module observed that a gene is chromosomal with no mobile-element context, which is consistent with intrinsic and does not establish it.

So: BacFlux produces supporting evidence for the intrinsic/acquired judgement. It does not produce the judgement. The confirmatory work is a species-wide distribution analysis, plus phenotypic testing (MIC against the relevant antimicrobial), per the decision tree in EFSA (2025) §3.2.1.

Where BacFlux does align well: EFSA (2024a) requires the AMR search to run against at least two maintained/curated databases, at ≥80% identity and ≥70% length coverage. BacFlux runs ten database queries across three independent legs — ABRicate over eight databases (argannot, card, ecoh, ecoli_vf, megares, ncbi, resfinder, vfdb), AMRFinderPlus, and a BBMap→CARD read-mapping leg that is immune to assembly collapse — with those thresholds applied on the ABRicate leg.

3. The EFSA Bacillus catalogue — what it is, and why we do not use it¶

It is not an AMR gene database. It is a prevalence catalogue: how often each of a set of AMR gene sequences occurs among published genomes of five Bacillus species. It answers the population question BacFlux structurally cannot.

How it was built¶

(EFSA 2024b, §3.)

Genomes. NCBI RefSeq, restricted to assembly level complete genome, snapshot 30 November 2023.

The query set is the unusual part. The AMR sequences searched for were not taken from CARD, ResFinder or NCBI as a database. They were extracted from dossiers submitted to EFSA by applicants up to December 2023 — the AMR hits those applicants had themselves reported in their production or active-agent strains. Identifiers arrived in four different currencies (NCBI protein ID, NCBI nucleotide accession, CARD ARO ID, UniProt accession) and were all normalised back to NCBI protein IDs. Hits that resolved to no database record were dropped.

So CARD and UniProt appear only as lookup services for resolving identifiers, not as the source of the gene set. The set is bounded by what applicants happened to report, not by AMR biology.

Species selection. Nineteen Bacillus and related species from the QPS list were screened; those with ≥30 complete genomes were analysed. Five qualified:

Species	Screened	Confirmed (ANI ≥95%)
B. velezensis	329	328 (99.7%)
B. subtilis	299	299 (100%)
B. amyloliquefaciens	76	57 (75%)
B. licheniformis	40	40 (100%)
B. paralicheniformis	26	26 (100%)

B. paralicheniformis was included below the threshold because many dossiers concerned it. Priestia megaterium had 40 genomes and was excluded anyway, because no dossier had ever been submitted for it. The catalogue is shaped by the regulatory caseload, not by the genus.

Pipeline. fastANI v1.32 (one-to-many) against each species' reference/type strain; genomes at ANI ≥95% retained; ANIclustermap for trees. A local BLAST database built from the retained genomes, with plasmid sequences separated from chromosomes by a custom script so plasmid hits are reported separately. Searches with tblastn (protein queries) and blastn (nucleotide queries), E-value 0.05. Then: deduplicate within a genome keeping the highest identity, apply ≥80% identity / ≥70% coverage, compute the frequency of strains matching each query, and the median identity and coverage. Python 3 with R 4.3.3 for plots.

Controls. Every table carries internal positive controls — rpoB, gyrA, gyrB and 16S rRNA from the species' own reference genome, which should be near-universal — and a negative control, fimA from Salmonella, which should be absent. In the published tables the housekeeping genes return ~100% of genomes and fimA returns 0% every time. The pipeline demonstrates its own validity in each result table.

Scale of the output (Annex A): 358 raw matches across the five species, 192 valid and non-duplicated, and 95 hits above threshold actually analysed.

The finding most relevant to BacFlux¶

The catalogue is a clear demonstration of threshold brittleness at 80% identity:

rphB in B. subtilis: found above threshold in 85.3% of genomes when queried as nucleotide, but 1.3% when queried as protein — because most genomes sit at 79–79.9% identity, just under the cutoff.
rphC behaves the same way in B. amyloliquefaciens (79.8% median) and B. velezensis (79.8%): absent by the rule, near-universal in reality.
satA in B. amyloliquefaciens: ~91% of genomes by nucleotide, ~12–21% by protein, with a median identity of 79.2%.

The same sequence is "present in nearly all strains" or "essentially absent" depending on whether it is queried as DNA or as protein. Since BacFlux applies the same 80%/70% EFSA thresholds on its ABRicate leg, the lesson transfers directly: a gene just under the cutoff is not absent, and a borderline call should be read from the actual identity value, not from presence/absence. This is a further argument for keeping AMRFinderPlus's per-gene curated cutoffs unmodified.

Also instructive: tetL sits in ~63% of B. subtilis chromosomes — neither "vast majority" nor rare, i.e. genuinely undecidable by the intrinsic rule. Genes originating in other species (ANT(4′)-Ib from an S. aureus plasmid, ermB from E. faecium, cat from a Streptococcus phage) appear in 1–4% of genomes, which is what a real acquired gene looks like.

Why it is not built into BacFlux¶

Coverage. It spans five species. On one environmental batch, fewer than a fifth of the isolates fell inside those five; roughly another third were Bacillus or Priestia species outside them, and the rest were other genera entirely. For most of a typical environmental collection it would return nothing.

It is a snapshot. RefSeq at 30 November 2023, dossiers to December 2023. It carries no update mechanism.

It answers a question BacFlux does not ask. Wiring it in as another ABRicate database would be a category error — it is not a detection resource. The only sensible use is the reverse: after AMRFinderPlus reports a gene in one of the five species, look up whether EFSA already regards that gene as widespread there. That is a manual cross-check, and it is how we treat it.

Recommended use: consult it by hand for B. subtilis, B. velezensis, B. amyloliquefaciens, B. licheniformis and B. paralicheniformis isolates when a regulatory question is actually in play. Note that the general-purpose pipeline behind it is published separately and can be run for any species with enough genomes — that, not the catalogue, is the reusable artefact.

Sourcing: standing, licence, and stated limitations¶

Not all of these carry the same weight, and the difference matters if any of this is quoted in a dossier.

Peer-reviewed EFSA guidance (highest standing)

EFSA Scientific Committee (2025). Guidance on the characterisation of microorganisms in support of the risk assessment of products used in the food chain. EFSA Journal 23(11):e9705. doi:10.2903/j.efsa.2025.9705. Adopted 24 September 2025. Open access (CC BY-ND). This is the current guidance, and its Appendix A lists the parts of the 2018 FEEDAP guidance it supersedes. Definitions quoted above are from its Glossary; the AMR decision tree is §3.2.1.
EFSA BIOHAZ Panel (2023a). Statement on how to interpret the QPS qualification on 'acquired antimicrobial resistance genes'. EFSA Journal 21(10):8323. doi:10.2903/j.efsa.2023.8323. The method for discriminating intrinsic from acquired.
EFSA (2024a). EFSA statement on the requirements for whole genome sequence analysis of microorganisms intentionally used in the food chain. EFSA Journal 22(8):e8912. doi:10.2903/j.efsa.2024.8912. Source of the ≥2-databases requirement and the 80%/70% thresholds.

Technical report (lower standing — not a peer-reviewed panel opinion)

EFSA (2024b). Catalogue of antimicrobial resistance genes in species of Bacillus used to produce food enzymes and feed additives. EFSA Supporting Publication 2024:EN-8931, 33 pp. doi:10.2903/sp.efsa.2024.EN-8931. Approved 3 July 2024. ISSN 2397-8325. "Reproduction is authorised provided the source is acknowledged."

It is an EFSA Technical Report, not an EFSA Journal guidance or opinion — a commissioned technical output. Its standing comes from being cited by the 2025 Guidance (as EFSA, 2024b) as an available implementation. Cite it as a supporting technical report, not as guidance.

Limitations the report states about itself: 95% ANI cannot separate B. velezensis from B. amyloliquefaciens, so their discrimination "only relies on the annotation by RefSeq, which constitutes a limitation in the analysis"; several NCBI accessions used were already superseded at publication.

Limitation we add: the query set is dossier-derived. It reflects what applicants reported, so absence from the catalogue is not evidence that a gene is absent from the species — only that no dossier raised it.

Non-EFSA

Siguier P, Gourbeyre E, Varani A, Ton-Hoang B, Chandler M (2015). Everyman's guide to bacterial insertion sequences. Microbiology Spectrum 3(2):MDNA3-0030-2014. doi:10.1128/microbiolspec.mdna3-0030-2014. Cited for IS size range (0.7–2.5 kb) and abundance.

Verified from software, not literature

The AMRFinderPlus organism list, the --organism refusals and the empty mutation tables were read from this repository's own run outputs and audit files, not from documentation.

Inference, not quoted from any source

That ABRicate-style BLAST is structurally unable to resolve point mutations. This follows from what the tools compute and is not a claim any of the cited documents makes.
That the threshold-brittleness seen in the catalogue transfers to BacFlux's own ABRicate leg. The mechanism is identical (same thresholds, same kind of search), but no source states it about BacFlux.