Issues for next release of spec/guidance

[katholt]

• if there are different breakpoints for multiple sites, how to code? e.g. cefuroxime for E. coli has breakpoints for iv and oral. current guidance says to code 2 rules and indicate in the note, but I think we need a proper format for this - e.g. a separate column like 'breakpoint condition'?

• if there is little/no evidence to support an interpretation of an acquired gene that has been found in a species, what to do? philosophically it is probably better to encode a rule with explicit evidence grade (weak) and limitations ('lacks evidence for this species', or 'lacks evidence for this genus') rather than no rule (in which case the choice of default interpretation is left to downstream reporting code)

• drug names should match CARD, but without the word 'antibiotic' at the end... this means lowercase, class terms are not pluralised. e.g. "penam" or "third-generation cephalosporin" not "penams" or "third-generation cephalosporins"; and "fluoroquinolone" not "fluoroquinolone antibiotic" => add into validation script

• addition of fourth option in clinical category, 'S (inducible R)' to cover e.g. inducible beta-lactamases in Enterobacter?

• how to encode clinical category for a core gene when there is an expert rule to report as 'R', but it is not in the expected resistance list? e.g. aac(6')-Iaa in Salmonella enterica... S. enterica isolates typically show MICs below the Enterobacterales breakpoints for these drugs, but clinical failure with these drugs has led to expert rules saying that Salmonella should always be reported R regardless of assay result.

Obviously the rule for this core gene should encode phenotype=wildtype, but the question is what should the clinical category be? 'S' would be consistent with the MIC but would contradict the expert rules, whereas R would contradict the MIC but be consistent with expert rules. This boils down to a philosophical question as to how should the clinical category be defined in AMRrules?

I would propose the best path is to follow the expert rules and define as 'wt R'... since the ultimate purpose of 'clinical category' is to predict treatment response NOT to predict in vitro assay result. (If the in vitro assay is not predicting clinical response well, there is probably a reason, such as this gene and/or the mdsAB efflux pump kicking in during infection and causing clinical failure. If we knew this, it would strengthen the rule and the call of R, but for now there is a lack of evidence on this).

Proposed solution: would be to define this as wt R for aminoglycosides. The breakpoint would be 'not applicable' as its a class, and the breakpoint standard would be 'EUCAST Salmonella Expert Rules v3.2 (2019)'. Essentially we would be treating these expert rules the same as Expected resistances. It should be 'aminoglycosides' as a class, rather than the specific drugs, because that is how the expert rules are expressed.

[katholt]

• how to encode essentially null rules for efflux genes? there are many in the databases, including some that are core so definitely need a rule. some are assigned in refgene/amrfinderplus to class 'efflux' but others are assigned to a specific class (e.g. oqxA to quinolones) and others are assigned to efflux in amrfinderplus but a specific drug in CARD (e.g. amvA in Acinetobacter), this seems to be more dependent on how they were initially reported in the publication and/or adopted by the community, rather than any consistent criteria for annotation. default interpretation for genes without rules will depend on the DB, and the annotations are not consistent between DBs/genotypers, e.g. if we assume no interpretation is needed for genotypes annotated with class 'efflux' then amvA would have no interpretation if using amrfp but assume erythormycin resistance if using CARD.
So it seems we should aim to encode rules for these.
But for what drug? Ideally each rule would be for a specific drug/class. So if the efflux gene has been reported as associated with a specific drug/class, or is annotated as such in refgene or CARD, then encode the rule as WT S for that drug. Otherwise can we leave the drug blank? Should we allow drug class = efflux? This doesn't really make sense. Do we need a separate field for efflux?

[katholt]

• how to specify accessions for nucleotide-based genetic variants? e.g. 23S or promoter variants. as these are relative to a nucleotide sequence not protein sequence, there is no nodeID, HMM, or refseq/genbank protein accession.

refgene specifies these as 4 fields: refseq nucleotide, refseq strand, refseq start, refseq stop - e.g. https://www.ncbi.nlm.nih.gov/pathogens/refgene/#ampC_C-11T

In AMRfp output, it is identified by the accession of closest sequence field, like this: "NZ_CP041538.1:1149245-1149489" ie refseq nucleotide:refseq start-refseq stop if strand=="+" and presumably refseq nucleotide:refseq stop-refseq start if reverse strand

So we can add a nucleotide column amongst the set of possible NCBI unique identifiers, and adopt this format for encoding the nucleotide accession and coordinates to uniquely identify genes like 23S.

This has been added to spec v0.6 draft

[katholt]

included in draft spec v0.6:

• new column for 'nucleotide accession'
• new column for 'breakpoint condition'
• addition of fourth option in clinical category, 'S (inducible R)'
• guidance on logical expressions for combinations (need to reference a specific reference syntax)
• noted expert rules as example of breakpoint standard that can be used (along with Expected resistance) to define a wt R
• added new ECO term "ECO:0006404 experimentally evolved mutant phenotypic evidence"

[katholt]

• removed fourth option in clinical category, 'S (inducible R)'
• decided to go with & | for logical expression syntax, as fewer characters and matches the TB grammar
• added guidance on how to represent coordinates along with nucleotide accession, in the 'nucleotide accession' field

[katholt]

Clarify guidance that if there is no formal breakpoint or ECOFF defined, you should

• use a breakpoint from another source, e.g. a paper, if you trust it (enter the breakpoint in breakpoint and indicate the source e.g. PMID:xxx in the field breakpoint standard
• indicate not available in breakpoint and leave breakpoint standard blank

[katholt]

• Context field - consider expanding to differentiate when mutations are found in a core gene copy vs acquired (e.g. plasmid-borne) gene copy [request from Acinetobacter group]
• Consider adding a field to explicitly indicate gene family [request from Acinetobacter group]
• How to encode rules for inhibitors? (could allow the inhibitor in drug field?) or disinfectants?

1. allow in 'drug' or separate column?
2. change column name to 'agent' and allow drug, inhibitor, or disinfectant?

[katholt]

re efflux rules - this is a live problem for Acinetobacter baumannii amvA. Interim solution is to encode with drug=blank and phenotype=wildtype, with no drug or clinical category.

[katholt]

Staph group noted they have unexplained expected resistance (ceftazidime) and coded this as gene=unknown, context=core, interpretation = ceftazidime WT R

Should we explicitly allow this (rules with unknown genes) for such cases? Should it also apply to expert rules to report as R, in addition to Expected phenotype R?

The reason for doing this within AMRrules would be to provide a more comprehensive catalog of the known mechanisms behind resistance, and assist with downstream phenotype prediction/inference. Also to provide a human-readable, curated explanation for why this is not yet known, which is useful for the community and also helps to explicitly highlight where knowledge gaps exist.

But there is a question of consistency - if we are to do this, then all AMRrules sets for all organisms should include rules for every expected resistance, even if there is no core gene to sensibly encode even a weak/null rule for.

Alternatively, one could argue that this kind of rule belongs in a downstream reporting logic, rather than in AMRrules. The benefit of that approach is that one can lookup separately the current expected/expert rules and apply them, rather than relying on AMRrules to keep up-to-date on this. If we are to include this in AMRrules, we would need to ensure that the rules are 'complete' in this sense, with every new update of expected phenotypes.

For now, as we want to provide an interpreted summary of S/I/R for relevant drugs, it does make sense to include these rules somehow in the internal AMRrules interpretation code logic, either through encoding in AMRrules (which has the benefit of transparency) or through consulting a separate list (accessible via the AMR package).

[natacha-couto]

During the last Staph subgroup meeting, a key question arose, should we encode rules for AMR genes not yet detected in Staph but that are located in MGEs? We know from the past, that certain genes (or variants of certain genes) are found in Enterococcus and only later are transferred to Staph (i.e., van genes). Should we preemptively encode rules for genes or variants of genes that we suspect might be acquired by members of the Staph genus?

[natacha-couto]

How to assign species specific rules if GTDB assigns all genomes to one species?
An issue with GTDB has been identified by the Bordetella subgroup because it doesn’t separate the 3 Bordetella species clearly. It might be because B. pertussis and B. parapertussis both evolved independently out of B. bronchiseptica (https://journals.plos.org/plospathogens/article?id=10.1371/journal.ppat.0010045).

The B. bronchiseptica strain RB50 (https://gtdb.ecogenomic.org/genome?gid=GCF_000195675.1) and the B. parapertussis strain 12822 (https://gtdb.ecogenomic.org/genome?gid=GCF_000195695.1) are both well known B. bronchiseptica and B. parapertussis reference strains respectively (https://www.nature.com/articles/ng1227) but have both been assigned as B. pertussis by GTDB.

B. bronchiseptica has a much larger genome size than pertussis (5mb vs 4mb) and parapertussis and there are also unique genes that separate B. parapertussis from B. pertussis. The issue with assigning the resistance for both bronchi and parapertussis as s__Bordetella pertussis would be misleading as they cause different clinical diseases/disease severity and have different host ranges. The resistance profile especially for bronchi is quite different to pertussis.

This issue needs to be discussed with the wider group to see if other groups have observed something similar with their species.