How To Optimally Combine Genotypic and Phenotypic Drug Susceptibility Testing Methods for Pyrazinamide

False-susceptible phenotypic drug-susceptibility testing (DST) results for pyrazinamide due to mutations with MICs close to the critical concentration (CC) confound the classification of pncA resistance mutations, leading to an underestimate of the specificity of genotypic DST. This could be minimized by basing treatment decisions on well-understood mutations and by adopting an area of technical uncertainty for phenotypic DST rather than only testing the CC, as is current practice for the Mycobacterium tuberculosis complex.


Rules for classification
Initial classification: association data from Miotto et al. 2017 It was not the purpose of this study to update the systematic review of categorical pDST data that resulted in Miotto et al. by rerunning the original search terms and including all studies published since the 30th December 2015 (1). Rather, we wanted to explore how the classification from Miotto et al. based on the corrected "interpretative best confidence values" (iBCVs) could be refined in light of the most important studies in this area, including those with types of data that were beyond the scope of the original review ( Figure S1). We, therefore, limited this study to the 636 variants originally identified in Miotto et al. and converted them to groups A-E according to Table S1 to yield the initial classification (see below for a discussion of synonymous mutations (1)).  Figure S1: Overview of additional data considered to refine the initial classification.
Second classification: association data from ReSeqTB We considered additional association data from the ReSeqTB platform (ReSeqTB public platform (http://www.reseqtb.org/, PFF_2019-01-24.csv, last accessed on 4th February 2019) (7). Whenever the corrected iBCV was from this dataset was more significant than the one from Miotto et al. (e.g. C138R was originally indeterminate whereas it was a high confidence mutation in ReSeqTB), the original result was updated to yield an updated corrected iBCV and the classification was updated accordingly (i.e. based on the same logic used for the initial classification). Moreover, we upgraded all in-frame coding indels from group C to group B as these likely abolish the function of PncA. Finally, we downgraded all mutations that were more than 40 nucleotides upstream of pncA from group C to group D as these are unlikely to confer resistance.
Third classification: direct and indirect MIC data Group C-E mutations in the second classification were upgraded to group B if these mutations might confer MICs close to the CC (i.e. if they were tested at least six times and yielded proportions of resistant and susceptible pDST results that were both above 25%). Moreover, all group C mutations that were tested at least five times and consistently yielded MICs50 µg/mL were moved to group D.
Fourth classification: homoplasy data Homoplastic group C-E mutations in the third classification were upgraded to group B as the independent evolution of the same mutation is typically as sign of positive selection (6). Homoplasic group B mutations in the third classification that had been group B or E mutations in the second classification and had MICs close to the CC were upgraded to group A as additional categorical pDST would not be helpful for these mutations (e.g. T47A).
Final classification: additional experimental evidence To err on the side of caution, strains with a single group A mutation in the fourth classification were downgraded to group B if these were regarded as neutral in at least one of the additional studies with experimental data or yielded inconsistent results in Yadon et al. (8). Mutations were not downgraded if their MICs were likely close to the CC (e.g. T47A). If one mutation was associated with resistance and the other one was neutral (e.g. for S18P+P54L in Walker et al. (9)), this combination was not downgraded. If a study provided information for only one of multiple mutations, the combination of mutations was downgraded, unless one of two was a LoF mutation (e.g. W119Stop+T168I was not downgraded, even though T168I was neutral based on Walker et al. (9)). Strains with a single group C mutation in the fourth classification were upgraded to group B if the mutation in question was associated with resistance in most studies with additional experimental evidence. The same rule was applied to strains with multiple mutations provided that these criteria were met for at least one of the mutations for each combination (e.g. g-71del+T100P was reclassified because of the evidence for T100P in Miotto et al. and Farhat et al. (1, 10)). By contrast, if a group C mutation (or all mutations in strains with multiple mutations) were consistently considered to be neutral and evidence was available from at least two studies (e.g. V45A), the mutation or combination of mutations was moved to group D.

Open questions
Synonymous mutations As part of our first expert rule, we assumed that synonymous mutations are neutral and that different nucleotide changes that result in the same amino acid change have the same effect. We found that the S65S (tcc/tcT) mutation, which is regarded as maker for a subgroup of lineage 3 (6) (1)). If confirmed, this would mean that our expert rule does not always apply, which might be due to codon usage effects.

Section: Categorical phenotypic drug susceptibility testing (red)
This section reports all pDST results for the genetic variants observed in pncA using mostly Sanger sequencing based on Supplementary Table S7 (14).  N of isolates ≤CC (100 µg/mL) (column AP): total number of isolates with MICs below or equal to the CC of 100 µg/mL.  N of isolates >CC (100 µg/mL) (column AQ): total number of isolates with MICs above the CC of 100 µg/mL.  Typing information (lineages): lineages assigned according to (7).  Homoplasy y/n: "y" shows that the mutation is homoplasic, whereas "n" indicates the mutation is not. NB: ReSeqTB accepts any WGS data, irrespective of the availability of associated pDST result. Therefore, the number of isolates with typing information may differ from the number shown in column Y.

Section: For comparison (grey)
This section provides other notable classifications for comparison.
-NIPRO V1 benign mutations (Ando 2010, PMID 19832709) (column BG): The Genoscholar PZA-TB ver. I (Nipro Corporation, Japan) was designed not to detect these mutations to avoid false resistance because they were considered to be neutral (16,17). This assay has been replaced by the Genoscholar PZA-TB ver. II kit, which is only designed to avoid false resistance due to three synonymous mutations (i.e. G60G, S65S and T142T) (18). Control and Prevention strain panel and two studies that had no strains in common with our study (8,9). This interpretation is used by Public Health England for its routine WGS-based diagnostic service (this information is current as of March 29 th , 2019). -Carter 2019 (BioRxiv doi.org/10.1101/518142) (Column BJ): This interpretation relies on a structure-based machine learning approach (14).