PCR for antibiotic resistance markers—not the whole story

In this month’s episode, we’re going to revisit a topic which this space has touched on before—not because things have changed much in the space (they haven’t), but because these types of tests are becoming even more commonplace. It’s worth reiterating for the end users of these test types what the associated test approach strengths and weaknesses are, and what caveats to interpretation should be borne in mind when reviewing results.

The tests we’re discussing are any molecular (DNA or RNA sequence based) method which purports to tell whether a specimen associated microbial organism will be susceptible or resistant to a given antibiotic (or usually, a family of structurally and functionally related antibiotics). Usually these approaches are PCR based but as followers of this column will know, there’s more than one way to amplify and/or detect a nucleotide sequence and our considerations here will apply to all of these.

Genotype doesn’t equal phenotype

Let’s start by reminding ourselves what a molecular assay detects (the presence, and perhaps the abundance of, a specific predetermined DNA or RNA sequence; in other words, a genotypic trait) and what defines antibiotic resistance in a microorganism (the ability of the microorganism to grow, more or less unhindered, when exposed to a set concentration of the antibiotic agent; that would be a phenotypic trait). The primary issue in a nutshell is that genotype does not equal phenotype; it equals the potential for a phenotype. What’s clinically relevant is the phenotype, so why do we test for genotype at all?

The answer of course is that phenotypic testing is slow compared to the possible pace of clinical outcomes. Actually isolating the organism(s) from a sample and getting MIC (Minimum Inhibitory Concentration) real phenotypic values to compare against CLSI breakpoints can take days, which the patient may not have. Presumptive application of antibiotics, particularly ones with wide spectrums of action, is rightfully frowned upon as it promotes increased frequency of antibiotic resistance. Molecular tests step into this challenge because they can be done in hours and have a high degree of correlation to phenotypic antibiotic resistance behaviour. Bottom line, and take home message number one for today: they’re fast, and very often correct, and so they provide a rational basis to support immediate therapeutic choices. That’s great but it’s a mistake to not recognize that this is still—by means of being a correlate measure and not a direct measure—an aid to empiric therapy.

If molecular test results are just a correlate to the phenotypic traits we actually care about, what are the ways that this correlation—or our assumptions which form the basis for this correlation—can fall down? Some of the mechanisms by which this happens include:

  • Phenotypes can arise from unrelated mechanisms. For example, one major family of antibiotics is based on what’s known as a “beta-lactam ring” molecule, and includes penicillins and cephalosporins. Some microorganisms develop enzymes (beta lactamases) which are able to break these down before they can act to interfere with cell wall growth; phenotypic resistance to beta-lactams is thus strongly suggested by detection of certain beta-lactamases in an organism. Just because an organism doesn’t have beta-lactamases though doesn’t necessarily mean it’s susceptible to beta-lactams. Other completely different mechanisms such as efflux pumps can lead to exactly the same phenotype.1 Take home message number two: we don’t always test for, or necessarily even know, all of the mechanisms by which an antibiotic resistance phenotype can occur.
  • Genotypic data may be incomplete. Let’s stick with our above example of testing for beta-lactamases. A first issue is that there are a lot of different versions of these genes, with differing sequences—and recall that a strength of molecular testing is its excellent specificity, which can be down to differentiating single nucleotide variations. That’s not so much a strength here as a limitation; a mutation under a priming site can reduce or even abolish amplification and thus detection of the gene. It’s also good to bear in mind that most molecular methods like PCR look at small portions of a gene, not the whole gene. Mutations elsewhere in the gene will not effect detection but might change activity of the protein product, leading for instance to a false impression of antibiotic resistance in a case where the gene is detected but has low activity (i.e. it would fall below MIC breakpoint in phenotype). In fact, mutations not even in the gene but in promotor regions might play a role, either increasing or decreasing phenotypic expression (tests based on expressed RNA levels would be one way to counter this effect, but the relative difficulties in handling RNA as opposed to DNA generally outweigh this benefit).

Of course, the companies producing validated assays in this application space have approaches to mitigate this issue, including selecting target gene regions with the lowest diversity (or biologically, the highest selective pressure against mutation), use of multiple primer sets to cover gene families, and even ongoing surveillance-by-sequencing of pathogenic organisms to detect novel resistance gene variants. In theory it’s possible to design and deploy additional or modified primer sets to cover novel target gene variants, but in practice this is a very slow process to clear regulatory approval; you need to know details of analytical and clinical performance (sensitivity, specificity, PPV, NPV) of the ‘modified’ assay before use, and the cost and time to assess, submit for approval, and release such ‘assay updates’ is generally prohibitive. Take home message number three: most molecular tests examine a fraction of target gene(s) as a surrogate for the whole, which may be misleading. Take home message number four: microorganisms trust in crowdsourced solutions to selective pressures and can develop and deploy selected variations rapidly and without bureaucracy or rules. That’s a challenge to keep up with, and molecular tests only look for what we know to look for.

  • Mixed cultures can be very misleading. We discussed above how molecular methods’ specificity is not always a benefit in this space; that also turns out to be true of its intrinsic sensitivity. If a clinical sample contains more than one organism—perhaps trace quantities of something else, which doesn’t necessarily even need to be viable—then molecular methods can come back with multiple organisms detected and multiple antibiotic resistance markers, but without ways to determine associations. In other words, which organism has which resistance markers? This can be handled to some extent by use of quantitative methods as opposed to purely qualitative detection, and then pairing up organism “identity signal strengths” with “antibiotic resistance marker strengths” to see which sets match. (In reality it’s not quite that simple, as issues like relative copy numbers of the antibiotic resistance gene(s) and the organism marker(s) come into play, their relative amplification efficiencies, and the like; also, one might get cases with signals of similar amplitude for all organisms and markers. In at least some cases though this approach can help correlate resistance markers to specific organisms detected. Another tool in unravelling the multi organism/multi resistance marker scenario can be empiric data, in as much as some organisms haven’t been previously observed to carry certain types of resistance markers, or do so in only rare cases; if you’re lucky, you may be able to assign a mix of organisms and resistances out of a mixture by this with an acceptably high degree of probability. Finally, there are for some organisms and some markers particular molecular assays which can unequivocally pair a resistance marker to its host organism—most readily done in examples where the resistance marker, if present, will reside at a known, genetically conserved organism genome locus and the test can be designed to flank both organism and resistance marker genetic components. Take home message number five: beware of specimens positive for multiple possible pathogens and multiple antibiotic resistance markers. Some assumptions are probably being made to decipher that mixture.

What does all of this mean for application of molecular testing in antibiotic susceptibility? It means that it remains a powerful, fast tool for helping to direct empiric therapy choices, and its application is greatly preferable to blind empiric therapy (that is, based purely on statistical past phenotypic susceptibility traits for a given organism). It also means however that it’s far from being a fire-and-forget solution to antimicrobial therapy choices. Emerging methods like mass spectrometry may prove to be even more useful here. Recall that this isn’t a “molecular method” by traditional definition of analyzing DNA or RNA – rather it has the ability to directly query a sample for the presence of characteristic fragments of key enzymes or intermediary metabolites which can relate to a particular antibiotic resistance mechanism. This then is one step closer to examination of final phenotype than is molecular testing, although its interpretation too may involve some of the same assumptions (and potential for being mislead) described above.

Clinical assessment of response remains key

While molecular susceptibility testing is a fantastic tool, it’s important for end users of these technologies not to be lulled into a false sense of complacency that they are somehow infallible. Clinical assessment of patient response and (where possible) actual traditional bench microbiology—manual or automated—in determining real phenotypic resistance is still the essential final step in evaluating whether the myriad assumptions inherent in taking a genotypic result and extrapolating to a phenotype hold true in each and every case. In the end of the day, what matters to the patient is the end result. In cases where clinical response isn’t what was expected, it’s good to bear in mind all of the chain of assumptions which underlay conversion of molecular data to therapy choices. While the examples covered here aren’t an exhaustive list of what those are, they’re at least some of the more likely points for discrepancy to occur.


  1. Pages JM, Lavigne JP, Leflon-Guibout V, Marcon E, Bert F, Noussair L, Nicolas-Chanoine MH. Efflux pump, the masked side of beta-lactam resistance in Klebsiella pneumoniae clinical isolates. PLoS One. 4(3):e4817 (2009)