This is a problem.

At the recent ASM 2017 Microbe meeting (the annual meeting for the American Society for Microbiology), there were many groups, academic, governmental, and industry, trying to use bacterial genomic sequence to predict which antibiotics would be effective against a bacterial infection. Lots of posters and talks on the subject. I’ve written about this before, but bacterial genomic sequencing has great potential to be cheaper than current, traditional methods for figuring which antibiotics might work–and depending on the technology–faster. And when someone is ill, speed is critical.

That said, I don’t think it’s quite there yet for many clinical bacteria. One of the key things I gleaned from the meeting is that the key issues are not either the wet lab or informatics–that is, both how the genome sequence is assessed (e.g., whole genome sequencing, targeted PCR/sequencing, microarrays) and what algorithm is used to figure out what those data mean don’t really affect the quality of results.

What does matter is the biological knowledge–the underlying database–used to make predictions. Interestingly, all those fancy machine learning tools don’t really seem to do much better than a cranky old biologist (or even a well-adjusted young one) for many antibiotics in many organisms, especially non-Enterobacteriaceae of middling resistance–which is precisely where we need help. It’s easy to accurately predict resistance and susceptibility to penicillin and its ‘relatives’ when something has a KPC-family carbapenemase (which typically confer resistance to most or all beta-lactams). The hard part is figuring out why a bacterial isolate is resistant to some cephalosporins and not others, as this can be variable depending on the genes (or alleles). And nothing did well in terms of clinical reliability when challenged with Acinetobacter and Pseudomonas (80 percent accuracy doesn’t cut it when faced with a sick patient).

One YOOGE problem is we lack knowledge of what individual resistance genes do. Many haven’t been characterized in the lab, but only identified by similarity (which works well, until it doesn’t*). The methods for characterization haven’t been standardized either, which doesn’t help. Other genes, especially those not discovered recently, haven’t been had their effects assessed against many antibiotics, as the ‘older’ genes were often discovered before some antibiotics were invented.

That last bit returns us to a paper we discussed a while ago–how some variants (alleles) of the KPC carbapenemase also can confer resistance to ceftazidime-avibactam–the short version is that no antibiotics typically used that start with cef-, ceph- or end with -cillin or -penem will work.

What I didn’t note–and is relevant to this post–is that one of the KPC variants that evolved within a patient had already been discovered. In 2008. It had even been assigned a formal name (‘KPC-8’). But, as best as I can tell, no one knew this variant also conferred resistance to ceftazidime-avibactam until 2017.

Admittedly, in 2008, ceftazidime-avibactam wasn’t even an option (I’m not even sure it was a serious concept nine years ago). But this should be humbling–and a reason to do more wet-lab characterization of resistance genes. Even in the genomics era (or are we on to the post-genomics era? I never can tell…), we still need good experimental work.

*These methods are usually pretty good, but not always.