Chemical vs. drugs (Part 2): How do you discriminate? / More on halicin

This newsletter is part of a series — here are the links to Part 1, (this one is Part 2), Part 3, Part 4, and Part 5. Also relevant is the 5 April 2024 newsletter entitled “48,015 → 0: Antibacterial Discovery Is Hard. Really, Really Hard.“

Dear All (wonkish note alert!),

The recent newsletter on chemicals, drugs, and halicin (link, see also a follow-up newsletter here) generated significant discussion and new insights that I thought I would share in a follow-up newsletter. Our theme in Part 2 is “Trust, but verify!” 

Let’s start by considering comments on the newsletter. A number of you wrote to add support for the idea that the specific antibacterial candidate identified by Stokes et al. (link, SU3227/halicin) is very unlikely to be a safe systemic human medicinal:

• As noted in the first newsletter, c-Jun N-terminal kinases are involved in many critical biological pathways and inhibitors of same (such as SU3227) would need to be highly selective.
• Indeed, the idea of seeking a novel antibacterial from a screen of human drug candidates from other therapy areas must by definition find compounds with effects on some aspect of human biology. Thus, such compounds would have two diverse activities (the antibacterial effect as well as the intrinsic effect on some human system) and hence be at risk for being toxic depending on the nature of the effect on that human system. It might be possible to make derivatives that would lose those effects, but the compound series starts with that liability.
• Finally, and going on a limb a bit as I am not a chemist, the specific molecule (see its structure as the inset to Figure 2H from the paper) contains a nitro group attached to an aromatic ring, something that I’ve been taught is a mutagencity risk (link to a comment by Derek Lowe entitled “I Want A New Nitro” and link to a longer in-depth 2019 paper by Kepali et al. in J Med Chem).

So, SU3227/halicin might be topical medicinal (and they do show that in their paper) but it is unlikely to go further than that for this particular molecule. That said, the comments also noted that the idea of applying machine learning to antibiotic discovery is noteworthy and we all hope that continued work will refine the approach and identify useful new areas of chemical space. 

So, we now come to the deeper question that is the focus of this newsletter: How can we distinguish toxic chemicals from possible new antibacterial drugs? And this may not be easy as we are asking something special of these molecules: we want them to kill a living organism, but to do so selectively!  So, there is a perspective from which we may need to be willing to consider molecules that carry some risk. That noted, the quip “It’s easy to kill bacteria — steam, fire, and bleach are consistently effective — but those aren’t drugs!” provides no help at all with finding a path forward. For example, what about the further 23 candidate antibacterial molecules identified by their screening of the 1.5 billion compound ZINC15 database (Figure 6D)? Those compounds do have antibacterial activity — might they be drugs?

There are no perfect tools here and the problem is complicated by the fact that antibacterial compounds are often not “drug-like” from the perspective of Lipinski’s Rule of 5 (link). Reinforcing the notion that antibiotics often defy typical medicinal chemistry rules, some antibiotics even have reactive elements (e.g., fosfomycin is an epoxide). So, what can be done? Well, we do have some practical suggestions to hand. 

Our initial Chemical-vs-Drug signpost comes from the Instructions to Authors for Antimicrobial Agents and Chemotherapy (AAC, link). As part of guideline #6 on the first page of the instructions, we have:

• “Manuscripts dealing with novel small molecular antimicrobials must provide at least some data showing that the proposed new agents or scaffolds have the potential to become therapeutic agents.
• “Appropriate demonstrations will vary but generally should be some combination of data on physical properties (solubility, protein binding, log P [logarithm of the ratio of the concentrations of un-ionized solutes in solvents]), pharmacological properties (Caco2 predictions of bioavailability, pharmacokinetics in an animal species), or tolerability (mammalian cell toxicity, likelihood of hepatic metabolism, potential for receptor interactions, potential for human ERG liability).
• “Initial presentations of compounds are not expected to address all these areas but rather to show an appropriate initial subset.
• “For example, the first publication of a novel compound or compound series might address selected physical properties plus mammalian cell toxicity.
• “Subsequent publications are expected to add progressively to the proof of the agent’s therapeutic potential.
• “For cationic peptide molecules, an additional expectation is that the molecule demonstrates a profile that is distinct from previously reported similar molecules and that it demonstrates reasonable pharmacokinetics or credible efficacy in an in vivo tissue model of infection (e.g., thigh or lung).”

Our second Chemical-vs-Drug signpost comes from the pragmatic screening approach for novel antibiotics that is being implemented by CO-ADD (https://www.co-add.org/, Community for Open Antimicrobial Drug Discovery), a global open-access screening initiative. Launched in February 2015 and funded for nearly 5 years by the Wellcome Trust supplemented by support from the Institute for Molecular Bioscience (UQ IMB) and The University of Queensland (UQ), CO-ADD provides free antimicrobial screening with no encumbrances (in particular, no impact on IP with the proviso that structures should ultimately be published in CO-ADD’s database) for compound collections from any interested academic researcher..

To date, CO-ADD has screened about 300,000 compounds from 47 countries by following this 3-step process (https://web.archive.org/web/20210417025918/https://www.co-add.org/content/free-screening):

• Step 1: Primary Screening
  • In the primary screening we test compound libraries against key ESKAPE pathogens, E. coli, K. pneumoniae, A. baumannii, P. aeruginosa, S. aureus (MRSA), as well as the fungi C. neoformans and C. albicans.
  • The screen will be at a single concentration (32 µg/mL) and will provide initial activity data to select compounds for further more detailed screening.  
• Step 2: Hit Confirmation
  • Active compounds from the primary screening will be tested in dose response antimicrobial assays to confirm their activity.
  • Active compounds will be screened for adverse effects, such as cytotoxicity, critical micelle concentration and membrane depolarization, as well as for their purity.
  • The Hit confirmation process will evaluate the basic drug qualities of actives, and which of the actives have the best potential for further development.    
• Step 3: Hit Validation
  • The next step of the CO-ADD process is to test the hit against a broader panel of microbes with multidrug-resistant (MDR) and pan-resistant bacterial strains and clinical isolates, with different co-factors (such as serum or lung surfactant).
  • The Hit validation will also include initial ADMET screening, including haemolysis, microsome and plasma stability and protein binding.
  • In addition, structural analogues could be included to evaluate any structure-activity relationship.

If you look at this sequence of tests, it is readily seen to parallel the ideas in AAC’s suggestions for initial data. Testing vs. both bacteria and fungi in Step 1 is a simple way to spot general toxins: if something kills both prokaryotes and eukaryotes, it is probably not a drug! Step 2 does simple tests to look for indiscriminate activity such as being a detergent. Finally, things get serious in Step 3 with a broader look at the activity of the compound series.

As a demonstration of the potentially utility of this approach to mine novel chemical space, CO-ADD has recently published a paper (Frei et al., link) in which they observe that metal-containing compounds have an unusually high validated hit rate. The authors conclude, “This work reveals the vast diversity that metal-containing compounds can bring to antimicrobial research. It is important to raise awareness of these types of compounds for the design of truly novel antibiotics with potential for combating antimicrobial resistance.”

I’m sure that the real chemists in our community can suggest other ideas (e.g., the idea of PAINS, link), but I think that’s enough for a general path. Putting it all together, the core idea is that we always need to remember that “It’s easy to kill bacteria — steam, fire, and bleach are consistently effective” and balance this with the adage “Trust, but verify.” 

Many thanks to all who wrote! All best wishes, –jr

PS: This newsletter has a follow-up here.

John H. Rex, MD | Chief Medical Officer, F2G Ltd. | Operating Partner, Advent Life Sciences. Follow me on Twitter: @JohnRex_NewAbx. See past newsletters and subscribe for the future: https://amr.solutions/blog/