FDA Workshop: Animal Models To Support Antibacterial Development (5 Mar 2020, post-meeting notes)

Dear All (detailed and wonkish note alert … lots of thinking out loud … get some coffee and settle in!),

I wrote previously (link) about FDA’s planned workshop on animal models to support antibacterial development and I was able to listen to the workshop via webcast. As brief context (and please be sure to look at the detailed background notes in the newsletter cited just above), this week’s workshop was a build on a 1 Mar 2017 workshop with both workshops focused on finding ways to generate data from animal models of sufficient quality to support registration of new drugs.

The idea here goes beyond typical PK-PD work used to establish target drug exposure levels and to set interpretive breakpoints. Rather, this second workshop set out to (i) review work that has been done since 2017, and then (ii) explore the question of whether the emerging models were sufficiently validated to become a more central part of the support for approval and labeling. Go here for the meeting’s home page … the agenda is posted — slides and a transcript will follow. 

As background, animal model data that go beyond basic PK-PD work have already started to work their way into some labels. Here are the examples of this that I can find — in each case, the text shown is in Section 12.4 of the FDA label (“Microbiology”), a section found in all antibiotic labels (and please let me know if I’ve missed an example … would happily update the version of this note on the website for future reference):

• The labels for the BL-BLIs (beta-lactam — beta-lactamase inhbitors) combinations of ceftazidime-avibactam (link), meropenem-vaborbactam (link), and imipenem-cilistatin-relebactam (link) all use animal data to document activity of the combination on bacteria resistant to the beta-lactam (BL) alone in a sub-section entitled “Activity against NAME_OF_BL-Nonsusceptible Bacteria in Animal Infection Models”:
  • “Avibactam restored activity of ceftazidime in animal models of infection (e.g. thigh infection, pyelonephritis, systemic infection induced by intraperitoneal injection) caused by ceftazidime non-susceptible beta-lactamase producing (e.g., ESBL, KPC and AmpC) gram-negative bacteria.”
  • “Vaborbactam restored activity of meropenem in animal models of infection (e.g., mouse thigh infection, urinary tract nfection and pulmonary infection) caused by some meropenem nonsusceptible KPC-producing Enterobacteriaceae.”
  • “Relebactam restored activity of imipenem/cilastatin in animal models of infection (e.g., mouse disseminated infection, mouse thigh infection, and mouse pulmonary infection) caused by imipenem-nonsusceptible KPC-producing Enterobacteriaceae and imipenem-nonsusceptible P. aeruginosa (imipenem-nonsusceptible due to production of chromosomal PDC and loss of OprD porin).”
• In the label for the anti-TB drug pretomanid (link), a drug that was approved specifically for use in combination with bedaquiline and linezolid for infections due to very resistant strains of TB, animal model data were used to show that pretomanid added to the effects of the two partner drugs. This labeling parallels the use noted above for the BL-BLIs … the difference is that the BL-BLIs are co-formulated for dosing whereas pretomanid is a standalone approval:
  • “In murine tuberculosis models, the 3-drug combination of pretomanid, bedaquiline, and linezolid reduced bacterial counts in the lungs to a greater extent and resulted in fewer relapses at 2 and 3 months post-therapy compared to 2-drug combinations of pretomanid, bedaquiline, and linezolid.”
• The label for cefiderocol (link) goes one step further than validating combinations by discussing outcomes with humanized dosing in (i) a neutropenic murine thigh model, (ii) an immunocompetent rat pneumonia model, and (iii) an immunocompetent murine UTI model. Here’s an excerpt:
  • “In an immunocompetent rat pneumonia model, reduction in bacterial counts in the lungs of animals infected with K. pneumoniae with MICs ≤8 mcg/mL and P. aeruginosa with MICs ≤ 1mcg/mL was observed using humanized cefiderocol dose.”
• Finally, and a very subtle extension of all of this, the label for micafungin (link, an antifungal indicated for a variety of forms of invasive candidiasis), includes text in Section 8.4 (Use in Pediatrics) noting that rabbit model data suggest that the standard dose is inadequate for CNS candidiasis in very young children:
  • “In a rabbit model of hematogenous Candida meningoencephalitis (HCME) with Candida albicans (minimum inhibitory concentration of 0.125 mcg/mL), a decrease in mean fungal burden in central nervous system (CNS) compartments assessed as the average of combined fungal burden in the cerebrum, cerebellum, and spinal cord relative to untreated controls, was observed with increasing micafungin dosages administered once daily for 7 days.
  • “Data from the rabbit model suggest that a micafungin dose regimen of 4 mg/kg once daily is inadequate to treat meningoencephalitis and that a dose regimen of approximately 10 to 25 mg/kg once daily may be necessary to lower fungal burden in the CNS in pediatric patients younger than 4 months of age [see Microbiology (12.4)].

With that background in mind, let’s go back to the workshop. As the basis for the day’s conversation, most of the workshop was spent on a detailed survey of the elegant work being done with mouse, rabbit, and pig models:

• As you might expect, murine models are used for initial work given their lower cost both in terms of $ and drug product (small animals just need smaller doses, and drug supply is usually limited in early development).
  • But, murine models can only go so far — in particular, it is often hard to mimic human PK.
  • And, there are limits on the endpoints that can be measured.
• On the other hand, the larger animal models permit very detailed biologic measures (imagine a rabbit or pig in an ICU!) during treatment of experimental infections with drugs dosed to produce a very close approximation of human drug exposures.
  • As you might guess, the larger animal models are expensive and not high throughput.
  • And, the models are tricky because secondary infections may set in and animal care issues are challenging (even juvenile pigs are large and physical care during sedation and intubation is challenging).
  • But, the models are very detailed and increasingly do seem to closely approximate human infections.

OK, so what about the question of what else these models could do to support drug approval? The examples from the approved labels cited above show recent use of animal models in the label to (i) demonstrate the value of a partner in a combination, and (ii) link humanized dosing regimens to microbiological effects against strains with specific MICs. Can we go further? Can we extrapolate into management of human infections?

While there was only limited conversation on this topic, the hint that lurked on the edge of the discussion was that labeling based on adequate and well-controlled clinical data on the organisms and infections that are possible to study in the clinic could be extended to include data from high-quality animal studies on rare pathogens, rare pathogen phenotypes, and (perhaps) even rare infections that are not possible to study in the clinic. 

With the caveat that I am now extrapolating, the concrete implementation that would seem to be a logical next step from the labeling above would be to consider an extension of the in vitro List 1 / List 2 data in Section 12.4. As a reminder, Section 12.4 of the FDA labeling for an antibiotic contains subsections reading as follows:

• “NEWDRUG has been shown to be active against most isolates of the following bacteria, both in vitro and in clinical infections:
  • A list of organisms for which we have clinical data for the approved indications follows in the text and is commonly called List 1.
• “The following in vitro data are available, but their clinical significance is unknown. At least 90 percent of the following bacteria exhibit an in vitro minimum inhibitory concentration (MIC) less than or equal to the susceptible breakpoint for NEWDRUG against isolates of similar genus or organism group. However, the efficacy of NEWDRUG in treating clinical infections caused by these bacteria has not been established in adequate and well-controlled clinical trials.”
  • A list of organisms which look susceptible but for which we do NOT have clinical data follows and is commonly called List 2.

As an example of the List 1 / List 2 boilerplate in action, look at any of the labels referenced above … pages 22-23 of the ceftazidime-avibactam label (link), for one example. 

So (and please note that I am now really extrapolating!), could we start having something you might call List 3 where we provide at least partial answers to questions along these lines:

1. For any/all approved indications, what is the predicted clinical efficacy of NEWDRUG vs. MDR/XDR strains of the species that were studied in man?
2. For any/all approved indications, what is the predicted clinical efficacy of NEWDRUG vs. specific species (presumably MDR or XDR) that are too rare to study in man?
3. For infections that cannot readily be studied in man (and hence are not approved indications), what is the predicted clinical efficacy of NEWDRUG vs. relevant pathogens?

In its fullest form, you might imagine a List 3 that summarizes data (i) from validated animal models where (ii) outcomes have been shown to correlate in some fashion with outcomes in man for both a new drug and one or more relevant comparators. The discussion of List 3 might read like this:

• “Although their clinical significance is unknown, the following in vivo data are available from validated animal models in which (i) pathophysiology and outcomes have been correlated to human infections and (ii) drugs have been dosed to produce exposures similar to human exposures.
• “When NewDrug is dosed in infection models that mimic (name of infection goes here) in humans, improved outcomes are seen for infections produced by NewDrug-susceptible strains of xxx, yyy, and zzz.
• “In comparative studies with OldDrug in these infection models, NewDrug improved outcomes relative to OldDrug for OldDrug-reistant strains of xxx, yyy, and zzz.
• “However, the efficacy of NewDrug in treating (these clinical infections) caused by (these bacteria) has not been established in adequate and well-controlled clinical trials.”

Inclusion of any amount of animal model data of this type in the label would be a very helpful step toward giving clinicians hints about the utility of new agents in situations for which clinical data can’t readily be generated. The limitation that we have to recognize is that labeling (at least at present) has to stay withing the bounds of the clinical indications that have been approved on the basis of adequate and well-controlled clinical data.

With this limit in mind, List 3 data addressing the first two questions above seems possible as it’s not hard to see extrapolation to rare species/strains for an approved indication. Indeed, the label for cefiderocol has already taken a step towards this by noting reduced bacterial loads with humanized exposures for List 1 bacteria and indications for which cefidericol has been approved. Stepping up from this to standardized outcomes seems within reach.

On the other hand, animal model data on pneumonia could not appear (at least not a present) in the label of a drug approved only for (say) skin infection or UTI. In many ways, this is the final hurdle as we have so little data outside of the major indications (go here for a list)! As a consequence, much of infectious disease practice is off-label relative to FDA approvals. There is a real tension here — we want high quality data in the FDA-approved label but our patients don’t always stay within label boundaries! Do I tell a patient with osteomyelitis (or meningitis or endocarditis) due to a rare pathogen that “Oh, sorry … you really need to come back when your infection conforms to current product labels”? Of course not: MD stands for Makes Decisions, and the ID community is used to extrapolating.

This need to provide even some hints for settings for which there are no current clinical data is a theme that also emerged at the 18-19 Nov 2019 (link), and you might want to review that prior note. And, the use of rabbit model data in the micafungin label is an example of a possible way forward here: I can only speculate, but FDA’s use of the rabbit data suggests that we can begin to recognize situations where being silent on insights from animal models does more harm than good.

OK, and getting off my soapbox, that’s enough for today. There is a lot to consider here and this workshop was a great next step in what will certainly be an iterative conversation. Overall, I am delighted to see how far we have come. Many thanks to FDA and the speakers for an instructive day!

All best wishes, –jr

PS: With coronavirus in the news, let me point to a detailed analysis posted on the McKinsey & Co. website (link). As you would expect from these consultants, the report is extensive and well-supported. Three cases are analyzed: an optimistic quick recovery (impact limited to Q1), a base case (global slowdown in 2020), and a pessimistic case (recession in 2020). Beyond good leadership and support of global initiatives, their core advice for all scenarios is to protect employees & customers, establish a COVID-19 response team for your business, stress test your financials, and maintain your supply chain. Addendum #1: BioCentury’s Steve Usdin provides an excellent commentary on Covid-19 (link) that summarizes the danger of placing political aspirations ahead of science. Get the facts before the facts get you! Addendum #2: The Johns Hopkins live dashboard (link to dashboard, link to LancetID paper about dashboard) is also an amazing resource.

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/. All opinions are my own.