Can We Optimize Phage Therapy Dosing?

0:05 Okay, today it's a great pleasure to have Gauri Rao from the University of Southern California, to talk to us today about PK/PD analysis and phages. So why don't you start by telling us about your background and how that led to an interest in phage therapy?

0:24 Yeah, thank you very much, Joe, for this opportunity to talk about PK/PD and phage, especially application to phage therapy. So I am a trained computer engineer. I also have a PharmD degree from the University at Buffalo, where I also did an additional masters in pharmacometrics. And my postdoctoral training that was also funded by Hoffmann-La Roche was using oseltamivir PK/PD data and looking into viral kinetics and host response.

0:57 So that was my first foray into a lot of what we call as quantitative systems pharmacology. Or looking at the interaction between the drug and the in this case, the viral pathogen, but also the host response. And then as I got my faculty position, my own wet lab was very focused on small molecule antibiotics, looking at their PK/PD using lot of in vitro in vivo systems and modeling.

1:28 And then almost six or seven years ago, I got a request for looking at phage data and looking at phage PK/PD data. And if I could kind of start modeling this data to help out this investigator. And so that got me really interested. And you know, I had heard about phage, but not really focused in that area. And as I started looking into it, the PK and the PD seemed very, very interesting, very different from small molecule PK/PD.

2:00 That I had done until that point, and it brought back all the viral modeling I had done. And so I was really excited about kind of applying a lot of my knowledge from my postdoctoral training. And also, you know, we don't get very often biologic kind of drugs in the anti-infective space. So I was really interested and that's how I got that got started with just kind of modeling the PK/PD of phage therapy.

2:32 And then I never looked back, and now, you know, my lab is very heavily focused on phage therapy.

2:42 Great, and maybe since the audience is often a lot of us phage people who might not be completely familiar with PK/PD analysis. Can you tell us exactly what it is and how it's used to help figure out how to come up with treatment regimens with traditional antibiotics and bacteria?

3:03 Yeah, absolutely, thank you for the question. As you know, I think a lot of our focus with any new traditional molecules is what kind. How can we dose so that we can reach a certain threshold to ensure that the exposure is adequate to help clear that bacterial infection? And a lot of this has been focused on earlier work done by Bill Craig and other folks in the field, where they looked at dose ranging studies.

3:36 Where they, you know, they try to figure out the relationship between the antibiotic exposure and what we refer to as minimum inhibitory concentration or MIC. So you have different PK/PD indices, basically kind of connecting exposure to response. This has been very relevant because that kind of helped us define targets. Whether a particular drug is exposure driven or for beta lactams, is a time above mic that you would like to see, or, you know, would the peak concentration be more informative about what kind of response you will see?

4:16 And so again, these are assuming that the MIC is going to remain the same over the course of treatment. And these, of course, do not take into account things like the exposure that you will achieve at the target site. So you do have, you know, in vivo studies that might be using the thigh infection model or bloodstream or lung infection to kind of get closer to ensure that the exposure to response relationship is really the exposure that you can achieve at the target site.

4:51 And so when you look at lot more modeling based approaches, you take that whole time course of exposure into account. It can be richly sampled, sparsely sampled, it could be an in vitro system, an in vivo system. But then we try to understand if that exposure would result in the response we want to see at the target site. And kind of to move it further. You can then start incorporating the variability in the infecting pathogen.

5:22 So even though we have an MIC determined for a certain pathogen, that might not be the MIC of the actual infecting pathogen. You might have different resistance, you know, kind of expressed in that pathogen, or it might be highly virulent and the host response might be very different. So, based on the baseline immune response in that patient, the amount of exposure the patient might require could be higher or lower.

5:55 If it is a renally cleared drug, you know, you may have to change the dosing. So, as we move towards more modeling based platform, it enables us to kind of consider far more number of variables to kind of get closer to the dosing. That would be appropriate for the patient. And it also enables us to do then what we call as therapeutic drug monitoring over the course of treatment.

6:22 I see, You just used the term therapeutic drug modeling. Can you expand a little bit on what that is?

6:36 Yeah, so, you know, in lot of our modeling approaches, and even when we know the target, we might calculate a dose for a particular patient. And then what we want to understand is when that dose, or that dosing regimen, is administered to the patient. If we had the opportunity to understand the PK, or what we call as get some blood samples and understand, you know, how is this patient responding to the drug?

7:05 And what kind of levels are they achieving? Then it tells the treat treating physician and the team if that dose is appropriate, or I, if that dose is not going to help the patient achieve the exposure that we expect to kind of be effective against that pathogen. So therapeutic drug monitoring you might be very familiar is done for, you know, drugs like aminoglycoside or vancomycin, so, you know, e- especially drugs.

7:34 What we call has that have a narrow therapeutic index. Where, you know, the window between what is efficacious concentration and what could be toxic is very close to each other. So we are very concerned that these patients, you know, don't go into the toxic zone.

7:52 right? Yeah, and I guess I'm just gonna dig in a little deeper. When you have a patient in a clinic and you're monitoring them, I imagine it's a combination of symptoms being relieved and colony-forming units of the pathogen going down. How is that balanced in the clinic?

8:23 You're right about that. So it is a combination of things. So absolutely, the physician is at the front because they have a sense of, you know, how is the patient responding? And in many cases, for example, if it is a lung infection or a deep seated infection, we may not have access to understand how the bacterial load is going down. But I think the patient symptomology, looking at lot of the other biomarkers, or looking at the, you know, lab works is a way that the physician uses in addition to the drug levels.

8:57 If it is possible, because, you know, the hospital or the clinic has to have a validated method of quantifying that drug in the sample.

9:09 great, and so one of the things that I think, you know, the phage community, maybe struggles with. I mean, I know when I once asked Chip Schooley how he decided how much phage to give Tom Patterson. He said, Well, we gave as much as we could without going over the endotoxin limit. And then, you know, Chip quickly mentioned that that's not the way he likes to do things.

9:35 But, you know, given the situation that Tom was in, you kind of did what you could do. And that sort of made sense. So that's a long way of asking you, what do you think PK/PD can bring to phage determining phage dosing regimens?

9:54 Yeah, no, I think that is the key question that all the regulators and everybody in the field have, right? About how do you dose phage therapy? And maybe just to give you a little more context, which you probably already know, but just to put it in the context. And how PK/PD modeling comes into the picture is first, you have two entities, right? So the phage and the bacteria, which are both growing.

10:23 And in traditional antibiotic PK, you call it as, you know, stationary PK, where once you dose the drug, the exposure is stationary and it's the bacteria that is responding to the drug. In this case, your therapeutic agent is also changing in response to the bacterial level. And that is, you know, let's say we, have a very a lytic phage that does not express any resistance mechanisms.

10:51 We have genomically characterized well, then you have two options either you dose it very high and hope for, you know, it will just kill all the bacteria at one time, so more like a passive system, or what you would ideally like is active. Where, you know, the phage can then amplify, self amplify, which is one of its key characteristics, and then continue to ensure that the bacterial population is suppressed.

11:21 And I think the dosing challenge is you have to have a guesstimate of the potential bacterial load at the target site because you need the bacteria above a so called proliferation threshold so that you have adequate bacteria. And then you have to have the phage above the inundation threshold. So that, you know, there is enough phage that can then adsorb to the bacteria and do its job.

11:49 Now, that being said, the phage also has to have characteristics where they can adsorb to that bacteria, so that's part of the selection process. But when once it injects its genomic material into the bacteria, is it very efficient in replicating? And what is the burst size? So, you know, what is the latent period, what is the burst size? How do you dose the antibiotic relative to the phage?

12:19 do you dose it together? do you dose it before after? So, you know, so those are the questions that come to mind. And then the last part I would say is how is the host? Because a lot of the phage clearance mechanisms are host driven, so what is the immune response within the host? Because that is also going to determine how the phage is going to cleared, and what could be potentially the bacterial load based on the host response.

12:49 And I think what we are trying to move towards, which, you know, I'm happy to kind of expand on, is kind of developing these models. Now, when I call on the phage exposure side, we are using physiologically based pharmacokinetic models, or they are referred to as PBPK where they take into account the specific phage clearance mechanisms like macrophages and where they may occur. And kind of then give you a more understanding of the phage exposure.

13:20 And the advantage of using these PBPK or physiologically based PK models is we can scale them from mice to rats to humans. We have done that in our recent publication. So have others. And so then that helps us to connect the exposure with the more mechanistic model that is kind of relating that exposure to how this would impact the bacteria. I'm happy to elaborate on any of these parts.

13:50 Going back to traditional antibiotics, one of the tools people often use is hollow fiber. Could you remind people what hollow fiber is? I know it's a powerful tool, at least in the early days of designing dosing regimens for traditional antibiotics.

14:25 And is that then, can you say, is it useful for do early stages with phages, or is do you have to do something different with phages?

14:34 So I think maybe just to explain what the hollow fiber system is. So, you know, it's very similar to a dialysis cartridge, because maybe people have a better image of the dialysis cartridge. The biggest advantage of the hollow fiber, as opposed to any animal model, is that you can pretty much simulate any drug exposure you want for a single drug, two drug combination, three drug combination.

15:02 You can take into account the renal function of the patient, you can take into account whether the infection site is the bone, it is the bloodstream, it is the lung, so it's a very powerful tool. You can take as many samples as you want for drug quantification, for bacterial load quantification to do genomic analysis. Hollow fiber systems can be run over seven days, fourteen days. They have them for TB studies over a month and months.

15:33 These are the kinds of things that we cannot do with the animal model. So that's a big advantage of the hollow-fiber system. The second part is that the bacteria are trapped in those fibers and the only way the bacteria can get cleared from the system is if the treatment is effective. So if the treatment is not effective, you will not see that response.

15:58 Unlike maybe people who are familiar with one compartment system, where based on the flow rate, the bacteria, you lose a certain amount of bacteria inoculum over time. And so, if you have a drug like meropenem with a very short half life of like one hour or two hours, depending on the flow rate, you might lose a lot of bacteria just with the flow rate. And you then cannot attribute that to the antibiotic, but in the hollow fiber, it can be a drug with a very short half life, very long half life, it doesn't matter.

16:32 You can simulate the exact PK, and I think now when you move to phage, there have been. There has been a study in the US that has been published in the hollow fiber and a study from France using the hollow fiber system. Some of the challenges are between the size of the phage and the pore sizes of the fibers and kind of. How do you address that, and how do you make sure that the phage and the bacteria can penetrate?

17:03 And is the system set up for having the phage be cleared from that system? Because, you know, you have the bacteria trapped in the fiber, so the phage interact. If there is a burst of phage from the bacteria that are lysed, are they cleared adequately? And so what? Going back to your question? We have done some preliminary studies in my group with the hollow fiber, and we have found that challenging to explain the clearance of the phage.

17:33 And a system that we have kind of moved towards is the CDC bioreactor system, where you don't have the phages trapped in the hollow fiber. And the distribution is not so much of a concern, but you have the phages interacting freely with the bacteria in within this bioreactor. And again, you can simulate the same drug exposure, drug clearance that you want. And in addition, you have these rods that are suspended within the bioreactor that have, you know, coupon, depending upon whatever material you want to simulate, where a biofilm could be formed.

18:13 So, at the same time, you are understanding the penetration of the phage in the planktonic cells and their impact on biofilm bacteria. So we have kind of moved towards that. But coming back to your question about the hollow fiber for the phage, I think more work needs to be done and we need to improve our understanding at the present time before we rule that in or out as a system.

18:38 I see, so maybe this next question will ask, will allow you to expand on a few things. But so I if someone comes to you, I don't know how often this happens. But you know, they have a patient and they're in, let's say they have a lung infection, and they're thinking about, and they found phages that kill the pathogen. What are the things that are most important for you to know?

19:10 If you're gonna advise a physician on the optimal dose to give a patient?

19:19 Yeah, no, thank you for the question. So one of the things we we've had these cases come not very often, and I will also be very clear that these are not critical cases. So these are cases that we have received where the patient has been refractory to traditional antibiotic therapy. And because most centers, including Pittsburgh that I'm involved with, they are all focused on compassionate use, because that's currently what we are allowed to.

19:47 So one of the pathways of workflow that we follow is that has been published by Andrea Kwa and her group from A*STAR in Singapore. Where we look at the phages that can target that particular infecting pathogen. This is ensuring that genomic characterization has been done. There is no resistant, you know, antimicrobial resistant genes expressed. These are lytic phages.

20:18 We then look at what we call as time kill curves. We want to understand if single two phage, three phage combination, how effective is it against this bacteria? And if we are lucky and we have a number of phages that can infect that pathogen, then we have the ability to design a cocktail that is very effective. The next, we also look at the antibiotics that we can pair it with in the same time kill, so if we can look at multiple antibiotics.

20:50 We then start looking at how low can we go on the dose to see that we get sustained activity. So this is something that we do over approximately two to three days. And we then at the same time move it into a biofilm plate assay to ensure that the selected cocktail also has activity against. And depending on how serious the patient is and how quickly does the clinic need this?

21:18 we also move it into a biofilm model to also understand over dynamic exposure for the phage and the antibiotic, will there be adequate response and can we predict that? And the last part I will say is that the phage exposure that I'm talking to is since we have an established now PBPK model, that works very well. We try to understand based on the phage family, based on the tail characteristics, because that dictates the clearance of the phage, we simulate that exposure.

21:53 And so using all this knowledge together, that kind of gives us the dosing frequency and the dose, for the patient.

22:03 And just, you know, one of the things that you touched on earlier, is the sort of, for lack of a better term, the concentration of bacteria in the patient's lung is that a determining factor? Do you want a higher dose when there's a higher? With antibiotics, my impression is that it doesn't really matter how many pathogens are there, but with phages, maybe that's different.

22:40 Is it? and if so, how?

22:45 Yeah, so the multiplicity of infection, right? The ratio of the phage concentration or the phage burden to the bacterial burden, right? And MOI is a lab construct, right? We use that to understand that dynamic relationship. So what you're kind of alluding to is, let's say we know that the bacterial concentration within this patient's lung is X at the time of treatment initiation. But then it is going to go through this dynamic relationship with the phage load, depending on how all the other things happening within the patient.

23:22 And then at the end, we want to understand whether the phage concentration that was dosed is effective. The biofilm model I'm talking about takes that into consideration. In our time-kill studies, we do look at different MOIs to kind of understand, you know, would the phage dose need to be higher or lower? But kind of going back.

23:50 We have seen that anything above 10^9 or 10^10 pushing the higher load, at least with the data we have at the present time, is not going to be that different. As long as that cocktail is very effective, or the individual phage is very effective and efficient. And the other part is, even if I came up with a high concentration, I think the challenges of manufacturing that high titer and ensuring that the titer is maintained.

24:21 You know, if this patient has to get this dose q twelve every day, you know, can we ship those kinds of cocktails, or phage at that titer, to the clinic and ensure the titer is maintained? There is another part of the problem. So our focus is not to just push the concentration or the titer very high, because many times it's really not clinically feasible and, again, it's very dependent on the phage itself.

24:51 I guess there are a couple of issues. One is, do you need to go that high? And is there a downside to going that high? There might be some toxicity. And I know of something that has happened in at least one or two cases.

25:12 There have been antibodies, neut- phage neutralizing antibodies generated. And I don't know if that is. It's possible to sort of do modeling to have an have any idea of whether you're likely to have a do your dose is likely to do that

25:30 Yeah, so we have actually just started working on that with the Mycobacterium abscessus phages, because we see more of that in that particular. So, based on the duration of treatment, what we see is that with each additional dose, you know, you are not getting that increased activity. And so to kind of understand if, you know, is it because the titer is going down because of increased host clearance, or because there are antibodies kind of being generated and those are making the phage inactive?

26:07 So we are kind of looking into that. So, yes, these models are very well capable of that.

26:14 great. So I guess we've sort of touched on this. But do you want to add anything else about what have been the biggest challenges you see going forward? For applying PK/PD analysis to phage therapy,

26:36 Yeah, I think, I mean, it is a challenge that we had with small molecules, small molecule antibiotics, too. But I think because we had newer antibiotics always coming, and maybe now that we are not seeing so many novel antibiotics. I think couple of things to consider is. We do not take the variability in the infecting pathogen into consideration. That's a less of a concern until we hit very highly resistant pathogens.

27:06 Second is we do not take the host response into consideration adequately, right? These are something which are not a part of our routine evaluation while dosing, especially new antibiotics, and I think with phages. My hope is we start kind of considering also the different phages that come into play. And because as we as this field is kind of now ballooning, and it's kind of it's getting a lot of momentum and attention.

27:38 We probably need to kind of share the genomics of these phages, because that plays a big role in how we understand their interaction with the bacteria. We also, I think, should be sharing a lot of the knowledge that we get, not just from individual patients who are either responded to treatment or not. But, you know, if they could measure some kind of phage exposure in these patients, or the animal data that is done, or the in vitro data is done.

28:07 Because I think, especially with phage therapy, we would have to move to more, community based modeling platforms, where we include a lot of knowledge about different phage families. Because what we have seen from our just, you know, small attempt from a single group where we kind of digitized all the available PK data from all the in vivo studies out there and published, we saw that phage PK kind of has a kind of shared knowledge, right?

28:40 So where, based on the understanding, the tail, the phage size, we could kind of start getting closer to the expected clearance. So I'm just saying that if we kind of gather this knowledge as a community and kind of build this framework. I think it would benefit everybody to kind of have this centralized repository of understanding how phage bacteria interact and kind of helping everybody understand then how to get closer to the dosing issues.

29:11 great. I had a question, I just forgot it, but I'm gonna, um, oh, I know what I wanted to ask you about. One of the things I was also struck by when I was reading PK/PD for traditional antibiotics was that. There was an aspect where you were looking for doses that would suppress the evolution of resistance during treatment, resistance to the antibiotic.

29:39 Where are we in terms of understanding how to suppress the emergence, or evolution of phage resistant bacteria? And designing dosing schemes that will limit that?

29:56 No, so I think that's a great question. So what I mentioned is in the workflow that we have adapted and happy to send you that paper. We look at, first of all, you know, is there benefit to single versus double, triple phage combination? So first that is that, but then also including antibiotics, and when I'm saying including antibiotics, we are looking at lot of reduced exposure with antibiotics, so that's a huge benefit.

30:25 And I think that is now integrated into the workflow where you look at the antibiotic exposure that might be appropriate with this phage single or cocktail. And if that would be effective in suppressing and kind of also evaluating them in these different systems, which we didn't do that earlier on with antibiotics. So yes, that's a part.

30:48 great, and then I'm guess I'm just gonna, I'm gonna go into asking you. These are hard questions to answer. But do you have an idea of where you think and hope you can be in five years and in fifteen years?

31:05 Ah, fifteen years. I hope I'm in the field, but five years for sure. I think I am really heartened to see, you know, with going to different phage specific conferences and also the recent meeting that NIH had conducted. It's really, I guess, heartening to see that. Lot of the governments, FDA, EMA, all are very invested in this therapeutic modality. I do understand there are phage skeptics out there, but I think it's also important for us to understand how this works.

31:40 How can we move towards engineered phage? So, I think I am very, I should say I'm very optimistic that we will learn a lot through this process about how to engineer phage, how to combine them with antibiotics, how to respond to, you know, resistance to phage therapy. And I think because the community is very involved and there are a lot of agencies involved, I think this is going to be a different push.

32:13 That we did not see a lot with small molecules, and we are seeing that also with small molecules. And I think I know everybody is talking about machine learning and AI, but I think in this case, you know, harnessing the knowledge about the phage genomics, the phage bacteria interaction, and there are a lot of good groups getting into machine learning. So what I see in five to ten years is integrating PK/PD approaches with the host response, with the knowledge or with these AI systems.

32:44 We already are doing that. I think that will be very important for us to kind of gain all the mechanistic insights into these phage bacteria and host interactions for kind of dosing phages. And also, you know, there are a lot of companies coming with very good engineered phages. And also kind of, let me just say one last thing is, you know, things like, infection in the back of the eye where no antibiotic can reach, CNS infections, bone infections, you know, large, wound infections.

33:21 I think we are just thinking about maybe lung and bloodstream, but there are other infections where even no matter how good our antibiotics are, they're not effective in penetrating and kind of getting there. And so I think phage therapy has a lot of promise in these infections that have anyway been very challenging to treat over the past several decades.

33:45 Great. Your answer made me think of something else. we're talking about machine learning, AI; call it what you will. And I know there's been a lot of people focusing on the important question of whether you can look at a phage genome and a bacterial genome.

34:08 And you know, the idea, the sort of the pie in the sky idea, which may become more and more possible, is to, like, say, oh, this one will infect this one without having done the experiment. But I guess, as you and others who are studying PK/PD learn more about the properties of, you know, in terms of latency and burst size that are make optimal phages for clinical use. I wonder, I are there any efforts that you know of AI of using AI or machine learning to predict the, you know, the latency and the burst size?

34:47 Is anyone feeding that in, that you know of?

34:51 I know of a few groups from the Viruses of Microbes meeting. We ourselves at USC are also working toward that. That is definitely where we want to be. The challenge is looking at the phage and bacterial genomes as well as whether the phage characteristics are well suited for treatment.

35:20 And I think that requires a lot of collaboration between experimental and computational and AI, so it's not just AI alone. Because I think we need to help it learn what the key characteristics are and what defines them. So we are kind of moving towards that, and we are not just the people, but there are a lot of groups that are doing that. And I think, you know, folks like Intralytix and JAFRAL, they're also investing very heavily in these systems.

35:52 Great. Well, I think I'm gonna wrap this up and thank you for taking the time to talk to me. I think people will find this very interesting. And I'll give you, if you have any final thoughts you want to impart, before I stop recording

36:11 No, thank you very much. This is my first ever podcast. I enjoyed it, thank you.

36:16 Great.