Skip to Content

Drug Assays

Pharmacokinetic Advice From Genentech

Here’s another solid pharmacokinetics paper, this one from Genentech, with advice on how to extend drug half-life (compare this other recent one). They’re specifically addressing the “make it less lipophilic” rule of thumb that many medicinal chemists have, and they demonstrate that this isn’t exactly a universal law. You have to take it case by case, unfortunately.

They’re working from Genentech’s large database of rat i.v. PK data, stability in rat heptatocytes, and measured (not calculated) logD values. That’s a lot of hard-won data (4767 compounds total), and I’m glad that they’re sharing conclusion from it. The plot at right is an example; each dot is a compound, and they’re shaded according to logD values (<0 to >4). What you see is that IV clearance and (unbound, steady-state) volume of distribution are strongly correlated, and that for the most part this graph is spread along compound lipophilicity. There are outliers (there are always outliers), but that’s the way to bet. As for half-life itself, the spreads in the data are pretty wide, but if anything, higher lipophilicity is correlated with longer i.v. half-life. The Merck paper linked to above also noted that increasing lipophilicity can be a way to extend half-life.

As in that paper, the Genentech team also did an extensive matched-molecular-pairs analysis, and they found a number of transformations that have a good probability of extending half-life (and they’re similar to the earlier paper). Nitrile to fluoroalkyl, methyl to fluorinated methyl, H to CF3, and tetrahydropyran to phenyl all have a good shot. In general, improving clearance in the rat hepatocyte assay has about a 67% chance of helping your half-life, and if you can do that without decreasing lipophilicity, you have an 82% shot. (I would guess that many of those are fluorinations and chlorinations). Just lowering the LogD, on the other hand, only works about 30% of the time.

“But wait”, you may be saying. “If I start halogenating my compound, its solubility will probably go to pieces and its off-target binding may well go up!” This, in those deal-with-the-devil stories, is the point where the mysterious stranger slips up and his business partner catches a glimpse of his cloven feet. That is exactly the sort of deal being offered by pharmacokinetics and compound property analysis: no free lunch has yet been delivered. You have to make these changes in full knowledge that helping one aspect of the compound’s behavior may well hurt another one, and proceed accordingly.

This analysis (as the paper notes) doesn’t deal with charged compounds; they’re a whole other subject. And it also doesn’t go into the other ways around these problems, such as slow-release formulations, etc. But discovery-phase projects generally aren’t thinking about those things either – they’re trying to get compound behavior that doesn’t need the extra work later on. So if you’re in that part of the business, the advice from this paper is to pay attention to hepatocyte clearance assays, don’t be afraid to keep your lipophilicity where it is (or even raise it a bit, while watching for potential trouble), and don’t assume that just making the compound more polar is going to fix much of anything. And pay attention to metabolic hot spots – those are a big reason why those halogenations keep working  in these analyses. It’s more work to do metabolite ID, and often more difficult, chemically, to address those problems than it is to make general polarity changes. But it’s time well spent.

22 comments on “Pharmacokinetic Advice From Genentech”

  1. Peter Kenny says:

    The authors will surely face the auto-da-fé for their heresy! Do they not know that lipophilicity is the work of Satan? How can they not revere the Sacred Property Forecast Index that was created by Pharma’s finest minds for the edification of The Great Unwashed and was validated by “clearer stepped differentiation within the bands”.

  2. Former Hogwarts School of Thermodynamics says:

    Mothers Against Molecular Obesity (MAMO) are much displeased. This heresy shall not stand! Re-queue “stepped differentiation within bands”, this time with scaling parameters for hepatocyte clearance, re-circle the wagons, and send out the internal edicts (memos).

    1. anon1 says:

      The GSK molecular obesity stuff is annoying as the term is fundamentally unscientific.

    2. What would the Mothers Against Molecular Obesity know of these matters? Meddlesome harridans who are fit for nothing but the ducking stool! Remember that Saint George slew the fire-breathing clearance dragon with an enthalpy-driven lance!

  3. Nate says:

    Please note – slow release / modified release formulations do not modify pharmacokinetic parameters. You can get some apparent profile changes but at the cost of a massively increased dose since the clearance is unchanged – to increase AUC you have to increase the amount of compound given.

  4. cynical1 says:

    You said that the analysis doesn’t deal with charged compounds and that they were a whole other subject. So in the analysis there are no amines or carboxylates that are charged to whatever extent at physiological pH?

    1. Peter Kenny says:

      To model distribution of ionized compounds, you need to address the logP versus logD question. It is logD that is actually measured (usually at a single pH) but I would argue that logP is more relevant to binding of acidic compounds to albumin. One way to think about this is to ask how you would expect increasing the acidity a compound to affect its affinity for albumin. The analogous example of bases binding to hERG is discussed in the article that I’ve linked as the URL for this comment.

  5. Sulphonamide says:

    I hadn’t come across Vd unbound until the first of the papers Derek highlighted recently…afraid this concept is escaping me (I thought by definition that Vd took account of all the non-specific binding – so what an unbound version of it may be is rather failing to click). I don’t suppose anyone has a nice succinct explanation or a suitable reference for amateurs (I’m sure there was no mention of it in the excellent Kerns and Di)? Thanks.

    1. Mous says:

      Thanks for asking this, I’ve been thinking the same. I hope someone can give a reply !

    2. Derek Lowe says:

      As I understand it, it’s volume of distribution / unbound fraction. That is to say, it’s volume of distribution with all that nonspecific binding corrected for. What’s the volume needed to accommodate the free faction alone, as opposed to the volume needed when you just calculate from the total concentration, where the plasma number is going to include protein-bound drug?

      At least, that’s how I have it in my head. I’ll be glad for corrections from people who are more into PK than I am!

    3. NJBiologist says:

      This is a bit of a guess, but Khojasteh et al may offer an option (that’s the Drug Metabolism and Pharmacokinetics Quick Guide; note that all three authors are also Genentech PK guys). Their equation 1.8 gives
      Vd = Vp + Vt(f_unbound_plasma/f_unbound_tissue)
      where Vp is the volume of plasma and Vt is the volume of tissue; the f terms are fraction unbound. This doesn’t get much explanation, but it’s easy to imagine using reference values for the volume parameters or a plasma protein binding result for f_unbound_plasma to solve for unknowns.

      Having said that, I’m still not clear on the utility of the concept–I’m with Sulphonamide.

      1. Calvin says:

        This concept confuses me. Vd (for) me has always been a good measure for whether your protein binding is going to cause a problem or not. For situations where you want to be in tissue (most cases) as long as you are above liver blood flow, then your protein binding is non-restrictive and while it may be bound is has a low affinity for the protein. Kinetics versus thermodynamics. The idea of Vd unbound just makes no sense. Somebody convince me that this is a useful term…..

  6. Calvin says:

    To be honest I’m somewhat surprised this got published. It’s not a bad paper, it’s just that I’m not seeing anything here that we did not already know. Try not to be too liphophilic but lipophilicity is not as bad as you think. Polarity is your friend but like a cat, only on it’s terms. Fluorine blocks metabolism. There really isn’t anything new here. There are numerous papers out there showing that a balance of fatiness and polarity is required to get decent PK.
    It’s the whole dinner plate spinning thing (a beautiful analogy given to me by my old boss; you know who you are). Can I keep all these plates spinning while people keep adding more plates? Which ones can I allow to break first?

    1. anon electrochemist says:

      Who needs data when you can hand wave?

      1. drsnowboard says:

        Who needs retrospective data when your managers hand wave their prejudices anyhow, and the data actually confirms your innate suspicions anyhow?

  7. John Wayne says:

    I’m also suspicious of the use of unbound volumes of distribution. Are actual measured values from animals not cool anymore? It has been my personal experience that unbound drug can be important or not important depending on your project. Be suspicious of anybody who tells you otherwise.

  8. Old Pump Kicker says:

    “In God we trust, everyone else bring the data.” — S. Moose

    I salute the experimental effort involved. I don’t have access to the paper, so maybe this is addressed: Given the large number of measured logD values, can this validate or improve the method for calculating logD. Derek has indicated that calculated logD is probably accepted as accurate more often than it should be.

  9. Curt F. says:

    The data in both these papers is impressive. The log-log plots definitively show trends, but they also sort of obscure the very large amount of scatter that is left unexplained by the correlations discussed in the paper. I think deep learning approaches, much ballyhooed by the commentariat here, would be very useful to apply to these data. Of course, doing so would required knowledge of all of the thousands of structures analyzed, which of course the papers don’t disclose. Nonetheless I hope internal teams at GSK and at Merck are doing this work. Even modest improvements in the accuracy of PK predictions would be a big advance!

  10. milkshake says:

    when Sutent was going into clinic and chronic animal toxicity studies were done, the PK guys were dismayed by the organ accumulation, nonlinear PK and a very long-lived cardiotoxic desethyl active metabolite responsible for QT prolongation. The volume of distribution of the drug was huge. The animals in long-term dosing study developed adrenal necrosis. In the clinic, the patients had to be given “treatment holiday” after few weeks of treatment, to help them clear off the accumulated metabolite.

    Now the chances are that the drug candidate with such a troubled PK profile would not have been advanced at a risk-averse pharma company, and even as Pfizer people hated the project when they acquired Pharmacia-SUGEN, and proceeded rapidly to dismantle the team behind it, the things were quite far along with Sutent in the clinic and they did not cancel it. Now it makes billions for Pfizer.

    As it happens, I worked on a follow-on analogs of Sutent with improved PK profile. We solved the problems with active metabolites and the organ accumulation, and our compound got into phase 1 and 2, after which it was shelved. It turns out that toxicities of Sutent were manageable, and the improved analog did not offer anything new in terms of efficacy. And it had to be dosed higher, bi-daily, because it did not accumulate and had reduced lipophilicity and smaller volume of distribution. In fact, it now seems that the organ accumulation of Sutent is important for the therapeutic effect of Sutent. So be careful what you are trying to fix.

    1. Barry says:

      Azithromycin was nearly shelved repeatedly because it has a huge Vss (29,000ml/kg) and levels in plasma were so low that it couldn’t be conveniently monitored. Eventually, someone persevered, learned to monitor levels in leukocytes instead of plasma. It made $billions, of course. The plasma compartment is only rarely the target compartment. It’s just the easy one to monitor

      1. Peter Kenny says:

        The unbound concentration in the plasma compartment is still important since, in the absence of active transport, it drives unbound concentrations in other compartments. Is azithromycin actively transported into leukocytes?

  11. Fabio Broccatelli says:

    Hi All, thanks for the comments and the interest, I will try to address what I can in the next few lines.
    The choice of using Vd and CL unbound comes from the intention of relating the 2 components of T1/2 to molecular properties that would not be confounded by the effect of plasma protein binding. Some might be familiar with experiments with albumin deficient (AD) mice showing that the same compound in WT vs AD show comparable unbound concentration, but different total concentration. Important to keep in mind that Vdu is derived from in vivo data corrected with an in vitro measurements (this is addressing the question about “actual measured values from animals” not being cool anymore; unfortunately measurements from animal have never been cool, but at this time and point of science are still necessary).
    A reference talking about VDu would be this:
    https://www.ncbi.nlm.nih.gov/pubmed/20891009
    I think some of the problems in visualizing all of these concepts come from the fact that PK parameters were originally defined with a reservoir and an extractor in mind, not necessarily a physiologic description of the body. After an IV dose, we can derived V from the concentration at time 0 and the Amount. C0=Dose/V. The volume is “just a term” relating the observed concentration to the amount. If we care about the concentration of the unbound drug (because we agree with most DMPK scientist that free drug is what matters in absence of major transporters effect or rate limiting permeability/koff from plasmatic protein) the volume that explains it is the Vu.
    I can also try to give it some physiological meaning as follows:
    Vtot=Vplasma+Vtissue * kp, kp describes partition constant between tissue and plasma, it equals fup/futissue. Let’s consider a drug with V>> than plasma volume. Vtot=Vtissue*kp, we can divide by fup both sides. Vdu= Vtissue/futissue. Important to keep in mind Vtissue is the refers to the volume of water in tissues. If we now take 2 compounds with same tissue binding, the compound with higher Vdu is the compound with higher ability to distribute in tissue water.
    A final “ideological” note: although we are saying that decreasing LogD to decrease CL is unlikely to fix T1/2 unless a MetID soft-spot is addressed, we are also still saying that T1/2 is to be optimized by decreasing CL, not increasing Vd. Infact AUC is also important, and if CL goes up AUC goes down.

Leave a Reply

Your email address will not be published. Required fields are marked *

Time limit is exhausted. Please reload CAPTCHA.