Principal Investigator National Center for Toxicological ... · The data we thought about was the daily dose, ... For example, alpidem and zolpidem, two對 drugs with high logP, greater
18
Minjun Chen, PhD Principal Investigator National Center for Toxicological Research, FDA Click to view Biosketch and Presentation Abstract or page down to review presentation
The “Rule of 2” – Do Drug Properties Predict Drug-Induced Liver Injury (DILI)?
Minjun Chen, Ph.D. Division of Bioinformatics and Biostatics,
The US FDA’s National Center for Toxicological Research [email protected]
March 19th, 2015
Views expressed in this presentation are those of the presenter and not necessarily those of the U.S. FDA.
Presenter
Presentation Notes
Good afternoon, everyone. First, thanks for inviting me here to introduce our work. I will talk today a little about the LDKB work. I am a toxicologist or I can say bioinformaticist, not a clinician. So, I will give you from my perspective whether drug properties can predict drug-induced liver injury.
Chen M, Suzuki A, Borlak, Lucena M, Andrade R, (2015) J Hepatol, submitted
Presenter
Presentation Notes
We have talked many times in the DILI field. One big challenge, I think is the lack of a reliable predictive model. Especially today, we don't have a good animal model predict to predict human effects.
Translational Challenges in DILI Risk Assessment
Preclinical toxicology often failed to predict DILI risk in humans “High-dosing of healthy animals” test identified < 50% DILI
liability in humans Conventional in vitro assays have very limited prediction for
DILI risk assessment No biomarker is available to identify the susceptible patients
prior to drug treatment
Kaplowitz N (2005) Nat Rev Drug Disov,4:489-99; Avigan M (2014) Seminars in Liver Disease,34:215-226; Chen M, Borlak J, Tong W (2014) Expert Rev Gastroenterol Hepatol, 8: 721-23.
Presenter
Presentation Notes
As we know, FDA still relies on the high-dosing healthy animal study. This study still can only identify 50 percent of DILI problems. This technology was developed more than 50 years ago, so we need some new technology to improve predictions today.
4
Liver Toxicity Knowledge Base (LTKB) Link to external resources, e.g., PubMed, Wiki, LiverTox
Excel-like spreadsheet
Drug specific data
Query interface
Chemical structure & similarity
search
Assess DILI risk of new chemicals
Search “LTKB” @ Google
Presenter
Presentation Notes
So, we developed a project called the liver toxicity knowledge base. And this database provides a better predictive model. We have put some of the collected data in the particular model to a public domain. We can either use a LTKB such as Google to find that.
- Japanese TGX Project >130 drugs 4 testing systems Multi-doses and times >20,000 arrays
Chen M, Zhang J, Wang Y, Liu Z, Kelly R, Zhou G, Fang H, Borlak J, Tong W (2013) Clin Pharmacol Ther, 93 : 409-412
~3000
Presenter
Presentation Notes
This slide gives you some more idea what data we have in our database. And basically, we have collected about 3,000 drugs. And basically these drugs, including almost all the academia drug, drugs that were pulled by the other agencies. Basically this we started to collect the human data and the non-human data or we collect part of the data. For the human data, we tried to collect all kinds of the DILI-related information, especially we have it noted as a DILI risk associated with the drug. For the drug property data, we also collected each drug from the chemistry property. DILI markedly related individual assay or some whole special biology risk poles using microRNA data or this other data.
6
How to Assess DILI Risk in Humans for a Drug?
• Three attributes of a drug are important for its DILI assessment: Causality: was liver injury caused by drug or other cause Incidence: how many case reports are considered significant Severity: elevated ALT; Hy’s law; disability and hospitalization,
liver failure; liver transplantation or death
• Risk = (How likely) x (How many) x (How severe)
This is opinion based !!!
Presenter
Presentation Notes
At the end of the day, we tried to correlate these drug properties with human data, build a particular model. This is our goal to do the project. To develop a particular model, we need to list the drug have known DILI positive and DILI negative. The amount of the DILI drug in this model is to develop all kinds of translational biomarkers. We tried all kinds of approaches. Finally we found that drug labels are good enough to serve our purpose. The drug label, basically, is an information tool. It provides certain data to the doctor and the patient. By the way, the FDA should inform the patients about the drug label.
7
Drug Labeling is not Perfect but Probably the Best and Mostly Consistent Information We Have
• Drug labeling is not perfect Opinion based, not based on pre-defined criteria Not according to a consistent or scientifically justified master plan “Guilty by Association”…carryover from other drugs
• Adverse Reactions: “This section must list the adverse reactions that occur with the drug and with drugs in the same pharmacologically active and chemically related class, if applicable”
• Can we use drug labeling to identify more dangerous drugs ?
The question is how to mitigate the inherent defects in labeling!
Presenter
Presentation Notes
We agree that the drug label is not perfect but it might be the most consistent, best information we can have to help us codify the drug.
8
Drug labels
Chen M, Vijay V, Shi Q, Liu Z, Fang H, Tong W (2011) Drug Discov Today, 16: 697-703
Presenter
Presentation Notes
We published a paper several years ago, describing our approach uinge drug labels to identify DILI drugs. The drug label has three sections to disclose a DILI risk: Box Warning, Warnings & Precautions, and Adverse Reactions. Dr. Temple discussed drug label a bit yesterday, so I don't want to repeat today. If you are interested, go to our 2011 paper (Drug Discov Today 16:697-703, and get more details.Tthis approach, classified each drug into most concern, less concern, and a non-DILI concern. After we had risk classification by labeling and we know the drug is a DILI drug or a non-DILI drug, we then go to our LTKB data.
-10
-8
-6
-4
-2
0
2
4
6
8
10
0.001 0.01 0.1 1 10 100 1000 10000 100000
logP
Daily dose (mg)
Most DILI-concern
No DILI-concern
The Rule of 2 (‘RO2’): High Lipophilicity (logP ≥ 3) + High Daily Dose (DD ≥ 100 mg) Predicts DILI
- Observed in 164 drugs - Verified by 179 drugs - Demonstrated on 5 drug pairs - Applied to co-medication
Chen M, Borlak J, Tong W (2013) Hepatology, 58(1): 388-396
DILI
Presenter
Presentation Notes
We tried to develop some predictive model based on our drug property data. The data we thought about was the daily dose, because most of the DILI drug we know was given -- but the daily dose alone basically is not predicting now because we know many signature, also given the 100 milligram. We thought about whether we could we find some other way to help. The LTKB database finally found that lipophilicity can also help for this purpose. If you could use, the DILI we are marking here, we found if the drug dose was more than 10 mg, then there was toxicity. Most non-DILI drugs got kicked out. Because of the rule of 2 there is a significant association with DILI risk. ___________ The “RO2” models is a rule we found in the LTKB research. Through the survey of 164 FDA-approved drugs, we found that an oral medication with high lipophilicity (logP > 3) and high daily dose (DD > 100mg) was signifcantly asssociated with severe DILI risk. Basically, in the area that logP > 3 and DD > 100, the red one is DILI drug and green one is non-DILI drugs, we found most of drugs located in these area were DILI drugs with very few false positives. This so called “RO2” rule were validated by an independent dataset with 179 drugs, and 5 drug pairs and some co-medications. The results has published in the hepatology, and got some positive comments from DILI experts. For example, Neil has comments that the “RO2” offered some added values to identify the potential of idiosyncratic DILI.
10
Assess ‘RO2’ Model by Drug Pairs
Compound Name
Daily dose (mg/day) AlogP ‘RO2’ test Label Section DILI Concern
I show you some more examples to demonstrate the Ro2, using drug pairs. Drug pairs are basically two drugs capable of causing the same or similar effect and have similar structures, but show toxicity differences. For example, alpidem and zolpidem, two drugs with high logP, greater than 3. but alpidem had a much higher dose. Now look at troglitazone and two other glitazones: troglitazone, has a larger logP greater than 3 but only troglitazone had a much higher dose than the pioglitazone or rosiglitazone. Another example is bosentan. Dr. Temple mentioned yesterday this drug was also a RO2-positive drug. Its daily dose is 400 milligram, and AlogP also greater than 3.
More of ‘RO2’ Successful and Failed Cases
11
Compound Name
Daily dose (mg/day) AlogP ‘RO2’ test Label Section DILI Concern
Entacapone 1000 1.66 Negative No mentioned No-DILI-concern
We also show that logP helps in other cases, for example, tolcapone and entacapone. Those are drugs that have high doses but only tolocapone has the much higher logP. The same applies to nefazondone and trazondone. But we don't say that RO2 always works. The RO2 only has limited sensitivity, about 30 to 35 percent. We have some false negatives, and false positives, for example, trovafloxacin, a drug we know was withdrawn. The daily dose is about 200 mg but logP is very low.
* 23 of 50 RO2 positives were assigned as DILI positives by at least one literature
Presenter
Presentation Notes
We wanted to know how to work on all FDA-approved oral drugs. So we collected all drugs approved by FDA before 2010, 748 oral drugs. And of these we had 168 drugs with most DILI-concern in labeling, but Ro2 identified only 72, about 43% sensitivity. Next, 193 drugs with no DILI-concern, of which only 11 drugs were ALT positive. That means that specificity was about 95%. There were 387 drugs of less DILI concern, but we only identified 13% as ALT positive.
Selected Failed Drug Development Due to Hepatotoxicity
Call for the sharing of the failed drugs!
* Equivalent human dose
Compound Name
Daily Dose (mg/day)
AlogP RO2 test Safety Concerns
NP260 3000* 5 Positive Liver toxicity were found in beagle dogCasopitant 150 4.73 Positive Liver toxicity were found in Phase III patientsTasosartan 200 3.051 Positive Phase III clinical trials showed elevated transaminases Fiduxosin 100 4.215 Positive Discontinued due to liver toxicityAlaproclate 100* 2.695 Negative Observed liver complications in rodent studiesPafuramidine 100 2.77 Negative Liver abnormalities identified in healthy volunteersAplaviroc 1200 1.386 Negative Severe hepatotoxicity observed in Phase III studiesPralnacasan 300 0.132 Negative Liver abnormality observed at one species of animalsFialuridine 15 -1.228 Negative Unexpected fulminant liver failure in patients
Presenter
Presentation Notes
We also wanted to know whether the Ro2 could help us identify drug failures in clinical trials or in drug development. Interestingly, in this model, Dr. Regev presented a drug with daily dose of 225 mg and AlogP of 3 to 4, a RO2-positive drug, a drug we discussed this afternoon. In this other drug, they had a daily dose of 120 mg and logP is 4.1, another RO2-positive drug. So, both drugs discussed today were RO2-positive. You can see some more examples here, collected from the literature. but some are RO2 positive, some negative. But anyway, it shows that RO2 can identify some of the hepatotoxic drugs during drug development. We also want to call industry to study the failing drugs more, to learn if they can help give us a better predictive model. We know RO2 has limited sensitivity and we are trying to incorporate some more related data.
Improve ‘RO2’ by Integrating High-Content Screening (HCS) Assays
14
Positive (N=49)
Negative (N=21)
Positive 27 1Negative 22 20
Positive 19 0Negative 30 21
Positive 13 1Negative 36 20
95% N/A
RO2-HCS 67% 55%
Human hepatotoxicity
RO2 47% 27%
57% 39% 100% 70
Accuracy Sensitivity Specificity
HCS
# drugs requiring HCS test
95% 56
Model Test result
Total positives (N=49) 6 13 7
HCS RO2
Chen M, Tung C, Shi Q, Guo L, Shi L, Fang H, Borlak J, Tong W, (2014) Arch Toxicol, 88:1439-49
Presenter
Presentation Notes
And finally, in this paper we use a high-content screen assay to improve sensitivity from 30 percent to 50 percent.
Drug Properties or Host Factors?
15
It is dose!
It is mitochondria!
It is HLA!
It is immune!
It is genome!
It is reactive metabolites!
DILI
Finally, they stop talking, start listening and collaborate to "see" the full elephant.
Presenter
Presentation Notes
Going to the question John asked me: Are drug properties or host factors predictive? I think this cartoon is a very good answer to the question. In this cartoon, there are blind people who want to know what an elephant looks like. The first time, they don't agree because they are concentrating on a different part of the elephant. But very interesting, at the end of the story, original story, these blind men stopped talking and they started listening and collaborating. And then they envisioned the whole elephant. So, we have some blind people discussing our chemistry. If we were to figure out what the data looked like, at least addressed, we proposed DILI basically an interaction between the drug property and the host factor. Drug properties and host factors work together to initiate cellular injury. In the individual patient, the host factors will contribute to the individual response and then finally determine the final outcome. So, I suggest considering in a DILI case not only the host factors but maybe also the drug properties, to help you understand what DILI is.
16
Take-home Messages • Along with host factors, drug properties also contribute to the
prediction of DILI
• LTKB provides a centralized repository of diverse DILI-related drug properties data.
• Several predictive models were developed The ‘RO2’ has added value in identifying idiosyncratic DILI In vitro assays can be enhanced by integrating the ‘RO2’
• Improvements have been made, but still a long way to go.
Presenter
Presentation Notes
Overall, we believe that drug properties and host factors together contribute to DILI prediction, DILI development. Although LTKB has collected diverse DILI-related drug property data, it can be helpful for understanding. We have developed a predictive model. A comment from Dr. Kaplowitz was that RO2 has added value to predict idiosyncratic DILI. We also believe if we incorporate more data. It can be improved. We still have a long way to go to make a better predictive model. _________ Finally, I give a summary for my talk today. Firstly, to answer the question we proposed, yes we believed the drug properties did significantly contribute to the prediction of DILI. Based on this belief, we developed a large drug database systematically collect the diverse DILI-related drug properties data and annotated with DILI potential in human based on FDA-approved drug labeling.
17
Acknowledgements LTKB interest group: • Eileen E Navarro Almario • John Senior • Marc Stone • Tina Burgrass • Ruyi He • Shashi Amur • Jane Bai • Crentsil Victor • Andrew Mulberg • Mark Avigan
Non-FDA collaborators: • Jurgen Borlak (Hannover Medical
NCTR: • Weida Tong • Gerry Zhou • Ke Yu • Jie Zhang • Yuping Wang • Hong Fang • Huixiao Hong • Joshua Xu • Halil Bisgin • Zhichao Liu • Chuchu Hu FDA Supports: • Critical Path Initiatives • Office of Women’s Health (OWH) • Chief Scientist Challenge Grant • Office of International Program
Presenter
Presentation Notes
Finally, I want to thank the many people who helped me on the LTKB project, and especially the LTKB interest group. And also we thanke many people in this room. Especially I want to thank our collaborator Dr. Jurgen Borlak from Germany and my colleagues at NCTR. Thank you so much.
