Cytomics in Drug Discovery and Development [construction icon]

Index

Guestbook for comments

Introduction

This webpage is meant as a humble contribution to the discussion about improving drug discovery and development, but it is still under construction. It did not go through the traditional peer review process of scientific publications and neither does it pretend to be a critical appraisal of medical evidence, so read it with care and with a critical mind. Feel free to comment and to criticise.

Why would drug discovery and development benefit from cytomics and what are the fundamental problems we face in drug discovery and development ? Cytomics and a coordinated research effort organised in a human cytome project aims at creating a better understanding of the cellular diversity of biological systems and reducing the great divide between present day reductionist models and clinical reality. Cytomics allows us to close the great divide between molecular research and the complexity of the intrahuman ecosystem. There is no clinical evidence to be found in a testtube, so the odds are against basic and applied research in our (pharmaceutical) laboratories and (pre-)clinical pipelines. Understanding the (heterogeneous) cellular level of biological organisation and complexity is (almost) within reach of present day science, which makes cytomics as a science and a human cytome project ambitious but achievable. A human cytome project is all about creating a solid translational science, not from bench to bedside, but from molecule to man. Even more than translational science, it is about transformational science as it transforms the molecular microcosm into the macrocosm of the intrahuman ecosystem and its physiology.

Although the present business model of pharmaceutical discovery and development is capable of coping with massive and late stage failures during the development life cycle of a new drug, it is not the way we should continue to work as the challenges ahead are even bigger than the ones we already faced in the past. The current business model is capable to deliver enough drugs to sustain itself, given the disease mechanisms drug discovery and development deal with, do not surpass a modest level of biological complexity. The present business model simply takes into account the high risk and failure rate of present day science. The scientific engine, from basic to applied research, of the overall process is not yet capable to predictably deliver new drugs which can stand the challenges of biological complexity. At any given moment in time the actual performance of the scientific engine of the process and its scientists has to be taken more or less for granted, and only cost (labor, headcount, process engineering, economy of scale, mergers and acquisitions) and income (price, health insurance, government pricing policies) are available for marginal improvements in process performance (efficiency) and productivity (effectiveness). Business engineering is somewhat easier than engineering the framework of science. We do what we can and what is possible, but not necessarily what is required. We look for solutions where we can find them, but not necessarily where we have to go.

We have to destroy too much capital and human effort to sustain the pipeline, which nowadays resembles a small tube in most companies (to be honest the past was not better either). We compensate for massive scientific failure rates by business model egineering. We have to build a business model on almost catastrofic failure rates compensated by high cost per product, which is only possible in medicine at this rate. But no matter how efficient we engineer our business processes, we cannot hide from the fact that we are running a pipeline process which has too much risk hidden underneath to feel comfortable with as the basis for delivering new treatments for the diseases of our society. Although the pharmaceutical process matrix is in itself consistently organised, it is is slightly out of touch with the complexity of clinical reality. This becomes visible at the end of the pipeline, when truth is finaly forced upon us and we can no longer hide from clinical reality, which extends beyond the frontiers of science. At the intermediate stages the confrontation with reality is only partial and resembles a democracy where the (clinical) opposition is (to be) excluded from voting. A safety test may prove that a candidate drug is safe, but leaves one blind on its efficacy. We cannot deal with all the biological complexity of the intrahuman ecosystem and so we have to withdraw into simplified representations of reality which are too simple as we have to acknowledge at the end of the pipeline in the vast majority of drugs being developed. Expansion of the frontiers of our knowledge is an interesting activity but the process of knowledge expansion in itself is an ineffective and inefficient process. At the intermediary steps we are capable to declare victory because we can set the rules less stringent than those which nature itself pushes through in a real world clinical situation at the end of the game. Life in the lab is therefore easier than life in the clinic, where you have to look the patient in the eye and fight diseases in reality and not in a test tube. Changing the overall process however is a Gargantuan endeavour as it is all about the core scientific engine in the first place, which is almost inert when it comes to paradigm changes.

We are doing our science right, but we are not doing the right science. The success ratio of science, when starting from basic research down to its successful application in the clinic is less than 1 percent. But as scientists have no clue which part of the 100 percent effort will succeed, society has no option but to waste 99 percent of the effort in order to succeed in 1 percent of all research (from basic research to successful application). The main problem is that our basic science is basic in dealing with biological complexity. We look at the intrahuman ecosystem through a keyhole and remain blind for most of its complexity. Although we increasingly automate the scientific process, the process in itself remains largely the same and the effectiveness does not change in the same order of magnitude as its automation which mainly drives the efficiency. We may proceed faster through some stages of the pipeline with our machines, but we still do not reach the goal in most cases. The amount of data being produced increases dramatically, but a pile of numbers does not equal a pile of true understanding of the perceived image of reality which our machines unfold to us providing the view through which we observe and perceive the disease process but still not the reality itself we would need to know and understand. Capacity does not equal performance in terms of results, activities based on the wrong assumptions do not help us in the end. Failing intermediary discovery and development stages leave us with massive failure at the final gate. We postpone the true confrontation with the complexity of nature until the end of the entire process and thereby fail to remedy the intermediary steps (pipeline science is what it is, just live with it). The moment of truth only becomes apparent during the final stages of drug development, buth then it is is already too late. We are good at scientific engineering within the boundaries of our models of biological reality, but rather inefficient in expanding the boundaries of true understanding of the dynamics of biological processes (i.e. the dynamics of biology or multilevel physiology, not just its molecular structure). Present day science, as its predecessors, is far from perfect or up to its clinical challenge, but it is all we have, so for any moment in time we just have to live with it and build our business on top of it.

The dramatic failure rate of science in the face of clinical reality is just one of the parameters in the equation of a business model. Business succeeds even when science fails massively. Pharmaceutical companies just take into account the massive failure rate of (basic) science in their business model. No matter the productivity of science, running a very profitable pharmaceutical business was, is and will always always be feasible, by compensating for the deficits of science by careful management and business engineering (which is somewhat easier than changing the paradigm of science). From the early days of industrial drug discovery and development, management techniques have been conceived and used to create a profitable business, regardless of the shortcomings of science to produce drugs which succeed in the face of clinical reality. Assembling a profitable pipeline can be done either within a traditional monolithic structure (big pharma) or a cluster of risk sharing companies each a part of the overall pipeline. These two business models establish a different risk exposure, but the overall scientific process (the actual engine which drives the pipeline) remains the same. Pharmaceutical management manages risks, which are near catastrophic in drug discovery and development. The science of conceiving the framework and managing a pharmaceutical company is even of a higher order than the scientific engine itself, as it is capable of generating profitability and bring new drugs to the people on top of a scientific process which has less chances to succeed than throwing a dice.

The question is of course, which direction we are capable to take in order to move forward; even more business engineering and automation (increase efficiency) or changing the paradigm and practice of basic and applied science (improve effectiveness), or both ? Although we are becoming increasingly efficient, we do not achieve more effectiveness in the same order of magnitude. Is academia and the industry capable to innovate its core activities ? Is changing such a complex process possible anyhow, while at the same time keeping the business in the air ? What is the value of getting more answers on the wrong questions ? What is the value of our work if the review by nature itself declines the results of our scientific effort ? Is there any hope of an improvement in the productivity of the scientific process itself ?

Do we just have to live with the as is situation of basic and applied science and make the best of it at any given moment in time ? Are we capable to improve the scientific process of knowledge expansion itself or do we continue crawling slowly at the borders of present day scientific boudaries instead of engineering the scientific process itself towards more productivity (truly succeed more often when facing the challenge of nature itself). Scientific practice is always done within the boundaries of the paradigms of the day but not along the lines demanded by nature itself. No matter how far we are today, there is something wrong with the productivity of the scientific method iself in the face of nature. The weakness is not the practice, but the process itself which is flawed towards its performance to find the right answers on complex questions. The process delivers what it is designed for by our science, but not what is required to deal with nature itself in the real clinical world in more than 90 percent of what we (try to) deliver.

As such this article is dedicated to all the patients hoping and waiting for new treatments of unmet medical needs and the improvement of existing therapies. It is also dedicated to all the scientists working in basic and applied research, working day and night to deliver these new drugs and treatments. It deals with process (technology, biology) and model deficits in drug discovery and development at various points throughout the pipeline. Pipeline analysis and engineering is a delicate process as it is required to change the bricks of the house, without destroying it.

The breakthroughs in basic research have not resulted in the creation of many new therapies for patients, which lead to the 'pipeline problem'. Improving drug discovery and development is not a simple endeavour, as we have seen in recent years. Although this article is critical about the (evolution of the) overall drug discovery and development process it also honours the individual contributions of scientists who have discovered and developed drugs which save and improve the lives of many people. The purpose of critical discussion is to advance the understanding of the field. While many are spurred to criticize from competitive instincts, "a discussion which you win but which fails to help ... clarify ... should be regarded as a sheer loss." (Popper). Let us look at the present with the future in our mind. Although this article may seem wide ranging and to some shows lack of focus, it is meant to be comprehensive and also show the lack of focus of many solutions which attempt to solve the pipeline problem by revolving around the core problem.
The problems with drug discovery and development are already leading to international initiatives. See also Innovation and Stagnation: Challenge and Opportunity on the Critical Path to New Medical Products - USA, Innovative Medicines Initiative (IMI) - Europe EU, New Safe Medicines Faster Project - Europe EU and the Priority Medicines for Europe and the World Project "A Public Health Approach to Innovation" - WHO.

Drugs have both a humanitarian value and a financial value. Pharmaceutical research and development contribute a major part of the research necessary to move new science from the laboratory to the bedside. Through academic and industry efforts, many new drugs and devices have been developed and marketed, which save and improve the lives of many people. However, the costs to bring new drugs to the market have risen sharply in recent years and the output of drug development and as such the Return On Investment (R.O.I.) has not kept pace. Drug development has always more or less resembled more of a lottery than a controlled process, due to the lack of basic understanding of biological complexity in the intrahuman ecosystem. In relation to the effort (cost involved), fewer drugs and biologics are making it from Phase I clinical trials to the marketplace, which has dramatically increased the cost of drug development (Crawford L.M., 2004). Late stage failure in Phase III clinical trials and NDA disapproval has risen from 20% up to 50% (Crawford L.M., 2004). From an economical perspective, the goal of improvements in drug discovery and development is to increase the Net Present Value (NPV also "fair value" or "time value") of pipeline molecules and to decrease the costs associated with pursuing failed projects. Basicaly the Net Present Value (NPV) is the worth of a good at the present moment and for investments the Net Present Value is an important indicator. Only an investment, that offers you a positive net present value, is considered to be worth to pursue. This has not been the case in recent years for many drug development projects, as up to 90% fail. The bottom line form an economical perspective is that in the end any change in the process or its (scientific) content should improve the Return on Capital Employed (ROCE).

This article aims at improving the probability of success in drug development (reduce late stage clinical development attrition) by using better disease models (higher predictive power) in drug discovery an pre-clinical development. This improvement should lead to bringing better drugs (more effective, less side effects) to the patients, both cheaper and faster.

The challenges which the pharmaceutical industry is facing:

  • Increasing competition and ending patents (the end of the blockbuster era).
  • R&D becoming more and more expensive and less productive (molecules instead of man).
  • Regulatory requirements becoming increasingly strict and heavy (risc aversion of society).
  • Challenges to keep a balanced product pipeline (managing decreasing predictivity).
  • Changing demographics and disease profiles (disease patterns are in a constant flux, increasing gap between molecular science and clinical reality).

From a business perspective, there are 2 sources of value creation by a more productive discovery processes ("clinical" quality of molecules). Both require a better scinetific engine in the first place:

  • Increased Net Present Value (NPV) of pipeline molecules because of higher likelihood of successfully reaching the market (more true positives).
  • Decreased costs associated with pursuing fewer failed projects (less false positives).
There are 4 levers for creating value in (pre-)clinical drug development (process improvement):
  • Increase the probability of success (POS) in drug development (better models early on).
  • Decrease the time in drug development (increase impact).
  • Decrease the cost of drug development (fail less often).
  • Maximize the income potential per product in the pipeline (increase positive impact).

What should we achieve for the overall drug development process in order to restore its productivity (principles):

  • Decrease (costly) late-stage attrition.
  • Reduce the overall time-to-market.
What should be the deliverables (targets, quantitative), however ambitious they may seem given the present state of science:
  • Up to 9 out of 10 drugs succeeding in drug development instead of failing.
  • A 50% reduction of the time-to-market.

The pharmaceutical industry has a history of initial innovative breakthroughs (first-in-class), followed by slower, stepwise improvements of such initial successes (best-in-class). How can we improve the Probability Of Success (POS) of the overall drug discovery and development process and as such improve both the quality and quantity of new drugs, both for innovative as well as stepwise improvements? Why do we need to learn more about the human cytome to improve drug discovery and development? How can cytome research help us to discover and develop better drugs with a higher success rate in clinical development? What is wrong with the drug discovery and development process as it is now, so its costs are soaring and its R.O.I. is declining? Why do we only succeed in improving success rates somewhat with biologicals and no longer with small molecules ?

Everyone managing the discovery and development of drugs has to ask a few questions about every new scientific idea or technology which pops up (and they do, all the time). Every scientific idea or technology to be applied to drug discovery and development must specify realistic and compatible goals and expectations. When we want to introduce a new scientific idea into drug discovery and development we must balance between good science and a credible business plan. We must be critical about the promises made. Is the claim or argument relevant to the overall subject? ('Subject matter' relevance). Is the argument or claim relevant to proving or disproving the conclusion at hand? ('Probative' relevance).

Personal note

My personal interest in cytomics, grew out of my own work on High Content Screening, as you can see in:

Scope of the article

This article deals with the analyis of several apects of the drug discovery and development process and the the weak spots and flaws within this process, which cause the high late stage attrition in the pipeline. The pharmaceutical industry has passed the threshold where only slight adaptive changes can restore its productivity and profitability. Reducing costs is not enough to restore the health of the industry, a paradigm shift is needed, but this will require a new vision on the fundamentals of the drug creating process and the way true knowledge and understanding is being built on the foundations of scientific discovery and through applied research.

However important business processes are, this article in itself is not about Business Process Improvement or Business Process Reengineering as this is outside the scope of Research Process Improvement and Research Content Improvement. The processes surrounding the drug discovery and development process require attention and optimisation too, but in the end success in drug discovery and development depends on bringing the right drug to the right patient. Both the drug discovery and development process and its scientific content require optimisation beyond their current state.

The scientific content of the drug discovery and development process, goes beyond business management principles and is more difficult to optimise than the process itself. It is the present state of basic (reductionistic) and applied science and technology itself, in relation to the complexity of biological systems, which still limits our chances for success. Inadequate understanding of basic science for certain diseases and the identification of targets amenable to manipulation is one of the major causes of failure in drug development. The endpoint of discovery is understanding a complex biological process, not just a pile of molecular data. The endpoint of development is therapeutic success in man (a complex biological system), not just a molecular interaction. The level of understanding at the end of drug discovery (and preclinical development) should achieve a knowledge level which is capable to predict success at the end of the pipeline much better than we do now. No matter what is the origin of the compound under evaluation, or how it came into being, a good description of its in vivo pharmacological properties is necessary to assess its drug-like potential. The sooner NCEs or NBEs evolve in a "rich" or lifelike biological environment (resembling the situation in a human population) the sooner we capture (un-)wanted phenomena.

There is a time-shift between the implementation of a new approach (linking genes almost directly to clinical diseases) and finding out about its impact on commercial success, which makes the feedback loop inefficient due to its long delay in relation to the quarterly and annual business cycle. From a business perspective, any process can be sped up and content can be sacrificed or complexity reduced. In a stage-gate process, the stages deliver the content for the decisions at the gates, so the stages should be informative and predictive. Processes and portfolio management can be optimised near perfection. This may be provide sufficient leverage for a (albeit complex) 'nuts and bolts process' (e.g. automotive industry), but not for processes in a biomedical context when our understanding of pathogenesis and pathophysiology is still very patchy and incomplete. We leave a large potential for improvement untapped. Improving a development process which still fails for 90% of all developmental drugs, is not optimised at all. With the current inefficient process we are, in most cases, unable to serve smaller patient populations.

If we ever want to reach the goal of personalized medicine, which is in my opinion is beter understood as succeeding in unraveling the molecular diversity of clinicaly similar disease manifestations, some conditions need to be fulfilled:

  1. A complete understanding of the molecular pathophysiology of a disease.
  2. Availability of higly reliable, highly predictive but cheap diagnostic tests.
  3. A very efficient and productive drug discovery and development process.
A lot of research and development will be needed to reach this goal, leaving aside the ethical minefield.

The complexity of intermediary modulation of gene-disease (un-)coupling was clearly underestimated in recent years. In the early stages of drug discovery, the data tend to be reasonably black and white. As you get to more multifactorial information and more complex systems later on in drug discovery and development, that becomes less true. Managing this complexity in a coherent way is a challenge we must deal with in order to be successful. So how can we facilitate (and understand) the flow through the pipeline, without generating empty downstream flow in clinical development? How to plug accelerators into the drug discovery and (pre-)clinical development pipeline which prove their value onto the end of the pipeline? How do we create a true Pipeline Flow Facilitation (PFF) process?

The first part of this article shows the problems of the drug discovery and development process. It shows the present problems of the pharmaceutical industry.
The second part of this article looks for the best way to improve the drug discovery and preclinical development process as these feed clinical development with drug candidates which should make it to the right patients.
The third part of this article deals with the problems with disease models in drug discovery and preclinical development and why they cause so much late stage attrition later on in clinical development.

Overview of related articles on this website

Personal interest and background where I provide som information how the idea for a Human Cytome Project (HCP) has grown over time.

References have been put together on one page.

Overview of problems and questions

Scientific background about the Human Cytome Project idea can be found here

The potential impact on the efficiency of drug discovery and development where I give an analysis of the reasons for the unacceptable high attrition rates in drug development which have now reached 9O%. Our preclinical disease models are failing, they look back instead of forward towards the clinical disease process in man.

Overview of solutions and suggestions

A proposal of how to explore the human cytome where I give an overview of the deliverables and the scientific methods which are (already) avalable.

How to deal with the analysis of the cytome in order to improve our understanding of disease processes is being dealth with in another article. The first part deals with the problems of analyzing the cytome at the appropriate level of biological organization.
The second part deals with the ways of exploring and analyzing the cytome at the multiple levels of biological organization.

A concept for a software framework for exploring the human cytome is a high-level concept for large scale exploration of space and time in cells and organisms.

Some thoughts on the pitfalls of applied research

The vulnerability of applied research, such as drug discovery and development, is hidden in the basics of scientific reasoning. In traditional Aristotelian logic, deductive reasoning is inference in which the conclusion is of no greater generality than the premises, as opposed to abductive and inductive reasoning, where the conclusion is of greater generality than the premises. Other theories of logic define deductive reasoning as inference in which the conclusion is just as certain as the premises, as opposed to inductive reasoning, where the conclusion can have less certainty than the premises. Scientific research is to a large extent based on inductive reasoning and as such vulnerable to overenthusiastic generalizations and simplifications. There is a lot more to say on the philosophy of science and its impact on research, but this is outside the scope of this article.
In addition the discussion about the problems of the drug discovery and development process is full of red herrings and other logical fallacies, which distracts our attention from the real question: does the treatment work in man (see also Organon from Aristotle). Ignoratio elenchi (also known as irrelevant conclusion) is the logical fallacy of presenting an argument that may in itself be valid, but which proves or supports a different proposition than the one it is purporting to prove or support (The promise to the pharmaceutical industry "do this" or "buy that" and you will deliver more and better drugs to the market). The ignoratio elenchi fallacy is an argument that may well have relevant premises, but does not have a relevant conclusion. The Red Herring fallacy is the counterpart of the ignoratio elenchi where the explicit conclusion is relevant but the premises are not, because they actually support something else. The complex relation between the input of the drug discovery and development process (manpower, methodology and technology) and its output (drugs which succeed) is underestimated, leading to unacceptable late stage attrition rates. However, there are no simple answer to complex problems, such as how to create a truly productive process, both effective and efficient. The truth is forced upon us during the late stages of clinical development, when we fail because of a lack of predictive power of discovery and preclinical development.

Like most opportunistic enterprises, pharmaceutical companies run the managerial risk of succumbing to the enthusiastic optimism of a pragmatic fallacy. Leading scientists and managers have to understand and systematically manage ambiguity in an increasingly complex environment. There is more ambiguity in clinical reality and economical reality than in an Eppendorf tube. You have to assess risk and benefits of decisions and anticipate the impact on drug discovery and development in the longer term, far beyond the short-term quarterly goals. Those who take responibility for strategic management have to grasp opportunities capable of generating new opportunities for improving drug discovery and development in a productive way. You have to scan the scientific, business and regulatory environment and think well ahead to identify things which may get in the way of meeting objectives - either obstacles or changes in the overall situation. Managers and scientists have to develop complex strategies which take into account the diverse interests across scientific domains, economics, rules and regulations. It does not help that scientists prefer the scientific excitement of reading Nature and Science, while managers prefer the Wall Street Journal and the Financial Times. There is a lack of cross-discipline understanding and colaboration throughout the entire process (not enough silo busters). The true challenge is to appreciate that the discovery, development, application, and regulation of the target to drugs pipeline has to be viewed as integral processes with each element having important, sometimes critical, implications on the other components with decisions weighed accordingly.

Drug discovery and development

Evolution of process and content performance

Evolution of sales
Figure 1: Evolution of sales for some big pharmaceutical companies.
Source: Yahoo finance and other sources
Evolution of earnings
Figure 2: Evolution of earnings for some big pharmaceutical companies.
Source: Yahoo finance and other sources
Evolution of earnings versus sales ratio
Figure 3: Evolution of earnings as ratio of sales for some big pharmaceutical companies.
Source: Yahoo finance and other sources
NYSE Pharma shares
Figure 4: NYSE Pharma shares of Merck &Co. (NYSE:MRK), Pfizer (PFE),
Eli Lilly (LLY), GlaxoSmithkline (GSK) and Bristol-Myers Squibb (BMY).
_DJI (Dow Jones Ind.), _IXIC (Nasdaq).
Source: Yahoo finance

The graph in Figure 1 shows the evolution of sales for some big pharmaceutical companies from 1999 until 2004. Pfizer, Johnson & Johnson and GSK show an increase ahead of the others. Merck suffered from the problems associated with Vioxx, which illustrates the fragility of commercial sucess. The graph in Figure 2 shows the evolution of net earnings from 1999 to 2004. Here there is less difference between the companies, no company is able to outperform the others in a dramatic way. The graph in Figure 3 shows the evolution of the percentage net earnings to sales. On average there is a steady decline from 19.5% in 1999 down to 14.0% in 2004.
The graph in Figure 4 shows that in recent years the growth of the pharmaceutical industry has slowed down. The pharmaceutical industry found themselves in a tight spot in the beginning of the 21st century. The sector has seen a decrease in financial performance following a boom period in the 1990s, fueled by a succession of drugs with sales over US$ 1 billion per year (blockbusters). As all drug companies improved their stock value, the cause is common to all of them and not to one company outperforming the other companies. Of all the knowledge required to develop a new drug, the most important component is a true understanding of the molecular basis of the disease process. This knowledge is mainly in the public domain and available for all companies, so no company is capable of outperforming its peers in the long run. Although it may take up to 15 years to develop a new drug, it may take up to 20-30 years to unravel the mechanism of a disease and this requires an effort on a global scale, not just of a single pharmaceutical or biotech company. Almost 90% of all science required to create a drug for a given disease, resides outside the pharmaceutical industry, but 90% of the risk is hidden within the inefficient process used within the pharmaceutical industry (90% failure in drug development). The industry faces the final challenge of proving that the ideas about a disease process truly work in the complex biosystem of man. The drug discovery and development process suffers from "particularism" and lack of "generalism" as the success with one drug does not lead to a consistent increase of performance. Working on improving the process as it exists today within the pharmaceutical and biotech industry is only leveraging 10% of the required knowledge-base. A failure rate of 90% for 10 or 50 drugs in the development pipeline, is not an example of process improvement, although the bigger pipeline will lead to a five-fold increase of drugs reaching the market.

How can the pharmaceutical industry get out of the current situation of spiraling costs and reduced R.O.I? There is no simple answer to this question, as a solution requires improvements in multiple domains. The management of research must ensure that the resources are directed to investigations consistent with the ultimate goal, the development of a successful drug. The management of research is full of uncertainty and complexity. Research has substantial elements of creativity and innovation and predicting the outcome of research in full is therefore very difficult. The costs and risks involved in developing, testing and bringing new drugs to market continue to grow, pharmaceutical companies are coming under increased pressure to make the discovery and development process more manageable and efficient. Today we need both better processes as well as better science to succeed in the disease jungle or the pathogen minefield. Success will go to those who can manage the hybrid activities between science, technology, and the market.

Although innovative and sound research is a prerequisite, it is ultimately the therapeutic success of the drug which results in sales and profits. And it is usually proprietary (patented) products which earn the highest returns, because they produce sustainable competitive advantage over a substantial period of time (e.g. patent lifecycle). We must keep in mind that it is the ability to produce proprietary products, not just interesting science, which leads to a profitable and sustainable pharmaceutical company.

The business process around the drug discovery and development process itself can be improved as well as the R&D process management itself (e.g. Business Process Improvement, process and portfolio management, ). Reducing R&D costs and shortening product development cycles will certainly contribute to an increase in profitability. But when the scientific substrate of the R&D process itself is not optimised too, we leave a huge potential for treating diseases cost-effectively and generating profit untapped. Both the process and its content require our attention. The recent gulf of mergers an acquisitions provides some short-term relief, but when we combine two companies with each a 90% attrition rate in drug development, we just get a bigger company with also a 90% attrition rate in drug development. This high attrition rate leaves little margin for dramatic improvement of overall productivity. At the moment the pharmaceutical industry is trying to generate some leverage by working on improving the development process for the 10% developmental drugs which make it through the pipeline. Business process engineering as such is less riskier than rethinking the overall discovery and preclinical development process. Business process improvement can be modeled on what was done in the automotive and aerospace industry when those sectors faced hard times. Due to the success of its blockbusters in the 1990s, the pharmaceutical industry only recently faced the same challenges.

I will focus on the content of the drug discovery and development process. How can we scrutinize the R&D projects earlier in the preclinical development process to help minimize the risks involved in clinical development of new drugs (now down to 10% success rates)? We need a better process content in relation to clinical reality, not only more content as such.

Drug discovery and development: an inefficient process

At the end of the drug discovery and development pipeline, there are patients waiting for treatments, company presidents and shareholders waiting for profit and governments trying to balance their health care budget. For pharmaceutical and biotech companies, the critical issue is to select new molecular entities (NME) for clinical development that have a high success rate of moving through development to drug approval. Finding new drugs (which can be patented to protect the enormous investments involved) and at the same time reducing unwanted side effects is vital for the industry. We must try to understand the reasons for failure in clinical development in order to improve drug discovery and preclinical development.

Evolution of Total Sales and Research & Development Spending
Figure 5. Evolution of Total Sales and R&D Spending
Source: Pharmaceutical Research and Manufacturers of America (PhRMA)
Pharmaceutical Industry Profile 2004 (Washington, DC: PhRMA, 2004)
Evolution of Total Sales and Research & Development Spending
Figure 6. Evolution of Total Sales and percentage of R&D Spending
Source: Pharmaceutical Research and Manufacturers of America (PhRMA)
Pharmaceutical Industry Profile 2004 (Washington, DC: PhRMA, 2004)
Evolution of Total Sales and R&D Spending
Figure 7. Evolution of Research and Development Spending Domestic and Abroad
Source: Pharmaceutical Research and Manufacturers of America (PhRMA)
Pharmaceutical Industry Profile 2004 (Washington, DC: PhRMA, 2004)
Evolution of R&D Spending and NDAs submitted
Figure 8. Evolution of Research and Development Spending and NDAs submitted
Source: Pharmaceutical Research and Manufacturers of America (PhRMA)
Pharmaceutical Industry Profile 2004 (Washington, DC: PhRMA, 2004)
FDA CDER NDAs received per year

The demand for innovative medical treatments is constantly growing as people in the wealthy developed world live longer with a concomitant increase in the burden of chronic diseases. At the same time, patient expectations about the quality of treatment and care they receive are rising and unmet medical needs remain high. There still are significant pharmaceutical gaps, that is, those diseases of public health importance for which pharmaceutical treatments either do not exist or are inadequate. What can modern society expect from its pharmaceutical industry to deal with the challenges arising? Let us take a look at the evolution in income of the US pharmaceutical industry and output over the last 30 years, from 1970 to 2003. The total sales of the US pharmaceutical industry has risen almost exponentially over the past 30 years (Figure 5). About 16% of sales income is spent on R&D (Figure 6), which makes R&D, after marketing costs, the second biggest item in the spending profile of large pharmaceutical companies. The percentage of sale income spent on R&D has risen from 9.3% in 1970 to 15.6% on 2003, a rise of 6.3%. The total amount of money spent on R&D has risen enormously since 1970, mostly in the US (Figure 7). In 2003, almost half of all R&D spending worldwide was made in the USA. However despite this almost exponential rise in total R&D spending, the number of NDAs approved by the FDA has not risen significantly (Figure 8). The money invested in R&D has not lead to an equal rise in output of the R&D process. Due to the increasing mismatch between rising R&D expenditure and decreasing R&D efficiency (Figure 8) the overall profit margins of the pharmaceutical industry are decreasing (Figure 3).

Whatever the phrmaceutical industry spends on R&D, it has a significant overhead of additional manpower to sustain. In 2000 the US pharmaceutical industry directly employed 247,000 people (down form 264,400 in 1993), with 51,588 of them working in R&D, which means only 21% of its workforce is directly involved in the drug discovery and development process (Kermani F., 2000; PhRMA 2002). In 2000 the European pharmaceutical industry employed 560,000 people of which only 88,200 worked in R&D, which is 16%. (EU source: The European Federation of Pharmaceutical Industries and Associations (EFPIA) and The Institute for Employment Studies (UK)). In 1990 the European pharmaceutical industry directly employed 500,762 people (76,287 in R&D or 15,2%). It took 10 years to increase total employment to 540,106 people (of which 87,834 in R&D or 16,3%, conflicting data), but then it took only 3 years to increase total employment up to 586,748 (of which 99,337 in R&D or 16,9%). The industry is not capable to reduce its overhead and to significantly increase its new drug generating workforce in relation to its total employment. From 1990 to 2003 the European pharmaceutical industry icreased its workforce with 85,986 , but added only 23,050 for R&D. The expenditures in R&D grow faster than its R&D workforce which indicates that money is being spent mainly on equipment (e.g. for HTS), but which fails to sustain the growth of productivity in the end (Figure 8). Ubiquity does not equal overall process efficiency and effectiveness.

Two elements which are often overlooked in the discussions about the increasing cost and duration of R&D: tax returns and the US Public Law 98-417 (the Hatch-Waxman Act) which was enacted in 1984. Pharmaceutical companies in general spend a certain amount of the revenues on R&D because of its impact on tax returns, so the cost is not the only driver. When sales increase tax deductions are important incentives to spend part of the revenue on investments in R&D (Figure 6). But in the end, the new investments have to support further growth, which is not always the case. The "Drug Price Competition and Patent Term Restoration Act" (1984) was intended to balance two important public policy goals. First, drug manufacturers need meaningful market protection incentives to encourage the development of valuable new drugs. Second, once the statutory patent protection and marketing exclusivity for these new drugs has expired, the public benefits from the rapid availability of lower priced generic versions of the innovator drug (Abbreviated New Drug Applications or ANDA). One aspect of the "Drug Price Competition and Patent Term Restoration Act", the "Patent Term Restoration" refers to the 17 years of legal protection given a firm for each drug patent. Some of that time allowance is used while the drug goes through the approval process, so this law allows restoration of up to five years of lost patent time. Under the Hatch-Waxman Amendments, patent protection can be extended (under certain conditions) for up to 28 years, about 11 years of extra protection compared to the 17 years originally granted by US law. The regulations governing the Patent Term Restoration program are located in the Code of Federal Regulations (CFR), Title 21 CFR Part 60.
The Uruguay Rounds Agreements Act (Public Law 103-465), which became effective on June 8, 1995, changed the patent term in the United States. Before June 8, 1995, patents typically had 17 years of patent life from the date the patent was issued. Patents granted after the June 8, 1995 date now have a 20-year patent life from the date of the first filing of the patent application. Although pharmaceutical companies suffer from longer development cycles, tax incentives and extended patent protection lessen the impact on their business results. The patients are the true losers of the game, because they have to wait longer for new drugs for unmet medical needs. Instead of creating a long-winded and inefficient process, medicine would be served better with a shorter and more productive process.

Evolution of R&D spending allocation
Figure 9. Evolution of R&D spending allocation
Source: Pharmaceutical Research and Manufacturers of America (PhRMA) and
Source: USA NSF Division of Science Resources Statistics (SRS)

Although overall R&D spending has increased over the years, there has been a remarkable shift in the allocation of R&D spending. Clinical development spending has increased significantly (Figure 9), while spending on applied research (i.e. preclinical) has decreased. Basic research spending shows an increase in recent years. The overall picture is an increased spending on clinical development, while there is less spending on the processes feeding clinical development with appropriate development candidates. Mainly the investments in applied research, which is the bridge between basic research and clinical development shows signs of neglect. The Early Development Candidates (EDC) were expected to require less preclinical validation than before?

The cost to develop a single drug which reaches the market has increased tremendously in recent years and only 3 out of 10 drugs which reached the market in the nineties generated enough profit to pay for the investment (DiMasi, J., 1994; Grabowski H, 2002; DiMasi JA, 2003). This is mainly due to the low efficiency and high failure rate of the drug discovery and development process. Pharmaceutical companies are always trying hard to reduce this failure rate. Indirect losses in drug development caused by a failure in drug discovery are among the most difficult to quantify but also among the most compelling in the riskmitigation category. Pharmaceutical companies want to find ways to bring down the enormous costs involved in drug discovery and development (Dickson M, 2004; Rawlins MD., 2004).

Only about 1 out of 5,000 to 10,000 drugs makes it from early pre-clinical research to the market, which is not an example of a highly efficient process. The focus of the pharmaceutical industry on blockbuster drugs is a consequence of the mismatch between the soaring costs and the profits required to keep the drug discovery and development process going. The blockbuster model now delivers just 5% return on investment and only one in six new drug prospects will deliver returns above their cost of capital. The "nichebuster" is now an emerging model for the post-blockbuster era.

Only diseases with patient populations large enough (and wealthy enough) to pay back the costs for a full blown drug development are now worth while working on. Research for new antibacterial drugs is being abandoned, due to an insufficient return on investment (R.O.I.) to pay for the development costs of new drugs (Lewis L, 1993; Projan SJ., 2003; Shlaes DM., 2003). If the industry cannot bring the costs down, it may as well try to raise its income by changing its price policy, but this shifts the solution for the problem from in- to outside the company and places the burden on the national health care systems.

Companies which were more successful in the past achieved a higher efficiency even without the availability of extensive genomic and proteomic data and new low-level disease models. The founder of Janssen Pharmaceutica, Paul Janssen, PhD, MD (1926-2003), in his early days achieved a ratio of 1 drug for every 3,000 molecules screened. Over the years he and his teams developed about 80 drugs (out of 80,000 molecules, so 1 drug for 1,000 molecules screened) of which 5 (6.3%) made it to the WHO Model List of Essential Medicines. He worked in fields as diverse as gastroenterology, psychiatry, neurology, mycology and parasitology, anaesthesia and allergy. As a scientist he has been one of the most highly productive and widely esteemed pharmacological researchers in the world for more than 45 years. He had a deep understanding of both drug discovery and drug development. Dr. Paul Janssen had always been the personification of a unique combination: on the one hand the brilliant scientist, and on the other the very successful manager. Let us take a look at his approach to active strategic management which requires active information gathering and active problem solving. Dr. Paul Janssen practiced Management By Walking Around (MBWA), which gave him access to all the research going on and allowed him to orchestrate the efforts of his scientists, from discovery up to Phase III, like a conductor and thereby avoiding silo development.

A deep understanding of a wide range of issues is required to bring a drug from early drug discovery to the market. Introducing new technology and generating more data alone are not sufficient to improve the drug discovery and development process (Drews J. 1999; Horrobin DF, 2003; Kubinyi H., 2003; Omta S.W.F., 1995). We need better content and understanding, not just more targets and data to be fed into the preclinical and clinical development process. As such the present-day discovery process, suffers from molecular myopia as it lacks the big picture understanding of disease mechanisms in man. In contrast the more traditional physiology based process, suffered from system-wide presbyopia as it lacked molecular resolution. The ideal approach would be the combination of both, which has the potential to improve both the quantity as well as the quality of the process. Quantity without a match in content quality (clinical relevance) leads to failure later on in the drug development pipeline. We have to look at drug discovery and preclinical development with clinical drug development and the patient in mind. Look back from clinical reality into the drug discovery and development process and analyse its failures. A process which in the end fails to prove its value in man should be changed.

The drug discovery and development process

Let us now take a closer look at the evolution of the output of the drug discovery and development over the years. How does the productivity of the process evolves? What is the cost/benefit ratio of the investments made and the overall outcome for drug discovery and development.

NDAs per calendar year
Figure 10. NDAs submitted over the years.
Source: FDA CDER NDAs received and NMEs approved.
INDs and NDAs per calendar year
Figure 11. Evolution of INDs and NDAs over the years.
NDAs left axis, INDs right axis.
Source: FDA.

The number of NDAs submitted does not show a significant increase in recent years (Figure 10), compared to almost 20 and 30 years ago in the days of physiology based drug discovery. The number of approved New Molecular Entities (NME) shows a sharp decrease in the early sixties, due to the more stringent regulations for drug safety testing because of the Thalidomide scandal. About 66% of the NMEs did not make it anymore when better testing was required by the FDA and the pre-Thalidomide productivity was never reached again. The number of NMEs was at its lowest at the end of the sixties (9 in 1969) and has slowly increased since the early seventies (Figure 10). NMEs are about 25% of all NDAs, before halfway the eighties it was on average less than 20%. The number of NDAs is not the only indicator of success for the pharmaceutical industry. Blockbusters generate higher sales per product, so both the number of NDAs as well as the sales per marketed drug are important indicators. Depending on a blockbuster makes a company vulnerable to problems (SAE) with a single drug and patent expiry of a blockbuster has a bigger impact. In figure 7 we can see that the number of INDs and NDAs submitted over the years, does not show a significant improvement. The larger than average number of approvals in 1996 reflects the implementation of the Prescription Drug User Fee Act (PDUFA). The number of INDs does not show a significant increase over the years, so the overall productivity of drug discovery and development has not improved, despite the high investments in Research and Development (Figure 7 and Figure 11). The number of active INDs shows an overall increase, but this only means that the drug development pipelines are filling up because the clinical trials take longer. The in- and outflow of drug development (INDs and NDAs) has not changed in a way to explain the increase in active INDs. The pharmaceutical industry itself expects that products will stay in phases longer than has historically been the case, lowering the probability of a product moving from one phase to another in a particular year. We have not seen a proportional increase in NDA submissions to the FDA, compared to the number of active INDs (Figure 10 and 11).
The main reasons for declining productivity of drug development are:

  • Tackling diseases with complex etiologies, which are not well understood.
  • Demands for safety, tolerability are much higher than before.
  • The proliferation of targets is diluting focus.
  • Genomics has been slow to influence day-to-day drug discovery.
  • A negative impact of mergers on R&D performance?

Cost of NCE Development in Millions of US Dollars
Figure 12. Sources: for 1976 Hansen, 1979; for 1987a Wiggins, 1987;
for 1987b Woltman, 1987c, for 1987c DiMasi, 1991;
for 1990a and b OTA pre-tax, for 2000 DiMasi, 2003.
Differences are also due to out-of-pocket versus capitalized costs.
FDA CDER Original INDs Received per Calendar Year
Figure 13. Source: FDA CDER INDs received per year
FDA CDER NDAs per Calendar Year
Figure 14. Source: FDA CDER NDAs approved per year
FDA CDER NDAs Standard vs. priority Review Perecentages per Calendar Year
Figure 15. Source: FDA CDER NDAs approved per year

Let us now take a closer look at the drug discovery and development process (clinical trials). Although different sources give different outcomes, the trend is one of increasing costs and reduced Return On Investment (R.O.I.). In 2000 it took about US$ 500 to US$ 802 million to develop a new drug and bring it to the market (DiMasi J.A., 2003), which is a significant rise since 1976 when it cost about US$ 137 million (all numbers in year 2000 US$) (Figure 12). These estimates include opportunity costs, which are lost profits that could have been realized if the money tied up in an enterprise had been invested elsewhere (DiMasi J.A., 2003). Almost half of the DiMasi (Tufts) US$ 802-million figure - $399 million - is comprised of this "cost of capital", leaving a figure of US$ 403 million for direct out-of-pocket expenses, most of which is expended in clinical trials. Whether you favor US$ 403 million or US$ 802 million, the cost of drug discovery and development is far too high. When we look into the diferent stages of drug discovery and development, the US$ 802M costs are divided over: Discovery and preclinical testing US$ 335M, Phase I: US$ 141.7M, Phase II: US$ 137.2M and Phase III: US$ 174M. The total cost for clinical development is US$ 452.9M. The cost for FDA Review/Approval: US$ 13.8M.

The basic numbers for time spent and costs made in drug discovery and development can be found in several documents published by institutes which generate reports about the pharmaceutical industry (Boston Consulting Group, Tufts Center for the Study of Drug Development, Pharmaceutical Research and Manufacturers of America (PhRMA), the Institute for Regulatory Science (RSI), CMR International, etc.). IMS is a source for pharmaceutical market information. The Association of Clinical Research Professionals (ACRP) and the Center for Information and Study on Clinical Research Participation (CISCRP) provide information on clinical trials.

To be complete, there are alternative views which criticize the calculation of the cost of drug discovery and development. Here are the Public Citizen and TB Alliance reports. A discussion of these reports and the Tufts study can be found here. Although an open and critical discussion is the only way to understand complex issues such as research and development costs, the discussion sometimes loses its focus and becomes tainted by sophisms to support political and personal agendas. I leave it to the critical reader to decide. The consequence of accepting the alternative views would be that the pharmaceutical industry would be losing money due to costs outside its core mission, which is even worse, because research and development can be improved, but this would not help in this case. The result is in each case, that drugs are only worth while to develop, if they have an enormous market potential (large numbers of wealthy patients), require as little as possible investments (me-too drugs and generics) and have the shortest development cycle possible (less complex diseases). Otherwise they do not earn back the money invested, when finally they reach the market. This leads to an increasing focus on typical Western "diseases" such as obesity or hypercholesterolemia, due to overintake of food and unhealthy living. Tropical diseases, if they do not make it to the wealthy world, are to be avoided. You cannot blame the pharmaceutical industry, because if they do not live up to the expectations of their shareholders, they are punished by a decreasing stock-value (see Figure 1).

The number of INDs coming out of drug discovery does not show a significant improvement since 1992 although overall costs have risen sharply.(Figure 13 and 14). The non-inovative drugs get a standard review by the FDA instead of a priority review and constitute about 75% of all NDA submissions (Figure 15).
About 10-20 % of the total costs are due to the drug discovery process, the rest is caused by drug development, production and marketing costs. Clinical development costs, on average US$ 467 million, which makes up more than half the total cost. The cost of a Phase I clinical trial is about US$ 15.2, for Phase II it costs about US$ 16.7 and Phase III US$ 27.1 (in 2000 US$, DiMasi J.A., 2003). The cost of a Phase III clinical trial ranges between US$ 4 million and US$ 20 million and you need at least two of them (Kittredge C, 2005). Study delays, such as slow patient recruitment, protocol amendments and review processes, are contributing factors. Every day that a drug is prevented from being on the market means a loss of sales, which in the case of blockbuster drugs can be as much as US$ 45 million per day.

Since pharma's "big wave of innovation" during the early 1990s, peaking in 1996 when 131 new drug applications (NDA) were filed and 53 new molecular entities (NME) were approved, R&D productivity has fallen by half. The larger than average number of approvals in 1996 reflects the implementation of the Prescription Drug User Fee Act (PDUFA). In 2003, 72 NDAs were filed and 21 NMEs approved - a 45% and 60% decline, respectively, since 1996 (Figure 13). The FDA provides the CDER Drug and Biologic Approval Reports. There is also a slow but steady increase in the relative number of drugs which appear to have therapeutic qualities similar to those of one or more already marketed drugs (FDA Standard Review procedure) in contrast to the drugs which show a significant improvement compared to marketed products in the treatment, diagnosis, or prevention of a disease (FDA Priority Review procedure)(Figure 14). The approval of NDAs shows a time shift in relation to the INDs submitted some years before, due to the time required for reviewing the data (compare Figure 13 and 14). IND peaks translate into smaller NDA approval peaks some years later, due to late stage attrition in drug development. About 90% of INDs do not make it to an NDA approval years later.

From about 8 years in the 1960s it now takes an average pharmaceutical company about 10 to 15 years to bring one new drug to the market. Of these 15 years about 6.5 years or 43% of the total time is spent in pre-clinical research. Development starts with candidate/target selection or the selection of a promising compound for development. Pre-clinical and non-clinical research involves necessary animal and bench testing before administration to humans plus start of tests which run concurrently with exposure to humans (e.g. two-year rodent carcinogenicity tests). About 7 years or 46 % of the total time is time spent in clinical research (1.5 years in Phase I, 2 years in Phase II and 3 years in Phase III). Phase I (First Time In Man, FTIM) of a clinical trial deals with drug safety and blood levels in healthy volunteers (pharmacology). Phase II (Proof of concept, PoC) deals with basic efficacy of a new drug, which proves that it has a therapeutic value in man (exploratory therapeutic). Finally Phase III deals with the efficacy of the drug in large patient populations (confirmatory therapeutic). It is easy to understand that the increase of the population used to study the effect has a dramatic impact on the complexity and the cost of the clinical trial.

To process a New Drug Application (NDA) takes the U.S. Food and Drug Administration (FDA) on average 1.5 years based on the results and documents provided by the pharmaceutical industry. The situation in Europe for the European Medicines Evaluation Agency (EMEA) is probably of the same order of magnitude. About 0.1 % of the original molecules screened in drug discovery enter phase I (5 out of 5,000 to be optimistic) and 0.02 % of the original molecules finally reach the FDA (1 out of 5,000). Of the 5 molecules entering phase I, about 4 out of 5 or 80 % fail to make it to a NDA. After approval by the FDA, the drug hits the market and enters phase IV of the clinical study process.

In the 1990s about 38 % of the drugs which came out of discovery research dropped out in phase I. Of those molecules which made it out of phase I, 60 % of those failed in phase II clinical studies. And now we get to the really expensive phase III in which 40 % of the remaining candidates failed. Of those drugs which made it out of phase III to the FDA 23 % of the ones that made it through the clinical trials failed to be approved by the FDA. All this translates to about 11 % overall success rates from starting the clinical trials (Kola I., 2004).

NDAs over INDs
Figure 16. Less than 10% of INDs make it to an NDA.
Source: FDA CDER NDAs approved per year and FDA CDER INDs received per year
More attrition - less success
Figure 17. Overall success of clinical development decreased from 18% to 9%,
worst decline in Phase II (effectiveness), from 46% to 28%
Source: Loew C.J., PhRMA, HHS Public Meeting, November 8, 2004
Evolution of attrition
Figure 18. Evolution of attrition from 1995 to 2004.
Source: Pharmaceutical Research and Manufacturers of America (PhRMA)
Attrition per therapeutic area
Figure 19. Trends in probability of success from 'first human dose' to market by therapeutic area.
Source: Pharmaceutical Research and Manufacturers of America (PhRMA)

As a rough indication of overall inefficiency we can compare FDA NDA and IND data five years different. If we take on average 5 years from IND (IMP in Europe) after 5 years of IND filing, less than 10% of INDs make it to an NDA (Figure 16). The evolution of NDA approvals also shows a decline over the years. In recent years overall success rates for clinical development decreased from 18% to 9% (Figure 17). This is mainly due to an almost 40% reduction of success in phase II clinical trials, which means a failure in exploratory treatment or clinical activity. A Phase II clinical trial is intended to determine activity, it does not yet determine efficacy, which is the goal of a Phase III clinical trial. Thus the outcome of Phase II is a decisive point in a drug's development. If we look at the evolution of attrition rates from 1995 to 2004, we see an overall increase in development candidates in preclinical development and an increase in Phase I and II development (Figure 18). There is no significant increase in Phase III clinical trials, as most developmental drugs increasingly fail in Phase II. The drugs show an activity in drug discovery and preclinical development, but no significant activity in a clinical situation on a real-life disease process. The increase in attrition is not the same for every therapeutic area (Figure 19). For alimentary and metabolic diseases the probability of success (POS) is even increasing and is about then times as high as for the nervous system (1999). The high success rate of anti-infectives is also caused by the fact that we are capable to make a valid representation of the entire target system (bacteria) in a "test-tube" or "petri-dish" early on in the process, and not only a billion-fold reduced representation of the human biosystem. As long as the dominant view on applied science remains that a set of molecules in a "test-tube" can represent the complexity of system (reductionism) our models will disappoint us at the end of the process when there is no escape from the complexity and variability of man and human population. Leaving too much of the original system out of an experiment brings too much flaws into the experiment. Elimination without confirmation of validity against the original condition gives flawed results (and late stage attrition).

The significance of increasing Phase II failures is a new evolution, as in the 1980s and early 1990s the failure rates remained relatively steady. The failure rate of new clinical entities (NCEs) remained relatively steady through the 1980s and early 1990s (DiMasi J.A., 2001). Among NCEs for which an investigational new drug (IND) application was filed in 19811983, approval success rates were 23.2%; 19841986, 20.5%; 19871989, 22.2%; and 19901992, 17.2%. This includes both self-originated and acquired NCEs. According to the FDA historically 14% of drugs that entered Phase I clinical trials eventually won approval, now 8% of these drugs make it to the marketplace, and that half of products fail in the late stage of Phase III trials, compared to one in five in the past (Crawford L.M., 2004). "...In the past, we used to see a 20 % product failure in the late stages of the Phase 3 trials. Currently, the failure ratio at this stage is 50 %. The reason for this unpredictability, in our analysis, is the growing disconnect between the dramatically advancing basic sciences that accelerate the drug discovery process, and the lagging applied sciences that guide the drug development along the critical path. ..." (Crawford L.M., 2004). Overall late stage attrition is on the rise, but how should preclinical development and Phase I clinical trials predict success or failure in Phase II or III, when they are not conceived or designed to do this? Each stage from discovery over preclinical development to clinical development is meant to provide an answer for a particular question, not for the question arising in the next stage of the discovery and development pipeline. Which elements or markers in preclinical development would allow us to predict events in clinical development. In order to achieve this we need a better understanding of the critical issues in the clinical disease process. The analysis of failures in Phase II should at least help us to understand the mechanism of these failures in order to feed those lessons back into preclinical development. The transition from preclinical development to Phase I and Phase I itself deals with finding a appropriate dosing scheme to start with (e.g. MTD Maximum Tolerated Dose), but not yet with clinical activity, which comes into play at Phase II.

There are some practical considerations to determine the clinical activity of a developmental drug, one of which is the sample size. The study design (case/control, cohort study, RCT, etc.) is the first decision, but sample size is a close second. An important issue is the power of the trial. Once the level of activity that is of interest has been decided on, one should design a trial that exposes the fewest possible patients to inactive therapy, e.g. by appplying the method of Gehan and Schneiderman (Gehan E.A., 1990). In general you need more patients when you want to find out about a smaller therapeutic effect. This is an important cause of the overall increase in patient numbers required, depending on what you want to prove. When we cannot achieve a dramatic therapeutic breakthrough with a diseases, a small improvment is what we want to prove. Instead of a revolutionary breakthrough, quite often therapeutic improvements are only incremental. Let me clarify this with an example.
When Louis Pasteur (1822 1895) developed a vaccine against rabies, the shortterm outcome was clear, either you died or you survived. Rabies is a viral disease with about 100% mortality, i.e. you almost always die when you get the disease. So the therapeutic effect was very simple to assess, which also made complicated analysis of the therapeutic results less necessary. There was also less consideration about possible side effects, as dying from rabies was a horrible disease process.
Let us now take a look at Alzheimer's Disease (AD) (named after Alois Alzheimer), a debilitating degenerative disease of which the pathological process is still not well understood. We cannot achieve a "restitutio in integrum" (restoration to original condition) and regrow the brain cells which are lost due to the disease process. So, now we can decide to wait until we know all about the process and then start developing a cure. This would mean that in the mean time we do nothing to help whatsoever. As you can understand, this is not a valid option. In the mean time, therapy is aimed at slowing down the process of mental deterioration. This however is a more subtle outcome than the short-term live or die outcome in the case of rabies. These less than 100% success rates make it harder to prove the success of a new therapy. The need to prove a small improvement, makes clinical trials more complex and much larger.

Sample size for comparing two means
Figure 20. Sample size (N) for comparing two means.
In addition to α and β, N only depends on Δ/σ, or the effect size.
α = 0.05 and 1 - β = power = 90% for a 2-sided test.
The graph shows N as a function of Δ/σ = difference in units of s.d.
Sample size for comparing proportions
Figure 21. Sample size (N) for comparing proportions (p).
In addition to α and β, N depends on p and Δ.
Let α=0.05, 1- β = power = 90%, 2-sided testing, p=0.5 (conservative estimate for variance).
The graph shows N as a function of Δ = difference p1-p2, e.g. 0.2 = 0.6 - 0.4

Designing a clinical trial is not a trivial endeavour as the days of Louis Pasteur are gone and the environment in which to develop new therapies has changed dramaticaly. A clinical trial requires careful design in order to be able to answer the research question (hypothesis) with some confidence in the answer. You want to prove that a new therapy works in a reliable way. Traditionally, H0 is the hypothesis that includes equality or the expectation that nothing will happen and the alternative hypothesis H1 that something significant will happen (Rosner B, 1995). A p-value is a measure of how much evidence we have against the null hypotheses. The significance test yields a p-value that gives the likelihood of the study effect, given that the null hypothesis is true. A small p-value provides evidence against the null hypothesis, because data have been observed that would be unlikely if the null hypothesis were correct. Thus we reject the null hypothesis when the p-value is sufficiently small. However life in clinical trilas is not that simple. There are two type of statistical errors you can make in a trial. A Type I error occurs when you reject H0 when H0 is true, i.e., you declare a significant difference when the result happened by chance (false positive - a drug will be used while it is not effective). A Type II error occurs when you accept H0 when H1 is true, i.e., you say there is no significant difference when there really is a difference (false negative - a drug will not be used while it has an effect). How do we deal with these issues? While we cant prevent the possibility of incorrect decisions, we can try to minimize their probabilities. We will refer to alpha (α) and beta (β) as the probabilities of Type I and Type II errors, respectively.

  • Alpha (α) is the probability of making a Type I error (rejecting the null hypothesis when the null hypothesis is true).
  • Beta (β) is the probability of making a Type II error (accepting the null hypothesis when the null hypothesis is false).
or
  • Significance level or α = P[Type I error] = P[Reject H0 | H0 true] .
  • β = P[Type II error] = P[Accept H0 | H0 false]
  • Power = 1 - β
An interesting element of a trial is the power of the trial. A study can have too little power to find a meaningful difference, when the sample size is too small. No significant difference is found and the treatment or method is discarded when it may in fact be useful. The alternative Hypothesis (H1 or Ha) is that there will be a significant (therapeutic) effect. The P(Type II error) = β and β depends on how large the effect really is. The power (P) of a test is the probability that we reject the null hypothesis given a particular alternative hypothesis is true and Power = 1 - β. Summarized: β = Probability(missing the difference) and Power = Probability(detecting the difference).

All this comes down to the overall rule that in order to prove a small decrease in disease progression we need a relatively large number of a patients. It is because of this kind of effect, the size of patients in clinical trials has risen dramaticaly in recent years. Also in the case of rabies, there was no effective treatment to compare with, so the comparison was straightforward and simple. In Figure 20 and Figure 21 you can see for two different types of trials, the effect of sample size required to detect an increasing difference. This is an important reason for having patient populations of up to 5,000 patients in Phase III clinical trials. If we could make a big difference with a treatment, then we would not need such large numbers of patients to prove our case. With a chronic degenerative disease, reducing the speed of progress of the disease with only 0.1%, could mean that in 20 years thousands of people would benefit (longevity in the Western world), but the problem is that you must prove this small difference within the scope of a clinical trial. This is one of the most important reasons for clinical trials to become increasingly global in nature and more complex in protocol design. The difference with the 19th century is also that we now have to compare with drugs which are already on the market and have a proven therapeutic effect. The pharmaceutical industry is increasingly challenging itself to improve against its own therapeutic success of the past. As such the pharmaceutical industry itself is the biggest problem for the pharmaceutical industry. There is a lot more to be told on on clinical trial design, but this is not within the scope of this article. The main issue is that in modern clinical development, the situation is more complex to evaluate than before.

In inductive research, applying statistics has to be done with care. Expanding a trial population beyond the boundaries of statistical relevance, may lead to spurious statistical significance but will not improve the correlation to clinical relevance. By doing this we increasingly feed the process with false positives and increase the pressure towards the end of the pipeline. The basic principles of probability (significance) and induction (relevance) should be taken into account when designing and performing experiments.

Why drug development fails
Figure 22. Despite a reduction in attrition due to pharmacokinetics issues, efficacy has not improved, since 1991.
In 1991 40% of PK failures were caused by poorly bioavailable anti-infectives,
when we remove these from the equation, then only 7% of failures in 1991 were caused by poor ADME.
Source: Pharmaceutical industry attrition profiles, evolution (Kennedy, T., 1997; Prentis RA, 1988).
Where drug development fails
Figure 23. The major cause for failure, efficacy, only becomes apparent late in development.
Source: KMR Group 1998 - 2000

What about the evolution of the basic reasons for attrition in drug development? Attrition due to a lack of efficacy of drugs in development has not improved since 1991 (Kennedy, T., 1997; Prentis RA, 1988) (Figure 22). Attrition rates due to poor pharmacokinetical profiles (PK) have dropped significantly, due to better preclinical in-vitro and in-vivo models. However about 40% of failures in clinical development were due to inappropriate pharmacokinetics of poorly bioavailable anti-infectives, if those were removed from the equation then ADME was only responsible for 7% of failures in 1991 (Kennedy, T., 1997). The basic numbers on attrition causes explain why attrition rates in Phase I clinical trials have declined less than those in Phase II. Drugs with unfavorable PK profiles are now increasingly stopped before they reach clinical development, so the ineffective ones now make it into Phase II in relatively larger numbers. The clinical development attrition trends also show an unfavorable evolution since 1991 (Figure 22). The disease models used in drug discovery and preclinical development fail to predict failure in clinical development in about 80 to 90% of the drugs which enter clinical development. And the combined predictive power of all clinical trials (Phase I to III) fails to predict failure in 1 out of four or 25% or even 50% of all drugs submitted to the FDA for approval.
The major cause of attrition, efficacy, also shows up late in development, as preclinical development and Phase I are unable to detect this failure. Preclinical development lacks the proper predictive models and Phase I is not designed to detect a failure in efficacy. Clinical safety issues increase with the number of people taking the drug, after it is on the market (Figure 23).

The reason for the up to 90% failure in clinical development is both related to the target (lack of efficacy, mechanism related toxicology) and to the compound (pharmacokinetics, chemistry related toxicology). A decade ago, the number of drugs failing preclinically due to poor pharmacokinetics was upwards of 40%, but improved in vitro and animal models have reduced that rate to about 10%. Failures due to toxicology, however, are still in the 30% to 40% range, making it the number one reason for preclinical attrition. "...The main causes of failure in the clinic include safety problems and lack of effectiveness: inability to predict these failures before human testing or early in clinical trials dramatically escalates costs. ..." ( Innovation and Stagnation: Challenge and Opportunity on the Critical Path to New Medical Products)

What can we learn out this numbers and what is being done in drug discovery? The role of absorption, distribution, metabolism, excretion (ADME) and toxicity (ADMET) is an important part of the drug discovery process as ADMET is an important cause of failure in drug development (Yan Z, 2001; Lin J, 2003; Nassar AE, 2004). Pharmaceutical profiling assays provide an early assessment of drug-like properties, such as solubility, permeability, metabolism, stability and drug-drug interactions (Di L., 2005). The drug discovery process (target identification, target validation, lead identification/optimization ) and preclinical development such as ADMET studies, fail to predict the failure of a drug in clinical development for 4 out of 5 or at least 80 % of the molecules which enter phase I. The rates of failure in expensive Phase III trials in oncology are the worst in the industry (Kamb A., 2005). Improving the predictive power of disease models in drug discovery, preclinical development and ADMET is an important issue to reduce the late stage attrition rate in drug development.

A new drug spends about 90 % or 13.5 years of his career within the discovery and development process, before it reaches the FDA for the last 10 % or 1.5 years. So the FDA does not account for the majority of the time it takes to bring a new drug to the market, nor does it account for the majority of failures which is only 20-25 % or 1 out of 5 or 1 out of 4 drugs which enter phase I or 1 out of 5,000 (0.02 %) if we start from the beginning of the process. Although the investments in the early stages of the drug discovery process have increased tremendously, this means nothing compared to the cost of failure in phase III of a clinical trial.

The manufacturing process

After the drug discovery and development is finished for a particular drug, the drug enters the market and is being manufactured. Making manufacturing more efficient is also an imperative for the pharmaceutical industry. The 16 largest drug companies spend more than twice as much on manufacturing as they do on R&D, according to a recent study by GlaxoSmithKline, Brentford, UK. These large companies spent $90 billion, or 36% of their expenses, on manufacturing in 2001, compared to some $40 billion or 16% on R&D. One of the most important reasons for horizontal mergers in the pharmaceutical industry is to reduce the operational costs of manufacturing.

Companies are under increased regulatory pressure for manufacturing, such as the Good Manufacturing Practice Guide for Active Pharmaceutical Ingredients (ICH Q7A), FDA Good Manufacturing Practice (GMP) and product labeling. The Good Automated Manufacturing Practice (GAMP) organization was founded in 1991 by pharmaceutical experts to meet the evolving FDA expectations for GMP compliance of manufacturing and related systems. Impending requirements being imposed by the FDA in the U.S. and the EMEA in Europe require companies to submit product labeling content in highly structured XML formats (Structured Product Labeling (SPL) in the US and Product Information Management (PIM) in Europe). Distribution of drugs is regulated by the Good Distribution Practice (GDP) of Medicinal Products for Human Use. New initiatives are being taken to improve the overall manufacturing process. Process Analytical Technology (PAT) provides a framework for innovative pharmaceutical manufacturing, control and product quality assurance. FDA Process Analytical Technology Initiative (PAT). The EUFEPS Process Analytical Technology Sciences.

The FDA wants to deal with the growing public health problem of counterfeit prescription drugs in the United States. Counterfeit drugs are not only illegal but are also inherently unsafe. A famous case of the withdrawal of a drug due to deliberate product tampering was the Tylenol murder case. The Tylenol murders occurred in the autumn of 1982, when seven people in the Chicago, Illinois area in the United States died after ingesting Extra Strength Tylenol medicine capsules which had been laced with cyanide poison. This incident was the first known case of death caused by deliberate product tampering. Johnson & Johnson was praised by the media at the time for its handling of the incident, although it cost the company about US$ 100M in lost revenues (see also Johnson & Johnson Credo). In the near future the FDA will require that the industry to implement full-scale RFID serialization (needed for closed-loop drug tracking) and electronic pedigree (ePedigree) applications (needed to find and prosecute violators) The Radiofrequency Identification Technology (RFID) is meant to monitor and protect the U.S. drug dupply chain. Radio Frequency IDentification (RFID) is an automatic identification method, relying on storing and remotely retrieving data using devices called RFID tags or transponders. In general, authentication systems that operate independently from the underlying data collection technology will help the drug industry secure the drug supply, protect valuable brands, and avoid legislation that will force costly compliance requirements that add little business value.

Inspection by the FDA are not to be taken lightly. Pharmaceutical, Medical Device, Biopharmaceutical and Generic Drug companies all face a common dread. The FDA has called, and they are coming to audit their manufacturing facility. At the "FDA's Electronic Freedom of Information Reading Room", the FDA publishes the findings of its inspections on-line: Warning Letters and Responses.

Vulnerability in Phase IV

The pharmaceutical industry depends on a relatively small number of active components. While there are around 10,300 FDA-approved drugs in the United States today, most of these are made up of some combination of only 433 distinct molecules. Half of these 433 molecules were approved before 1938, and at least 50 are "me too" drugs, a slightly modified form of a compound already on the market. Finally, there are only eight major, chemical "scaffolds" upon which all the 433 molecules are based.

Due to the difficulty and inefficiency of the drug discovery and development process, pharmaceutical companies rely on only a few drugs for their income and profit. This makes them extremely vulnerable for massive income loss when one of the drugs encounters problems after it is on the market. Serious problems with a drug after it has been on the market in general means lawsuits against the company and a serious blow to its reputation (e.g. pharmacovigilance or Phase IV trial). Each year about 17,200 Adverse Events (AE) and 800 Serious Adverse Events (SAE) are typically reported to the FDA for newly approved drugs (Source: FDA). Seven of the 303 (2.3%) new molecular entities (NME) approved by the FDA between January 1994 and April 2004 were withdrawn from the market due to safety concerns. Although 97.7% of NMEs do not cause such safety problems, the 2.3% which do, bring the pharmaceutical industry in trouble. Older drugs can also be a major cause of hospital admissions, such as with aspirin (Pirmohamed M, 2004). The perception that new drugs are less safe than older ones is not always true. However the accompanying harm to patients and the billions spent developing and marketing the drugs are a big problem for the industry. No amount of testing can guarantee to find all of the possible side-effects for every person who may take a medicine. A reaction which occurs at a rate of 1 in 100,000 people or even at a higher rate of 1 in 10,000 for instance, may not be seen until very large numbers of people use the medicine. Even the largest clinical trials are underpowered to detect rare events before a drug hits the human population at large. The increasing attention to chronic diseases for which there is no "restitutio ad integrum" possible, but only long-term treatment also increases the exposure of individual patients to the drug, which we cannot foresee during clinical trials before the drug hits the market. Even with these odds, no pharmaceutical company wants to be in the news with Serious Adverse Events (SAE) about a drug already on the market. Being the CEO of a pharmaceutical company is not something for the faint of heart. One day a company is praised for a new breakthrough drug, the next day it has its name in the news associated with lethal side effects of another drug. Some recent events have shown that pharmacovigilance principles and procedures are in need for improvement. The EU European Risk Management Strategy (ERMS) of 2002 is an example of such an initiative. It aims at strengthening the EU Pharmacovigilance System (see also EudraVigilance).

Current methods in pharmacovigilance often use monitoring and simple analysis of safety signals after they have been detected in the postmarketing process. Sometimes Phase IV clinical trials (postmarketing) reveal important side effects which were not discovered before. This was the case for Vioxx according to the APPROVe study by Merck. The cost of missing a safety signal or not detecting it before it affects the general population is huge. The withdrawal of a drug from the market has serious consequences both due to the loss in revenue for the company and the financial consequences of lawsuits. As was the case with Vioxx, a pharmaceutical company may turn to unethical and unlawful pressure on scientists and doctors to protect its commercial interests (Moynihan R., 2009; Cahana A, 2006). Manipulating opinion about a drug as was the case with Vioxx in the ADVANTAGE seeding trial of Merck and guest authorship and ghostwriting is not unusual in the pharmaceutical industry (Hill KP, 2008; Ross JS, 2008).

The cost of an adverse drug reaction on an average per patient basis is about € 2800 (approximately US$ 3,360) in hospitalization costs alone (Gautier, 2003). The total losses to a company can reach billions of dollars from the loss of reputation and revenue and from medical and litigation expenses.

Some examples of Serious Adverse Events (SAE) over the years give an indication of the impact on the lives of people, society and the pharmaceutical industry. An inadvertently toxic preparation of sulfanilamide had a central influence on the US Food and Drug Administration (FDA). A preparation called "Elixir Sulfanilamide" contained diethylene glycol as a solvent, which is toxic. This preparation killed over one hundred people, mostly children, and led to the passage of the 1938 Food, Drug, and Cosmetic Act (the 1937 Elixir Sulfanilamide Incident). Thalidomide (Softenon) was withdrawn from the market in the sixties when thousands of babies were born with deformities as a result of their mothers taking Thalidomide during pregnancy (McBride WG, 1961). Thalidomide never made it to the USA in the sixties, mainly due to Dr. Frances Oldham Kelsey of the FDA, who refused to authorize thalidomide for market when she had serious concerns about the drug's safety. In the USA the Thalidomide case lead to the Kefauver-Harris Drug Amendments (1962) to be applied retroactively to the Federal Food, Drug, and Cosmetic Act (1938). In Europe the Thalidomide case lead to the first European Community pharmaceutical directive issued in 1965, namely Directive 65/65/EEC1. No medicinal product should ever again be marketed in the EU without prior authorisation. On 16 July 1998, the FDA announced the approval of Thalidomide for Hansen's Disease (Leprosy) for erythema nodosum leprosum (ENL). This imposed unprecedented authority to restrict distribution (Thalidomide Education and Prescribing Safety oversight program- S.T.E.P.S).

Several notorious cases of adverse events have been widely publicized in recent years. One is the case of cerivastatin (Baycol, a popular cholesterol-lowering drug) from Bayer. In 2001 cerivastatin (Baycol) was removed from European and USA markets because of the risk for rhabdomyolysis (Bayer, 2001; Furberg CD, 2001; Davidson MH., 2002; Kind AH, 2002; Ravnan SL, 2002; Staffa JA, 2002; Maggini M, 2004). In 2001 when the drug was recalled, there were approximately 700,000 users of the drug. The initial cost of the recall was US $20 million in refunds for active prescriptions. (Eakin, 2003) An additional US $705 million in lost operating earnings and more than US $150 million in out-of-court settlements magnified the negative financial impact.

Prepulsid was withdrawn form the market due to cardiovascular adverse effects (Griffin JP., 2000; Wilkinson JJ, 2004). In late 2003 there was the SSRI case, concerning the antidepressant medicines known as selective serotonin reuptake inhibitors (SSRI). The SSRIs were associated with an increased risk of suicidal behavior (Fergusson D, 2004; Gunnell D, 2005). In 2004 the COX-2 inhibitor rofecoxib (Vioxx) was withdrawn because of cardiovascular adverse effects (Dyer C., 2004; Juni P, 2004).

The sales and marketing process

The sales and marketing of drugs is also highly regulated. The Federal Food, Drug, and Cosmetic Act (the act) requires that all drug advertisements contain (among other things) information in brief summary relating to side effects, contraindications, and effectiveness. In the US, the FDA Office of Medical Policy, Division of Drug Marketing, Advertising, and Communications (DDMAC) takes care of this. One of the most important reasons for horizontal mergers in the pharmaceutical industry is to reduce the operational costs of Sales and Marketing.

Improving the process

"...If biomedical science is to deliver on its promise, scientific creativity and effort must also be focused on improving the medical product development process itself, with the explicit goal of robust development pathways that are efficient and predictable and result in products that are safe, effective, and available to patients. We must modernize the critical development path that leads from scientific discovery to the patient..."
Innovation and Stagnation: Challenge and Opportunity on the Critical Path to New Medical Products, FDA ( March 2004)

In 2000, EUFEPS established the New Safe Medicines Faster Project, the ultimate goal of which would be to contribute to effective development of medicines for the benefit of the European citizens. In a Workshop, held on March 15-16, 2000, in Brussels, ideas and suggestions for research topics, methodologies, techniques and other means of promoting the drug development process were identified, put together and published in the Workshop I Report. In the future, it was sugested, identifying new technologies, capable of more effective selection, development and approval of new, innovative and safe drugs; using such technologies to increase the capacity and speed of the pharmaceutical development process; and cultivating a pan-European interdisciplinary network to bridge the existing gap between industry, academia, health care and regulatory authorities; would to be of paramount importance.

Budget spending as percentage of total budget
Figure 24. Budget spending as a percentage of total R&D budget (US$ 935M)
Source: Life Science Insights, Ernst & Young, Tufts CSDD and Boston Consulting Group, July 2004.
Budget spending as percentage of total budget
Figure 25. Time spending as a percentage of total R&D time (14.5 years)
Source: Life Science Insights, Ernst & Young, Tufts CSDD and Boston Consulting Group, July 2004.
Burnrate
Figure 26. Budget spent and remaining as a percentage of total R&D budget (US$ 935M)
Source: Life Science Insights, Ernst & Young, Tufts CSDD and Boston Consulting Group, July 2004.
Burnrate
Figure 27. Time spent and remaining as a percentage of total R&D time (14.5 years)
Source: Life Science Insights, Ernst & Young, Tufts CSDD and Boston Consulting Group, July 2004.
Burnrate
Figure 28. Burnrate of budget for each individual phase.
Source: Life Science Insights, Ernst & Young, Tufts CSDD and Boston Consulting Group, July 2004.
Burnrate
Figure 29. Cumulative burnrate of overall process. Additional impact of a phase on overall burnrate.
Source: Life Science Insights, Ernst & Young, Tufts CSDD and Boston Consulting Group, July 2004.

Every project or process has a time, cost and quality, which are important parameters when we want to improve its performance. When we look at drug discovery and development, we look at a process which is applied on particular R&D projects. Do we apply the right process on our individual projects? Let us now take a look at the cost and time of the overall R&D process, which nowadays starts with target identification and target validation. We already know that the output of the overall process is low (90% attrition in clinical development). When we take a look at our budget (US$ 935M), we spend about 18% on target identification, qualification and prioritization, 22% on target validation and we spend about 22% on Phase III clinical trials (Figure 24).
When we take a look at the most time consuming phases, we spend almost 21% of our time on preclinical studies and about 21% on Phase III clinical trials (Figure 25).
By the time we are finished with lead identification and optimization, we have spent about 40% of the R&D budget. By the time we reach Phase III of clinical development, we have spent about 80% of our budget. From target identification to preclinical studies it takes about 66% of our total R&D budget, which leaves us with 33% for clinical development (Figure 26).
When we take a look at the time, we spend about 60% of our time from target identification to preclinical studies, which leaves us with 40% of our time for clinical development (Figure 27).
Let us now take a look at the burnrate of our budget per unit of time. At its start the process resembles a fighter jet taking of full throttle forward, afterburners glowing and racing towards the sky. When we reach preclinical development the process resembles a caravan of mice and men crossing the (pre-)clinical-desert until we reach Phase III (Figure 28 and Figure 29).
Compared to the "primitive" physiology based (empirical) process we have added a target identification and validation step in-front of the process, which consumes about 40% of our budget and 20% of our time, but we have neglected to balance this investment with the quality of its predictive power in relation to the clinical outcome of the process (75% failure due to biological reasons). Improving a process requires a balance between cost, time and quality. Target identification and validation should be done with an in-depth patho-physiological understanding of the biological process at a molecular level and not only the target on itself. Try to understand the system of biology and the biology of the system, not only the mechanics of target-drug interaction.

In order to improve the drug discovery and development process, where should we try to optimize it? We have to balance time, cost and quality. Adding more steps in front of the process as with target identification and validation is not an issue anymore. Instead we should do things different and improve the time, cost and quality of what we are doing in a balanced way. A critical path defines the optimal sequencing and timing of interventions by all stakeholders involved in a procedure (Coffey RJ, 1992; Kost GJ., 1983; Kost GJ., 1986). Critical paths have to be developed through collaborative efforts of basic and applied scientists, managers and others to improve the quality and value of drug discovery and development. Unbalanced changes in a project process (scope, time, cost, quality), lead to a disproportionate decline in performance. Quality should be measured against the impact on clinical success and not only on the next step in the process. After about 7 or more years in pre-clinical research, a new drug is ready for filing an initial new drug application (IND) after which the FDA's Center for Drug Evaluation and Research (CDER) monitors the clinical studies. The CDER monitors the study design and conduct of clinical trials to ensure that people in the trials are not exposed to unnecessary risks.
The Center for Biologics Evaluation and Research (CBER) is the Center within FDA that regulates biological products for human use under applicable federal laws. Biologics, in contrast to drugs that are chemically synthesized, are derived from living sources (such as humans, animals, and micro-organisms). The FDA monitors the participants of clinical trials (FDA/ORA Bioresearch Monitoring Information Page). In Europe the European Medicines Agency (EMEA) is a decentralised body of the European Union with headquarters in London. The Committee for Medicinal Products for Human Use (CHMP), deals with medicinal products for human use. In Europe the EMEA is the bridge between the pharmaceutical industry and the national "Competent Authorities". In Europe an Investigational Medicinal Product (IMP) is the name for a drug in clinical development. In Europe a Development Medicinal Product (DMP) is a medicinal product under investigation in a clinical trial in the EEA, which does not have marketing authorization in the European Economic Area (EEA).

The clinical trials, from phase I to III are highly regulated and a company can only optimize the flow of events, but up to a large part it cannot decide freely what needs to be done in these stages of the process (e.g. ICH E6 Good Clinical Practices). The ICH develops guidances for harmonisation of drug development on Quality (Q), Safety (S), Efficacy (E) and Multidisciplinary (M) topics. Once a drug hits a regulatory authority, such as the FDA (CDER) or the EMEA strict rules need to be followed for the approval and failure to comply will only delay this process. The European legislation on pharmaceuticals can be found in EudraLex - The Rules Governing Medicinal Products in the European Union

So it is by improving the quality and shortening the process in drug discovery an preclinical development, a pharmaceutical company can make the most significant difference. But this has proven to be a dauting challenge up to now, as attrition rates in clinical trials remain high. A reduction of more than 60 % in time and about 50 % of the costs could be achieved by implementing a well-designed e-Clinical process (people, process, technology and proper change management), buth this does not yet deal with the fact that about 9 out of 10 INDs (USA) or IMPs (EU) do not belong in clinical development at all.

A lot of money is being lost in drug development and clinical trials because there are too many drugs in clinical trials which should have never reached this stage. Every approved NDA carries the burden of all the other INDs which failed and with 9 out of 10 INDs faling, this burden is very high. This shows that the gatekeepers of (pre-)clinical drug development are failing, which should not happen (in such high numbers) in a well-established stage-gate process. The stages provide the information for the gatekeepers to decide, but when the predictive power of stage-based data is too low, the decisions at the gates are of limited power. The results in drug discovery and preclinical development are biased towards overestimating the chances of success in clinical development. Efficacy is overestimated and adverse effects are underestimated. There is a need for a broader strategy to support go-no go decisions at each stage-gate. The failure to stop 90% of candidate drugs before IND filing, only becomes visible years later in drug development. Late stage attrition in drug development is due to early stage failure of disease models in drug discovery and preclinical development.

Improving drug development

Evolution of the overall process

Discovery and Development Process
Figure 30. Evolution of discovery and development process
A. 1950s and 1960s, B. 1980s, C. and D. 199Os and present.
Modified from Ratti E., 2001.

The drug discovery and development process has changed considerably over the past 50 years (Figure 30). The discovery process had several steps added in-front which were meant to reduce uncertainty and make the overall process more predictable. Clinical development was divided in multiple stages, but the true proof of therapeutic improvement for a given therapy compared to either placebo or competing therapies is still at the end of the pipeline, now in Phase III. The discovery and preclinical development stages cannot answer the questions of clinical development. What happened up-front is that we moved further away from man and moved down to the single molecular level. We still cannot model the complexity of man, but we can model a molecule. We reduced complexity and increasingly introduced false positives and poor data quality. The latest developments are to bring man, the ultimat model organism, back into the process in an earlier stage (e.g. Phase 0, microdosing). Much work remains to be done to improve the predictive power of those early stages. The early stage predictions of success and failure in relation to late stage development should capture more of the complexity of pathological processes in man into the models employed. What happens can be compared to what is going on in the poem "The Blind Men and the Elephant" by John Godfrey Saxe. A lot of detail, but no understanding of the complexity of the overall behavior of the drug in relation to its place in the "ecological" system of the "biotope" man. The focus on molecular targets in recent years, now resembles the situation in the poem "Der Zauberlehrling" from Johann Wolfgang Goethe. The molecular "Sorcerer's Apprentice" can no longer control the spirits that he called and now needs help to master the deluge of new and unvalidated targets.
We added inner resolution (molecular instead of system level), but at the same time we reduced the outer resolution (molecular resolution instead of system-wide overview). Man is not a pile of molecules, but a complex ecosystem.

Process performance

The process does not perform at the same historical success rates anymore as attrition has now reached 92% Preclinical and clinical development is a process driven endeavour where the improvements can be made by improving the process management, both in management approach as well as with better project management tools. Model improvement in preclinical development is a crucial issue. The main reason for failure in clinical development is due to the failure of preclinical models.
The current bottlenecks in drug development are:

  • Predictive pharmacology (PK/PD).
  • Predictive toxicology (Tox).
  • Lack of validated biomarkers.
  • New clinical trial designs.

Pharmacokinetics (PK) describes the kinetics of a drug, or how the body handles a specific compound. Generally, it involves the absorption of the compound, where the compound goes in the body, how the compound is changed, and how it is eliminated: absorption, distribution, metabolism, excretion (ADME) (Bohets H, 2001; Caldwell G.W., 2004; Parrott N, 2005)
Pharmacodynamics (PD) or drug metabolism (DM) describes the impact that the drug has on the body, i.e. what are the drugs effects on the body? Pharmacodynamics (PD) studies the relationship of the time course of a drug (and metabolites) in the body and its effects, it describes the action of a specific compound with regard to its uptake, movement, binding and interactions at its site of activity. A general way to consider these is pharmacokinetics (PK) is what the body does to the drug, and pharmacodynamics (PD) is what the drug does to the body.

Reactions involved in drug metabolism (DM) are often classified as Phase I (activation) and Phase II (detoxification) reactions. Enzymes catalyzing Phase I reactions include cytochrome P450 enzymes. Enzymes catalyzing Phase II reactions include the conjugation enzymes UDP-glucuronosyltransferases (UGT), glutathione S-transferases (GST) as well as other enzymes that protect the cell from toxic damage due to oxidative stress. Phase I and Phase II enzymes acting in concert, convert hydrophobic compounds to more hydrophilic compounds that can be readily eliminated in bile or urine.

Preclinical development

Once a chemical lead is discovered, it is subjected to preclinical testing to assess biological activity. Preclinical studies are conducted both in vitro- in cell cultures and tissues- and in vivo- on live animals such as dogs, monkeys, and pigs. In addition to establishing the drug's pharmacological effects, these studies also identify acute and subchronic toxicology, teratogenicity, and carcinogenicity risks. How to find out if a discovery lead has the physical and chemical, as well as the biological, properties to be a valid drug development candidate? Many disciplines are involved in hit-to-lead transition and lead development. From determining Quantitative Structure Activity Relations (QSAR) to in-vivo assays in model organisms. The process of lead optimization is an iterative process where many scientific disciplines are involved of which I only mention a few. The problems with late stage attrition in clinical development has its cause in the decisions made at the transition from "model to man". We are unable to predict clinical success from preclinical disease models in 90% of all drugs in clinical development.
Six scientific disciplines are involved in preclinical compound characterization:

  1. Analytical and bioanalytical methods
  2. Pharmacology (e.g. therapeutic ratio, Mode Of Action)
  3. Nonclinical formulation
  4. Pharmacokinetics (PK, ADME)
  5. Pharmacodynamics (PD, DM)
  6. Pathology and toxicology (Path/Tox)

This is the traditional matrix of techniques involved in preclinical assessment of a drug candidate (pharmacokinetics (PK) and pharmacodynamics (PD) are of course also being studied in patients during clinical development). The final decisions concerning the usefulness of a drug are the domain of experimental and clinical pharmacology (Burger A., 1987). Bioavaiability of a drug is an important issue, as so elegantly captured in Lipinski's rule of five and can be used as a rule of thumb to indicate whether a molecule is likely to be orally bioavailable (bioactive) (Lipinsky CA, 1997). However this has not been able to reduce the late stage attrition rate in clinical development. Most of the tools used for toxicology and human safety testing are decades old and may fail to predict the specific safety problem that ultimately halts development or that requires post authorization withdrawal. Each aspect of preclinical safety studies (pharmacological screening for unintended effects; pharmacokinetic investigations in species used for toxicology testing; single- and repeat-dose toxicity testing; and special toxicology testing (such as mutagenicity) has not been rigorously tested by a robust analysis of its predictive power. Preclinical development was never designed to make up for the pathophysiological deficit of target-based drug discovery. Physiologically unvalidated development candidates were mainly screened for pharmacokinetic (PK) properties and pharmacodynamic (PD) properties, but this does not validate their clinical therapeutic efficacy in a clinical patho-physiological environment.

One reason for this stagnation of inovation in preclinical development is the fact that many of these experiments are required and highly regulated by regulatory authorities on IND and NDA filing (Hayashi M, 1994; Legler UF., 1993; Spielmann H, 2001). The Organisation for Economic Co-operation and Development (OECD) has provided many guidelines, such as the Reproduction/Developmental Toxicity Screening Test (OECD Guideline 421). With limited resources, a company must focus on those tests which it has to perform to get the clinical development candidate accepted in the first place. There is a trend to improve the preclinical evaluation of drugs, such as performing Phase 0 tests. Regulatory authorities are also aware of the fact that something has to happen to reduce attrition rates in clinical development. The focus is on optimising the interface between late preclinical development and early clinical drug development by utilising modern in vitro - in vivo extrapolation techniques. The industry has to improve the current wasteful and uninformative system for testing drug candidates, and shift to research methods that use biomarkers to predict drug side effects and benefits (derived from a speech given by FDA acting deputy director Janet Woodcock). Biomarkers could help us to to improve the predictive power of drug discovery and early drug development (Fowler BA., 2005; Kola I, 2005). Verification that a biomarker assay is specific for its intended purpose poses a formidable challenge.
We need (validated) biomarkers for preclinical and clinical development in order to:

  1. Treat diseases more effectively:
    We currently lack predictive biomarkers to stratify patients with similar diseases as well as accurately measure disease susceptibility, presence and progression.
  2. Verify the impact of novel drugs on targets/pathways:
    We currently lack the ability to determine the ability of a novel drug to bind the desired target and whether this binding actually leads to changes in the desired pathway.
  3. Avoid Serious Adverse Events (SAE):
    About 1.5 million people are hospitalized each year due to adverse effects of prescription drugs.
  4. Increase drug development predictability:
    Only 1 in 10 drugs entering Phase I ever reach the market with the great majority of compounds failing in Phase II.

We really need drug candidates (NCEs, NBEs) which make it to the market at much higher rates than with the current 10% overall success rate from IND to NDA. This will require a significant paradigm change in the assessment of potential investigational drug candidates (Apic G., 2005; Caldwell GW, 2001; Schadt EE, 2005; Shaffer C., 2005).

Highly sensitive techniques, such as Accelerator Mass Spectrometry (AMS) and PET allow for the detection of biomarkers. Accelerator mass spectrometry (AMS) is a mass spectrometric method for quantifying rare isotopes, which is being applied to biomedical and toxicological research (Barker J, 1999; Brown K, 2006; MacGregor JT, 1995; Turteltaub KW, 1990). AMS can be used to study long-term pharmacokinetics and to identify biomolecular interactions in neurotoxicology and neuroscience (Palmblad M, 2005). AMS enables compounds and metabolites to be measured in human urine and plasma after administration of low pharmacologically or toxicologically relevant doses of labelled chemicals and drugs (White IN, 2004).

Clinical development

For many scientists working in drug discovery and preclinical development, the environment in which clinical trials happen is still something mysterious. I want to clarify some of these issues, mainly the regulatory (and scientific) framework in which these complex and expensive trials have to be performed. Drug discovery and preclinical development should be done while keeping an eye on clinical development. Maybe this helps to understand the fears and nightmares of those scientists and their collaborators, every time they start a development track and face the fact that up to 90% of their efforts are in vain.
A clinical trial (also clinical research) is a research study in human volunteers to answer specific health questions. Carefully conducted clinical trials are the fastest and safest way to find treatments that work in people and ways to improve health. Interventional trials determine whether experimental treatments or new ways of using known therapies are safe and effective under controlled environments. Observational trials address health issues in large groups of people or populations in natural settings. Each clinical trial starts with the definition of a primary question (study hypothesis). From that hypothesis the best way to confirm or reject it, can be designed. There are several designs possible, which can be found in the literature on clinical trials. A randomised, double-blind, crossover or factorial, multi-site Phase III clinical trial is a complex endeavour and a drug which fails at this stage has cost an enormous amount of money (Figure 31). Not to mention the shattered hopes of the patients it was intended to provide with new hope. A Phase III clinical trial may involve up to 5,000 patients distributed over numerous clinical trial sites, which gives you an idea of the logistic complexity to manage such a trial. It is important to keep in mind that a Phase I trial does not deal with the outcome of a Phase II trial and so on. Each trial type is designed to answer a different type of question and it only provides an answer to this question(s), not to the ones being dealt with in another type of clinical trial. Clinical trials are becoming more expensive and even more regulated. A Phase I trial costs about US$ 8,000-15,000/subject, a Phase II costs about US$8,000-15,000/patient and finally a Phase III trial costs about US$4,000-7,500/patient. Improving a proces is a matter of methodology and technology.

The overall profile of therapeutic indications and adverse events, leads to the labeling of the proposed drug. The "sponsor" of a new drug must obtain approval from the FDA by specifying both the medical conditions the drug is effective against and the patients groups for whom the drug has been shown to be effective. This information is contained in the proposed "label" submitted by the developer or sponsor. It is the sponsor's responsibility to assemble all the evidence that would support the uses proposed in the label (preclinical and clinical development). With the wide gap between molecular targets and clinical diseases, this has become increasingly complex and risky. This also explains the problems with "first-in-class" drugs and the reduced risk once knowledge and understanding build after years of widespread use. There is also the potential safety problem of "off-label" use of a drug, besides the problems with reimbursement.

Clinical trials
Figure 31. Clinical trials from Phase I to Phase IV. Phase II can consist of a Phase IIa and IIb.
Phase III can also consist of a Phase IIIa and IIIb.

A basic clinical trial process consists of several stages:

  1. Hypothesis formulation, primary and secondary endpoints
  2. Protocol development
  3. Investigator/site selection and trial preparation
  4. Subject identification and enrollment (causes most of the delays)
  5. Collection, monitoring and processing of data (CRF, PRO, Lab)
  6. Clinical trial management
  7. Data analysis and reporting of results (SAS)
  8. Submission for review by a regulatory agency (FDA, EMEA, ...)

A "ram-it-through paradigm" in clinical trials readily produced beta-blockers, H2 blockers, nonsedating antihistamines, and other big classes of drugs. The development of predictive models to assist the decision process to enter the next Phase of clinical development is an interesting path taken to reduce late stage attrition (Albert JM, 1994; De Ridder F., 2005; Hale M, 1996; Holford NH., 2000). There is a transition going on from empirical to causal models for deriving evidence of effectiveness. There is also a transition from empirical to causal models for deriving evidence of safety. Clinical development is changing from a reactive empirical model to a proactive Model Based Drug Development (MBDD) process (pharmacometrics). Instead of rushing through clinical development, a learn-and-confirm approach allows for a more dynamic and adaptive process (Sheiner LB., 1997). Phase I and IIa are the learning phases and Phase IIb and III are the confirming phases. There is a clear trend to learn more earlier in the drug development process. Phase I is no longer just for establishing safety and dosing levels, Phase I research is playing an increasingly important role obtaining more data about the potential success of a drug. The emphasis is increasingly on mechanistic early phase clinical trials to maximise the chances of obtaining clinical data to make sound go/nogo decisions. Model-based drug development includes exposure/response assessments in the form of pharmacokinetic/pharmacodynamic (PK/PD) modeling. Pharmacokinetic (PK) and pharmacodynamic (PD) modeling and simulation (M&S) are powerful tools that enable effective implementation of the learn-and-confirm paradigm in drug development (Chien JY, 2005; Gobburu JV, 2001; Grasela TH, 2005). One of the major prerequisites for the successful application of PK/PD-modeling, however, is the availability of response measures such as biomarkers that provide an immediately accessible link between pharmacotherapeutic intervention and clinical outcome and allow to easily assess variations in desired and/or undesired drug effects in response to changes in dose, dosage regimen, dosage formulation, administration pathway, or external factors affecting drug response. Biomarker-based PK/PD modeling can become the basis for a scientifically driven, evidence-based, streamlined drug development process.

The problems with drug discovery and development are not limited to scientific and technical issues alone. Discovery and preclinical research may be a scientific and technical minefield, but with clinical development we in addition enter a moral and ethical minefield. Life in the lab may not be easy, but life at the bedside isn't either. Improving the clinical development process is not easy, as we have to operate in a highly regulated environment, which limits the freedom to change the process. In a highly regulated environment you cannot do different things, but you have to do (regulated) things differently, e.g. a new statistical approach (e.g. Bayesian methods), modeling and simulation, learning and confirming (Maurer W., 2005). In the early stages of drug development we should be able to extract more predictive information from our research, to reduce the late stage attrition in Phase II and III. More knowledge and understanding of complex processes earlier on, would allow for better predictions and less failure later on in the process. Applied science within regulatory constraints is the only way to bring basic science into clinical reality. Just to give you an idea about all the regulatory issues involved, I will give an overview of some guidelines (Grunfeld GB., 1992). As clinical trials involve human experimentation, they are to be conducted according to high ethical standards. Historical events lead to the adoption of ethical guidelines for the conduct of research on human subjects. The Nuremberg Doctors' trial, officially United States of America versus Karl Brandt, et al., lead to the Nuremberg Code. The Nuremberg Code (1947) laid down 10 standards to which physicians must conform when carrying out experiments on human subjects.
In summary, the Nuremberg Code includes the following guidlines for researchers:

  • Informed consent is essential.
  • Research should be based on prior animal work.
  • The risks should be justified by the anticipated benefits.
  • Research must be conducted by qualified scientists.
  • Physical and mental suffering must be avoided.
  • Research in which death or disabling injury is expected should not be conducted.
A physician participating in a clinical trial is bound by the Physician's Oath put forward in the Declaration of Geneva (Adopted by the 2nd General Assembly of the World Medical Association, Geneva, Switzerland, September 1948). The ethical dimension of human trials is guided by the Declaration of Helsinki (first version adopted by the 18th WMA General Assembly, Helsinki, Finland, June 1964). The World Medical Association (WMA) has developed the Declaration of Helsinki as a statement of ethical principles to provide guidance to physicians and other participants in medical research involving human subjects. Some examples of extreme abuse lead to new legislation in the US: the Tuskegee Syphilis Study (1932-1972, Public Health Service Syphilis Study), Jewish Chronic Disease Hospital Study (1963) and trials performed at the Holmesburg Prison in Philadelphia (from mid-1950s to mid-1970s). The US National Commission for the Protection of Human Subjects of Biomedical and Behavioral Research, published the Belmont report on 18 April 1979. The Belmont report deals with Ethical Principles and Guidelines for the Protection of Human Subjects of Research. In 1987 the IND Rewrite Regulations (FDA Title 21 CFR Parts 312, 314, 511, and 514) were created, to ensure FDA's ability to monitor carefully the safety of patients participating in clinical investigations, while also facilitating the development of new beneficial drug therapies.

Government agencies take care of enacting the laws and regulations, such as the US Food and Drug Administration (FDA) (e.g. CBER Guidances), The European European Medicines Agency (EMEA) and the Japanese Ministry of Health, Labor and Welfare. On 1 April, 2004, Japan's Pharmaceuticals and Medical Devices Agency (PMDA) began conducting NDA reviews for the Ministry of Health, Labor and Welfare. The Council for International Organizations Of Medical Sciences (CIOMS) develops principles and proposals for the collection, evaluation, and reporting of safety information obtained during clinical trials to all appropriate stakeholders. The CIOMS I workgroup (1990) dealt with international reporting of Adverse Drug Reactions (ADR) and created the "CIOMS I reporting form" for standardized international reporting of individual cases of serious, unexpected adverse drug reactions.
As regulatory authorities increase their standards on drug development, they put the hurdle higher for all the players in the field. So, all companies face the same challenge to improve their process. One of the positive consequences has been that the US pharmaceutical industry is now the most competitive in the world, as they had to adher to increasing quality standards. The US pharmaceutical market is still the most profitable of all, but also one the most regulated.

As clinical trials in Phase III are large-scale projects, streamlining data exchange procedures can make a significant contribution to cost reduction and shortening the duration of a trial, e.g. for a Computer Assisted New Drug Application (CANDA) of the FDA. Standardization of data formats is one aspect of process improvement for which many attempts have been made over the years, e.g. Drug Application Methodology with Optical Storage (DAMOS), Multi Agency Electronic Regulatory Submission (MERS), Market Authorization by Network Submission and Evaluation (MANSEV) and Soumission Electronique des Dossiers d'Autorisation de Mise sur le Marché (SEDAMM). The need for international standardization in clinical trials is being dealth with by organizations such as the International Conference on Harmonisation of Technical Requirements for Registration of Pharmaceuticals for Human Use (ICH), e.g. Common Technical Document (e-CTD for e-submission). There is the ICH E3 guideline on the Structure and Content of Clinical Study Reports. The Clinical Data Interchange Standards Consortium (CDISC) develops XML-based standards for data exchange in clinical trials (e.g. ODM, SDTM,...). eSignatures should allow for exchanging digital data in a secure way within the pharmaceutical industry. The Secure Access For Everyone (SAFE) initiative, is a set of standards for digitally signed transactions that create a trusted community for legally enforceable data exchange.

Clinical trials are process driven and as such amenable to process management improvement. Improving clinical trials is an ongoing effort, driven by the tendency to new clinical trial designs and e-Clinical trials. The traditional Randomized Controlled Trial (RCT) is still the standard, but new designs are being applied. Besides a new design, new technology can help to improve the clinical trial process. The Internet provides a means to improve the process for large multicenter clinical trials (Paul J, 2005). However clinical trials from Phase I to III (and IV) are a highly regulated process, (FDA Title 21 CFR part 11, HIPAA, GCP, GMP,...) which limits the freedom for improvement (e.g. Computer System Validation). Clinical trials gain speed by using Cinical Trials Management Software (CTMS), Clinical Data Management Software (CDMS) and automating the process of data capturing (RDC or Remote Data Capture and EDC or Electronic Data Capturing), such as for the Patient-Reported Outcome (PRO) and Case Report Forms (CRF). The usage of computer systems in clincal trials is guided by the FDA Guidance for Industry on Computerized Systems Used in Clinical Trials.
The Biomedical Research Integrated Domain Group (BRIDG) Model is a comprehensive domain analysis model representing biomedical/clinical research. It was developed to provide an overarching model that could readily be comprehended by domain experts and would provide the basis for harmonization among standards within the clinical research domain and between biomedical/clinical research and healthcare. Work is going on to develop a Clinical Trials Object Model (CTOM) and a Structured Protocol Representation, from which e-Clinical trials could be designed, developed and managed.
The procedures for data submisson are being improved by Electronic Regulatory Submissions and Review. The review process is being regulated by Good Review Practices (GRPs). The FDA JANUS project will combine information from numerous clinical trials into a single data warehouse. JANUS will give the FDA the ability to cross-analyze data from multiple trials, identify systemic deficiencies across similar products, and detect potentially dangerous drug interactions. So, a lot of improvements are going on in the drug review process, but this will be of limited success if the drugs (NCE, NBE) themselves do not show better quality profiles.

Process improvement is not only a matter of using ICT, but also requires a change in project management principles in order to succeed. Clinical trial process improvement is for 80% a matter of people and process and for 20% a matter of technology (ICT, biomarkers,...). An overall improvement from protocol (e.g. eProtocol) to NDA submission is required, for instance by shortening the time from protocol completion to First Patient First Visit (FPFV), First Patient In (FPI) to Last Patient Out (LPO), Last Patient Last Visit (LPLV) to DB lock and finally from DB lock to NDA. A reduction of more than 60 % in time and about 50 % of the costs could be achieved by implementing a well-designed e-Clinical process (people, process, technology and proper change management). An e-Clinical trail also allows for better process monitoring (performance metrics system).

Improving drug discovery and preclinical development

The modern pharmaceutical industry faces an increasingly widening gap between, on the one hand, growing numbers of potential drug targets and lead compounds and, on the other hand, a lack of reliable methods to identify those molecular targets unequivocally linked to disease pathophysiology and lead compounds with the best chance of success in (pre)clinical development. How should we proceed to improve drug discovery and preclinical development? We have seen an enormous investment in research at the infra-cellular level, such as High Throughput Screening (HTS), genome based and proteome based disease models in the past ten years and at the same moment have witnessed a disproportional decline in the productivity of research and development in drug discovery (Horrobin DF, 2000; Horrobin DF, 2001; Noble D., 2003; Bleicher KH, 2003, Betz UA, 2005; Bilello JA., 2005). The pharmaceutical industry has yet to find a way to reduce its high attrition rates (Kola I., 2004). Late stage attrition rates in oncology are among the highest in the industry, alhough the need for better cancer treatment is high (Kola I., 2004; Kamb A., 2005; Saijo N., 2004; Suggitt M, 2005).
The consolidation in the pharmaceutical industry will not solve this problem in the long run, as it only reduces the costs (mainly of sales and marketing and manufacturing) but does not improve scientific productivity; it only postpones the moment of truth. The scientists themselves will have to find new ways to improve their productivity; management cannot do this in their place. Society tries to protect itself against the adverse effects of new drugs, such as with Thalidomide in the sixties (McBride WG, 1961). This is done by increasingly stringent regulations but the currently used methods in the discovery process for new drugs cannot keep pace with these new requirements. However, as we can see, increasingly strict regulations do not explain all the problems pharmaceutical research is facing today.

We can summarize the (r)evolution in the overall drug discovery and development process as such:
Traditionally we have a "trial and error" approach:

  • Blind trust on high-throughput technologies.
  • Limited success rates of new biological targets.
  • Separated scientific disciplines, functional orientation ("silos").
  • Sequential approaches in "new" biology and chemistry.
  • Low degree of specialization in chemistry and biology ("generalists").
This will have to evolve into a "cognitive" chemical biology approach:
  • Focus on selected families and a systems biology or biology of systems approach.
  • Accumulation of knowledge on chemical and biological structure spaces, learning curves.
  • Interdisciplinary problem solving.
  • Parallel processes, information driven and integrated technology, bioinformatics.
  • Teams across scientific disciplines.
  • Technology platforms and demand for more specialized skill sets.
  • Networks of knowledge & partnering.

Disease models in drug discovery and preclinical development

Scientific progress from basic science to its applications

Basic and applied research achieve results through brilliant ideas and hard work. "Adde parvum parvo magnus acervus erit" (Ovidius, By adding little to little there will be a great heap). You cannot get much done in a short time, but once you start to string together work for several years or decades, you have the start of a body of work that puts a mark on the world.
The concern and goal of these articles is to clarify the problems with bringing basic research to clinical applications, not to criticise basic research as such. When we look at scientific methodology not from the perspective of understanding fundamental biological processes, but for their predictive power to generate results which facilitate the application of basic science to clinical reality, a different picture emerges. I want to look at scientific methodology with its impact on treating the pathological process in man as a reference.
Keep in mind
Occam's Razor: "Numquam ponendo est pluritas sine necessitate", which we should translate as "One should use the explanation that is simple enough to explain all there is to explain, but nothing simpler". Quite often our explanations fail to capture the entire biological phenomenon, which is why we fail in almost 90% of all drugs in clinical development. 75% Of all failures in clinical development are caused by lack of efficacy (30%), adverse efects in man (13%), animal toxicity (20%) and pharmacokinetics (12%).
I will take a look at several aspects of the discovery and development process to provide wide ranging information to those who are interested in improving the pharmaceutical R&D process. I want to avoid "stovepiping" by skipping information levels. In 2003 Seymour Hersh wrote an article in the New Yorker titled, "The Stovepipe". Hersh defines "stovepiping" as taking a request for action arising from intelligence "directly to higher authorities without the information on which it is based having been subjected to rigorous scrutiny." Stovepiping results in intelligence failures when conclusions are allowed to pass rapidly from the lowest levels of the intelligence gathering apparatus, the ones with their hands directly on new information as it comes in at ground level, up to the decision making authorities many levels above without passing through the normal many-layered time-intensive vetting and checking processes in-between. This can also be applied to a request for improvement of the "war against disease".

Failing disease models in drug discovery and preclinical development

Anyone working in science must realize that theories, models and approximations are powerful tools for understanding and achieving research and development goals. The price of having such powerful tools is that not all of them are perfect. This may not be an ideal situation, but it is the best that the scientific community has to offer. Each individual must try to attain an understanding of the nature of our descriptions of the physical world and what results can be trusted to any given degree of accuracy (see also Models, Approximations and Reality).
Currently used disease models fail to predict the outcome of clinical development and are incapable to reduce attrition rates in drug development. The reasons for failure can be summarized as follows. Drugs fail in development and beyond due to:

  • Man - understanding of pathophysiology is faulty.
  • Efficacy - no significant effect on a clinical disease process.
  • Toxicity long term safety is still totally unpredictable.
  • Bioavailability and half life half life cannot be predicted, only guessed.
  • Metabolism drug/drug interactions; parent or metabolite.
What should we do to improve this?
  • Improve our ability to explore and understand human disease processes.
  • Better target identification and validation.
  • Improve the predictive power of toxicology.
  • Achieve a more precise drug metabolism and pharmacokinetics (DMPK) assessment.

There is no single disease model which allows us to predict clinical success in all cases. Bridging the gap between molecule an man is a delicate process, which requires careful consideration and design. Quite often there is not enough consideration of building and validating the path from clinic to model and back. In the end it is clinical reality which decides on the fate of new drugs and not the technology or disease models used to create them. The successful identification of drug targets requires an understanding of the high-level functional interactions between the key components of cells, organs and systems, and how these interactions change in disease states (Butcher EC, 2004; Noble D, 2003; Stumm G, 2002). Pre-clinical studies and early clinical trials should pay more attention to both the pharmacology of the drug as well as the (in-vivo) biology of the target (Newell DR., 2005). Both basic and applied research have have made truly enormous contributions to the health of mankind, but the challenges ahead are no less than they were in the past. Only with an open mind and a thorough analysis of the present situation we will be able to analyze the problems of present day research and development.

Validating the predictive power of our disease models is not an easy task as we operate in a complex pathophysiological environment.
"...One must rely heavily on statistics in formulating a quantitative model but, at each critical step in constructing the model, one must set aside statistics and ask questions. ... without a qualitative perspective one is apt to generate statistical unicorns, beasts that exist on paper but not in reality. ... it has recently become all too clear that one can correlate a set of dependent variables using random numbers as dependent variables. Such correlations meet the usual criteria of high significance. ..." (Hansch C, 1973). We fail to predict clinical success with our current disease models, which translates itself in a high (up to 90%) attrition rate in clinical drug development.
The problem of prediction in relation to a model system (van Drie JH., 2003) (free from the "Kubinyi Paradox"):

  • Inside the model: trivial
  • Outside the model: wrong
  • At the edge: 50/50

There are two approaches to drug discovery, historically a physiology based approach was used, while nowadays a target based approach has become more popular. Most of the problems of the drug discovery and development process can be traced back to its early stages, namely target validation and lead selection. With respect to the targets this is, among other factors, due to a lack of definite clues with regard to the (dis)regulation of cellular pathways that underlie diseases. This gap in basic knowledge is even further widened by the lack of adequate animal test models and disease markers to monitor the particular disease process as well as the outcome of therapeutic interventions. With respect to lead selection there is a poor insight in the fundamental relations between the physico-chemical characteristics and the biological properties of the drug candidates, including mechanisms of cellular action and toxicity, processes of drug disposition (ADME) and the aspect of drug safety on a longterm basis. In modern drug discovery the early stages of drug discovery involve the identification and early validation of a disease-modifying target (Lindsay MA., 2003; Schneider M, 2004). Failing to make the right decision at the important step of hit to lead transition has costly time and resource implications in downstream drug development (Alanine A., 2003; Kuhlmann J, 1999). Why do the early stages of drug discovery fail so often and why are they the cause of a huge efficiency deficit later on in the drug discovery process?
Assessing the drugability of a target is only one of the important criteria to consider. Drugability of a target is the feasibility of a target to be effectively modulated by a small molecule ligand that has appropriate bio-physicochemical and absorption, distribution, metabolism and excretion properties (ADME) to be developed into a drug candidate with appropriate properties for the desired therapeutic use. The weak spot is the definition of therapeutic use, which quite often does not mean we are working with a clinicaly validated target.

Since the early successes of compound screening against isolated molecular targets in the 1970s, the industry moved away from physiological based screening to a target-based screening (Luyten W, 1993; Herz JM, 1997). In the beginning Target-based screening was initially used to improve the drug-like properties and selectivity of pharmacologically active products. Target-based drug discovery has been very successful when applied to already physiologicaly validated targets of existing drug. Later on the hope was that sequencing the human genome would generate a wealth of new targets, and the hope of almost directly linking 'genes-to-drugs' was embraced by the industry in the 1990s. When the drug discovery process moved beyond historical targets, however, it became apparent that the target-directed approach was flawed: without solid biological validation, target-based drug discovery has proven very disappointing (Williams M., 2003; Butcher E., 2005). Improving target identification requires a dramatic improvement of our understanding of cellular pathways underlying pathogenesis and/or pathophysiology. Improving lead selection requires an hypothesis-driven approach in which, out of the multitude of potential targets, a rigid selection of druggable targets is made.

There is a fundamental problem with studying disease-relevant mechanisms in the current disease models as the pharmaceutical industry has been investing heavily in studying the 'bricks', instead of looking at the 'building' (cytome, organism) as a dynamic unified pathophysiological system. Finding a gene or a target does not equal understanding a clinical disease process in man. The genome has yielded a series of novel molecules that do not have 20 or 40 years of biology behind them for us to understand exactly what they do and where to apply them. The emphasis in recent years has been on increasing quantity while at the same moment sacrificing the quality of correlation with clinical reality. Accepting the perceived truth that ubiquity equals utility, organisations often initiate efforts to automate processes, assuming that improvement will be a natural consequence of automation. This avalanche of data does not lead to an improvement of understanding. "Le savant doit ordonner; on fait la Science avec des faits comme une maison avec des pierres; mais une accumulation de faits n'est pas plus une science qu'un tas de pierres n'est une maison." Henri Poincaré (1854-1912). We increased our capacity in genomics and proteomics, but we did not improve the quality of the preclinical physiological disease process evaluation in the same way. Technologies such as genomics and proteomics have produced an explosion of new poorly validated targets that actually increase the rate of compound attrition and the costs of R&D.
You could also think of it as a pointillist painting, of which we have been looking at the individual dots, instead of looking at the entire painting. Another analogy is that we are trying to explain the tidal patterns of the oceans, by studying a water molecule and ignoring the moon. We have to look for the picture that is in the puzzle early on, not just giving the 'pieces' to the next process down the line. We have to look at biological phenomena at several scales of integration and from a functional point of view in order to get a grip on the development of pathological processes. Crosslinking and crossreferencing biological organisational levels in order to understand the 'web' of biological interactions in a pathophysiological process. We should try to understand the dynamics of disease processes also at a higher level of biological integration, closer to the clinical reality, than only the genome or proteome. An integrated cellular and organism-level approach is needed to study disease processes (Lewis W., 2003).

When we modify a gene, e.g. by creating transgenic animals, we must try to understand the dynamics of the pathways we are modifying. The gene products are part of a delicate web of intertwined pathways where subtle changes can have unpredictable effects. Quite often transgenic animals or animals with gene knock-outs do not show the expected phenotype, because of a different genetic background and the highly dynamic interplay of metabolic pathways and environmental influences on the final phenotype (Sanford LP, 2001; Pearson H. 2002).

The (early stage) disease models we use don't work as they should do and do not provide enough overall predictive power in relation to clinical reality. One can study cellular components, like DNA and protein as such, but this will not reveal the complex interactions going on at the cellular level of biological integration or in other words, the cytome. Both medicine and pharmaceutical research would benefit from using more (primary) cell oriented disease models (in-vivo and in-vitro) and even higher-order models, instead of using infra-cellular models to try to describe complex pathological processes at a molecular level and getting lost in the maze of molecules which are the building blocks of cells. Keep in mind that cytome-oriented research is not the same as cell-based research.

An important moment in the drug discovery and development pipeline is the transition from discovery research to clinical development. Different approaches to develop gatekeepers have been proposed to reduce the failure rate in drug development on both sides of the transition (Lappin G., 2003; Nicholson J.K., 2002; Pritchard J.F., 2003). Translational medicine is emerging as a gatekeeper for evaluating drugs when they traverse the "great divide" between "bench and bedside" (Fitzgerald GA., 2005). Translational Medicine can be defined as an interactive process between preclinical and clinical investigation. Translational biomarkers and molecular profiling should assist in increasing success in clinical development. However, translational medicine alone will only bring bad news earlier in the process, so it should be combined with a concomitant improvement of disease models in drug discovery and preclinical development.

Drug discovery and preclinical development should improve the quality of drugs they allow to enter clinical development and clinical development should be able to protect itself from drugs likely to fail in phases I to III. A better quality of drugs entering drug development is needed, not just more quantity (drastic reduction of false positives). Failing in larger numbers will not bring the solution to create a better process from discovery to phase III an IV. If we just drive more drugs into clinical development, but keep failing at a rate of up to 90%, we are not helping ourselves.

A highly defined oligo-parametric infra-cellular disease model used in High Throughput Screening (HTS) which in its setup ignores the complexity of higher order cellular phenomena, may produce beautiful results in the laboratory, but fails to generate results of sufficient predictive power to avoid considerable financial losses later on in the drug discovery pipeline (Bleicher KH, 2003). A living (primary) cell may be a less well defined experimental environment for the biochemist, but it will provide us with the additional modulating influences on our disease models which are lost in lower-order disease models. The cytome can be analyzed either in-vitro in well designed cell-based assays or in-vivo by using for instance molecular imaging.

Metabolic variation in disease models

Preclinical models should help to identify factors that are important determinants of intersubject variability in man. We need to make a clear distinction between our inability to elucidate the overall molecular mechanism of a disease process and population variability in metabolic activity. Genetic variability is only one part of the picture, as there are multiple determinants of intra- and interindividual variation:

  • Demographic: age, body weight or surface area, gender, race
  • Genetic: target variability and metabolism, e.g. CYP2D6, CYP2C19
  • Environmental: smoking, diet
  • Physiological/Pathophysiological: renal (Creatinine Clearance) or hepatic impairment, disease state
  • Concomitant Drugs
  • Other Factors: meals, circadian variation, formulations

Real life variability should be incorporated in preclinical disease models and not excluded. Nowadays the first stages of drug discovery and development use genetically homogeneous disease models, which as a result do not show the same metabolic heterogeneity of patient populations. In-vivo variation is not an artefact of life, but a fact of life. Variability in drug response is a complex and multifactorial phenomenon, of which genetics is only a part. Genetic and metabolic heterogeneity is now seen as reason to exclude potential patients from treatment, not as a consequence of the failure of drug development. We need to make a clear distinction between our inability to elucidate the true molecular mechanism of a disease process and population variability. Molecular diversity in a clinical disease process leads to treatment failure because of the wrong action is taken to modify a disease process which we do not understand. Metabolic variation in drug metabolism is another reason for treatment failure, but is more easy to deal with as these problems involve shared properties of drugs.

If we cannot develop drugs which will work in a genetically and metabolically heterogeneous environment, we try to reduce the patient population until it fits our abilities. However this micro-management of patient populations leads to a level of complexity in disease treatments the pharmaceutical industry, physicians and society cannot deal with in the end. Also, excluding people from treatment, because we are unable to develop drugs which benefit a large population of people poses ethical problems. Dose optimisation however is an interesting path to minimize side-effects. At present there are still too many roadblocks to achieve the goal of a better and finely tuned disease treatment. Personalized medicine will remain a distant dream, if we do not succeed in achieving a much better understanding of molecular pathophysiology and at the same time dramaticaly improve the drug discovery and development process. Pharmacogenomics is being used to explain differences in drug metabolism during drug development, such as with Cytochrome P450 (Dracopoli NC., 2003; Halapi E, 2004; Kalow W., 2004). Toxicogenomics and genotyping are used as a tool to identify safer drugs, worthwhile to enter clinical development (Guzey C, 2004; Koch WH., 2004; Yang Y, 2004).

There are many causes for variability in drug metabolism. this variability can lead to an Induced Error Of Metabolism (IEOM), just as an Inborn Error Of Metabolism caused by a genetic defect. Variability of a metabolic change due to a drug can have many causes:

ΔMetabolic change = ΔUptake + ΔDisease state + ΔHost state + ΔElimination

In the case of a pathogen, the variation in virulence is a cause of variation, as well as the host defense system status (e.g. the Varicella Zoster virus which causes Varicella or chicken pox, and herpes zoster or shingles). Metabolic variation due to variability in drug uptake and elimination can be a serious cause of trouble. The Cytochrome P450 enzymes are an important cause of metabolic variation in the metabolism of drugs (Slaughter RL, 1995). Many drug interactions are a result of inhibition or induction of cytochrome P450 enzymes (CYP450). The CYP3A subfamily is involved in many clinically significant drug interactions, including those involving nonsedating antihistamines and cisapride, that may result in cardiac dysrhythmias. CYP3A4 and CYP1A2 enzymes are involved in drug interactions involving theophylline. CYP2D6 is involved in the metabolism of many psychotherapeutic agents.
Variability in drug metabolism can also be caused by food. Grapefruit juice affects the pharmacokinetics of various kinds of drugs, the major mechanism being considered to be inactivation of intestinal cytochrome P450 3A4, a so-called mechanism-based inhibition (Bailey DG, 1991; Fuhr U., 1998; Saito M, 2005).

The EMEA road map and the FDA Critical Path identify pharmacogenomics (PGx) as the emerging technology that will enable efficient and successful drug development. Pharmacogenomics is not yet used to design or use early stage disease models with sufficient genetic heterogeneity to select drug molecules which will hold their activity in a metabolic heterogeneous environment. Genetic heterogeneity, epigenetic modulation and metabolic variation are not taken into account in the first stages of the drug discovery process. Optimizing a drug molecule for binding to one particular genetic variant, imminently leads to failure in a genetically heterogeneous patient population. Randomization in experimental design to counteract a systematic bias in ones results involves more than sample unit randomization patterns.

Biological variation in heterogeneous cell or animal populations may be an unpleasant fact of life, but it correlates better to the real conditions of the genetically and metabolically heterogeneous patient populations. Ignoring biological variation in drug discovery will cause failure in drug development. Using pharmaco-genomics only to exclude slow metabolizers, etc., from clinical trials and thereby homogenizing the trial population can lead to a dramatic reduction in potential patient population and a decline in profit generation potential. Finding sites to manipulate metabolism which are less sensitive to genetic variation would improve our overall success rates. The important phase of a drug life cycle starts when it hits the market and we better take care that it will spend its full life cycle to generate enough revenue to fuel the company.

Hypo- or Subcellular disease models

When molecular biology entered drug discovery in the 1980s and 1990s the dominant view on the relation between genotype and phenotype was derived from the simple dynamics of prokaryote genetics. The preeminent French scientist and 1965 Nobel laureate Jacques Monod, said in 1972 "Tout ce qui est vrai pour le Colibacille est vrai pour l'éléphant" ("What is true for Escherichia coli is also true of the elephant"). However, the outcome of the Human Genome Project has revealed that the processing of our genetic information is much more complex than in Prokaryotes. We have seen an increase in capacity of DNA and RNA expression techniques, but their information still delivers data up to the level of the expressed protein, but not beyond. The quantitative chain of functional causation stops at the protein level. Higher order spatial and temporal dimensions of cellular dynamics are beyond the reach of these techniques. Gene expression studies do not tell you about the functional outcome of protein dynamics and enzymatic activity in the different cellular compartments. Up and down-regulation of gene expression, does not inform you about the functional interrelation of the encoded proteins and their spatial and temporal dynamics in the cell. Molecular pathways do not exist as parallelized unrelated up-and down regulating patterns, but are highly dynamic and intertwined modular networks (Sauer U., 2004).

Gleevec or imatinib or STI571 (signal transduction inhibitor number 571) is a good example of a drug aiming at a molecular target for which the entire chain of models, from molecule to man was thoroughly explored. The disease mechanism was well understood and the molecular biological mechanism was well-embedded in a pathophysiological understanding (Jones RL, 2005; Mauro MJ, 2001). This endeavor required the integration of a number of disciplines, including structural biology, computational chemistry, structurally directed medicinal chemistry, array screening assays, and molecular and cellular biology (Druker BJ, 2000).

Publications on the association of the Philadelphia chromosome and leukemia can be traced back to the early sixities of the twentieth century (Benson ES, 1961; Tjio JH, 1966). In 1982 it was found that in chronic myelogenous leukemia (CML), c-abl sequences are translocated from chromosome 9 to chromosome 22q- (de Klein A, 1982). During the eighties of the twentieth century the molecular mechanism of the gene product, a tyrosine kinase was investigated (Pendergast AM, 1987; Maxwell SA, 1987; Lugo TG, 1990). More than 15 years of work by scientists from all over the world was needed to understand the molecular mechanism of the disease (this costs far more than 800M US $). Once the target was identified and the disease process understood, the selection for drug candidates could start (Druker BJ, 1996). Finding the gene is not enough, the hard work is to find out about the molecular mechanism of the disease and this has not changed even now the Human Genome Project has been completed. The work on the c-abl tyrosine kinase predates the Human Genome Project with more than 20 years (the same goes for Herceptin). With the current state of technology and science we may expect to see the results from the new targets coming out of the Human Genome Project in 10 to 20 years.

Let us now look at some methods used in molecular biology, not for their value in basic research, but for their predictive value for preclinical development (a bit unfair, I know). We must be aware that studying a basic molecular mechanism is still far away from understanding the clinical disease process as such. Southern, Northern and Western blots may show the quantitative sequence of gene expression up to protein concentration (Alwine JC, 1977; Alwine JC, 1979; Howe JG, 1981; Hinshelwood MM, 1993). DNA microarrays give a quantitative indication of gene expression (Barbieri RL, 1994; Schena M, 1995; DeRisi J, 1996; Jeong JG, 2004; Kawasaki ES., 2004). However finding a positive correlation between the pattern of gene expression and a given disease state is not the same as finding a causative relationship between (a) gene(s) and the causation matrix of a disease (Miklos GL, 2004). Moving up to the level of the dynamics of protein expression already demands a higher degree of sophistication in both assay design and data analysis (Kumble KD., 2003). However, without a functional assay on in-vivo dynamics of protein function and studying its spatial and temporal expression patterns (process flux) in the cell (compartments) and tissue, the functional impact on the cell remains unclear (Kriete A, 2003; Young MB, 2003; Egner, A., 2004).

Studying subcomponents of cellular pathways ignores the functional unity of the biological processes in the cell and the functional interactions between pathways. Studying an isolated drug target ignores important off-target interactions which become a cause of failure too late in drug development. Studying proteins in isolation and uncoupled from the intracellular molecular oscillating clocks, ignores the importance of temporal patterns (Okamura H., 2004). Without a better understanding of the phenotypic and functional outcome in the cell, the failure rate of the drug discovery process will remain high and very costly. There is a predictive deficit in the current oligo-parametric disease models used in pharmaceutical research which necessitates complex and expensive studies later on in the drug development pipeline to make up for the predictive deficit.

A simple homogeneous binding assay, will fail to capture important aspects of functional protein heterogeneity. G-protein-coupled receptors (GPCRs) represent by far the largest class of targets for modern drugs, but we have not yet unraveled the subtle dynamics of their function (George SR, 2002; Kenakin T., 2002; Kimple RJ, 2002; Ellis C., 2004; Kristiansen K., 2004). Heterodimerization enhances the complexity of ligand recognition and diversity of signaling responses of heterotrimeric guanine nucleotide-binding protein-coupled receptors (GPCRs) (Foord SM., 2003; Liebmann C., 2004). Heterogeneity of protein interactions that underlie both cell-surface receptor expression and the exhibited phenotype are caused by interactions with proteins which modify the activity profile of the GPCR, such as activity modifying proteins (RAMPs) (Christopoulos A, 2003; Fischer JA, 2002; Morfis M, 2003; Sexton PM, 2001; Tilakaratne N, 2000; Udawela M, 2004). These protein functions and interactions in different cells and cell types which we do not take into account in our subcellular models pop up rather unpleasant in the drug development process.

The popular techniques to explore and analyze low-dimensional data at high speed are based on the idea that this would provide all the data with sufficient predictive power to allow for a bottom-up approach to drug discovery. The current High Throughput Screening (HTS) and other early stage methods allow gathering low-dimensional data at high speed and volume, but their predictive power is too low as they lack depth of descriptive power (Perlin MW, 2002; Entzeroth M, 2003). We are just clogging the drug development pipeline with under-correlating data in relation to clinical reality. A bigger flow of unmanageable data does not equal a higher correlation to clinical reality.

The knowledge gathered at the infra-cellular level has to be viewed in its relation to the (living) cell in its native environment and the biological and non-biological processes influencing its function and health, which requires a top-down functional and phenotypical approach rather than a bottom-up descriptive approach. Complex disease processes cannot be explained by simple oligo-parametric low-level models. A high-speed oligo-parametric disease model does not equal high predictive power. It is not the ability to study a simplified disease model at high speed which will allow us to succeed, but we must study and verify the functional outcome of the in-vivo disease process itself.

A game of chess is not described by naming its pieces, but by the spatial and temporal interaction of both players or in other words the flow of actions and reactions, described in a space-time continuum and if we add the color it is a spatio-spectro-temporal flow of events. The individual pieces or moves do not explain the final outcome of the game, only when the entire process is analyzed from a positional and functional point of view we can understand and predict the reason why one player wins or loses. You have to study a game of chess at the appropriate organizational level in order to understand it or you will fail to find an explanation for the outcome of the game.

Isocellular disease models

Cellular disease models have already allowed us to study disease processes in geat detail, but they need to be dealth with carefully. Validation of a cellular model and a thorough understanding of its strengths and weaknesses is required. Using cellular disease models in more detail is not a trivial endeavor. Cellular disease models need to be related to at least the in vivo cellular disease process we want to study, so a validation of this correlation is very important (Gattei V, 1993; Thornhill MH, 1993; Lidington EA, 1999; Dimitrova D. S., 2002).

We now know that metabolic pathways show complex interactions and that gross genetic rearrangements can impair entire parts of cellular metabolism. The cellular models used in research should be validated for their functional and phenotypical representation of in vivo, in-organism processes. However many popular cell lines are not selected for their close linkage to clinical reality, but for their maintainability in the laboratory, lack of phenotypical variation, ease of transfectability, etc. . It is assumed that those cellular models are a valid representative of the disease process, but almost never a thorough assessment is being done. The phenotypic background of a cell has an important impact on the structure and function of cellular proteins (Tilakaratne N, 2000; Kenakin, 2003).

Primary cell lines cells in general require a more complex tissue culture medium than most popular cell lines. Cancer cells (and transformed cells) can usually grow on much simpler culture medium. Replicative senescence and varying behavior at each passage (which may necessitate a change of cell lines for long term experiments) also make primary cell lines less popular, as they necessitate a change of cell lines and variability in experimental data. Reduction of unpleasant variability in experiments by choosing a specific disease model may create nice results, but of a reduced predictive value. Quite often results obtained with one cell line, cannot be confirmed by using another cell line, without even talking about primary cells (Kenakin T, 2003).

CHO cells (Chinese Hamster Ovary, Cricetulus griseus) are used in many assays, but they are not derived from a human cell and are aneuploid (Tjio, J. H., 1958). HeLa cells are derived from an aggressive cervical cancer; they have been transformed by human papillomavirus 18 (HPV18) and have different properties from normal cervical cells (Gey, G.O., 1952). The U-2 OS osteosarcoma cell line is easy to maintain and transfect (Ponten J, 1967). The PC12 cell line which responds reversibly to nerve growth factor (NGF) has been established from a rat adrenal pheochromocytoma, it has a homogeneous and near-diploid chromosome number of 40 (Greene LA, 1967). HEC cells are derived of a human endometrial adenocarcinoma cell line and are also very popular (Kuramoto H., 1972). Two cell lines are very popular for epithelial barrier studies: Caco-2 cells and Madin Darby canine kidney (MDCK) cells. Caco-2 cells are the most popular cellular model in studies on passage and transport, they were derived from a human colorectal adenocarcinoma (Kirkland SC, 1986; Hidalgo IJ, 1989). Caco-2 cells are being used as a model to evaluate small intestine transport. The interpretation of Caco-2 transport data is often confusing, and is not always in agreement with in vivo observations, even when P-glycoprotein (P-gp) is blocked by specific inhibitors (Hu, M., 1999; Hunter, J., 1993; Lennernas H, 1994). Heterogeneity in the Caco -2 cell line, depending on passage number and origin of the cells, leads to differences in transepithelial transport (Walter E., 1996). Madin Darby canine kidney (MDCK) cells were isolated from a dog kidney (Gaush, C.R., 1966). They are currently used to study the regulation of cell growth, drug metabolism, toxicity and transport at the distal renal tubule epithelial level. MDCK cells are also used as a cellular barrier model for assessing intestinal epithelial drug transport (Cho, M.J., 1989). We quite often use the models which will grow in vitro, just to have at least something, although we know that this is a highly uncertain approach. Availability of a particular cellular model system does not equal predictability of the system.

Some popular cell lines may correlate with themselves and not with the complex dynamics of the physiological process in man they are supposed to represent. Studying the dynamics of the involvement of a protein in a disease in patients and transforming this knowledge into a disease model in a particular cell line requires a careful assessment before embarking on a drug discovery process. Functional cell model drift should be verified at regular intervals and taken into account.

Even within individual cell lines there is not always homogeneity in phenotype and function. Cancer cells show genetical and chromosomal instability as they tend to lose parts of chromosomes (Duesberg P., 1998; Lengauer C, 1998; Duesberg P, 2004). Using cell lines derived from cancers poses a correlation risk in relation to clinical reality on research done by using these types of cell lines. Continuous sub-cultivation of cells and an increase in the number of passages may lead to chromosome rearrangements and loss of functional reactivity (Dzhambazov B, 2003). Loss of function destabilises a cell when critical parts of pathways are lost, although cell cycling may continue in parts of the cell culture, but this will cause a drift and on experimental results.

Many of the most popular cell lines lack parts or even entire chromosomes and therefore large chunks of metabolic pathways. A drug molecule can not interact with the proteins which are not present in the cell line and an adverse or even positive effect will go unnoticed. Functional loss of proteins and enzymes in cancer cell makes them unresponsive to drugs if the protein(s) which are the target of a drug are lost without killing the cell as such.

Even when a protein is successfully expressed in a cell as shown on a Western blot, this does not equal functional success. Western blotting tells you how much protein has accumulated in cells. Even knowing the rate of synthesis of a protein by Radio-Immune Precipitation (RIP) does not predict the functional outcome of protein expression. Protein function is also depending on the metabolic background of the cell in which the protein is expressed and its spatial and temporal organisation. If the enzymatic and structural background of the cell does not meet the prerequisites to put a functional protein in the right location, embedded in the right functional environment, nothing appropriate will happen. An appropriate functional assay is required to validate proper function of the expressed protein.

Conclusions drawn from phenotypically uniform and simplified cell lines do not show a reliable or constant correlation to in-organism cellular dynamics. A functional comparison between isolated native cardiac myocytes and cloned hERG demonstrates the advantages of cardiac myocytes over heterologously expressed hERG channels in predicting QT interval prolongation and TdP in man (Davie C, 2004). The involvement of heterotrimeric G proteins in cell division was only discovered by looking at the native protein in its natural environment and not by using a trasfected system where the physiological gene regulation is disabled (Zwaal RR, 1996; Kimple RJ, 2002).

The location of glycosyltransferases involved in N- and O-glycan chain elongation was traditionally found to be confined to the Golgi-apparatus in phenotypically simple cells, such as HeLa cells. However, localization studies conducted in primary cell cultures often reveal ectopic localizations of glycosyltransferases usually at post-Golgi sites, including the plasma membrane (Berger EG., 2002). This shows that the prototypical or average cell does not exist in drug discovery as a valid broadrange cellular model. Cells are not just interchangeable containers with molecules in which we can pour reagents but highly dynamic environments.

In vivo enzymatic reactions are not linearly correlated to protein concentration or of zero order. The intracellular environment causes a more complex functional pattern for a given protein, such as bell shaped relation between protein concentration and function. A blunt on/off expression in a transfected cell does not correlate well to the physiological condition in a primary cell. When the appropriate metabolic environment is not present when studying a protein in a cellular disease model, predictivity of the disease model may be low compared to physiological conditions.

A traditional (homogeneous) cell culture in the laboratory may not yet mimic the physiological conditions in an entire organism, so our approach to cell-based research (and beyond) requires some redesign also. Creating a virtual organism, by differential screening of a multitude of cell type representing the main cell types in the human body (cardiomyocytes, hepatocytes ) could help us to improve the predictive value of cellular disease models. We need to study cell-to-cell and cell-type-specific pathway dynamics in more detail, as is the case for nuclear factor-kappaB (NF-KappaB) (Schooley K, 2003). Studying Biologically Multiplexed Activity Profiles (BioMAP) directly in primary cells provides us with results which are closer to the situation in man. BioMAP profiling can allow integration of meaningful human biology into drug development programs (Kunkel EJ, 2004).

Metabolic pathways in cells do not exist in a void, but are interconnected and highly dynamic processes. Blocking a pathway has far-reaching consequences for the intracellular environment. The upstream metabolites will either find their way through other metabolic pathways or pile-up. Some inborn errors of metabolism are an example of this principle (PKU ). Drugs blocking pathways also cause a distortion of the delicate balance in metabolic processes and may cause upstream effects by metabolites which are normally metabolized before they can cause any harm. The kinetics of the pharmakon may be documented, but the change in cellular metabolism and pathway-network distortion are less well understood. Upstream metabolites may become processed by other pathways and unexpected adverse effects may show up. Adverse effects on cellular metabolism are only present in those cells which have an intact metabolic pathway and not even all cell types activate the same pathways at all times.

Cell cycling and circadian oscillators are a source of periodic variation in our cell based experiments (Okamura H., 2004). Oscillations of cellular functions are to be explored and taken into account, as they act as a dynamic background or reference level representing a dynamic reference for studying cellular structure and function. Oscillations which are not accounted for act as periodic noise, disturbing our measurements and masking subtle, but possibly important events in our experiments.

Differential multiplexing in (high-content) cell based screening could help us to acquire more information about the spatial and temporal dynamics of cellular processes. Developments are going on towards experimental multiplexing and up-scaling of the capacity of quantitative cellular research (Perlman ZE, 2004). Techniques such as High-Content Screening (HCS) or multiplexed quantification can be applied to cellular systems to study intra-cellular events on a large scale (Van Osta P., 2000; Taylor DL, 2001; Van Osta P., 2002b; Abraham VC, 2004; Van Osta P., 2004). Subcellular differential phenotyping is already possible on a large scale by using human cell arrays. We can use multiplexed molecular profiling in cells to obtain more information (Stoughton RB, 2005). Light-microscope technology is used to explore the spatial and temporal dynamics in cell arrays in great detail (Ziauddin J, 2001; Bailey SN, 2002; Baghdoyan S, 2004; Conrad C, 2004; Hartman JL 4th, 2004).

Analyzing a large number of tissues for candidate gene expression is now greatly facilitated by using Tissue MicroArray (TMA) technology (Kononen J, 1998; Simon R, 2002; Braunschweig T, 2004).

From individual cell to cytome

Studying cell function and drug impact at the level of the individual cell is called cellomics (Russo E., 2000). However, the concept of cellomics does not take into account the supra-cellular heterogeneity which is present in every cellular system, such as a cell culture or an organism. By studying cells while ignoring their diversity we make the same mistake as the statistician who drowned crossing a river that on average was just three feet deep.

Due to the heterogeneity of cell types and differences between cells in a healthy and disease state, we need to take this heterogeneity into account. Cytomes can be defined as cellular systems and the subsystems and functional components of the body. Cytomics is the study of the heterogeneity of cytomes or more precisely the study of molecular single cell phenotypes resulting from genotype and exposure in combination with exhaustive bioinformatics knowledge extraction (Davies E, 2001; Ecker RC, 2004b; Valet G, 2003; Valet G, 2004).

In order to get the broader view on pathological processes, we should move on to the phenotypical and functional study of the cellular level or the cytome in order to understand what is really going on in important disease processes. Although the genome and proteome level have their predictive value in order to understand the processes involved in disease (and health), the cytome level allows for an understanding of pathological phenotypes at a higher level. By integrating the knowledge from the genome and proteome, we could give guidance to the exploration of the cytome, which was not possible before this knowledge was available.

The cytome level will also provide guidance to focus the research at the genome and proteome level and so creating a better cross-level understanding of what is going on in cells (Gong JP, 2003; Valet G, 2004; Valet G, 2004b). Some would see this as taking a step back from the current structural and systematic descriptive approach, but it is mainly a matter of integrating research at another level of biological integration and looking in a different way to the web of interactions going on at the cellular level. Biological processes do not exist in a void, but they are a part of a web of interactions in space and time, rather than being an island on their own. A cell is a multidimensional physical structure (3D and time) with a finite size, not a dimensionless quantity. We cannot ignore the spatial and temporal distribution of events, without losing too much information.

In recent years the tools have matured to start studying the cellular level of biological integration, but the tools are still used in the same way as if they were derived from low-content high-throughput phenomena as this is still the dominant research model. The tools to generate and explore a high-dimensional feature space are still scattered and not brought into line with the exploration of the cytome.

Functional processing in cellular pathways

The interconnection of genome, proteome and cytome data will be necessary in order to allow for an in-depth understanding of the processes and pathways interacting at the cellular level. A monocausal approach will have to be replaced with a poly- and pluricausal approach in order to understand and explain the phenomena going on at the cellular level. Pluricausal means causal contributions at different levels, such as genes, other cells and environmental influences. Polycausal means multiple causal contributions at the same biological level, such as polygenic diseases or multiple agonistic and antagonistic environmental influences. The concept of a multithreaded, multidimensional, weighed causality is needed in order to study the web of interactions at the cellular level. A drug modulates cellular function, but changes can be studied at different levels of biological integration:

Disease outcome = drug x (a x clinicaln + b x physiologicalp + c x cellularq + d x geneticr )

Disease models should incorporate mixed and nonlinear effects. Diagnosis and drug discovery merge if we take parallel models for both. The clinical diagnosis or para-clinical diagnosis of a disease should show a high correlation with the disease models used to study its possible treatment. A cause (e.g. a single gene defect, a bacteria) can have multiple consequences and as such be poly-consequential, which is the mirror situation of a single consequence being caused by multiple causes (co-causality or co-modulation) acting either synergistic or antagonistic (e.g. a disease with both a genetic an environmental component). In reality, a pathological condition is a mixture of those extremes (e.g. a bacterial or viral infection and the hosts immune system) and as such a simple approach is not likely to succeed in unraveling the mechanism of a disease. With the current systematic and descriptive approach however, we get lost in the maze of molecular interactions. We are looking at too low a level of biological integration and we get lost in a maze of structures and interactions. The cell is the lowest acceptable target, not its single components, like DNA or proteins.

We are looking at the alphabet, not even words or sentences, nature is not a dictionary, but it is a novel. We should study the flow of events in a cell with more power, not only the building blocks. As an example, Mendel did not need to know about DNA in order to formulate his laws of inheritance and he did not know that the discovery of the physical carrier of inheritance, DNA, would confirm his views later on, but his laws are still valid as such. Certainly physics was not at the stage it was in the 20th century when Newton formulated the law of gravity, but his observations and conclusions were valid. When Einstein formulated his relativity theory, he did not have modern physics at his disposal. His theory does not fit well to the quantum level, but does explain phenomena at a higher level of functional integration and as such is an appropriate model.

The value of a scientific model does not lie in the scale of phenomena it describes, but in its predictive correlation to the reality it tries to capture. The more we may try to exclude elements from reality, the better we may be able to build a model which holds in a tightly controlled situation in our laboratory, but fails when challenged by full-blown reality in the outside world.

What we find should not be in contradiction to what lower level structural descriptive research discovers, but we should not wait for its completion to start working on the problems we are facing in medicine and health care today.

Epicellular disease models

Epicellular models are of great importance for both efficacy testing as well as for toxicity testing. Organoids, parts of organs, isolated organs and animals are being used as epicellular disease models. The advantages of in vitro systems in toxicity testing are numerous. In vitro tests are usually quicker and less expensive. Experimental conditions can be highly controlled and the results are easily quantified. However, the relative simplicity of nonwhole-animal testing results in limitations as well. Cells or tissues in culture cannot predict the effect of a toxin on a living organism with its complex interaction of nervous, endocrine, immune, and hematopoietic systems. In vitro systems can predict the cellular and molecular effects of a drug or toxin, but only a human or animal can exhibit the complex physiological response of the whole organism, including signs and symptoms of injury.

Rats, mice, dogs, (non-human) primates, rabbits, cats, guinea pigs, hamsters, miniature swine, goats, farm pigs, etc. are some of the species which are available for preclinical and non-clinical evaluation. The main reasons for wanting to predict human sensitivity are that most failures in the clinic are due to safety problems and lack of effectiveness. The inability of animal models to predict these failures before human testing or early in clinical trials dramatically escalates costs. The choice of the relevant animal species for preclinical studies is an important consideration. The goal of pre-clinical safety assessment studies is ultimately an estimate of safety in humans. A critical assumption is that the toxicity responses seen in various animal models will be reflective of those in humans (e.g. comparative toxicogenomics). Assumptions require in-depth validation, some of which we will discuss in this section.

Animal models are an important part of the drug discovery an development process and they have made a significant contribution to our understanding of disease processes. The correlation of the animal model to the actual process (efficacy, toxicity) in man is an important issue to consider (Hondeghem LM, 2002; Huskey SE, 2003). Inter-species differences, can have a significant impact on the interpretation of results in an animal model for a human disease. Earlier successes do not relieve us from continuing critical assessment of every model. With the increasing complexity of the diseases being studied, subtle interspecies differences become increasingly important. In the past twenty years a lot has changed in the use of animal models to study human disease processes and develop new drugs.

A first example of using an epicellular screening model for drug discovery was the discovery of Arsphenamine (Salvarsan) by Paul Ehrlich. Paul Ehrlich (1854-1915) started the first screen in search of compounds effective against syphilis, resulting in the world's first "blockbuster" drug Salvarsan. Sahachiro Hata discovered the anti-syphilitic activity of this compound in 1908 in the laboratory of Paul Ehrlich, during a survey of hundreds (606) of newly-synthesized organic arsenical compounds. This was the first organized team effort to optimize the biological activity of a lead compound through systematic chemical modifications, the basis for nearly all modern pharmaceutical research.
What was different from modern-day High-Throughput Screening (HTS) is that their mouse model was not only a valid representation of the disease Syphilis in man, but also in one pass provided information about ADME and Toxicity for which their mouse model provided sufficient predictive power. Essentially we now perform screening in test-tube enabled models, even if we now call them multiwell-plates. Only disease processes which can be represented in a test-tube or agar-plate are amenable to ultra-high automation, but in many cases they are a "reductio ad absurdum" of the full blown human disease process (the premise that the test-tube assay represents the disease with enough validity to start from here is often false). The validity of simple-model screenings to represent the complexity of a disease process in man is of secondary importance as we hope to compensate for this later on in the process (we do, but at an unacceptable cost).

Another historical example of a drug discovered by using an animal model for a disease, is the discovery of the first successful oral antibiotic, Prontosil (Sulfamidochrysoidine). Prontosil was developed by Gerhard Domagk, working in the Bayer sector of I. G. Farbenindustrie. Sulfamidochrysoidine seemed to have no effect on bacteria in vitro, but Domagk went ahead and tested it on 26 mice injected with streptococci. Fourteen were kept as controls, and 12 were treated with Prontosil. All of the controls died within a few days, while all of the treated mice survived. Later Gerard Domagk received the Nobel Price for his discovery, which saved the lives of many people.

Much has changed since Gerard Domagk's discovery of Prontosil. Animal models are being used for evaluation of efficacy and toxicity. We can now use genetically modified animals to study gene regulation and cell differentiation in a mammalian system (Gordon JW, 1980; Isola LM, 1991; Brusa R., 1999). Transgenic and gene-deleted (knockout) mice are used extensively in drug discovery (Rudmann DG, 1999). The pioneering work of Mario Capecchi, Martin Evans, Oliver Smithies and others has enabled the construction of increasingly sofisticated animal models (Smithies O, 1984, Evans MJ., 1989). Homologous recombination between DNA sequences residing in the chromosome and newly introduced, cloned DNA sequences (gene targeting) allows the transfer of any modification of the cloned gene into the genome of a living cell (Capecchi MR., 1989). The challenges now are to model the complex multifactorial diseases, instead of simple monogenic diseases (Smithies O., 1993; Smithies O., 2005). Simple knockouts are usually designed to lead to loss of protein function, whereas a subset of cancer-causing mutations clearly results in gain of function. Mored dynamic transgenic mouse systems are now avaialable (Sauer B., 1998; Maddison K, 2005). The original Cre and FLP recombinases have demonstrated their utility in developing conditional gene targeting, and now other analogous recombinases are also ready to be used, in the same way or in combined strategies, to achieve more sophisticated experimental schemes for addressing complex biological questions (Garcia-Otin AL, 2006).

Although knockout technology is highly advantageous for both biomedical research and drug development, it also contains a number of limitations. For example, because of developmental defects, many knockout mice die while they are still embryos before the researcher has a chance to use the model for experimentation. Even if a mouse survives, several mouse models have somewhat different physical and physiological (or phenotypic) traits than their human counterparts. An example of this phenomenon is the p53 knockout. Gene p53 has been implicated in as many as half of all human cancers. However p53 knockout mice develop a completely different range of tumors than do humans. In particular, mice develop lymphomas and sarcomas, whereas humans tend to develop epithelial cell-derived cancers. Because such differences exist it cannot be assumed that a particular gene will exhibit identical function in both mouse and human, and thus limits the utility of knockout mice as models of human disease (Pray L., 2002).

There is not only interspecies variation, but also intraspecies variation. The biological variability is also present in genetically modified animals. We expect a specific phenotype from a specific genetical modification. In genetically modified mice however, the observed phenotype is not always the direct result of the genetic alteration (Linder CC., 2001; Schulhof J, 2001). The effect of the genetic modification is not completely straightforward, due to variations in the genetic background of the animals (Crusio WE., 2004). Transgenic mice containing the same genetic manipulation exhibit profoundly different phenotypes due to diverse genetic backgrounds (Sigmund CD., 2000; Sanford LP, 2001; Holmes A, 2003; Thyagarajan T, 2003; Bothe GW, 2004).

The relevance of a particular model is linked to the similarity of a process in the model animal and man. There is however considerable variation between species which complicates the use and evaluation of animal models (Hucker HB., 1970; Smith CC., 1967; Smith RL., 1974; Hengstler JG, 1999; Chiu SH, 1998; Nelson SD., 1982; Fry JR., 1982). A required part of drug development is the "Chronic Bioassay". Hundreds of pharmaceuticals have been reported to give a positive result in the standard "Chronic Bioassay", which consists of an 18 to 24 month daily administration of the test compound in mice and rats. This is in contrast with 20 pharmaceuticals, which are known to be carcinogenic to humans (Van Deun K, 1997). Interspecies differences are an important issue and require in-depth validation (e.g. comparative toxicogenomics). Animals are popular models for studying G.I. absorption for oral drug uptake, but in addition to metabolic differences, the anatomical, physiological, and biochemical differences in the gastrointestinal (G.I.) tract of the human and common laboratory animals can cause significant variation in drug absorption from the oral route (Kararli TT., 1995).

A classical example of inter- and intraspecies variability is Thalidomide. The mouse and rat were resistant, the rabbit and hamster variably responded, and certain strains of primates were sensitive to thalidomide developmental toxicity. Different strains of the same species of animals were also found to have variable sensitivity to thalidomide (Neubert D, 1988; Bila V, 1989). Although the drug was marketed in 1957, reproductive studies on thalidomide in animals were not started until 1961, after the drug's effects on human fetuses had begun to be suspected (MacBride, 1961). Initial studies on rats and mice revealed some reproductive abnormalities, notably reduction in litter size due to resorption of fetuses; however, only when the compound was tested in the New Zealand white rabbit did abnormalities similar to those noticed in human babies occur (Cozens DD., 1965). Studies on monkeys revealed that they were almost as sensitive as humans to the deformative effects of the drug (Delahunt CS, 1965).

Mouse models require careful evaluation and validation. Selection of mouse models of cancer is often based simply on availability of a mouse strain and a known compatible tumor. As a consequence cancer models in mice quite often fail to predict success in clinical development later on (Kelland LR., 2004; Kerbel RS., 2003; Peterson JK, 2004; Schuh JC., 2004; Voskoglou-Nomikos T, 2003).

Using inbred mouse strains reduces variation in genetic background, but also reduces the correlation of the disease model to real-world genetic and metabolic variation encountered in human populations. Finding a strong correlation in an inbred laboratory animal population is no guarantee that the correlation will hold in an out bred natural population (Mackay TF., 1996; Macdonald SJ, 2004). Do we want nice results with a low standard deviation (SD), or do we need results highly correlating with clinical reality? If one wishes to obtain the optimal mouse model for a human disease, one needs to choose the correct genetic background as well as the correct mutation (Erickson RP., 1996). Careful validation of a transgenic mouse model is required, such as with models for Alzheimer's Disease, for example by using biomarkers and PET (Bacskai BJ, 2003; Klunk WE, 2004; Klunk WE, 2005). Validation of a model for a complex disease requires a thorough validation.

We still do not have an in-depth understanding of the delicate spatial and temporal interplay in metabolic pathways in cells, organs or entire organisms in transgenic animals. Introducing or removing a gene without a clear understanding of its spatial and temporal expression pattern, leaves us with a correlation deficit in relation to the disease process in man.

When we modify a gene, we modify a pathway-web with upstream and downstream consequences for cellular metabolism in different cellular compartments (nucleus, Golgi ). The metabolites which (dis-) appear due to the modification will modify a highly dynamic network of metabolic interactions. In-vivo spatial and temporal variation in protein structure and activity profiles will add to the complexity of unravelling the functional impact of modified gene expression.

When we want to bridge the gap from "model to man" we must identify and verify the process in its native environment the cell and the entire organism (e.g. with a biomarker) and compare and validate what happens in a mouse to what happens in man (Lee JW, 2006). The use of biomarkers can be added to preclinical studies to help justify the choice of animal species. The ultimate animal model for compound selection is still man himself and with the recent introduction of extremely sensitive detection methods such as accelerator mass spectrometry (AMS) and Positron Emission Tomography (PET) it is now possible to evaluate the absorption, metabolism and elimination of compounds in man as part of the pre-clinical selection process (e.g. microdosing).

Scientific evolution and when Biology meets Chemistry

Before we enter the road towards the development of a new drug we still lack a lot of knowledge and understanding of the complex biology of the disease pathway. The long way towards the application of a new drug or biological for a human disease starts nowadays at the lowest possible level of biological complexity; the confrontation of a single molecule with a single molecule.
We remain blind for the complexity of the intrahuman ecosystem until very late into the development cycle of new medication which is why there is a bias towards false positives built into the process. At the moment we just accept this and we still succeed against all odds to bring new medication to the market. The overall system works, but is not an example of an efficient and effective process, we just accepted to live with these low odds of success.

The fundamental mechanism by which science progresses is the same as how nature evolves by means of natural selection. In science the role of mutations is performed by new paradigms which break through the frontiers of established science and provide a comptitive edge though their improved ability to explain reality. However more than 99 percent of all experimetns are performed within the safe environment of present day scientific principles. Science evolves by means of scientific selection, which is somewhat different from natural selection as the selection pressure is partly exercised by man in an artificial laboratory environment instead of nature itself (e.g. peer review, commercial pressure, present day scientific theory and paradigms.
The scientific selection process therefore tends to produce a bias towards false positives, which becomes clear at the end of the process when confrontation and selection by nature itself can no longer be avoided. This happens when we have to face clinical reality itself and not only the artificial environment of the laboratory. The quality of the selection pressure in the end determines the potential speed and direction of scientific evolution.

Biology meets Chemistry meets Biology
Figure 32. Biology and Chemistry meet at the point of minimal system complexity.
The bandwidth of system selection pressure is minimal at the moment of confrontation.
Relative understanding
Figure 33. Proportion of understanding achievable within chemical and biological system space.
Biology is more complex than molecular chemistry, we do not master each one completely.
Evolution of certainty
Figure 34. Evolution of system selection pressure during the R&D process.
Selection pressure evolves in a non-linear way.
Evolution of certainty
Figure 35. Bandwidth of uncertainty during the R&D process.
What is not present at the moment of decision, does not contribute to the decision.

Working with models at different levels of biological complexity, has its consequences. We arrive at the point of confrontation between biology and chemistry at a moment when we have reduced system complexity to its bare minimum (Figure 32). The long an winding road towards unraveling the disese process and arriving at a screenable target has made us strip away man from the target molecule. Almost nothing is left of the intrahuman ecosystem at the moment when we confront chemistry with biology. Less than 0.01 % of the full blown selection pressure is present at the moment of confrontation between biology and chemistry. We are as such 99.99 % blind for interfering phenomena. The molecule surviving the initial selection process evolves in an almost pristine environment compared to the metabolic jungle of man. Unable to increase our understanding of biological processe in the intact human ecosystem, fast enough to keep productivity of the R&D process sustainable, we retreated into simplified models in the hope to catch up later on in the process.

When we slowly retreat back to the confrontation with the full blown intrahuman ecosystem, we make our decisons at each stage gate within an environment which is only partly representative for the ultimate situation (Figure 33). Instead of symplifying at each step (as is the case during unraveling a disease, left side of Figure 32), we are confronted with a system of higher complexity after passing a stage gate, for which the previous step provides less than ideal predictive power. Even if we extract all information before arriving at a stage gate, our predictive power is flawed in relation to increased complexity awaiting us after the gate. We see what we observe, but not what we need to know. The retreat is also not following the same trajectory as we used for approaching the biological-chemical confrontation, but a twisted path biased by regulations and interspaced with gaps, caused by model discontinuities.

The relative power of our understanding in relation to the system we are working in, is higher in the early chemical environment than in the later biological environment (Figure 34). As we can witness in the high late stage attrition rates, the overall chance of succes of Phase III now equals about 50 %, or it is as predictive as tossing a coin.

The selection pressure, both positive and negative reaches a high level only at the end of the pharmaceutical R&D process, because it is only then the full blown complexity of man and human population is able to exert its effect (Figure 34). The consequence is that the power of a decision at a stage gate is highly uncertain during most of the process (Figure 35). Increasing the capacity of under-predicting processes early on in the pipeline, will not bring down attrition rates.

Less questions, better answers

Failure rate
Figure 36. A drug discovery and development process with a right choice of
93% at each step still only performs at about 50% overall. By Phase III we spend
50% of our efforts on failures.
Bayesian inference
Figure 37. Searching for needles in a haystack leads to very high false positive rates.

The overall drug discovery and development process is Data Rich, but Information Poor (DRIP), which explains its low overall efficiency. Within major pharmaceutical companies, there are numerous functional silos generating different data from different technologies with different levels of error. A process as complex as the present-day drug discovery and development process requires a success rate (true positives and true negatives) of more than 93% at each step to perform above 50% overall (Figure 36). Even if we make the right choice in 93% of all cases at each stage-gate and make a false positive or false negative choice in only 7%, this means that by the end we will spend 50% of our efforts on failures in Phase III. The wrong choices we make in the beginning we take with us through the entire process. Every stage succeeds or fails at its own stage-gate, but the overall process fails because the decisions at each stage-gate are under-correlating to the true end-point of the process (cure a disease in man, not in a test-tube). We confirm or reject the past (up to the stage-gate), but do not (enough) predict the future state of the process and so we add-up correlation deficits (false positives) with each step. At the moment the choice of target and lead candidate is flawed, due to a lack of knowledge about the disease process and the interaction of the chemical substance with the human biological system. We fail at the beginning, but pay the price at the end. As we were incapable to understand molecular processes in complex biosystems, we added some simplified, but high-speed environments upfront to the process. We increased quantity, but in the process sacrificed the quality of correlation with the reality of disease in man.

False positives are a problem in any kind of test: no test is perfect, and every the test will incorrectly report a positive result in certain cases. The problem of false positives lies, however, not just in the chance of a false positive prior to testing, but determining the chance that a positive result is in fact a false positive. Using Bayes' theorem, if a condition is rare and only a minority of molecules show activity in a High Throughput Screening (HTS) campaign, then the majority of positive results will be false positives, even if the assay for that condition is (otherwise) reasonably accurate (Figure 37).
Blind screening of a random chemical library (e.g. generated by combinatorial chemistry) against a biological target leads to an excessive number of false positives, putting more pressure on the secondary screens and beyond to weed out these false positives. A simple example, based on
Bayesian inference illustrates the phenomenon. An assay is being used which has a 99.99% true positive rate. The graph shows the percentage of false positives for three different true negative rates 95%, 97% and 99%. As you can calculate yourself (see Bayesian inference), the true negative rate has a more dramatic impact on the number of false positives when the population frequency (hit rate) is very low. By using a chemical library whose chemical space is badly matched to the biological space of the target we create an high proportion of false positive results. As the high attrition rates in clinical development show (e.g. Phaze II) show, we do not get rid of these false positives until very late in the overall process. However when the population frequency of chemical structures matching with the validated target increases, the rate of false positives decreases exponentialy (Figure 37). Designing molecules with an understanding of their potential interaction with a truly validated biological target leads to an increase in process efficiency which surpasses all other attempts made to improve the drug discovery and development process. Mistakes made at the beginning are harder to compensate for later on in the process. It is at the start of the process when the path towards a succesful drug in man heads in the right direction or not. We first need to understand what happens in the cytome of man, not only in an Eppendorf. Now the tools are becoming available to achieve this goal:

  • Investigate the molecular physiology of a disease process in individual cells and in man.
  • Design chemical structures which match the validated drug target(s).

The idea of a Human Cytome Project is about the improving ability to understand system-wide phenomena at a (sub-)cellular resolution (molecular physiology). This leads to being able to ask the question about effectiveness and toxicity in one step in a complex biological environment, in model organisms and in man (ask less questions, but get better answers). Molecular observation and answers to system-wide questions leading to better models generated from improved observations.

Human Cytome Project - the way to go

Exploring the organizational levels of biology
Figure 38: Exploring the organizational levels of human biology for the benefit of medicine.
A linked and overlapping cascade of exploratory systems
each exploring -omes at different organizational levels of biological systems
in the end allows for creating an interconnected knowledge architecture of entire cytomes.
The approach leads to the creation of an Organism Architecture (OA)
in order to capture the multi-level dynamics of an organism.

Practical issues on how to to explore the human cytome and a concept for a software architecture can be found on these pages:
The article "How to explore and find new directions for research", discusses ways to explore the cytome, ranging from digital microscopy, High Content Screening (HCS) to Molecular Imaging. The cytomic microcosm, both in-vivo ans well as in-vitro is accessible for highly detailed detection and analysis. New experimental techniques allow for an in-depth exploration of the (patho-)physiology of the human cyctome. The human cytome is now more accessible than ever for exploration and as a consequence a leap forward in our understanding of the complex machinery of the cytome in health and disease.
The article "A framework concept to explore the human cytome on a large scale", presents some concepts on the way towards large scale exploration of the human cytome system.

Final remarks

The war against human diseases is not an easy one to win. It spans the globe and stretches through the ages. The trenches of our war against disease reach from bench to bedside and require a strong and united effort in order to succeed.

Innovation is much more than discovery of new drug targets. There is no substitute for knowledge and understanding. More rational, more efficient and more informative preclinical and clinical drug development is required. We have to get away from a trial and error approach to a "cognitive" chemical biology approach, matching biological and chemical space. We have to do a better effort to build up understanding about the target (system) and active compounds.
Drug discovery and development show a decreasing efficiency in relation to the amount of money being invested. The level of understanding at the end of discovery (and preclinical development) should achieve a knowledge level which is capable to predict success at the end of the pipeline much better than we do now. We need to improve our preclinical disease models and our understanding of clinical reality. The quantity of what we can achieve with automation of drug discovery needs to be matched with the predictive quality of the science underlying the automated procedure. There is an urgent need to improve the productivity of drug discovery and development, but this time we should evaluate the underlying process better than we did with the advent of target-based drug discovery and High Throughput Screening (HTS). A high statistical significance, does not guarantee a concommitant clinical relevance, as correlations derived from inductive reasoning may be flawed in relation to the clinical reality they are meant to represent within the experiment.

Our disease models should capture more of the high-order complexity of biology, beyond the genes and proteins which are the current focus of research. Higher-order disease models should place genetic and protein research results in a broader perspective and complement them with information about high-order interations in space and time. We should study cells with taking into account their in-cytome differentiation and their real-life behavior. The sooner NCEs or NBEs evolve in a "rich" or lifelike biological environment (background, population variability) the earlier we capture (un-)wanted phenomena. It is all about improving the Probability Of Success (POS) for the overall process. There is no escape from the demands for better treatments from patients, society and shareholders. Scientists are working day and night to develop new treatments for unmet medical needs, but their effort should become more efficient and effective, so more of the candidate drugs reach the patients waiting for new and improved treatments. The weight of higher-order biological exploration should increase in the overall discovery and development process. More of man's diseases should be captured in our (pre-)clinical disease models.

A possible way out is to start drug discovery at an intermediate level of biological complexity, the cellular level (the cytome, taking into acount physiological cellular heterogeneity), with analysis of phenotypical and functional parameters (Cellular Physiology or Phenotype Based Screening). Deconvolve back towards molecular targets, which now have been confronted with drug candidates in an already complex biological system.

In the end process improvement should lead to a dramatic decrease in false positive results in preclinical development, while at the same time avoiding an increase in false negatives. This should lead to a better performance of the overall drug discovery and development pipeline in which more and better drugs reach the patients. I want to end this article with a quote from Dr. Paul Janssen: "For many sick people, there are still no drugs, and it is our job to develop good medicines".

Acknowledgments

I am indebted, for their pioneering work on automated digital microscopy and High Content Screening (HCS) (1988-2001), to my former colleaguas at Janssen Pharmaceutica (1997-2001), such as Frans Cornelissen, Hugo Geerts, Jan-Mark Geusebroek, Roger Nuyens, Rony Nuydens, Luk Ver Donck and their colleagues.

Many thanks also to the pioneers of Nanovid microscopy at Janssen Pharmaceutica, Marc De Brabander, Jan De Mey, Hugo Geerts, Marc Moeremans, Rony Nuydens and their colleagues. I also want to thank all those scientists who have helped me with general information and articles.

References

References can be found here

Cytome Research

Additional Information

History of Human Cytome Project set of articles

Previously posted versions, sorted by date

Copyright notice and disclaimer

These webpages represent my personal interests, opinions and ideas, not those of my employer or anyone else. I have created these web pages without any commercial goal, but solely out of personal and scientific interest. You may download, display, print and copy, any material at this website, in unaltered form only, for your personal use or for non-commercial use within your organization.
Should my web pages or portions of my web pages be used on any Internet or World Wide Web page or informational presentation, that a link back to my website (and where appropriate back to the source document) be established. I expect at least a short notice by email when you copy my web pages, or part of it for your own use.
Any information here is provided in good faith but no warranty can be made for its accuracy. As this is a work in progress, it is still incomplete and even inaccurate. Although care has been taken in preparing the information contained in my web pages, I do not and cannot guarantee the accuracy thereof. Anyone using the information does so at their own risk and shall be deemed to indemnify me from any and all injury or damage arising from such use.
To the best of my knowledge, all graphics, text and other presentations not created by me on my web pages are in the public domain and freely available from various sources on the Internet or elsewhere and/or kindly provided by the owner.
If you notice something incorrect or have any questions, send me an email.

This article was created by Peter Van Osta.
Email: pvosta at gmail dot com
Mobile: +32 (0)497 228 725

A first draft was published on Monday, 1 December 2003 in the bionet.cellbiol newsgroup. I posted regular updates of this text to the bionet.cellbiol newsgroup.

Latest revision on 10 May 2014