Whitcomb, David C. MD, PhD
Personalized medicine—deciphering disease mechanisms in individual patients to predict the effect of interventions—is an exciting, futuristic concept for which few successful examples of systematic approaches and implementation exist. In this article, I describe a unique and highly effective academic program created at the University of Pittsburgh School of Medicine and the University of Pittsburgh Medical Center (UPMC) in the Division of Gastroenterology, Hepatology and Nutrition to rapidly develop personalized medicine for digestive diseases. The program has required a paradigm shift away from traditional biomedical approaches that are based on the germ theory of communicable infectious diseases and scientific approaches designed to identify a single factor responsible for a complex disease, as articulated in Koch's postulates (stated in List 1). The need for this shift is predicated on the view that the current bed-to-bench-to-bed process for developing effective treatments for patients with complex inflammatory diseases is comparatively inefficient, ineffective, time consuming, and expensive.1,2
As an alternative to traditional statistical approaches to understand complex diseases, we have taken a reverse engineering approach to subdivide complex chronic inflammatory syndromes into scientific disciplines (where statistics are applied to test mechanistic models) and then integrate the findings into predictive disease models. Similar chronic inflammatory diseases in different organs (i.e., pancreas, intestine, liver) require similar scientific approaches, and for these types of diseases, detailed patient disease information and outcomes are organized in parallel. We approached this conceptually by aligning the names of the organ systems into rows and the names of the scientific disciplines into columns, which produces a grid. We reasoned that when phenotypic and biomarker information corresponding to each cell of a grid is collected on a patient, and the information from many patients is combined, it becomes a matrix. The advantage of the matrix is that it allows for highly organized patient information to be evaluated from multiple perspectives. This is invaluable for higher-level analysis of the interactions between genes, genes and environment, and disease modifiers linked to systemic processes instead of only the specialized cells of an organ. This level of information and organization is required for the development and implementation of personalized medicine for complex disorders.
I developed this matrix strategy in 1999 while negotiating to become division chief based on a long-term vision of personalized medicine for complex inflammatory disease, a recognition that academic physicians and scientists need specific niches for recognition and promotion, and a conviction that an academic medical center had a responsibility to solve complex medical problems. On the basis of the scientific strengths of the University of Pittsburgh and the patient populations served by UPMC, we proposed the recruitment of a series of faculty members who had specific disease interests and scientific expertise to fill in the cells of a grid defined by seven complex digestive disorders (rows) and seven scientific disciplines (columns). Each faculty member was to serve as a “translator” for specific types of medical problems within a complex disease and recommend data elements to be collected on patients within his or her domain. Working together, groups of faculty could provide all relevant medical and scientific perspectives on complex problems and provide more comprehensive phenotyping of each patient. Phenotypes, genotypes, and biomarkers could then be integrated into personalized predictive models designed to improve outcomes at much lower cost for each individual patient.
In the rest of this article, I first review modern Western medicine, using examples to illustrate why applying methods designed to identify single factors responsible for a syndrome, as in an infection, is comparatively ineffective in resolving complex disorders, where the etiology and character of the disorder require the interaction of multiple factors. I then describe a new vision of addressing complex digestive disorders, using a matrix academic division (MAD), and conclude with an example of how our institution's MAD approach to chronic pancreatitis (CP) is leading to the successful implementation of personalized medicine.
Modern Western Medicine: The Germ Theory of Disease
Modern Western medicine owes much of its success to adoption of the germ theory of communicable infectious diseases, first proposed in the early 1800s, and the scientific methods needed to diagnose and prove effective treatment. The underlying hypothesis is that an infectious disease is caused by single pathologic entity, such as microbes. The method to prove that a single agent (e.g., Bacillus anthracis) causes a disorder (e.g., anthrax) was developed by Koch and articulated by Loeffler in the 1890s3 (see List 1). Since then, many striking advances in medicine, both in terms of diagnosis and treatment, have been seen in infectious diseases, while slow progress is also being made in diabetes, heart disease, cancer, and other, more complex disorders.
However, the effectiveness of Koch's method breaks down when the disease is a syndrome with multiple etiologies producing identical phenotypes, when multiple factors are required to initiate a disease (e.g., a two-hit model), and when genetic and environmental factors strongly affect disease features and response to therapy. This problem distinguishes inflammation caused by germs from idiopathic inflammatory diseases; the latter results from different combinations of factors in different people that contribute to etiology, susceptibility, severity, and progression and require a new diagnostic and treatment approach. The new approach, which is the topic of this article, relies on the development and testing of disease models integrating detailed phenotypes, genetics, and biomarkers to predict disease pathways and effective treatment, which is the heart of personalized medicine.
Translational Research in Complex Disorders
Complex inflammatory disorders seem to occur through the interaction of genetic, metabolic, and environmental factors that may not be independently pathologic but can combine and interact to initiate disease and alter disease severity, progression, and complications. Furthermore, complex inflammatory disorders are not congenital but develop after an injury or some type of stressor, have organ specificity and involvement of multiple types of inflammatory cells, and variably affect the nervous system or vascular system, altering healing or tissue regeneration and increasing the risk of cancer. The mechanistic etiology of simple disorders can be resolved using standard statistics when there is an order-of-magnitude more of data than of variables. However, with the highly complex problem of idiopathic, genetic-variant-associated complex diseases, there is an order-of-magnitude more of variables than of data (e.g., millions of genetic variations in the human genome and thousands of subjects). Examples of complex inflammatory disorders in the field of gastroenterology include liver cirrhosis, CP, and inflammatory bowel disease (including Crohn disease and ulcerative colitis).
Because complex inflammatory disorders have multiple etiologies and variable outcomes, no single approach will provide all of the answers regarding their causes or provide a target for effective treatment. Instead, each component of the disease must be studied within an appropriate context (e.g., comparing well-matched patient subsets with and without the specific biomarker or end point). Thus, rather than using a global, unbiased, null-hypothesis significance test (to obtain P values) to identify the components of complex etiology and then build a statistical classifier model, the reverse engineering approach focuses on the components of a larger system, develops plausible models based on knowledge of biology, and uses statistics to guide the development of predictive models. One of the many advantages of the reverse engineering–predictive modeling approach is that it markedly limits the number of possible variables to be considered in each component model and directs the selection of biomarkers to monitor disease progression and response to treatment. The mechanistic insights gained from testing the components of a complex disorder are then integrated into a more holistic overall disease model. However, model building is much more difficult than null-hypothesis significance testing and requires ongoing contributions of many different types of faculty.
The term translation in the context of research means different things to different people.4 In complex inflammatory disorders, that term defines effective communication between members of multiple clinical and scientific disciplines, as illustrated in Figure 1. In that illustration, a group of clinicians must be conversant with a group of clinical or basic scientists and pull together the complete story for each complex disorder. The basic scientists should watch for common threads across multiple organ systems identified through their conversations with the different clinical specialists. For translation to occur, there must be a translator with expert knowledge of both the disease and the science who can understand those common threads. Once that happens, the elements needed for personalized medicine can be ascertained.
Structuring a program that can support such a matrix organization and deliver effective personalized medicine to individual patients with complex disorders has been a challenge to academic medical centers. Below, I illustrate how to structure a clinical MAD to facilitate translating research to personalized medicine.
Structuring a MAD for Personalized Medicine
After recognizing that unique combinations of risk factors may give rise to a particular complex inflammatory disease syndrome in a specific patient, my colleagues and I at the University of Pittsburgh School of Medicine reorganized our traditional academic GI division with a clinical arm, a research arm, and an educational arm to be a MAD division with the specific objective of providing the “right treatment to the right patient at the right dose at the right time.” We began with our own postulates:
1. Complex inflammatory diseases describe a multistep process.
2. The organ or system where the disorder originates is not “normal” but, rather, has intrinsic defects in a specialized group of cells of the organ.
3. Chronic inflammatory disorders are not congenital but, instead, require environmental stressors to become manifest.
4. Complex inflammatory disorders involve both abnormalities in specialized cells and also abnormalities in components of the immune response to injury, repair, or healing. These secondary effects may be shared in different organs with chronic inflammatory processes.
Teasing out the necessary elements to solve a complex inflammatory disorder, communicating the unsolved problems to the scientists, retrieving possible mechanistic answers, and applying them to clinical practice require several critical components:
* A large patient population with diverse phenotypes and genotypes
* Immediate and continued access to experts from a wide variety of basic research disciplines
* The recruitment, training, and cooperation of dedicated physician–scientists (translators) who can link all relevant sciences to the specific syndrome and communicate with the scientists
* Biology-based disease models for organizing and testing clinical and experimental data
* Mathematical representations of general physiologic models that are then adapted to individual patients based on their individual risk, disease activity, and predicted outcome
* Sustained financial support
Fortunately, this combination of conditions exists at UPMC, and we hope that our MAD division can serve as an example of a new paradigm in action.
Establishing a translational culture
A major burden of digestive diseases derives from chronic inflammatory conditions with inflammation, fibrosis, pain syndromes, and inflammation-associated cancer risk. Beginning in 2000, we began restructuring the entire GI division with the ultimate goal of resolving the etiology of complex digestive disorders and their complications. Understanding the etiologies of these disorders, the risks of acquiring them, and various complications would provide the foundation for prevention and treatment.
With the assistance of an organizational development group at the University of Pittsburgh, we created an implementation plan to maximize efficiency and effectiveness as defined by clinical service opportunities, disease burden, economy of scale, and opportunities for strong collaboration in critical areas of research. In addition, the academic needs of the physician–scientist were considered and facilitated because our expert faculty are our most valuable resource. The result was the novel MAD, as illustrated by a grid in Chart 1, for which seven centers of excellence (COEs) were created (their names are shown to the left of the chart's seven rows).
From an operations perspective, the COEs serve as a target for clinical referrals and are ideally staffed by seven or more physician–scientists (horizontal axis of the chart). Seven intersecting scientific disciplines (vertical axis) were likewise identified to focus on specific types of problems that run through multiple inflammatory diseases. The particular faculty member with expertise at the intersection of a specific clinical disorder and a specific scientific discipline has the responsibility as a translator to represent the physicians working in a COE in discussions with clinical or basic scientists of the corresponding scientific discipline within the university or elsewhere. In some cases, a scientist with a strong clinical interest and support from the COE physicians functions as a translator. Such a design is powerful from an organizational standpoint because it allows each physician–scientist translator to be supported by six clinical colleagues focused on the same type of complex disease (the other six physician–scientists in a COE) and six scientific colleagues working in the same discipline (the other six scientists working in the same discipline) but focused on parallel complex diseases served by other COEs.
Within the grid, faculty translators maintain two dimensions of communication: one with the clinicians in the COE, and the other with clinical or basic scientists. In some cases, the key faculty may have primary appointments in another department, but they typically have joint appointments in the Division of Gastroenterology, Hepatology and Nutrition and are integrated into MAD. Of note, there are two specialized programs with multiple faculty members sharing complementary interests within the same cell of the grid: the Center for Pain Research with a basic science focus, and the Visceral Inflammation and Pain Center with a clinical focus. The clinical faculty translator's primary responsibility on the research side is to identify the best clinical or basic scientists to address a very specific mechanistic question. Stated in a general way, the question is “Why are we unable to predict the outcome of this subset of patients with these characteristics?” The collaborations are one-on-one and are very goal oriented.
The second dimension of communication is horizontal, which occurs in two structured formats. The physician–scientists of a COE meet in a multidisciplinary clinic on a weekly basis with other physicians, surgeons, pathologists, and support staff. Each week, a COE sponsors one clinical case conference (a multidisciplinary conference) and a translational research conference/working group meeting to facilitate the clinical and translational research effort. This type of interaction is more of a collaborative think tank and facilitates higher-level integration.
By organizing the COEs around parallel complex inflammatory disorders, we also developed a critical mass of faculty with similar scientific interests, which justified the establishment of core research groups and core laboratories. When possible, we integrated faculty into existing basic science groups (e.g., cell biologists) or formed new ones to address a need raised by physician–scientists. For example, 10 years ago, no faculty focused on visceral pain, so core discipline leaders were recruited and, in cooperation with the Departments of Neurobiology and Anesthesiology and with the support of the dean, the Center for Pain Research was created.
Resources available to both clinical and basic scientists are also critical, and not just the traditional core research infrastructure and equipment. Administrative staff, data management, and other costs have been paid by incorporating components into the grants that will use specific resources, by funds from philanthropy, and by seed money support from UPMC. Accurate phenotyping by dedicated physicians and the ability to create “virtual cohorts” through a robust electronic medical record system and flexible data management have been essential to our success.
Finally, our division has a monthly faculty meeting to discuss routine business, but, more important, we continually remind the faculty of the goals, components, and interactions of our MAD approach and continually celebrate remarkable advances by the faculty. Research retreats, listservs, and newsletters also facilitate communication among such a large and diverse faculty.
Using the MAD grid to generate discovery
The 7 × 7 part of the grid shown in Chart 1 can also be viewed as an outline for phenotyping multiple systems in a single patient. The grid of information from a single patient becomes a matrix as thousands of patients are evaluated over the same features using standardized data collection tools and biomarkers linked to each organ as viewed by a COE and each scientific discipline. The matrix then becomes a hyper-matrix as measures relevant to the 49 phenotypic components of each patient are monitored over time, with further degrees of information based on response to treatment. Although the data and biospecimen management systems are beyond the scope of the current discussion, they are critical for functionality.
From a mechanistic discovery perspective, understanding all the elements of a complex disorder, which is the purpose of each COE, occurs when a series of physician–scientist translators can apply insights from their respective scientific fields in the greater disease model. Collecting the pieces of information related to normal and abnormal responses to stresses and interventions, adding detailed genetic and environmental information, organizing the data sets, and incorporating them into algorithms and predictive models constitute the role of computer scientists and biostatisticians. The proof comes from the application of the new knowledge in sequential patients with different variations of the disease.
Working Example: Pancreatitis
The Division of Gastroenterology, Hepatology and Nutrition has emerged as one of the leading groups with pancreas expertise in the world, which we believe is a direct result of our MAD organization. Our pancreas group has published hundreds of original reports and reviews over the past decade, with some of the major advances related to the modeling of a complex inflammatory disease: CP.
CP is one of the complex inflammatory diseases that challenges the germ theory. In 1896, Chiari5 suggested that the inflammation did not come from infection but, instead, resulted from autodigestion of the pancreas by pancreatic digestive enzymes that became active while they were still inside the pancreas. Since then, CP has been defined by the amount of scarring and other complications,6 with the assumption that the etiology was alcoholism because no other factor could be identified. However, most alcoholics do not develop CP, and most patients with CP report little alcohol use, even under intense interrogation. In 1995, and despite years of traditional investigation, a New England Journal of Medicine review concluded that “chronic pancreatitis remains an enigmatic process of uncertain pathogenesis, unpredictable clinical course, and unclear treatment.”7
Resolving the etiologic problem involved the study of two uncommon but striking genetic disorders that result in CP. The first is hereditary pancreatitis, which targets acinar cells, and cystic fibrosis, which targets duct cells. With the cells of disease origin defined, and examples of disease mechanism illustrated, we were able to begin teasing out the influence of other genetic and environmental factors that modify disease susceptibility, severity, and complications. The discovery process involved different scientific disciplines, and each was chosen based on the specific problem being studied.
Acinar cell pathophysiology in CP
Our first major breakthrough came from the study of some unusual cases. In 1996, gain-of-function mutations in the cationic trypsinogen gene (PRSS1) were identified in a study of large families as the cause of hereditary pancreatitis.8 This unusual form of pancreatitis begins with episodes of acute pancreatitis in childhood, CP in a subset of patients by early adulthood, and a 30% to 40% risk for pancreatic adenocarcinoma later in life.8 Studies of this rare condition revealed that trypsin was a key molecule in the pathogenesis of acute pancreatitis and CP.9,10 Evaluation of the molecular structure of trypsinogen (molecular modeling) in relation to the disease-causing genetic mutations (genetics) also linked calcium dysregulation with pancreatitis and localized the site of dysfunction to the acinar cells (cell biology). Because trypsin is primarily expressed in the pancreas, the discovery explained why this chronic inflammatory disorder was located in the pancreas (susceptibility and organ targeting). However, it did not explain disease severity, clinical course, or the likelihood of complications.
Duct cell pathophysiology in CP
Cystic fibrosis is an autosomal recessive disorder caused by severe mutations in the cystic fibrosis transmembrane conductance regulator gene (CFTR), which is an epithelial cell anion channel involved in fluid secretion and absorption. In the pancreas, CFTR is expressed in the proximal duct cells, not the acinar cells. Infants born with cystic fibrosis have abnormal sweat chloride, CP, and occasionally meconium ileus, with lung disease developing later.
In 1998, two genetics groups noted an excess of heterozygous severe CFTR mutation in CP patients.11,12 Although the explanation remained obscure, the CP phenotype was identical to alcoholic CP, hereditary CP, and idiopathic CP. This finding also demonstrated that initiation of injury could occur in the acinar cells (PRSS1 mutations or calcium dysregulation) or the duct (CFTR mutations in the duct cells or duct obstruction) and that, regardless of injury location, it led to the same chronic inflammatory phenotype. Thus, the mechanism of injury determined the organ (susceptibility), but the immune response defined the general phenotype (scarring) and determined various complications (organ dysfunction, pain, diabetes, cancer development).
Working model of CP
In 1999, we proposed the sentinel acute pancreatitis event model of CP9 involving a “two-hit” model in which the first (sentinel) episode of acute pancreatitis activates the immune system, whereas the second “hit” involves stressors that would not be sufficient to cause direct injury alone but can drive the inflammatory response toward fibrosis rather than healing.9,10 Potential stress factors that increased the risk of progressing to CP include alcohol use, smoking, metabolic stressors, and recurrent acute pancreatitis.
We used this conceptual framework of a mechanistic process to design the North American Pancreatitis Study 2 (NAPS2), the largest prospective, cross-sectional cohort study of recurrent acute pancreatitis and CP ever conducted in the United States (1,000 patients, 695 controls).13 Using broad inclusion criteria (two or more episodes of acute pancreatitis or any imaging evidence of pancreatic injury or scarring), we then classified patients according to the major known etiologic factors6 and family history. We quantified alcohol use and smoking throughout their lifetimes as environmental stressors. We defined specific, quantified end points (fibrosis, calcifications, maldigestion, pain patterns, diabetes mellitus, cancer), collected detailed data on clinical history and medication use and effectiveness, and carried out serial imaging studies as biomarkers of disease activity over time.
Surprising findings thus far include the discovery of a threshold amount of alcohol (≥60 grams of ethanol per day) that must be exceeded before CP risk increases above that for nondrinkers14; that smoking is a strong, independent risk factor for the development of CP; and that a combination of alcohol and smoking confers the greatest risk.14,15 Overall, alcohol abuse accounted for only about 25% of the incidence of disease, which forced us to reconsider how we look at this disease!
Insights into idiopathic CP
We quickly asked ourselves, “If most cases of CP are not caused by alcohol, as had been assumed for decades, what is triggering the process of fibrotic injury in the pancreas? ” The answer is the pancreatic secretory trypsin inhibitor (gene symbol SPINK1), an acute-phase protein produced in the pancreatic acinar cells that is up-regulated during inflammation as a feedback inhibitor of activated trypsin. In 2000, Witt et al16 found that, in children, homozygous SPINK1 N34S mutations were associated with pancreatitis. Also, we showed that heterozygous SPINK1 mutations were associated with early-onset CP.17 However, 2% to 3% of many populations carried the high-risk SPINK1 haplotype17 suggesting that SPINK1 variants acted as disease-severity modifiers in patients with recurrent trypsinogen activation within the pancreas.
We then envisioned two pathologic pathways.18 One links environmental stressors to trypsin activation, followed by failed SPINK1 feedback protection, and then immunocyte activation (macrophages and pancreatic stellate cells), to cause CP.18 The other pathway involves immunocyte activation and CP independent of trypsinogen activation and SPINK1 variants. Surprisingly, the frequency of SPINK1 mutations was significantly lower in alcoholic CP than in idiopathic CP, indicating that following activation of the immune system, alcohol could drive inflammation and fibrosis independent of trypsin activation. This explains why the risk of alcohol, smoking, and alcohol plus smoking is similar in CP and liver cirrhosis.14,18 On the other hand, it also indicated that idiopathic CP was often linked to recurrent trypsin activation.
Separately, we developed a mathematical model of how CFTR functions in pancreatic duct cells19 and predicted that mutations causing selective defects in bicarbonate secretion would lead to recurrent trypsin activation. We tested this hypothesis in subsets of the NAPS2 cohort and in a familial pancreatitis cohort and discovered that the common CFTR R75Q variant, which is not associated with cystic fibrosis, was strongly associated with CP, especially in the presence of SPINK1 mutations.20 Using site-directed mutagenesis and electrophysiology in a cell system, we proved the existence of a selective defect in CFTR bicarbonate conductance.20 This defined the phenotype of a new high-risk class of CFTR variants (proposed Class IVb) that do not cause cystic fibrosis but do cause CP.
Our findings have launched a revolution in how chronic inflammatory diseases of the pancreas are viewed. Figure 2 summarizes the current distribution of CP etiologies in the NAPS2 cohort, with alcohol use responsible for a minority of cases. Although the observed phenotype is the same, multiple etiologies are responsible for initiating the process in the pancreas, while common factors seem to drive the process. Thus, although the disease “looks” the same in the clinic setting, different therapeutic approaches are required, depending on the etiology. Providing an appropriate prescription for each patient represents our implementation of personalized medicine.
To better understand complex inflammatory diseases of the pancreas, we and others are designing carefully focused clinical trials that will target subsets of patients with genetically defined risk and biomarkers of disease activity. However, as demonstrated by our faculty grid (Chart 1), there is the potential to rapidly apply discoveries in the pancreas to parallel genetically or phenotypically defined complex inflammatory syndromes in the intestines and liver. This approach will allow small but well-powered studies to be designed by minimizing subject variance and more accurately defining the biomarkers and end points.21 More important, it provides a framework for rapid integration of new and complex data sets generated by next-generation sequencing, molecular discoveries, and “omic” results (i.e., genomics, proteomics, metabolomics). Finally, the future will involve better design and use of electronic medical records and mechanistic mathematical models capable of integrating multiple risk factors in multiple domains over time to provide accurate predictions about the cause of the condition, the state of the condition, and the future of the condition as modeled with or without potential interventions.
The author wishes to thank Jean Ferketish, PhD, Michelle Kienholz, and Christopher Langmead, PhD, for critical review of the manuscript and helpful suggestions, and Jessica LaRusch for developing Figure 2.
The studies described in this article were funded in part by grant number NIH R01 DK061451 (D.C.W.), the Frieda G. and Saul F. Shapira BRCA Cancer Research Program (D.C.W.), the Wayne Fusaro Pancreatic Cancer Research Fund (D.C.W.), and UL1 RR024153 from the National Center for Research Resources (NCRR), a component of the National Institutes of Health (NIH), and the NIH Roadmap for Medical Research.
Dr. Whitcomb has been a consultant for Abbott and Axcan pharmaceuticals, is a stockholder in Ambry Genetics, and has a patent for genetic testing of patients for hereditary pancreatitis.
The contents of this article are solely the responsibility of the author and do not necessarily represent the official view of the NCRR or the NIH.
This article is based on a lecture, “Bedside to bench,” by Dr. Whitcomb, given at the 51st Indian Society of Gastroenterology's annual meeting on November 21, 2010, Hyderabad, India.