Important Points Highlighted by the Speaker
• Genomics as a high-throughput, hypothesis-generating research method can complement traditional low-throughput, hypothesis-testing-based research to improve patient care.
• With demonstrated clinical utility, genomic medicine will, over time, be incorporated into clinical practice.
• Challenges for the field of large-scale genomic medicine include establishment of the infrastructure needed to generate, store, distribute, and interpret genomic data; increasing the sensitivity of sequencing; and assuring patient privacy.
• The study of rare diseases offers a way of implementing the tools and procedures that will later be used in more widespread applications of genomic medicine.
Leslie Biesecker of the National Human Genome Research Institute said that patients come into the health care system to get answers to three simple questions: What is wrong with me? What caused it? What can I do about it? Even if there are no known treatments for a disease or disorder, patients and physicians find significant value in having a diagnosis, Biesecker said. Genomic medicine can help shorten the diagnostic odyssey for patients, direct therapeutic intervention when available, and also build the knowledge base for developing treatments in the future.
Historically, Biesecker said, hypothesis testing in basic research, clinical research, and clinical practice research has relied on methods that produced relatively little data. This is changing, however. Biesecker said that in contrast to the low-throughput assays that have been employed to test a single hypothesis with a single assay and that are both expensive and time-consuming, genomics is emerging as a high-throughput, hypothesis-generating research1 paradigm that can address these limitations. As a hypothesis-generating research tool, genomics will be used to narrow the focus to answer research questions with more classical, hypothesis-testing validation experiments.
Biesecker described four clinical scenarios in which the hypothesisgenerating paradigm could be applied to improve patient care. First, a newborn’s genome could be tested for life-threatening congenital or metabolic disorders, and any genetic abnormality would be followed up with a specific test. Second, if an older patient presented with symptoms of a certain disorder, such as asthma, genomic data could be used to determine the disease subtype and to identify relevant pharmacogenomic information for treatment. Third, when two people are interested in conceiving a child, the genomes of both partners could be tested for recessive alleles for serious congenital disorders. Finally, during adulthood a patient may also be interested in learning about disease risk for particular cancers, such as breast or ovarian. These are lofty ideas about what a clinician could do with genomic data and a set of analytic tools to affect patient health, said Biesecker.
Altering the way that clinical medicine is conducted to include genomic analyses will take some work, Biesecker said. A major problem is that genomic analyses generate much more data than any clinician or patient can use, and clinicians are generally not likely to order tests that provide more data than they are seeking. Additionally, as more tests are performed, the likelihood of a false-positive signal increases. For example, many physicians resist doing a full panel of blood tests when only a specific test is needed. However, clinical laboratories routinely perform multiple tests when a single test is ordered because it is more cost-effective. If one of the non-requested tests on the resultant panel reaches a “panic” value, this is
1 Hypothesis-generating research is defined as an exploratory approach to science whereby an initial experiment, designed with a broad question in mind, produces a large amount of data or observations that can be used to formulate hypotheses that can be tested by directed experiments in the future.
reported to the clinician. Similarly, Beisecker said, a genetic “panic” value would indicate that action is needed.
Overcoming these obstacles is not a new challenge. Tandem mass spectrometry for newborn screening generates hundreds of peaks corresponding to metabolites, but much of this output can be filtered based on the analytes that are known to be useful. Clinicians and researchers always have the option of using data or not. Filtering or reserving genomic data for future use is not a radical change, Biesecker said.
Clinicians should not have to become geneticists, Biesecker said. Clinicians need to be able to use the majority of genomic data for routine clinical care and be able to recognize situations involving critical abnormalities that require the involvement of specialists, such as genetic counselors and clinical geneticists. But “most of the data in the genome should be eventually used by the general practicing physician, not by a clinical geneticist,” Biesecker said.
The clinician’s role will need to change from selecting tests based on clinical insight to integrating data from multiple sources in order to answer questions. Clinicians will have to adapt to large-scale, pre-differential testing. Clinicians will also become bioinformaticians. “This is a radical notion,” Biesecker said, “but if you think about it, our lives are becoming much more information intensive in everything that we do. We are much more inclined to gather large amounts of information, mostly through the Internet, to make decisions in our everyday lives. There is no reason to presuppose that clinicians cannot do the same thing. And once it becomes clear to the clinician that the tools can work, they—like us when we are shopping or trying to answer a question on the Internet—will use largescale information sources to answer those questions.”
Clinicians are by nature conservative in their practice, Biesecker said, but once they are shown the utility of a new approach, many will quickly change their practices. Still, changes of this magnitude will take several years to be put into practice widely and effectively.
As the clinical utility of genomic data increases, there will be practical considerations that need to be taken into account. Biesecker discussed four main areas that present significant challenges. First, the infrastructure necessary to generate, store, and distribute these kinds of large-scale data is not yet available. Second, although it may eventually be more cost-effective to regenerate genomic data when needed, at this point whole-genome analysis will probably be cost-effective only once a person’s genomic information is evaluated on four or five different occasions. “You can generate it once and re-use it in multiple occasions,” Biesecker said, “and the cost of that test is distributed over the lifetime of the patient.” Third, data need to be securely
stored and readily accessible, and patients need to be confident that their data will be used for their benefit. Biesecker discussed a system in which patients have access “keys” to their data so that the data cannot be shared without consent. (More patient privacy and other ethical and legal issues are discussed in Chapter 6.) Finally, the development of robust databases that correlate genetic variants with phenotypes will be a major challenge, Biesecker said. (This issue is discussed further in Chapter 8.)
Despite the challenges involved with bringing genomic medicine into the clinic, the tools to implement changes that will improve care with genomic data are available, Biesecker said. Using these tools to improve medicine will require gathering data, educating clinicians and patients, and developing the infrastructure. Biesecker said that additional clinical and translational research will be needed to move research results into the clinic and that information systems, such as analytic software, will need to become much more sophisticated and clinician-friendly. He added that clinical trials will need to compare the utility of sequence-driven algorithms to current practice. In particular, he said, the sensitivity of sequencing technologies needs to be improved. The sensitivity of current technologies ranges from 88 to 92 percent when examining all of the genes in the human genome; the sensitivity will need to exceed 95 percent to be useful for detecting specific diseases.
As a way of moving forward, Biesecker suggested using rare diseases as an entry point for genomic medicine. The many thousands of rare diseases and disorders result in a major expenditure of health care resources as patients go from provider to provider and test to test in what has been called the “diagnostic odyssey.” A system that could efficiently diagnose patients with rare diseases would provide an opportunity to build infrastructure and also would provide experience in how to handle incidental findings. Sequences that reveal rare metabolic disorders would produce many other findings of interest, which clinicians could then use in routine care. Specialists applying informatics tools to sequence information in this context could in turn teach generalists and make the tools more user-friendly.
As an example of an infrastructure-building project, Biesecker described a cohort study called ClinSeq, a large-scale pilot sequencing study involving about 1,000 participants that Biesecker and his colleagues began in 2006 to evaluate candidate genes associated with cardiovascular outcomes such as coronary artery calcification. The design of ClinSeq allowed for the
evaluation of pilot technologies, assessment of challenges, and exploration of how best to interact with patients over the course of the study. ClinSeq was based on the assumption that common diseases consist, in substantial measure, of individually rare phenotypes. The many thousands of functional genes in the human genome provide pathways to similar phenotypic endpoints that are grouped together and interpreted as distinct diseases. According to this hypothesis, the underlying heterogeneity of common diseases explains the variation routinely seen in the disease progression, severity, therapeutic response, and side effects of therapies. By dividing the molecular pathophysiology into finer groups, it should be possible to understand this variation and predict the phenotypes.
Although the initial goal of ClinSeq was to study atherosclerosis, the patients in the study were asked to consent only if they agreed to return for additional tests if data from the genome sequence raised particular questions. This made it possible to perform hypothesis-generating research instead of studying only those phenotypes identified a priori. As a result, Biesecker and colleagues were able to use the ClinSeq cohort data to test for the presence of a mutated gene that, in a separate study, had been determined to cause combined malonic and methylmalonic aciduria (CMAMMA), a rare recessive metabolic disorder. With the ClinSeq data, Biesecker said, he was able to identify and diagnose a patient in the ClinSeq cohort with a homozygous mutation for CMAMMA who did not present clinically with the neurological manifestations and normal vitamin B12 levels found in typical adult patients. However, upon examination of frozen blood and urine samples, the patient had the characteristic pattern of elevated methylmalonic acid and malonic acid without a vitamin B12 deficiency (Sloan et al., 2011). “We thought we understood the spectrum of that disorder,” he said, “but when you use a genomic approach, you dip into a database and you find a patient who has a phenotype that you didn’t predict even existed. You identify that phenotype, and that expands your understanding of the disorder.”
Biesecker also gave an example of how the ClinSeq study data aided in disease prevention. An analysis of the genome of a middle-aged man revealed a pathogenic mutation of BRCA2 (breast cancer 2, early onset), previously undiscovered in his family. In such a case, Biesecker said, it is more logical to test for breast and ovarian cancers proactively instead of waiting until multiple family members are affected by the disease. Thus, this is a situation in which the pilot study revealed important information that would not have ordinarily been discovered because the patient did not have classic risk factors of disease.
Biesecker said that he shies away from the idea that genomics is going to revolutionize medicine. “Revolutions aren’t terribly pleasant affairs to be involved with,” he said. “They are very disruptive. A lot of people get harmed by them, and they have manifold unanticipated consequences.” Instead, he said, he prefers to think of the changes going on today and anticipated for the future as evolutionary. “We take what works in medicine and evolve it toward [doing] what we already know how to do better.”