Genome sequencing in the clinic: the past, present, and future of genomic medicine

Abstract

Genomic sequencing has undergone massive expansion in the past 10 yr, from a rarely used research tool into an approach that has broad applications in a clinical setting. From rare disease to cancer, genomics is transforming our knowledge of biology. The transition from targeted gene sequencing, to whole exome sequencing, to whole genome sequencing has only been made possible due to rapid advancements in technologies and informatics that have plummeted the cost per base of DNA sequencing and analysis. The tools of genomics have resolved the etiology of disease for previously undiagnosable conditions, identified cancer driver gene variants, and have impacted the understanding of pathophysiology for many diseases. However, this expansion of use has also highlighted research’s current voids in knowledge. The lack of precise animal models for gene-to-function association, lack of tools for analysis of genomic structural changes, skew in populations used for genetic studies, publication biases, and the “Unknown Proteome” all contribute to voids needing filled for genomics to work in a fast-paced clinical setting. The future will hold the tools to fill in these voids, with new data sets and the continual development of new technologies allowing for expansion of genomic medicine, ushering in the days to come for precision medicine. In this review we highlight these and other points in hopes of advancing and guiding precision medicine into the future for optimal success.

THE PAST

Road to Personalized Medicine

Many books and reviews have highlighted past achievements in understanding what DNA and genes are, how to sequence this DNA, and finally into the genomes era of the late 1990s and early 2000s, all culminating into what is now the “Genomics Era for Medicine” (Fig. 1). From the rediscovery of Mendel defining modes of trait inheritance, the Morgan laboratory defining trait linkage and mutations, to the identification of DNA as the genetic material (5), genetics grew as a field of biology in the 1900s. The development of sequencing technologies in the mid to late 1970s (97) opened up the door to sequence the first genes (78) and genomes (96). The beginning years of sequencing for clinical impact were mostly based on targeted gene approaches, with power to resolve only common variants within disease patients. Among these first gene variants to disease associations, found using genetics, were the polymorphisms of beta-globin genes for sickle cell disease in 1978 (59), the F508- of CFTR in cystic fibrosis in 1989 (60), CAG repeat length in the HTT gene in Huntington’s disease in 1993 (71), and the 1990 discovery of BRCA1 variants for breast cancer risk (45). Shortly following the completion of the first human genome in the early 2000s (51, 114), projects were proposed to take genomics into the medical field for personalized healthcare, such as the Personal Genome Project (19), with hopes of identifying genetic events for disease etiology in a broad range of diseases. In 2005, publications started solidifying genomics into an understanding of disease at a genome level through the development of genome-wide association studies (GWAS), with the first published study addressing the genomics of age-related macular degeneration (62).

Fig. 1.

Timeline for the discovery of DNA/genes into the genomic era for medicine.

Population Genetics and GWAS

One of the initial GWAS studies contained 116,204 single nucleotide polymorphisms (SNPs) and only looked at 96 cases with 50 controls, at the time a ground-breaking paper in Science (62). GWAS continued to advance from that initial point, and several reviews of those first few years have previously been published (18, 115). As the cost of performing the genotyping assays of GWAS decreased, in combination with increased density of variants on the assays, larger cohorts were collected for other indications such as neurological (81), cardiovascular (2), metabolic (108), renal (63), hepatic (15), and reproductive (85). Studies have grown to include more than 100,000 individuals, such as the body mass index association studies in 249,796 European individuals (103) and the 126,559 individuals studied for educational attainment (94).

As of the writing this article, nearly 70,000 genetic associations have been curated from ~3,000 publications within the National Human Genome Research Institute-European Bioinformatics Institute GWAS database (116) consisting of ~2,500 separate publication-measured traits (https://www.ebi.ac.uk/gwas/, accessed on March 23, 2018). Following the initial growth phase of GWAS from 2005 to 2010, there has been a stable ~300 papers per year with quality GWAS data (Fig. 2A). These studies have revealed the most associations for traits such as asthma, body mass/obesity/height, schizophrenia, breast cancer, coronary artery disease, and Type 2 diabetes (Fig. 2B). Several chromosomes (chr) such as chr6 and chr19 return a higher than average phenotypic activity from GWAS relative to the chromosome size (Fig. 2C). Surprisingly the sex chr and mitochondrial genome are poorly associated with disease in GWAS, due not to biological association but to limited assessment within genome studies. The X-chromosome in males is highly associated with hemizygous disease risk and in females with random X-inactivation, resulting in ~50% of cells with hemizygous loss of function mutations (68, 90). In addition there is suggestions these sex chromosomes, particularly the Y-chromosome, have high phenotypic impact within high-throughput animal phenotyping studies in genetic crosses (88, 89).

Fig. 2.

Current statistics from the National Human Genome Research Institute-European Bioinformatics Institute (EBI/NHGRI) database. The EBI/NHGRI genome-wide association study (GWAS) catalog (https://www.ebi.ac.uk/gwas/docs/file-downloads, downloaded 3/23/2018) was assessed for publications per year (A), traits with the most associations (B), most associations per chromosome Mb (C), and associations per ethnicity (D).

As the GWAS database continues to increase in size, it contains an overwhelming number of associations for individuals of European ancestry, at an astonishing ~63% (42,571/67,857 as of March 2018) of all associations (Fig. 2D), far exceeding the actual global population composition. This is a result of many disease areas having predominantly larger European populations within the studies, therefore increasing statistical power for the one ethnicity (Fig. 3). Addressing this lack of diversity to ensure just distribution of benefits from genomic technologies is one of the leading ELSI (Ethical Legal and Social Implications of Genetics) concerns for the future of genomic research and clinical practice (20). This represents a new iteration of a long-existing problem in biomedical research, resulting in the National Institutes of Health (NIH) Revitalization Act of 1993 mandating inclusion of minorities in all NIH-funded research (16). In the past, not only minorities but women and children have been underrepresented in clinical trials, resulting in the need to incentivize inclusion of more representative populations through the 1997 Food and Drug Administration Modernization Act and the 2002 Best Pharmaceuticals for Children Act (74). In short, convenience and ready availability are threatening to leave some populations out as the beneficiaries of scientific advancement; and at least one policy working group found that skewed representation in which higher socioeconomic populations were overrepresented made assessing the risks of genomic testing difficult, as higher socioeconomic populations may be immune or less vulnerable to many common concerns about the risks of testing (such as employment or insurance concerns) (21).

Fig. 3.

Population structure in current GWAS studies. Each search term was queried against the GWAS studies on November 1, 2016 from the EBI/NHGRI GWAS catalog. Populations were extracted and plotted as box and whisker plots. Caucasian/European = cyan, African American = red, Hispanic = gray, South American = yellow, Middle Eastern = Blue, Asian = green, African = dark blue.

In the past few years, GWAS approaches have expanded into Epigenome-wide association studies (EWAS) (14, 91), an approach where the methylation of cytosines in the DNA are studied at a genome-wide level for disease association. These experiments have expanded to such a point that meta-analyses are now being performed on the generated data (69), much like those done for GWAS (117). Moreover, as medical health record information is more reliable to be data-mined and correlated to genomic information, phenome-wide association studies (PheWAS) (25) have been increasing in potential utility for genomic medicine (24, 48, 84). Even as the development and expansion of these GWAS, EWAS, and PheWAS tools increase and impact medicine, they still require extensive ability to segregate phenotypes, test causality of genotype-to-phenotype, and are underpowered for rare variants. Arguments can be made for both rare and common variants in disease etiology (39); thus, many of these genomic tools are underpowered for full clinical utility. Other genomic tools such as animal models of disease and the expansion of whole exome/genome sequencing are beginning to overcome these limitations.

Genotype to Phenotype: from Animal Models to Cells for Human Health

Human genetic studies have many limitations, mainly resulting from the difficulty in deciphering one variant among a landscape of thousands of other unique variants; therefore, connecting a genotype to a phenotype with high confidence is very difficult in humans. Important questions arise once a variant is identified, such as does the variant alter the function of the gene, does the functional change result in disease or another phenotype, and is that associated disease or phenotype relevant to the specific clinical condition in the tested individual? These questions are hard to answer with current scientific tools within the index subject. Thus, the dissection of genetic elements that contribute to traits and diseases has always needed models within other organisms. Very early in genetics, animal models served this very function, with the fruit fly (Drosophila melanogaster) serving as one of the first and most successful examples of genotype-to-phenotype relationships (105). This included groundbreaking work by the Morgan laboratory mapping trait inheritance (79), Sturtevant defining trait linkage mapping (106), Muller defining mutation due to environmental exposures (80), studies of homology between different genera (107), and finally having complete genomes from 12 of the Drosophila species (104). Yet the fly is still very distantly related to humans, with 7 chr composing ~149 Mb with 30,443 proteins (https://www.ncbi.nlm.nih.gov/genome/?term=txid7227[orgn]) compared with the humans’ 23 chr pairs (22 autosome pairs and 2 sex chromosomes) with ~2,991 Mb and 79,074 proteins (https://www.ncbi.nlm.nih.gov/genome/?term=homo+sapien). The rat and mouse genomes are more similar to humans and thus provide a more reliable testing model organism for genotype-to-phenotype relationships. While primate models are even more similar to human, they have disadvantages ranging from increased time of study, expense, and ethical concerns for performing experiments. For the rat we have provided a table of the major milestones for genetic understanding and community resources that have relevance to understanding genotype-phenotype relationships (Table 1), many of which can also be found within the mouse community as well.

Table 1.

Progress of genomics in the rat to create a model organism for studying human disease and available resources for the community

Data Type	Reference	Year	Accomplishment
Sequencing	(38)	2004	Brown Norway (BN) rat sequenced: combining bacterial artificial chromosome, end sequencing, whole-genome shot gun, BAC fingerprinting mapping
Sequencing	(44)	2008	copy number variation found in inbred and outbred strains: computational and array-based comparative analysis of hybridization signals
Genome modification	(37)	2009	first use of Zinc-finger nucleases to begin humanizing the rat strains
Sequencing	(3)	2010	spontaneously hypertensive rat (SHR) sequenced: Illumina paired-end
Sequencing	(100)	2012	genomic variation in 2 founders (SHR/Olapcv and BN-Lx/Cub) of recombinant inbred panel and liver RNA-Seq data, paired-end library, SOLiD and Illumina
Sequencing	(42)	2013	DA and F344 strains sequenced: Illumina paired-end read technology
Sequencing	(92)	2013	8 founders of the rat heterogeneous stock (46) sequenced: SOLiD
Sequencing	(4)	2013	comparative analysis of genome sequences for 25 inbred strains and substrains
Genome modification	(67)	2013	1st use of CRISPR-Cas9 in rats
Expression	(122)	2014	generation of RNA-Seq library from the F344 strain for 11 tissues at 4 time points, in both males and females
Sequencing	(7)	2014	sequencing of the SHR Y-chromosome known to be involved in blood pressure regulation (31): first Y-chromosome sequence added in the Rnor6.0 genome release
Sequencing	(49)	2015	comprehensive inventory of 40 sequenced rat strains and substrains
Sequencing and phenotyping	(89)	2016	sequence analysis and consomic phenotyping in multiple strains of rats for all nuclear chromosomes including the Y-chromosome for the 1st time
Resources	https://rgd.mcw.edu/		major source for rat genomic, genetic, phenotype, strain data, quantitative trait loci, genetic markers, correct nomenclature, functional annotations for rat, human, mouse derived data, pathway analysis (99)
Resources	http://www.ensembl.org/Rattus_norvegicus/index.html		Ensembl Rat Browser
Resources	https://rgd.mcw.edu/rgdweb/models/gerrc.html		source for models of gene knockout, knock-in, conditional expression, transgenic and Cre-recombinase in rat.
Resources	http://www.rrrc.us/		long-term preservation and distribution of rat models from USA
Resources	http://www.anim.med.kyoto-u.ac.jp/NBR/		long-term preservation and distribution of rat models from Japan
Resources	http://pga.mcw.edu		physiological characterization of 2 panels of consomic rat strains (SS×BN, FHH×BN) and multiple inbred strains for cardiovascular, pulmonary, renal, vascular, and blood systems

In the mouse (101) and rat (23), entire chromosomes (consomic) can be introgressed from strains of animals that have disease or particular phenotypes relative to “healthy” control animals, or vice versa, to determine the chr that contribute to disease. Combining the generation of every chromosome cross between two strains with the measuring of hundreds of phenotypes one can suggest chr with the highest impact on phenotypic traits (64, 73). For example, two complete rat consomic panels were developed by introgressing the Brown Norway (BN) to Fawn-Hooded Hypertensive (FHH) or the Salt Sensitive (SS) rat genome background. Surprisingly, per DNA base, the Y-chr contributes to a vastly elevated number of phenotypes relative to all other chr and also has twice the elevated rate of variation between strains (89). As the sex chr where already mentioned to be incredibly underpowered in GWAS (Fig. 2), animal models can have power to explain human phenotypes contributed by sex chr (88) that current techniques used in humans struggle to illuminate these roles of the sex chr. For example, animal models can be used in explaining the recently reported impact of the loss of Y-chromosome (LOY) as a major disease contributor (30, 35). To resolve the impact of specific regions of chr on phenotype, researchers utilize F1 crosses, heterogeneous stock breeding, and congenic mapping, all of which allow the association of regions/genes on chr with phenotype. This information can then be used in the human to map and test variants that may translate across organisms.

The development of technologies such as Zn fingers, TALENs, and now CRISPR/Cas9 has revolutionized our ability to characterize clinical variants found from whole genome sequencing in human (29, 33). As these technologies can be applied to nearly any species, particularly those that have had their genomes sequenced, it is possible to study a gene or variant in diverse species such as the fruit fly, Caenorhabditis elegans, zebrafish, mouse and rat. An easier approach, and what most groups have used to date, for characterization is either the deletion or insertion of an entire gene of interest. Yet these are based on all or none impacts, limiting the scope for potential gain of function or gene dosage contribution to disease. Utilizing these genome editing technologies, one of the growing approaches in animals is the generation of humanized species, such that the model organism is susceptible to human disease when it normally is not; therefore, opening the door for studies such as viral/bacterial infections and human tumor growth (26, 28, 32, 54). With more advanced humanized species, ethical concerns raised in the past from chimaera research return, and the NIH has only recently sought public feedback concerning plans to lift a moratorium forbidding federal funding, which is currently pending review by ethics panels (50, 58). More recently groups have begun to insert variants using donor sequence after cutting the genome with CRISPR/Cas9, thus generating single variants of interest. While these approaches are still difficult, they should continue to grow in the future and become more common practice for variant-to-phenotype associations.

While animal studies have significant scientific promise, they do result in multiple practical issues in addition to the ethical issues identified above. Primarily, their genomes are not the same as the human genome, and not all traits are shared between species. Therefore, many groups have moved into human cell culture characterization, using either isolated disease tissue from a patient, primary or stable cell lines, or generating organoids in culture. Moreover, human cells can be manipulated in many ways to define the biology of disease progression. Tumor biopsies can be sequenced and grown in culture, or they can be delivered to humanized models of mice and rats to study drug responses based on genetic variability of the specific tumor (43, 112). Germline cells can be obtained, sequenced, and then converted into inducible pluripotent stem cells (83, 109, 121). These cells can be manipulated with genome editing tools such as CRISPR/Cas9 (124) and then differentiated to many linages including neural (27), liver (102), fat (87), and heart (36, 125) to characterize phenotypic changes with the hope of potentially being used to treat disease. The question remains whether current research tools and costs can keep pace with the ever-growing number of human variants identified from clinical sequencing, or whether new methods are necessary.

Clinical Sequencing and Impact on Patient Treatment

Oncology and cancer research has long been at the forefront of using genomics in understanding disease etiology, including large-scale projects like COSMIC (34) and The Cancer Genome Atlas (9, 11). As mentioned earlier, sequencing initially was used for identifying common variants in conditions like cystic fibrosis, and GWAS was used to reach significance in large populations. With knowledge of the entire genome, including multiple ethnicities from the 1000 Genomes Project (1), new approaches [whole exome sequencing (WES)] were designed to sequence just those regions of the genome that code for proteins (~1% of the genome), opening the door for identifying potential pathogenic variants without spending the vast resources needed for the entire genome (6, 17). Some of the first large-scale sequencing projects, such as those done in cancer and the National Heart, Lung, and Blood Institute Gene Ontology (NHLBI GO) Exome Sequencing Project (111), began to highlight the full spectrum of variation using whole exome sequencing, showing a large number of rare variants that might influence disease. Around this same time, studies began to be published that showed discovery of rare diseases caused by rare variants (55). Among these first cases was a young boy whose life was saved through the identification of a C203Y variant in XIAP causing his inflammatory bowel disease. The variant was linked to a phenotype that primarily was an immunologic defect associated with a X-linked inhibitor of apoptosis deficiency. Based on this finding the clinicians caring for this patient applied a bone marrow transplant to prevent the development of life-threatening hemophagocytic lymphohistiocytosis. Here, WES not only provided the diagnosis but also suggested the previously unconsidered treatment option of a bone marrow transplant, bringing precision genomic medicine into the public spotlight for the first time (118).

With WES becoming more established in the clinic, the number of patients benefitting from these tools has rapidly grown (40, 120). It is estimated that the use of WES allows ~25% of rare disease cases to reach a molecular diagnosis (56, 119). With the robust yield and the continual decrease in cost of sequencing, many groups have begun the transition to whole genome sequencing (WGS) from exome (123). WGS can resolve protein-coding variants at a similar or higher rate as WES (72, 98) while also allowing for better coverage of the exons and identification of variants that might alter gene expression, gene splicing, and chromosome replication. With a limited knowledge of noncoding regions of the genome, investigators still interpret many variants with ambiguity, but with future research the role for noncoding regions may be better understood. Sequencing an entire genome just once allows for additional bioinformatics analysis to interpret future knowledge that whole exome does not provide, allowing for reassessment of variants instead of resequencing of the patient with new technology. Genome sequencing is still in its infancy, and identification of variants that are rare (such as n = 1 cases) does create issues that arise and warrant caution in reporting, with defined standards now becoming available (41). Constraints also exist as only some health insurance providers have started reimbursing the ordering institution for WES but few do for WGS.

THE PRESENT

Current Cost of Sequencing

As sequencing continues to become more routine (61), the larger numbers of sequenced individuals will further increase the utility and promise of genomics (57). In 2014, the threshold of a $1,000 cost for sequencing an entire whole genome (47), i.e., the raw cost to generate millions of short reads of ~100 bases for a genome, was crossed. Yet this is still a bit deceiving. Taking those 100 base reads and combining them into a genome by alignment to a reference and sorting through the hundreds of thousands of variants, many which are unique, means the actual cost depends on the depth of sequencing, the tools for computational analysis, and person time to analyze the data. All of this cannot be done for just $1,000. Sequencing a genome with lower depth, by decreasing the average of how many reads align to any site of the genome, reduces the cost but also decreases the confidence of variant interpretation, particularly for heterozygous variation. The choice of sequencing at low depth (~10×), average depth (30–40×), or high depth (~90×) should be made based on the circumstance to optimize cost/benefit.

While sequencing for most of the 2000s was performed as research grade, an increasing number of companies and academic institutes have begun offering certified sequencing through the College of American Pathologists (CAP) and Clinical Laboratory Improvement Amendments (CLIA) for clinical use. One of the biggest obstacles for genomic medicine to scale is the growing of sequencing infrastructure capacity to meet the clinical demand. Larger companies and academic institutes have expanded their sequencing capacities in the USA such that hundreds of thousands of genomes can be sequenced each year. The cost of bioinformatics for the growing amount of sequencing is also a hurdle, but as more genomes are completed and more advanced analytical pipelines are developed, these costs will decline. It should be noted that clinical-based sequencing must be ordered by a qualified clinician, but as the population continues to accept and embrace these approaches, there will be pressure on other physicians to learn and adopt these genomic tools to make them more common practice in patient care.

Assembly of Large Genomic Data Sets for Disease and Controls

With the decrease in sequencing cost, many individuals have now been sequenced, mostly at the research level. Large-scale projects have begun to publish these data sets, giving researchers a larger knowledge of variation within human populations. The genome Aggregation Database (gnomAD), updated in 2017, has grown to contain 123,136 exomes and 15,496 genomes from multiple ethnicities (66), providing a searchable format for genes and variants within each population (http://gnomad.broadinstitute.org/). NIH has invested heavily in the past few years in sequencing, moving projects like Trans-Omics for Precision Medicine (TopMed) into databases with >100,000 whole genome-sequenced individuals, many with detailed phenotype and clinical records (10). The Human Longevity, Inc. database (HLI) has made its first 10,545 30–40× whole genomes sequenced publicly available through the Open Search By Human Longevity, Inc. (110). With companies such as HLI announcing they intend to generate one million genomes or more, it is uncertain how much of this information will be released to the public for research in the future vs. being held within private siloes. The larger the public genomic data sets grow, especially with integration of other data, the more power will be available to interpret each individual genome. Sharing of information is vital for precision genomics medicine to reach its full utility.

The generation of genomic data also brings with it many ethical and legal questions. Unlike most traditional medical interventions, genomic testing often has direct implications for individuals and populations beyond the individual tested (75). Recent controversies raised in genomic research involving the Havasupai Tribe in Arizona illustrate the myriad indirect effects of such research, including potential stigmatization and threats to sacred cultural values and beliefs (77). Significant thought must be given to potentially unintended consequences that result from testing and the proper scope of information that should be obtained (Should we attempt to identify “the gay gene”? What sex- or ethnicity-correlated genetic dispositions might be useful or harmful? Can genetics be used for designer babies or preimplantation screening? Do we create stratifications in costs for health care based on the genetics of ethnicity? Etc.) (76).

Large-scale clinical sequencing projects are also extending to pharmacogenomics, with companies such as AstraZeneca launching a 2 million genomes project for their own clinical trials. They aim to incorporate genomics into clinical trials and revolutionize pharmacogenomics, with many additional companies and studies following in their footsteps. While the potential benefit of these efforts are great, they do come with additional ethical considerations that must not be ignored, particularly issues of patient data protection, oversight of pharmaceutical companies use/patenting/marketing of the data, stratification of groups, and differences in costs for treatments based on genomic data (8, 53, 70).

From Variant to Function

As the number of genomes continue to rise, interpreting the millions of variants that will be found for potential clinical impact becomes even more essential. In Fig. 4, we provide a sample workflow (our group utilizes) for analysis of variants, particularly for variants that appear promising but have not been associated with the disease of interest. Taking known disease causing variants from databases such as ClinVar (65) and filtering against a variant list from any genome is relatively simple, allowing a clear connection from variant to disease. However, when variants are rare or fall within a gene previously not identified with disease, additional tools are needed. For rare diseases, trio sequencing (i.e., the sequencing of mom and dad along with child/proband) can be used to resolve variants that are found in an affected child. For tumors, normal tissue and the tumor can both be sequenced to identify the somatic mutations that may give rise to cancers. These analyses, in combination with the level of conservation of a site throughout vertebrate evolution, can help filter impactful variants. Tools created to combine many aspects of evolution and protein changes can precompute every single possible variant that could be found, therefore decreasing computational costs when analyzing a single genome (52). Yet frequently these approaches identify genes not previously established for disease. These variants are referred to as variants of uncertain significance (VUS) or within genes of uncertain disease significance (GUDS) and are not clinically actionable (93). With additional experimentation, however, these variants can be reclassified as likely pathogenic (93). One approach is to identify additional patients with similar phenotypes and nominated genes using tools of the Global Alliance for Genomics and Health (GA4GH), such as Matchmaker Exchange (86), connecting individuals across the globe that share variation within genes. Tools such as MyGene2 (https://www.mygene2.org) even expand this idea directly into the patient’s hands, connecting researchers, clinicians, and patients on the basis of genes and phenotypes. Identifying a large number of similar patients combined with some basic biochemical, molecular, and animal experimentation can further resolve a variant’s role in disease etiology (Fig. 4).

Fig. 4.

Sample workflow for clinical sequencing and interpretation of variants.

To speed up these analyses, detailed knowledge of every gene within the genome provides tools to interpret variants with higher resolution at a faster pace; however, we are currently far from understanding every gene. For example, querying annotated publications for the 20,120 high-confidence proteins or 2,030 transcription factors from the UniProt database (Fig. 5) shows a clear skew to our knowledge of proteins throughout the genome. A few proteins represent a very large amount of the curated work within UniProt. Around 15% of human proteins currently have 0 publications curated, which we refer to as the “Unknown Proteome.” Data sets such as the Human Protein Atlas (113) show the expression profile for the genes of the Unknown Proteome is 8% ubiquitously expressed and 17% tissue specific (primarily from the testis). Of the ubiquitously expressed genes, most of these are found in a wide range of vertebrate species, suggesting they greatly affect cell function and are highly conserved, yet we have minimal knowledge about their biology. This highlights the fact that many variants falling within the poorly defined regions of the genome will be missed from current bioinformatics, not because of limitations of bioinformatics but because of limited studies of these genes.

Fig. 5.

Curated protein publication bias with the UniProt database. A: number of genes binned into various number of publications as a percent of total publications. Shown in black are the values for all 20,120 validated human proteins of UniProt and in red are the values for 2,030 proteins annotated in UniProt as transcription factors. B: the same data from A but plotted as the total number of publications represented by each group.

While projects such as the Encyclopedia of DNA Elements (ENCODE) (31a) and RoadMap Epigenomics (95) have begun elucidating noncoding regions within the genome, significant progress must still be made. It is likely that many of the disease-causing variants remaining unidentified fall within these noncoding regions. From GWAS, it is suggested that the majority of variants identified for trait association are noncoding and likely expression quantitative trait loci (eQTLs) (82), meaning the variant does not alter the function of the protein but more likely changes the expression of the protein. GWAS has not resolved the causal variant for the majority of identified loci, so additional tools and detailed analysis of GWAS mechanism to disease will be needed to address rare variants. In addition, the field of EvoDevo hypothesizes that speciation events often happen through changes in gene regulation timing through alteration in the noncoding DNA (12, 13). Therefore, evolutionary conservation is also not able to resolve many potential impactful noncoding variants within the genome. Future tools are needed to resolve the role of noncoding variants at a cellular level if we want to fully understand all the genetic contribution to disease development and progression.

THE FUTURE

Where Can Sequencing Go?

As these large-scale projects that involve millions of genomes return results, we expect to gain a clearer view of what regions within the genome have variation and those that do not, including noncoding control regions. As we perform cell-level ENCODE-style experiments at higher resolution, we can also correlate these conserved sites with functional enhancers and promoters. With pharmaceutical companies analyzing large numbers of WES, we expect pharmacogenomics to expand continually. With the number of genomes being sequenced ever growing, and with integration of other data sets such as medical health records, imaging, expression profiles, and mobile health data (such as Fitbit), the field is moving into the big data age. Each individual could be represented by terabytes of information sitting in datacenters, which need to be secured for patient protection while being available for analysis for correlations among other patients by research centers. This level of data will require massive datacenters with fast data transfer speeds as well as high security to protect private patient information. Research databases will continue increasing in size, such that maintaining NIH-funded data sets in multiple locations will not be feasible, but cloud storage and computing must increase to disseminate the information into the hands of many scientists. If social media data can be leveraged in marketing and even politics, the power of medical and genomic information can be seen as even more of a risk. The genome by its nature is difficult to deidentify due to its uniqueness, making it more difficult to keep genotype-to-phenotype associations stored within datacenters and necessitating a delicate approach as genomic medicine advances. Technologies will continue advancing in the coming years as well, with technologies moving the cost of whole genome sequencing potentially to a few hundred dollars with shorter turn-around time, opening the possibility for a human to be routinely sequenced or even building multicellular human genomes, yielding even more data per individual.

Even as WGS becomes the standard method for researchers in the USA, GWAS studies need to continue around the world, particularly in areas such as Africa, Asia and South America where populations are underrepresented and understudied. From functional studies, many groups will continue to break down the linkage disequilibrium block of GWAS to resolve the functional units and define transcription factor networks involved in disease progression, which may unlock new tools for deciphering the role of rare variants in the noncoding portion of the genome. To decrease the time needed for classification of every protein-coding variant associated with disease, we need to develop fuller databases for protein knowledge. A larger focus and more resources need to be applied to characterizing poorly studied proteins if we want to understand the importance of clinical variants anywhere within the genome at a speed compatible with real-time clinical care.

Potential Issues and Conclusions for the Future

Moving genomic medicine into the science of individuals, known as n of 1 science, we must make clearer associations of variants that are either common or rare with changes and mechanisms of biology. As the databases of sequenced genomes continue to grow, we will better understand the dynamics of the human genome, the acceptability of variation at any one site within it, and the degree of structural changes in the genome. Yet resources need to be put into the functional outcome of genetic variants, developing cheaper and quicker tools for variant characterization. With continual identification of VUS and GUDS, time and money will be needed to test and reclassify variants. Much of the future for genomic medicine relies more on phenotype associations and knowledge of genes and the proteins those genes code and less on further advancements of sequencing technologies. As we try to understand the functional effects of the genome, a better understanding of the interactions with the metabolome and microbiome will further those goals in humans. The use of animal models, cell culture, microdevices mimicking organs, and even basic biochemistry of proteins will facilitate these advances. We must begin addressing our knowledge of proteins in the genome, starting with elimination of the Unknown Proteome. We need to start paying more attention to the sex chromosomes (X and Y) and also the mitochondria for phenotype association in humans, as animal models have suggested these regions to be drivers for phenotypic diversity. The role of noncoding variants need to be moved out of standard cell lines and human tissues and into individual cells of the human tissues to understand normal cellular control and the diversity of cell types within any human tissue. Any one variant may not act alone, further complicating our understanding of biology; therefore, we must also consider how to model and test the role of epistasis and gene modifiers within the genome. Above all, for genomic medicine to become a major factor in the future, we must remember that genomics, as powerful as it is, cannot explain all disease risk, including those risks that are passed from generation to generation. Lifestyle choices, environmental exposures, the microbiome, and cultural upbringing need to be merged with genetics to create a full view of diversity of human biology. These integrations will move our genomic understanding to phenotypic outcomes through precision medicine and revolutionize health care.

GRANTS

Supported by National Institute of Environmental Health Sciences Grant K01ES-025435 (to J. W. Prokop).

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

J.W.P. analyzed data; J.W.P. and J.L. prepared figures; J.W.P., T.M., K.S., S.M.B., C.B., S.R., E.A.W., and J.L. drafted manuscript; J.W.P., T.M., K.S., S.M.B., C.B., S.R., E.A.W., and J.L. edited and revised manuscript; J.W.P., T.M., K.S., S.M.B., C.B., S.R., E.A.W., and J.L. approved final version of manuscript.