The Personal Genome Project Canada: findings from whole genome sequences of the inaugural 56 participants

Abstract

BACKGROUND:

The Personal Genome Project Canada is a comprehensive public data resource that integrates whole genome sequencing data and health information. We describe genomic variation identified in the initial recruitment cohort of 56 volunteers.

METHODS:

Volunteers were screened for eligibility and provided informed consent for open data sharing. Using blood DNA, we performed whole genome sequencing and identified all possible classes of DNA variants. A genetic counsellor explained the implication of the results to each participant.

RESULTS:

Whole genome sequencing of the first 56 participants identified 207 662 805 sequence variants and 27 494 copy number variations. We analyzed a prioritized disease-associated data set (n = 1606 variants) according to standardized guidelines, and interpreted 19 variants in 14 participants (25%) as having obvious health implications. Six of these variants (e.g., in BRCA1 or mosaic loss of an X chromosome) were pathogenic or likely pathogenic. Seven were risk factors for cancer, cardiovascular or neurobehavioural conditions. Four other variants — associated with cancer, cardiac or neurodegenerative phenotypes — remained of uncertain significance because of discrepancies among databases. We also identified a large structural chromosome aberration and a likely pathogenic mitochondrial variant. There were 172 recessive disease alleles (e.g., 5 individuals carried mutations for cystic fibrosis). Pharmacogenomics analyses revealed another 3.9 potentially relevant genotypes per individual.

INTERPRETATION:

Our analyses identified a spectrum of genetic variants with potential health impact in 25% of participants. When also considering recessive alleles and variants with potential pharmacologic relevance, all 56 participants had medically relevant findings. Although access is mostly limited to research, whole genome sequencing can provide specific and novel information with the potential of major impact for health care.

Rapid technological advances are enabling a view of human genetic variation in ever-increasing detail and at plummeting costs.¹ Until recently, analysis has been targeted largely to defined genes, but pan-genomic approaches, such as microarrays, gene-panel testing and exome sequencing, have become mainstream. Now, whole genome sequencing can capture all of the genes (about 1% of the whole genome) and most of the rest of the genome in a single experiment, with the potential to recognize all types of genetic variation and thereby usurp the less comprehensive technologies (Box 1).² Information from whole genome sequencing can already identify the molecular causes of suspected heritable conditions and cancer;^2–7 however, we anticipate that genomic analysis will become a standard component of proactive health care, given its potential to identify predisposition to medically actionable conditions, explain uncharacterized disease and reveal carriers for recessive disorders and predictors of medication safety and response.⁸ Interpretation of sequence data remains challenging, with unknown clinical utility and predictive value among the general population.⁹

Box 1: Human genome variation.

The genome is the complete set of genetic material (DNA), contained in the cell’s nucleus and mitochondria. Genes are functional units that instruct the cell to produce specific proteins. They are segmented into exons (coding units) and introns (noncoding spacers), with regulatory sequences at either end and at intron/exon junctions. Noncoding DNA between genes includes various regulatory or structural elements but is largely uncharacterized. Each of 2 versions of a gene (1 maternal and 1 paternal) is called an allele. The Human Genome Project provided the initial draft reference DNA sequence (23 pairs of chromosomes encompassing about 25 000 genes) against which to compare future genome sequences. Despite much similarity, each person’s genome is unique — from variations in the DNA sequence, copy number of genes, its organization and epigenetic changes. Some variations may be inconsequential, contribute to the differences among healthy humans or provide protection against environmental challenges; others have health-related consequences. Genome interpretation involves distinguishing among these. Variant alleles may be null, missense, nonsense, splice variants, deleted, duplicated, disrupted, etc., depending on their effect on the related gene products. Their impact on characteristics of the individual (the phenotype) are described as recessive, semidominant, codominant or dominant. Some traits or diseases result from single-gene variants, with outcomes that are predictable using principles of classical Mendelian genetics. Most involve much more complex interactions among gene variations, with epigenetic and environmental influences. Risk alleles are found more often among people with a particular condition than among those without. Few alleles are deterministic; most have variable expression. Penetrance reflects the proportion of individuals with a particular underlying genetic variant who display a given trait. Mosaicism occurs when a variant arises postfertilization, so that not all cells in the individual have it. Similarly, mitochondrial genomes in each cell may not all be identical, and a variant in only a subset is called heteroplasmy. The size of genetic variants can range from 1 nucleotide pair (bp), into the thousands (kb) or millions (Mb). Canada’s Genetic Non-Discrimination Act S.C. 2017, c.3, which received royal assent on May 4, 2017, prohibits anyone from requiring individuals to undergo a genetic test or disclose the results of a genetic test.

The Personal Genome Project Canada was launched in 2007, and shares the guiding principles and open consent policy of the parent project in the United States.¹⁰ It aims to develop a public data set of fully annotated genomic information, connected with human trait information. It can provide control data for other studies, but it also aims to forecast effects of integrating DNA-derived knowledge into routine clinical practice. The project will evaluate the utility of such information, and how best to gather and apply it within Canada’s provincially administered, publicly funded health care system. Participants in this ongoing project are highly motivated to promote genomic research and explicitly forego privacy commitments. We report the data and experiences from whole genome sequencing and medical annotation of genomes of the first 56 participants in the Personal Genome Project Canada.

Methods

Study participants

Information about the Personal Genome Project Canada was posted online (www.personalgenomes.ca) and disseminated through newspaper articles, by word-of-mouth and through Medcan Health Management Inc. Registered volunteers from across Canada underwent an in-person (n = 54) or phone (n = 2) interview and entrance exam (Figure 1), to ensure that they were aware of the potential risks associated with participation and that research results should not substitute for clinical diagnostic testing. To enrol in the project, participants must be over the age of 18 and state their intention to share their genomic data publicly. Self-reported baseline trait data included birth month/year, medications, allergies, vaccines, personal medical history, ethnicity/ancestry, blood pressure, height and weight. We did not exclude individuals based on known health conditions. Blood was drawn at the Medcan clinic (n = 54) or at a community laboratory (n = 2). Participation in the project is an ongoing process, both for the participants described here and for additional volunteers.

Figure 1:

Personal Genome Project Canada (PGPC) workflow. Potential participants in the project must meet eligibility criteria and undergo an entrance examination. Consent is sought and reaffirmed at stages throughout the process. Research ethics board protocols and consents, and genome data files are available at www.personalgenomes.ca. For adjunct analyses, we also assayed 55 samples using high-resolution microarrays (Affymetrix Cytoscan-HD) (Appendix 1), and we generated induced pluripotent stem cell lines for 3 individuals (Figure S4, Appendix 1).

Data generation and interpretation

We used the Illumina HiSeq X system to sequence DNA extracted from whole blood (median sequence depth of 38× across all 56 samples (Table S4, Appendix 1, available at www.cmaj.ca/lookup/suppl/doi:10.1503/cmaj.171151/-/DC1).

We compared each genome to the Genome Reference Consortium (https://ncbi.nlm.nih.gov/grc) human reference sequence (GRCh37/hg19). Aiming for variants with substantial health impact, we gave first consideration to single nucleotide variants (SNVs) and small insertion/deletion variants (indels) (Figure 2) that are rare (frequencies < 5%) in control cohorts (Supplementary methods, Appendix 1). Preliminary reports described alterations of genes listed in the Clinical Genomic Database (https://research.nhgri.nih.gov/CGD/) where the variant would likely eliminate gene function, and others reported to be disease associated by the Human Gene Mutation Database or ClinVar (Supplementary methods, Appendix 1).^2–4,11,12 We returned these reports to participants and offered a genetic counselling session to contextualize the information.

Figure 2:

Analysis and interpretation of whole genome sequencing data. We analyzed data from WGS for variants in the nuclear and mitochondrial genome: single nucleotide variants (SNVs; alternate single bases), insertion/deletions (indels; small segments of DNA that are missing or replicated), structural variants (SVs; variations involving larger segments), including copy number variants (CNVs; deletions/losses or duplications/gains), as well as other rearrangements (inversions or translocations). We also analyzed the data for 391 variants in 14 pharmacogenes (Table S2, Appendix 3). *Two variants (in MUTYH and PCDH15) were both recessive pathogenic and dominant risk factors. Note: B = benign, CGD = Clinical Genomic Database, HGMD = Human Gene Mutation Database, P = pathogenic, LB = likely benign, LP = likely pathogenic, RF = risk factor, VUS = variant of uncertain significance, WGS = whole genome sequencing.

Data were also analyzed for copy number and other structural variations (copy number variants [CNVs]/structural variants [SVs]) by comparison to the Database of Genomic Variants,¹³ and for variants in mitochondrial DNA. We also extracted information on 14 pharmacogenes from the whole genome sequencing data (Appendix 1), based on guidelines by the Clinical Pharmacogenetics Implementation Consortium, Dutch Pharmacogenetics Working Group and Canadian Pharmacogenomics Network for Drug Safety, and US Food and Drug Administration label recommendations.^14,15 To gain further insight into the spectrum of genomic variation, we assessed the disease-causing potential of all disease-associated variants in accordance with guidelines of the American College of Medical Genetics and Genomics. Variants were sorted into categories of standard terminology: “benign,” “likely benign,” “variant of uncertain significance” (VUS), “likely pathogenic” or “pathogenic,”¹⁶ by applying specified information from the published literature and various disease- and population-based databases. We expanded the American College of Medical Genetics and Genomics classification to include rare “risk factors,” and predicted how the putative disease-causing variants would influence the health management of the participant. Using a protocol from The Hospital for Sick Children’s Genome Clinic,^17,18 a team of researchers, health care professionals and clinicians reviewed the available information and provided input on ambiguous observations, to reach consensus for interpretation. If consensus could not be obtained or the supporting evidence was not sufficient, we designated variants as being of uncertain significance. Those findings that were deemed relevant to health were discussed with participants by a genetic counsellor.

Ethics approval

The study was approved by the Research Ethics Board at The Hospital for Sick Children (REB no. 1000053640). Acknowledging that publication of genomic data is associated with both anticipated and unforeseeable risks, such as the potential for re-identification, consent is sought and reaffirmed at stages throughout the process to ensure that participants have the necessary time and information to make informed decisions about their involvement in the project (Figure 1).

Results

Personal genome sequencing and medical annotation

We report on the genome analyses of the first 56 consecutive participants (Table 1). Whole genome sequencing found an average per participant of 3.7 million high-quality SNVs and indels (1198 rare coding) and 491 CNVs (2.3 rare coding) (Table 2). We assessed all rare genomic variations (SNVs, indels, CNVs and SVs) that were either predicted to eliminate the function of genes listed in the Clinical Genomic Database or reported to be disease-associated by the Human Gene Mutation Database or ClinVar (Figure 2). These included 1606 variants: 1591 SNVs or indels, 13 deletions of 2 kb to 1.9 Mb, 1 mosaic X-chromosome loss and 1 large paracentric inversion. Most were interpreted as VUS (919/1606, 57.2%) or benign/likely benign (503/1606, 31.3%) according to the guideline from the American College of Medical Genetics and Genomics. We classified 175 SNVs/indels and 9 CNVs as pathogenic/likely pathogenic or risk factors (Table 2; Table S1, Appendix 2, available at www.cmaj.ca/lookup/suppl/doi:10.1503/cmaj.171151/-/DC1). These represented an average of 3.3 disease-associated alleles per individual (range 0–8), of which most (172; 3.1 per individual) were associated with autosomal recessive or semidominant inheritance. For example, 5 participants carried single-copy pathogenic variants in CFTR (the gene for cystic fibrosis). We found 94.6% (53/56) of participants to be carriers of at least 1 single-copy pathogenic allele.

Table 1:

Demographic characteristics of enrolment and participants in the Personal Genome Project Canada

Characteristic	No. of participants†
Potential registered*	1123
Enrolled in study*	63
Included participants
This WGS study	56
Median age at enrolment (range), yr	51 (25–81)
Female:male	25:31
Common health conditions
History of cancer	7
Cardiovascular disease	7
Neuropsychiatric disorder	11
Self-reported ancestry
European	51
Middle Eastern	3
Canadian Indigenous	1
East Asian	1

Note: WGS = whole genome sequencing. Potential participants registered online (www.personalgenomes.ca). Enrolment was a multistep process (Figure 1). Ancestry clustering data are available in Appendix 1 (Figure S1, S2).

^*Reference date: Sept. 1, 2017.

^†Unless specified otherwise.

Table 2:

Nuclear and mitochondrial results of whole genome sequencing analyses

Type	No. of variants, mean ± SD	No. of variants, median (range)	No. of participants	Sum (1–56)	Size range (kb)
Sequence-level variants (SNVs, indels)
All	3 708 264 ± 366 708	3 851 444 (2 491 596–4 102 372)	56	207 662 805
Rare* coding	1198 ± 115	1188 (939–1457)	56	67 101
Rare potentially disease-associated†
All	28.41 ± 8.45	29.5 (11–43)	56	1591
Autosomal dominant	6.77 ± 3.24	6 (1–15)	56	379
Semidominant	6.79 ± 3.35	7 (0–14)	55	380
Autosomal recessive	13.68 ± 4.71	14 (4–25)	56	766
X-linked	0.79 ± 1.06	1 (0–6)	30	44
Mitochondrial‡	0.39 ± 0.68	0 (0–3)	17	22
Rare (likely) pathogenic§ or risk factor
All	3.04 ± 1.68	3 (0–8)	53	175
Autosomal dominant	0.05 ± 0.30	0 (0–2)	2	3
Semidominant	0.20 ± 0.48	0 (0–2)	9	11
Autosomal recessive	2.77 ± 1.63	2.5 (0–8)	52	155§§
X-linked	0 ± 0	0 (0)	0	0
Mitochondrial	0.02 ± 0.13	0 (0–1)	1	1
Risk	0.11 ± 0.31	0 (0–1)	6	6§§
Copy number variants and other structural variants
All CNVs
Deletion	431 ± 45.90	438 (275–521)	56	24 140	1–1919 (+ mosaic X loss)
Duplication	60 ± 4.80	60 (40–73)	56	3354	3–863
Rare¶ coding** CNV
Deletion	1.23 ± 0.96	1 (0–4)	41	69	2–1919 (+ mosaic X loss)
Duplication	1.09 ± 0.93	1 (0–4)	42	61	7–863
Rare coding CNV, covering CGD genes
Deletion	0.25 ± 0.51	0 (0–2)	13	14	2–1919 (+ mosaic X loss)
Duplication	0.29 ± 0.45	0 (0–1)	16	16	15–863
Rare (likely) pathogenic§ or risk factor CNV
Deletion	0.16 ± 0.42	0 (0–2)	8	9¶¶	4–1919 (+ mosaic X loss)
Duplication	0 ± 0	0 (0)	0	0	0
Rare inversion	0.02 ± 0.13	0 (0–1)	1	1	8630
Clinically relevant pharmacogenomics diplotype††
All	3.86 ± 1.09	4 (1–6)	56	216
Serious adverse drug reaction‡‡	0.23 ± 0.43	0 (0–1)	13	13

Note: ACMG = American College of Medical Genetics and Genomics, CGD = Clinical Genomic Database, CNV = copy number variant, HGMD = Human Gene Mutation Database, kb = kilobase pairs, SD = standard deviation, SNV = single nucleotide variant, WGS = whole genome sequencing.

^*Rare sequence level variants are defined as those with allele frequencies < 5% in control databases (Supplementary methods in Appendix 1).

^†Variants in CGD genes, either predicted null alleles or disease-associated in HGMD or ClinVar (Supplementary methods in Appendix 1).

^‡All mitochondrial variants detected at > 5% heteroplasmy.

^§Interpretation according to the ACMG guideline.¹⁶

^¶Rare CNVs are defined as those with < 50% overlap with all gold standard variants in the Database of Genomic Variants (http://dgv.tcag.ca/) and occurring at a frequency of < 1% among all WGS-CNVs in the unrelated parents in the Autism Speaks MSSNG cohort (http://www.mss.ng/).

^**Coding CNVs are defined as those overlapping coding exons.

^††Haplotype pairs on homologous chromosomes that are associated with risk for altered drug efficacy or adverse drug reactions.

^‡‡HLA-A*3101 and HLA-B*5701 associated hypersensitivity reactions; TPMT- poor/intermediate metabolizers with myelotoxicity risk.

^§§MUTYH variant was interpreted both as likely pathogenic (autosomal recessive) and a risk factor for colorectal cancer.

^¶¶One likely pathogenic deletion (25 kb) was identified by microarray only; 3 likely pathogenic deletions (4–7 kb) were identified by WGS only.¹⁹

Based on expert consensus, we considered 19 variants in 14 of 56 participants (25.0%) to have overt health implications (Table 3). Six variants in 5 participants were pathogenic or likely pathogenic (4 SNVs, 1 CNV). In a 65-year-old man, we identified a pathogenic BRCA1 variant, which is reportable according to the recommendations by the American College of Medical Genetics and Genomics.⁵⁰ The participant was of Ashkenazi Jewish descent, a population with higher frequencies of pathogenic BRCA1 and BRCA2 variants compared with the general population. His father had died of prostate cancer in his 70s, but the participant had limited knowledge of the medical history of his extended family.

Table 3:

Rare variants with potential health impact identified in the study, by participant ID no.

Participant ID no.	Gene (accession no.)	Variant, zygosity	Associated disease, inheritance	Interpretation: evidence*	Clinical follow-up	Management implications
PGPC-43	BRCA1‡ (NM_007300.3)	c.68_69delAG, p.(Glu23Valfs*17), het	Breast and ovarian cancer, AD	Pathogenic: PVS1, PS3, PS4, PP520	FHx of prostate cancer (70–80 yr)	1, 2, 3
PGPC-16†	ELN (NM_001278914.1)	c.455–1G > A, p.?, het	Supravalvular aortic stenosis, AD	Likely pathogenic: PVS1, PM221	No relevant PHx/FHx, normal heart CT	1, 2, 3
PGPC-25†	LMNA§ (NM_170707.3)	c.1748C > T, p.(Ser583Leu), het	Lipodystrophy, familial partial, AD/AR	Likely pathogenic: PM1, PM2, PP1, PP322	PHx of hypercholesterolemia, hyperpigmented patch	1, 2, 3
PGPC-16†	LZTR1 (NM_006767.3)	c.774delT, p.(Phe258Leufs*93), het	Schwannomatosis, AD	Likely pathogenic: PVS1, PM223	No relevant PHx/FHx	1, 2, 3
PGPC-40†	SLC7A9 (NM_001243036.1)	c.614dupA, p.(Asn206Glufs*3), het	Cystinuria, AD/AR	Pathogenic: PVS1, PS4-M, PP524,25	No relevant PHx/FHx	1, 2, 3
PGPC-27†	Multiple	Seq[GRCh37] Xp22.33q28(1_155270560)x1[0.7]	Mosaic Turner syndrome	Pathogenic26	No obvious clinical manifestation	1, 2
PGPC-02, PGPC-27†	CHEK2 (NM_145862.2)	c.470T > C, p.(Ile157Thr), het	Cancer susceptibility, AD/AR	Risk factor: PS4, PP527,28	27: FHx of breast, prostate, throat cancer	1, 2, 3
PGPC-25,† PGPC-29	F2 (NM_000506.3)	c.*97G > A (20210G > A), het	Thrombophilia, AD/AR	Risk factor: PS3, PS429	29: No relevant PHx/FHx	1, 2, 3
PGPC-24	LPL (NM_000237.2)	c.953A > G, p.(Asn318Ser), het	Dyslipidemia, AD/AR	Risk factor: PS3, PS430–32	PHx/FHx of hypercholesterolemia, FHx of CAD (60 yr)	1, 2, 3
PGPC-36	MUTYH¶ (NM_001048171.1)	c.892–2A > G, p.?, het	Familial adenomatous polyposis, AR; colorectal cancer risk, AD	Risk factor (AD), pathogenic (AR): PVS1, PS3, PP5, BS133,34	FHx of gastric cancer (70–80 yr)	1, 2, 3
PGPC-48	PCDH15 (NM_001142769.1)	chr10:g. [55741501_57660800del], het (1.9 Mb)	Neuropsychiatric disease risk, AD; deafness/Usher syndrome, AR	Risk factor (AD), likely pathogenic (AR): PVS1, PM235–38	FHx of ADHD	1, 3
Variants of uncertain significance
PGPC-40†	ANK2 (NM_001148.4)	c.4373A > G, p.(Glu1458Gly), het	Long QT syndrome, cardiac arrhythmia, AD	VUS: PS3, PP1, BS139–41**	No relevant PHx/FHx	1, 2
PGPC-40†	CDH1 (NM_004360.3)	c.2343A > T, p.(Glu781Asp), het	CDH1-related cancer, AD	VUS: PM242,43	FHx of gastric cancer (75–80 yr)	1, 2
PGPC-19	CHMP2B (NM_014043.3)	c.85A > G, p.(Ile29Val), het	Amyotrophic lateral sclerosis, Frontotemporal dementia, AD	VUS: PS3, PP5, BP444–46	No relevant PHx/FHx	1
PGPC-14	KCNE2 (NM_172201.1)	c.29C > T, p.(Thr10Met), het	Long QT syndrome, Atrial fibrillation, AD	VUS: PP5, BS147,48	ND	1, 2
Uncertain significance, genetic counselling recommended
PGPC-32	MT-TV	m.1659T > C, 7% heteroplasmy	Childhood neurologic disorder	Likely pathogenic††49	ND	1, 4
PGPC-22	Multiple	Seq[GRCh37] inv(20) (q11.23q13.12), chr20:g. [35583655_44214109inv], het (8.6 Mb)	Likely minimal risk for unbalanced aberrations in family members	Large, rare structural variant	ND	1
53 of 56 PGPC participants	172 recessive carrier variants (e.g., 5 pathogenic CFTR alleles)	Multiple	Potential disease risk in family members	Risk factor	ND	1, 4

Note: AD = autosomal dominant, ACMG = American College of Medical Genetics and Genomics, ADHD = attention-deficit/hyperactivity disorder, AR = autosomal recessive, CAD = coronary artery disease, CT = computed tomography, FHx = family history, het = heterozygous, LEP = limited evidence for pathogenicity, Mb = megabase pairs, ND = no data, PGPC = Personal Genome Project Canada, PHx = personal history, VUS = variant of uncertain significance. Management implications: genetic counselling = 1, screen/monitor for medical complications and anticipate risk prevention = 2, identification of family members at risk = 3, Family planning = 4.

^*Interpretation according to the ACMG guidelines.¹⁶

^†Participants with 2 or more variants.

^‡Gene is included in the ACMG list.⁵⁰

^§Gene is included in the ACMG list but associated to a different phenotype (dilated cardiomyopathy).

^¶Gene is included in the ACMG list when biallelic.

^**This variant was recently interpreted as likely pathogenic in another healthy cohort.⁵¹

^††Pathogenic at a higher level of heteroplasmy.

In addition, we identified a likely pathogenic ELN splice acceptor variant in a 63-year-old healthy man, which is predicted to cause in-frame skipping of a well-conserved exon implicated in intermolecular cross-linking of the tropoelastin polymer.²¹ Elastin dysfunction is associated with incompletely penetrant supravalvular aortic stenosis and other vascular lesions, none of which were found on examination using cardiac computed tomography. In the same participant, we detected a likely pathogenic frameshift variant in LZTR1, associated with increased risk for schwannomas.²³ Penetrance for the disease is uncertain, and the participant had no personal or family history of schwannomas.

Our analysis also determined that a 70-year-old man (with self-reported hypercholesterolemia) had a heterozygous rare variant in LMNA; the same variant was reported to cause semidominant partial lipodystrophy and metabolic abnormalities, with cardiovascular risk factors particularly pronounced in obese individuals or carriers of 2 pathogenic alleles.²²

We also identified a single pathogenic variant in SLC7A9 in a 49-year-old man. The same variant is associated with dominant cystinuria in the stone-forming range;^24,25 stone formation is often prevented by adequate hydration and dietary modifications.

Finally, we recognized the mosaic loss of an X chromosome — in about 70% of the blood cells — in a 54-year-old woman with no obvious clinical presentation of Turner syndrome. This could create eligibility for screening for potential late-onset cardiac or endocrine manifestations.

We identified 7 risk factor variants in 5 genes (CHEK2 (×2), F2 (x2), LPL, MUTYH and PCDH15; Table 3). Among those was a CNV deletion (1.9 Mb) of exons 1–23 of PCDH15, which has been reported as a risk factor for neurobehavioural disorders,^35–38 in a 44-year-old participant with a family history of attention-deficit/hyperactivity disorder. Despite an extensive literature review, we concluded that 4 other variants — in ANK2, CDH1, CHMP2B and KCNE2 — had uncertain clinical significance (Table 3). We found a missense variant in ANK2 in a 49-year-old man that had been reported previously in a large French pedigree with long QT syndrome, sinus node dysfunction and sudden death,³⁹ and was associated with convincing functional studies in mouse cardiomyocytes. ⁴⁰ This variant was recently interpreted as likely pathogenic in another healthy cohort.⁵¹ However, it is as frequent as 0.1% in some populations (http://gnomad.broadinstitute.org/), which suggests that the variant is either unrelated to disease or functions with incomplete penetrance. Additional variants, as observed in CDH1, CHMP2B and KCNE2, have been published as disease alleles, with functional support, albeit with discordance in the literature and databases.

A 65-year-old woman had a likely pathogenic variant in MT-TV in 7% of her mitochondria,⁴⁹ and a 65-year-old man had a paracentric inversion (8.6 Mb) on chromosome 20q11.23q13.12. Neither seemed disease-associated in the respective participant but could be relevant to other family members (Table 3).

We found multitudes of other data that were potentially relevant to health. For example, there were 172 recessive alleles in 137 disease-associated genes (some have been identified in Canadian studies^52,53), and 8 large CNVs(> 100 kb) of uncertain significance but involving genes (e.g., a duplication affecting 16 genes in 1 participant (PGPC-56) (Table S1, Appendix 2). Most variants were interpreted as of uncertain significance or likely benign. Recognition of novel variants declined with each sample analyzed, in particular for the number of variants classified as likely benign or of uncertain significance (Figure 3). Therefore, the burden of variant interpretation becomes lighter with each additional genome interpreted.

Figure 3:

Decline in the number of novel variants as additional samples were analyzed. The burden of new variants drops with each sample analyzed, in particular the number of variants classified as benign/likely benign and uncertain significance. For each of the 55 genetically unrelated participants (excluding the child in a mother–father–child trio) in the Personal Genome Project Canada, single nucleotide variants, insertion/deletions and copy number variants that overlapped genes from the Clinical Genomic Database were classified as benign/likely benign, variant of uncertain significance or pathogenic/likely pathogenic according to American College of Medical Genetics and Genomics interpretation guidelines.¹⁶ We then performed 1 million simulations; in each simulation, we randomly assigned the order of the samples, and the number of variants found in sample i that were not found in any of the samples 1, 2, …, i − 1 was calculated for each i, 1 ≤ i ≤ 55. The lines indicate the number of new variants for each value of i (averaged over the million simulations), whereas the shaded areas represent ±1 standard deviation from the mean, for each of the 3 variant categories: benign (green), variant of uncertain significance (blue) and pathogenic (red).

Pharmacogenomics

In addition to risks of genetic disease, we assessed the data for variants in 14 pharmacology-relevant genes (“pharmacogenes”), according to clinical guidelines.¹⁵ Participants had an average of 3.9 diplotypes (range 1–6) that were associated with risk for altered drug efficacy and/or adverse reactions (Table S2, Appendix 3, available at www.cmaj.ca/lookup/suppl/doi:10.1503/cmaj.171151/-/DC1). We found 13 participants (23.2%) at risk for severe potentially life-threatening adverse drug reactions (i.e., HLA-A*3101- and HLA-B*5701-associated hypersensitivity reactions, intermediate or low thiopurine methyltransferase activity with myelotoxicity risk). One participant (PGPC-28) had relevant findings in 6 pharmacogenes. These could compromise metabolism of drugs by CYP2D6 or CYP2C19, reduce ability to metabolize thiopurines, increase risk for simvastatin-related muscle toxicity and have implications for initial warfarin dosing. Three participants (PGPC-09, PGPC-16, PGPC-32), who were identified with CYP2C19-metabolizer status indicating favourable response to proton pump inhibitors, had self-reported use of such drugs.

Interpretation

The Personal Genome Project Canada constitutes a public resource of data from the population at large that supports evaluation of whole genome sequencing and its utility for personalized medical practice in Canada. Although most variants identified by whole genome sequencing were of uncertain significance or likely benign, most participants (53/56) carried at least 1 disease-associated allele (mean 3.3/individual [SD 1.7]). From a clinical research perspective, at this stage, we considered findings to have personal health implications for 14 participants (25%). An additional 172 pathogenic alleles were associated with autosomal recessive or semidominant inheritance patterns (mean 3.1/individual [SD 1.7]), which is close to empirical estimates of the carrier burden for recessive diseases.⁵⁴ Participants also carried an average of 3.9 pharmacorelevant diplotypes associated with the metabolism of about 50 drugs. This highlights the potential of whole genome sequence data to be used pre-emptively for precision medicine, to reduce risk of adverse drug events or therapeutic failure. In general, we believe our interpretations to be conservative.

Given the variety of potentially relevant findings (Table 3), whole genome sequencing will likely become part of mainstream health care in the foreseeable future. General health care providers will be involved in interpreting and delivering genomic information in the context of personal and family histories. This requires awareness of the spectrum of potential findings, and the technical strengths and limitations of the underlying tests.

Unlike lower-resolution genomic tests, such as karyotyping, microarrays and exome sequencing, whole genome sequencing captures the entire compendium of variation in 1 experiment. Whereas most earlier studies, as well as direct-to-consumer genetic testing, have focused on SNVs and small indels,^51,55–57 our study exploited the full potential uncovered by sequencing the entire genome, including copy number, other structural and mitochondrial variants, several of which would not have been detected by other methods (Table S5 in Appendix 1).

For use in the context of clinical diagnostic sequencing, the American College of Medical Genetics and Genomics compiled a list of 56 (revised to 59) genes associated with “actionable” phenotypes^50,58 for which functional variants should be reported as “secondary” (incidental or unanticipated) findings. Using these criteria, our analysis identified only 1 pathogenic variant (in BRCA1), a number expected from the rate of incidental findings in larger cohorts.⁵⁹ However, we found additional variants that we deemed to have health implications (in ANK2, CDH1, ELN, KCNE2, LMNA, LZTR1 and PCDH15; Table 3), and yet others that could be relevant for family planning or newborn screening (in participants PGPC-22 and PGPC-32), or for decisions about appropriate therapies or medications.

Even when using established analysis guidelines,¹⁶ variant interpretation is sometimes subjective, and requires considerable manual curation and critical review of the underlying evidence, which may be fraught with discordant interpretation⁶⁰ and misclassifications.⁶¹ Further challenges arise when the a priori probability of disease is low or findings are associated with variable outcomes. All of these issues will become more relevant as the focus for sequencing shifts from diagnostic to predictive/preventive genomics. Improved control data, and even machine-learning approaches (both for variant calling and interpretation), ⁶² should mitigate some of the subjectivity. However, given the phenotypic spectrum associated with most variation and the influence of environment, concordance is a distant goal.

Once on file, genome sequence data can be reanalyzed as informatics tools improve and novel disease associations emerge.⁶³ Also, new medical concerns, exposures, treatment needs or previously unnoticed familial risks may warrant reinterpretation. Along with the massively increased identification of informative variants by whole genome sequencing, come ever more uncertain findings. Larger collections of genomes, interpreted in the context of thorough and evolving personal and family histories, will help to shift the proportion of VUS into known benign or pathogenic classifications, and enable risk predictions for unbiased cohorts.

Canada’s Genetic Non-Discrimination Act was passed just as we were informing this initial cohort of results and seeking their final consent for publication. Anecdotally, prior perceived limitations to participation seemed to be somewhat relieved once protection afforded by the Act was assured.

Limitations

The 56 inaugural participants of the Personal Genome Project Canada are a small cohort of volunteers, both highly educated and idealistic regarding genomic research. These volunteers do not reflect the diverse Canadian ethnicities, but we explicitly aim to expand diversity as the sample size increases, including participation from Indigenous and recent immigrant peoples. Early personal genome sequencing cohorts were suggested to be enriched for individuals with perceived risk or subtle symptoms of genetic disease;⁸ although we did not enrol participants who were explicitly seeking genetic information for suspected heritable conditions, neither did we exclude participants with known health conditions (Table 1).

Because of challenges in interpretation, many potentially disease-causing variants were disregarded by our initial analysis (such as novel missense variants, synonymous and noncoding variants, variations in genes not yet associated with a phenotype or variants with allele frequencies > 5%). Certain types of pathogenic alleles are not detected reliably at present through the short-read whole genome sequencing method we used (e.g., those in regions on the Y chromosome and telomeres^64–66 or trinucleotide repeat expansions). We analyzed 1 variant at a time and did not consider genetic networks.⁶⁷ This approach will continue to be appropriate for those genetic variants with substantial discrete impact on phenotypes. However, the full impact of personal genomics in precision medicine will emerge as we recognize those variants with incremental effects and complex interactions, all influenced by recent human adaptation.^68,69

Conclusion

In 14 of the 56 participants (25%) — a relatively mature and ostensibly healthy cohort — we identified genomic variants with potential implications for health management of the individuals but also for their families and future generations. We also added recognition, in 100% of participants, of pharmacogene variants, carrier status for recessive alleles and/or copy number variants (some involved in mental health). Coupled with growing knowledge of how such genomic variation relates to health, disease and treatment options, these findings suggest that whole genome sequencing can benefit routine health care in Canada’s future. Despite a considerable burden of uncertainty, and the possibility that false-positive findings may engender follow-up investigations⁵¹ and a “worried well” population,⁷⁰ incorporation of sequence-based family history should serve to enhance personalized patient care.

See related article at www.cmaj.ca/lookup/doi/10.1503/cmaj.180076