Reviews

Using human genetics to improve

safety assessment of therapeutics
Keren J. Carss   1, Aimee M. Deaton   2,9, Alberto Del Rio-Espinola3,13, Dorothée Diogo4,
Mark Fielden5,10, Diptee A. Kulkarni6, Jonathan Moggs3, Peter Newham   1,
Matthew R. Nelson7, Frank D. Sistare8,11, Lucas D. Ward   2,9 ✉ and Jing Yuan2,12
Abstract | Human genetics research has discovered thousands of proteins associated with
complex and rare diseases. Genome-wide association studies (GWAS) and studies of Mendelian
disease have resulted in an increased understanding of the role of gene function and regulation
in human conditions. Although the application of human genetics has been explored primarily
as a method to identify potential drug targets and support their relevance to disease in
humans, there is increasing interest in using genetic data to identify potential safety liabilities
of modulating a given target. Human genetic variants can be used as a model to anticipate the
effect of lifelong modulation of therapeutic targets and identify the potential risk for on-target
adverse events. This approach is particularly useful for non-clinical safety evaluation of novel
therapeutics that lack pharmacologically relevant animal models and can contribute to the
intrinsic safety profile of a drug target. This Review illustrates applications of human genetics
to safety studies during drug discovery and development, including assessing the potential
for on- and off-target associated adverse events, carcinogenicity risk assessment, and guiding
translational safety study designs and monitoring strategies. A summary of available human
genetic resources and recommended best practices is provided. The challenges and future
perspectives of translating human genetic information to identify risks for potential drug
effects in preclinical and clinical development are discussed.

Genome-wide association
studies
Studies of phenotypes
performed by testing
correlation of genotyped
markers with a phenotype in
a large population of unrelated
individuals. Results are usually
relatively common coding or
noncoding variants associated
with relatively weak effects
on the phenotype.
Mendelian disease
A typically rare disease caused
by variants in a single gene.
Studies of family members
with these diseases usually
uncover relatively rare variants
with high penetrance.
✉e-mail: lward@alnylam.com
https://doi.org/10.1038/
s41573-022-00561-w
Human genetic research has discovered thousands of
proteins associated with complex and rare diseases.
Genome-wide association studies (GWAS)1 and studies of
Mendelian disease2 have resulted in an increased under-
standing of the role of gene function and regulation in
human conditions, which has been leveraged to discover
new therapeutic targets and provide evidence for ther-
apeutic hypotheses of targets already being pursued3,4.
Retrospective analyses of drugs in clinical development
demonstrated that drugs with supporting human genetic
data for their intended indication were twice as likely to
be approved as drugs that lacked such validation5,6.
In addition to identifying new targets for drug effi-
cacy, the study of human genetics also has the potential
to identify risks for drug toxicity and associated adverse
events (AEs). Currently, preclinical safety evaluation
of pharmaceuticals relies mainly on animal models.
Although the use of animals has made huge contribu-
tions to safety assessment, there remains uncertainty
regarding the extent to which preclinical findings can
be extrapolated to humans owing to species differences
in biology7. This contributes, in part, to the high failure
rate of drugs in clinical development due to both lack of
efficacy and safety concerns; a survey of 246 terminated
clinical programmes showed that 37% failed owing
to non-clinical or clinical safety issues, versus 18% due to
lack of efficacy8.
This challenge is particularly evident for the ther-
apeutic modulation of novel targets that have limited
biological characterization, where there is a lack of
pharmacological activity in animal models owing to
poor cross-reactivity of drugs to the orthologous disease
target, or where animal models have poor translatability
for human disease or safety. Even when a suitable animal
model exists, some safety liabilities may emerge only in
the chronic animal toxicology studies (typically 6 months
in rodents and 9 months in non-rodents, or in 2-year
rodent carcinogenicity studies), or manifest only during
developmental periods of exposure, which are often not
studied until later stages of clinical development. As a
result, there is significant interest in the application of
human genetics to improve our understanding of the
potential safety liabilities associated with modulating
the activity of intended and unintended drug targets3,9,10.

NaTure RevIeWS | DRuG DISCOveRy volume 22 | February 2023 | 145

0123456789();:
Reviews
Phenome-wide association
One of the first ways that human genetics was applied
study to drug safety was through pharmacogenetics, the study
A study of a genetic variant of the genetic basis of variation in drug response (Box 1).
performed by testing the Although pharmacogenetics is a crucial component in
correlation of the variant with
many phenotypes, typically
improving care and advancing precision medicine, we
all of the phenotypes available extend this Review to broader applications of human
in a biobank or electronic genetics, making inferences from entire populations
medical system. instead of studying variation only between recipients of
a particular drug.
Genetic variants that increase or decrease protein
expression or function may serve as a model for lifelong
activation or inhibition of a drug target. Such ‘experi-
ments of nature’ can inform the likelihood of safety lia-
bilities following chronic drug exposure, which has been
illustrated in a recent phenome-wide association study
(PheWAS) for selected drug targets11,12. In addition, a
retrospective large-scale evaluation recently showed
a correlation between the organ systems affected by
genetic variation in drug targets and the organ systems
in which adverse effects were observed in clinical trials13.
Collectively, these results demonstrate that human
genetic data can be used to help predict the effects
associated with therapeutic manipulation of drug targets.
In this Review, we provide an overview of how
human genetic data can provide valuable insights into
potential safety liabilities of modulating drug targets —
both intended targets (on-target effects) and unintended
targets (off-target effects), including carcinogenicity risk.
We provide an overview of available human genetic
resources and illustrate their utility for safety assess-
ment through several published case studies. Finally, we
discuss some challenges and limitations of using such
data, as well as suggested best practices for how human
genetics data can be used to enhance safety assessment
of novel drug targets and modalities.
Application of genetics to drug safety
A wide variety of techniques can be used to obtain
human genetics data that can then be applied to the
drug development process (Fig. 1 and Table 1). Genetic
evidence can be applied to predict potential AEs during
various stages of the drug discovery and development
pipeline, from target selection to clinical development.
Human genetics can potentially be leveraged to decide
whether to pursue or abandon a programme, to invest in
early mechanistic de-risking investigations, to improve
the design of preclinical safety assessment studies or to
Author addresses
1
AstraZeneca, R&D, Cambridge, UK.
2
Amgen, Cambridge, MA, USA.
3
Novartis Institutes for BioMedical Research, Basel, Switzerland.
4
Takeda, Cambridge, MA, USA.
5
Amgen, Thousand Oaks, MA, USA.
6
GlaxoSmithKline, Collegeville, PA, USA.
7
Deerfield Management Company, L.P., New York, NY, USA.
8
Merck & Co., West Point, PA, USA.
9
Present address: Alnylam Pharmaceuticals, Cambridge, MA, USA.
10
Present address: Kate Therapeutics, San Diego, CA, USA.
11
Present address: 315 Meadowmont Ln, Chapel Hill, NC, USA.
12
Present address: Pfizer, Cambridge, MA, USA.
13
Present address: GentiBio Inc., Cambridge, MA, USA.
146 | February 2023 | volume 22 www.nature.com/nrd
0123456789();:
inform the design of clinical trials, either by excluding
patients at high risk of the AEs or by enabling specific
AE monitoring during the trial. It should be noted that
genetic information can be used at all stages, ranging
from target discovery to lead discovery, non-clinical
studies, clinical development, life cycle management and
post-marketing pharmacovigilance. Accordingly, the
starting point for these investigations will vary: in some
cases it will be instigated by consideration of a novel
target based on biology alone; it may be prompted by
findings that arise from testing of a preclinical lead com-
pound in in vitro systems or animals, or from the need to
design such tests; or it may be triggered by AEs observed
during clinical trials or through post-marketing reports.
Target validation and on-target safety
Human genetics, including the study of Mendelian
disease as well as common and complex traits, can be
used to predict potential drug-induced target-related
AEs. Examples of clinically approved or failed drugs
illustrate the potential application of human genetics
to predict the safety of drug targets. A recent study13
evaluated the link between diseases and phenotypes
associated with genetic variants in drug target genes
and AEs reported for the corresponding drugs. Using
evidence from both Mendelian genetics and GWAS, the
authors demonstrated that AEs are more likely to occur
in organ systems in which there is genetic evidence that
links the drug target to a phenotype in that organ sys-
tem. For example, endocrine-related side effects were
noted among 12% of drugs for which the target has no
association with endocrine-associated phenotypes, while
that rate increased to 28% for drugs for which the tar-
get has associations with endocrine-related phenotypes.
These enrichments were evident even after controlling
for genetic effects related to the intended therapeutic
indication and related organs.
Mendelian disorders, in which rare, highly pene-
trant loss-of-function (LOF) or gain-of-function (GOF)
variants in a target gene cause a disease, can point to
potential AEs of target inhibitors or activators, respec-
tively. Examples include drugs that target diacylglycerol
O-acyltransferase 1 (DGAT1) and sepiapterin reduc-
tase (SPR). DGAT1 inhibition has been proposed for
the treatment of type 2 diabetes and obesity14. Clinical
trial results reported dose-related gastrointestinal
AEs, including diarrhoea, which dramatically lim-
ited the therapeutic index15. In 2012, after the results
of the first-in-human study was published, children
with DGAT1 deficiency were reported in a family of
Ashkenazi Jewish descent16. These children, who carried
DGAT1 LOF mutations, presented 3 days after birth with
severe intractable diarrhoeal disorders. These genetic
findings provided strong support for the hypothesis that
the severe gastrointestinal AEs observed in clinical trials
were target-related and highlight the value of investigat-
ing the phenotypes of individuals with such variants at
an early stage of consideration of a new drug target.
Similarly, inhibition of SPR, the final enzyme in the
tetrahydrobiopterin (BH4) de novo synthesis pathway,
has been investigated for the treatment of neuropathic
pain. Despite the genetic and non-genetic evidence
 Reviews

Box 1 | Genetics of patients: the state of pharmacogenetics

Pharmacogenetics, the study of human heritable inter-individual genetic
variation in drug responses, can add valuable information to the risk
assessment of a drug and guide drug therapy decisions. Over the decades,
the number of gene variants that have been found to influence drug
responses has steadily increased. Some variants were found to make
drugs toxic; others rendered certain drugs ineffective. These genetic
variations are most extensively characterized in genes involved in drug
absorption, distribution, metabolism or excretion (ADME), such as genes
that encode cytochrome P450 and transporters. Pharmacogenetic
variations can also occur in genes that underlie immunity such as the
human leukocyte antigen/major histocompatibility complex (HLA/MHC)
locus or in the drug target itself or its pathway.
Understanding the pharmacokinetics of a drug can help anticipate
adverse events (AEs) that could be caused by variations in drug-
metabolizing enzymes, and vice versa: discovering associations between
AEs and patient genetics may reveal unexpected aspects of ADME.
Cytochrome P450 enzyme activity is known to be highly variable
owing to genetic variabilities. For example, the antiplatelet agent
clopidogrel requires activation by CYP2C19 to become bioactive.
Patients with loss-of-function CYP2C19 alleles (c.681G>A; rs4244285
or CYP2C19*2/*2) present with poor metabolization and relatively low
CYP2C19 activity, thus higher on-treatment platelet aggregation
compared with normal carriers (CYP2C19*1/*1)223. In 2010, the FDA
added a boxed warning to the clopidogrel drug label about higher rates
of cardiovascular events in patients undergoing percutaneous coronary
intervention who are poor metabolizers of CYP2C19 substrates (these
cardiovascular events are due to lack of efficacy, so are not strictly AEs by
some definitions).
Genetic variants in other enzymes involved in metabolism pathways
have been implicated in drug AEs. Thiopurine-related haematotoxicity is
clinically relevant and therapy-limiting in treatment of patients with
acute lymphoblastic leukaemia (ALL) or inflammatory bowel disease (IBD).
Severe haematotoxicity in patients under thiopurine therapy has been
associated with polymorphisms in thiopurine S-methyltransferase
(TPMT)224, and TPMT genetic testing is clinically implemented for dose
individualization225.
HLA genetic predisposition has been widely studied for immune-
driven drug toxicities. These investigations are particularly valuable
for idiosyncratic toxicities that cannot be predicted preclinically and
can only be addressed during clinical development. An interesting
example is carbamazepine-induced Stevens–Johnson syndrome (SJS)
and related toxic epidermal necrolysis (TEN), whereby an association
with HLA-B*1502 in the Chinese Han population226 led to the FDA
recommendation for genotyping in Asian populations227 and exclusion
of patients with HLA-B*1502 from treatment, preventing SJS–TEN cases
in Taiwanese patients228 (HLA-A*3101 is the relevant allele in European
populations229).
Pharmacogenetics has been reviewed extensively, from drug
discovery230–232, drug efficacy233, drug safety234–238 and regulatory
perspectives239–241. Currently, variants in around 20 genes provide
useful predictions of reactions to 80–100 drugs — about 7% of drugs
approved by the FDA242. Although pharmacogenetic discoveries could
be beneficial to many patients, up-front screening has not been widely
adopted, and genetic testing is often not used until after a patient
has had problems with a medication. This is mainly due to the lack of
supporting randomized controlled trials. Clinical use of pharmacogenetic
testing has been largely based on observational and retrospective trials
conducted in targeted patient populations and for research purposes.
Recently, Ubiquitous Pharmacogenomics (U-PGx) has initiated a large,
randomized controlled trial to study a panel of 13 genes for clinical
outcomes of 8,000 patients in seven hospitals in seven European
countries243. Meanwhile, initiatives on pharmacogenetics passports such
as guidelines developed by the NIH-funded Clinical Pharmacogenetics
Implementation Consortium (CPIC) and The Dutch Pharmacogenetics
Working Group (DPWG) will facilitate the implementation of pharma­
cogenetic testing for any upcoming treatments244 or clinical trials using
pharmacogenetics-guided dose, such as for warfarin showing lower risk
of major bleeding245.

identifying BH4 synthesis as a key modulator of periph-
eral pain17, LOF variants in SPR and other genes in the
pathway have been linked to neurological disorders18,19,
raising potential safety concerns associated with SPR
inhibitors penetrating the blood–brain barrier. In 2017,
Quartet discontinued its SPR inhibitor development
efforts owing to the observation of concerning neuro-
logical effects in toxicology studies, attributed to central
nervous system exposure20.
Mendelian genetics can also point to important cel-
lular mechanisms that may indirectly predict AEs. One
such example is IL-2 receptor-α (IL-2RA) deficiency,
which causes immunodeficiency with lymphoprolifer-
ation and autoimmunity and is characterized by defects
in T cell regulation21–23. Imbalance between regulatory
T cells and effector T cells in the pancreatic islet has been
implicated in breakdown of self-tolerance and develop-
ment of type 1 diabetes (T1D)24. The immune cell profile
of the patients with IL-2RA deficiency, together with its
reported biological link to T1D, is consistent with
the new-onset diabetes cases reported as a side effect
of the IL-2R antibody baziliximab25.
Mechanistic information can also come from
studying the Mendelian genetics of proteins that are
known to directly interact with a target of interest.
Lipopolysaccharide-responsive and beige-like anchor
protein (LRBA) is known to regulate and colocalize
with the cytotoxic T lymphocyte-associated 4 (CTLA4)
protein26. LRBA deficiency is characterized by immuno­
deficiency and variable autoimmune presentations,
including severe enteritis26–29. Patients with LRBA defi-
ciency show severe decreases in CTLA4 levels, and treat-
ment with abatacept, a CTLA4–IgG1 immunoglobulin
fusion drug, can substantially improve enteritis in these
patients26,29. These observations demonstrate a func-
tional link between CTLA4 and the gastrointestinal
symptoms observed in patients with LRBA deficiency,
and is consistent with severe, potentially life-threatening,
colitis reported in cancer patients treated with ipili-
mumab or tremelimumab, two anti-CTLA4 monoclonal
antibodies30. Another example is deficiency of erythro-
cyte protein 4.2 (EPB42), which causes spherocytosis31;
loss of EPB42 results in depleted CD47 in the erythro-
cyte membrane, due to their interaction32. Inhibitors of
CD47 for cancer immunotherapy exhibit anaemia and
thrombocytopenia as adverse effects due to the role of
CD47 on normal erythrocytes33.
In addition to analyses of rare Mendelian syndromes,
analyses of variants segregating in populations provide
valuable information for drug safety (Fig.  2). In the
past decade, GWAS have identified tens of thousands
of variants associated with complex traits and enabled
the investigation of causal gene–phenotype relation-
ships through integration with functional annotations

NaTure RevIeWS | DRuG DISCOveRy volume 22 | February 2023 | 147

0123456789();:
Reviews

Novel therapeutics
O OH
O

Anti-X drugs

Target protein X

a Indication Y Side effect Z

b c d
X–/–

X X
Gene X Gene X Phenotyping individuals Studying evidence for target Guiding secondary
Association studies for with rare biallelic loss of in tumour suppression pharmacology
on-target liabilities function in gene X

Fig. 1 | Overview of the applications of human genetics to drug safety. This scenario illustrates evaluation of the
risk of a novel therapeutic, for example, an inhibitor of protein X. a | Population-scale association techniques, such as
phenome-wide association studies (PheWAS), genome-wide association studies (GWAS) and Mendelian randomization
(MR), can be used to infer putative on-target liabilities; in this example, an inhibitor of protein X may treat indication Y
while potentially causing side effect Z. b | The existence and phenotyping of individuals with very biallelic loss-of-function
variants (in this example, in gene X) can be used to discover liabilities or de-risk a target. c | Evidence of a role for on-target
or off-target effects in cancer from high-throughput sequencing, transcriptomics and perturbation screens can be used
to investigate carcinogenicity risk. d | Information from human genetic associations or syndromes can be used to design
secondary pharmacology screens to contextualize and prioritize off-target liabilities.

Expression quantitative
trait loci
Variants, typically common and
noncoding, whose genotype
correlates with expression
of a gene, usually nearby,
in a given tissue or cell type.
and causal inference methods. Retrospective examples
illustrate the potential of GWAS to inform drug safety.
A notable example is the association at the CTLA4 locus
discussed above with several autoimmune diseases
that are also reported as common AEs of ipilimumab
or tremelimumab, including vitiligo 34 and alopecia
areata35,36. GWAS have the potential to provide valua-
ble information for safety assessment, particularly in
the ideal scenario where the target gene is clearly estab-
lished as the gene driving the GWAS association, and
the functional effects of the causal variant, including
tissue or context-dependent gene expression changes or
GOF or LOF of the protein encoded by the target gene,
are well understood. In particular, when a suitable
genetic variant mimicking the effect of the drug has
been identified, methods such as PheWAS or Mendelian
randomization (MR) can be used to investigate pleio-
tropic effects and predict potential adverse outcomes,
as described below11,34,37–44.
In PheWAS, a variant or group of variants is tested
for association with a large number of phenotypic end
points. Data are typically derived from electronic health
records (EHRs) or comprehensive questionnaires in
disease-agnostic cohorts45,46, although, in principle,
PheWAS can also be performed using full summary
statistics from published GWAS. A recent PheWAS
interrogated functional variants linked through GWAS
with 19 candidate drug targets for common disease
indications11. The study found several associations
that suggest possible on-target AEs. One example is a
functional variant in PNPLA3 (encoding patatin-like
phospholipase domain-containing protein 3) that is
associated with an increased risk of nonalcoholic fatty
liver disease (NAFLD, the intended indication) and is
also associated with a slightly decreased risk of acne, high
cholesterol, gout and gallstones. Hence, therapies for
NAFLD that act through reducing or inhibiting PNPLA3
may result in increased risk of these AEs. This study
also reported association of a partial IFIH1 (encodes
interferon-induced helicase c domain-containing pro-
tein 1) LOF variant with lower risk of several autoim-
mune conditions (the intended indications) and with
higher risk of asthma. How these observations would
translate into AEs for a drug that targets these proteins
depends on multiple factors, as we discuss below.
Another recent analysis demonstrated the benefit of
integrating additional information to PheWAS, such as
gene expression data and expression quantitative trait loci
(eQTLs), to predict AEs47. After identifying potential
on-target AEs through PheWAS, follow-up analyses
including colocalization tests on genetic risk variants48,49
are required to validate the causal link between the vari-
ant tested and the associated phenotype (or phenotypes),
which ultimately would be confirmed in randomized
clinical trials with the drug product. Exome sequencing
data combined with detailed EHRs and quantitative trait

148 | February 2023 | volume 22 www.nature.com/nrd

0123456789();:

Endogamous populations
data from the UK Biobank50–52 is proving to be a par-
Populations in which ticularly rich resource in this space. Several large-scale
reproduction is restricted PheWAS have recently been published, and the results
to relatively small social made available for the community via resources such as
groups or extended families,
resulting in a higher degree
the AstraZeneca PheWAS Portal53 and GeneBass. These
of consanguinity between databases empower the scientific community, including
partners and a higher those without the statistical and computational resources
incidence of homozygosity to perform such studies themselves, to use PheWAS data
of alleles.
for their own research, including safety assessment of
Gene constraint
candidate drug targets.
Reduced nucleotide diversity In MR, a genetic variant, or a combination of genetic
in the coding sequence of a variants, is used as a proxy for exposure to assess the
gene resulting from negative causal effect of an exposure on an outcome (which can
selection on deleterious
variants.
be related to efficacy or safety54–57). MR has been used to
investigate unintended effects of drugs, including statins,
proprotein convertase subtilisin/kexin-type 9 (PCSK9)
inhibitors, ezetimibe, IL-1 inhibitors and bempedoic
acid39,41–44.
PCSK9 inhibition has been studied extensively
by MR, because PCSK9 GOF variants cause familial
hypercholesterolaemia58–60 and rare-to-common LOF
variants are associated with decreased low-density
lipoprotein (LDL) levels and decreased coronary heart
disease risk61–64. Predicted LOF variants in PCSK9 that
are present in more than 2% of people of African origin
have no apparent deleterious health consequences61,65.
This was used as evidence to support the probable safety
of pharmaceutical PCSK9 inhibition. MR studies of
PCSK9 and other lipid-lowering drugs have variously
reported associations with Alzheimer disease and type 2
diabetes42,66–69; however, inhibitors such as evolocumab,
alirocumab and inclisiran are effective and well toler-
ated, with no clinical evidence of these risks70–72. The
discrepancy between MR analysis and pharmacolog-
ical inhibition was confirmed by the comparison of
PCSK9 MR–PheWAS and pooled clinical trial results69.
This inconsistency was explained by observations in
knockout mice, where downregulation of pancreatic
Pcsk9 expression is linked to β-cell cholesterol overload
and apoptosis, causing impaired glucose metabolism,
whereas downregulation of liver or circulating PCSK9
protein is linked to plasma LDL-cholesterol (LDL-c)
reduction without altering glucose homeostasis 73.
Alternative explanations could be loss of PCSK9 in
Table 1 | Applications of human genetics to drug safety
Application Techniques
Target identification and GWAS and PheWAS in large populations
validation, and evaluation of
Study of Mendelian disease
on-target safety
Phenotyping of individuals with very rare biallelic
loss-of-function variants
Evaluation of target-mediated Study of somatic mutations, copy number variants
carcinogenicity and gene expression in cancer tissues
Secondary pharmacology and Genetics-guided prioritization of off-targets included
off-target toxicity in screening assays
Genetics-informed mechanistic toxicity
investigations
Pharmacogenetics GWAS or candidate-gene studies among patients
treated with a particular drug
GWAS, genome-wide association studies; PheWAS, phenome-wide association studies.
NaTure RevIeWS | DRuG DISCOveRy
0123456789();:
Reviews
development versus pharmacological inhibition in the
adult, incorrect variant selection74, differences in patient
population between genetic studies and clinical trials,
concomitant medications (for example, statins) or lim-
ited duration of clinical follow-up. These results high-
light the complexity of MR analysis and interpretation.
Recently, MR–PheWAS has been proposed to conduct
hypothesis-free MR analyses, which could enable MR
to be more broadly used for hypothesis generation in
safety assessment.
Investigations of phenotypes associated with LOF
variants in a target gene are increasingly being used as
evidence for probable safety of pharmaceutical inhibi-
tion of that target if the association is statistically robust
with no disease relevance. Recent examples where LOF
variant analyses contributed to gene safety assessment
are hydroxyacid oxidase 1 (HAO1), leucine-rich repeat
kinase 2 (LRRK2), chemokine CC motif receptor 5
(CCR5) and mitogen-activated protein kinase kinase
kinase 15 (MAP3K15).
A therapeutic agent that inhibits HAO1 is approved
for the treatment for primary hyperoxaluria type I75. A
study in the Genes and Health cohort identified a healthy
adult with predicted complete HAO1 knockout due to
a rare homozygous predicted LOF variant, supporting
the likely safety of this therapeutic approach76.
LRRK2 is a promising therapeutic target for the treat-
ment of Parkinson disease77–79, and heterozygous LOF
variants are not associated with reduced life expectancy
or with any specific disease80, supporting the safety
profile of LRRK2 inhibition to 50%.
Similarly, CCR5-Δ32 variants, which prevent func-
tional expression of the CCR5 chemokine receptor,
were identified in HIV-1 infection-resistant people or
slow-progressing patients81,82, leading to the develop-
ment of CCR5 antagonists for HIV-1 infection83. The
lack of overt phenotype for homozygous CCR5-Δ32
has provided confidence in safety for development of
CCR5 antagonists and genome-edited autologous T cell
transplants84.
Finally, MAP3K15 inhibition has been proposed as a
therapeutic opportunity for diabetes85. Using a PheWAS
approach, researchers estimated a low safety risk for
selective therapeutic inhibition of MAP3K15 in humans.
Such examples are likely to become more frequent,
facilitated by the increasing availability of data sets that
combine genetic and rich phenotypic data86,87. Population
genetics phenomena, such as the increased frequency of
alleles that are rare elsewhere owing to founder effects
(such as deCODE in Iceland and FinnGen88,89), and
increased likelihood of homozygosity of rare alleles in
endogamous populations (such as PROMIS90 and Genes
and Health86), will facilitate the identification of individ-
uals with heterozygous and homozygous LOF variants
for safety assessment.
Although the investigation of individual LOF or GOF
variants has demonstrated value to predict target-related
AEs, gene constraint has not proved conclusively to be
a useful metric to inform safety risk. High constraint
indicates selective pressure against variants, which is
suggestive of functional importance of that sequence.
Genes that are depleted of coding variation, particularly
volume 22 | February 2023 | 149
Reviews

a
Cases Controls

Statistical method adjusting

or harmonized for population
stratification and selected • OR =(a/c)/(b/d)
confounders (age, sex, etc.) OR > 1: minor allele increases risk
OR < 1: minor allele decreases risk
• P value

Minor allele count Major allele count

Cases a b
Controls c d

b GWAS c PheWAS
chr1:1234424_A/G Phenotype 1
chr1:9234424_G/C Phenotype 2
Phenotype
chr5:5678998_A/T Phenotype 3
(disease, quantitative Variant
measurement, chr12:17568998_G/A chr19:10463118_G/C Phenotype 4
medication use, etc.)
chr19:10463118_G/C Phenotype 5
chr22:86568998_T/G Phenotype n
Indication Y Side effect Z
16 OR < 1
Association with side effect Z

Association driven OR > 1

by variant X
10
affecting gene G Significance

variant X in gene G
Significance

Association with
8 cut-off
cut-off
6

0
1 2 3 4 5 6 7 8 9 10 12 14 16 18 2022X Phenotype
Chromosome number

d Population Mendelian randomization

Random allocation of alleles
Variant
affecting
Genotype Genotype target gene G
group A group B

Decreased Increased
protein P protein P
levels levels Exposure
(e.g. biomarker,
protein levels)
Higher Lower
rate of rate of Causal
outcome O outcome O inference Confounder

Pharmacological modulation of
protein P encoded by gene G Outcome
may cause side effect O (e.g. disease)

Haploinsufficient genes
Genes in which heterozygous
loss of function causes a
phenotypic change; that is, the
remaining (haploid) functional
copy is insufficient to maintain
the wild-type phenotype.
LOF variants, are consistently found to be enriched for
certain classes of genes including those required
for essential cellular functions, haploinsufficient genes,
mouse knockout genes that result in early lethality and
genes associated with dominant Mendelian diseases91–93.
Current constraint metrics mainly capture systemic
tolerance to variation at early developmental stages,
which may limit their utility to model the effect of later,
more tissue-specific inhibition of a gene. Additionally,
their resolution is affected by the sample size and
ancestry of the population on which they were calcu-
lated. Nevertheless, their value for disease gene inter-
pretation and discovery suggests that they may also
be informative for safety assessment of drug targets —
that more constrained, essential genes may be less safe
to inhibit. However, there is currently limited evidence to
support including gene constraint metrics as a substan-
tial part of any framework for using human genomics
data in target safety assessment, and the targets of several
successful drugs are highly constrained (for example,

150 | February 2023 | volume 22 www.nature.com/nrd

0123456789();:

◀ Fig. 2 | Population genetics methods overview. Population genetics refers to
rare-to-common variants found in the general population, which may be tested for
predisposition to a disease or a trait. a | Genetic association studies use statistical methods
to test whether the frequency of a variant differs between individuals with different
phenotypes. The phenotype may be binary, comparing cases against controls, or a
measurable trait with a continuous variation; in the illustrated example, the frequency of
the variant alleles is compared between cases and controls. Potential confounding factors
that can (partially) explain the (lack of) association, including but not limited to differences
in ancestry, age or sex, should be carefully considered. The association analysis provides
an effect size, which estimates the direction and magnitude of the association of the allele
tested, and a P value indicating the significance of the association. The effect size can be
expressed as a β-coefficient (for a quantitative trait) or as an odds ratio (OR) (for a binary
phenotype). A corrected significance cut-off is often warranted to account for the
number of tests performed in the analysis and to differentiate true-positive from
false-positive associations. In the context of safety assessment, considerations
about false-negative associations, including statistical power, may also be important.
b | In a genome-wide association study (GWAS), millions of variants are tested for
association with a phenotype. c | In a phenome-wide association study (PheWAS), one
variant (or multiple aggregated rare variants) can be tested for association with a large
spectrum of phenotypes; in the illustrated example, the tested variant associates with
decreased risk of disease Y, but also associates with an increased risk of a possible side
effect phenotype Z. d | When a significant robust association between a variant and
a phenotype is identified, Mendelian randomization methods can be applied to infer
the causal mechanism. Mendelian randomization tests whether an association between
a variant and a phenotype, often a disease, can be directly attributed to an effect of
the variant on an intermediate, quantitative, phenotype. The intermediate phenotype
can be a biomarker or protein measurement, for example. Mendelian randomization
mimics a randomized control trial and takes advantage of the non-modifiable nature
of germ­line genetic variants, preventing reverse causality (left panel). However, it relies
on an assumption that the variant mechanistically affects the outcome only through the
same mechanism as the drug, and not through other pathways (red crosses, right panel).
3-hydroxy-3-methylglutaryl-coenzyme A reductase
(HMGCR) and statins)47,94,95.
Lastly, selective targeting of variant alleles, such as
pathogenic germline or somatic GOF variants, can
prevent liabilities linked to loss of wild-type allele, as
shown by the small interfering RNA (siRNA) TD101
for pachyonychia congenita96, two antisense oligonucleo­
tides for Huntington disease97 and the EGFR inhibitor
osimertinib in lung cancer98.
Off-target drug screening
A key consideration for drug safety is not only the biol-
ogy of the intended target, but also interactions with
‘off-targets’, or secondary pharmacology. The process of
evaluating secondary pharmacology can be less straight-
forward, because unlike the primary target, the identity
QT prolongation of the secondary targets and the dynamics of their inter-
A delay in ventricular actions with the drug are not easy to anticipate before
repolarization during the collecting experimental data. Genetics can be used to
cardiac cycle, visible by understand the biological effects of these off-target inter-
electrocardiography and
a common drug-induced
actions both prospectively, to guide counter-screening
adverse event. during drug development, and retrospectively, to
help identify off-targets responsible for preclinical or
PROTACs clinical toxicities.
Heterobifunctional small
Small-molecule drug candidates are commonly
molecules composed of
two ligand binding moieties assessed for their ability to bind or modulate various
connected via a linker. off-target proteins using in vitro secondary pharmacol-
One moiety binds to the target ogy assays99,100. These secondary pharmacology panels
protein, while the second are not mandated by regulatory authorities but, ideally,
moiety engages with cell E3
ubiquitin ligase complexes to
they should consist of proteins in which inadvertent
induce proteolysis of the target modulation would cause serious safety issues. Many
protein. ‘anti-targets’ on secondary pharmacology screening
NaTure RevIeWS | DRuG DISCOveRy
0123456789();:
Reviews
panels are selected by industry on the basis of biolog-
ical links between these proteins and adverse drug
reactions99. Human genetics can help to predict pro-
teins liable to perturb vital organ systems and repre-
sents a potentially valuable but underused resource for
selecting off-target proteins against which to screen. An
examplar is the potassium voltage-gated channel sub-
family H member 2 (hERG) (encoded by KCNH2)101.
This is a known mechanism through which drugs
can cause QT prolongation and increase risk of cardiac
arrhythmias102 or seizures and is predicted by genetics,
as KCNH2 LOF variants cause long QT syndrome103.
Further support comes from the observation that
patients with certain non-pathogenic variants in KCNH2
are more sensitive to the arrhythmogenic effects of phar-
macological hERG channel inhibition104,105. Although
some other proteins commonly included on secondary
pharmacology screens have human genetic support —
for example, genetic evidence implicates the sodium
voltage-gated channel α subunit 5, Nav1.5 (encoded by
SCN5A) in various cardiac arrhythmias and gastrointes-
tinal symptoms106 — genetics may be useful to identify
additional proteins that could more broadly inform sec-
ondary pharmacology screens100,107. Importantly, as with
on-target safety, the strength of genetic evidence must be
critically evaluated.
An additional consideration for off-target screening
is deciding which phenotypes are of the greatest safety
concern. ICH guidelines (S7A) highlight the cardiovas-
cular, respiratory and central nervous systems as vital
organ systems and consider them the most important
ones to be assessed in safety pharmacology studies.
Therefore, genes with high-impact variants that affect
these systems should be prioritized for screening. A
recent systematic assessment of Mendelian disease genes
identified several proteins that could be added to current
off-target screens based on genetic evidence implicating
them in nervous system and cardiovascular disorders107
(Table 2). Genetics should be considered for use in com-
bination with the more traditional pharmacological
approaches to guide selection of secondary pharmacol-
ogy screens, and increasing amounts of genetic data will
likely highlight more proteins that should be considered
for off-target screening.
Lastly, the secondary pharmacology experimental
design should take into account the drug modality.
A typical screen for binding or enzymatic inhibition
by a small molecule is inappropriate for screening the
off-target effects expected from technologies such as
RNA interference (RNAi) or proteolysis targeting chime-
ras (PROTACs), and the universe of potential off-targets is
different for these technologies108.
In addition to broad assessments of drug specificity,
a counter screen against a protein related to the drug
target is usually used during the development of both
small- and large-molecule therapeutics. Several his-
torical examples are informative (Table 3). One notable
example in which interactions with a protein related to
the drug target caused preclinical toxicity is that of the
β-secretase 1 (BACE1) inhibitors. Ocular toxicity caused
by these inhibitors mirrored symptoms seen in neuronal
ceroid lipofuscinosis — a retinal disease caused by LOF
volume 22 | February 2023 | 151
Reviews

Table 2 | Selected proteins that could be added to current off-target screens

Protein Gene Mutation-associated OMIM Category
phenotype identifier
Cathepsin F CTSF Ceroid lipofuscinosis 615362 Nervous system
Cathepsin C CTSC Papillon–Lefevre syndrome, 602365 Immune system
Haim–Munk syndrome
Glutamate metabotropic receptor 1 GRM1 Spinocerebellar ataxia 614831 Nervous system
Hyperpolarization-activated cyclic HCN4 Sick sinus syndrome, Brugada 163800, Cardiovascular system
nucleotide gated potassium channel 4 syndrome 613123
High-affinity choline transporter SLC5A7 Neuronopathy 158580 Nervous system
(solute carrier 5 member 7)
Thromboxane A synthase 1 TBXAS1 Ghosal haematodiaphyseal 231095 Cardiovascular system
syndrome
Transient receptor potential cation TRPM4 Progressive familial heart 604559 Cardiovascular system
channel subfamily M member 4 block
Selection is based on a systematic analysis of Mendelian disease genes107. OMIM, Online Mendelian Inheritance in Man.

Seed hybridization
Interaction of a short sequence
at the 5′ end of a micro RNA
(miRNA) or a small interfer­
ing RNA (siRNA) with the
3′ untranslated region of a
mRNA, resulting in targeted
degradation. Although this is
the natural behaviour of
endogenous miRNA, it can
be an unintentional off-target
activity of exogenous siRNA
therapeutics.
variants in CTSD, which encodes cathepsin D, an aspar-
tyl protease related to BACE1 (refs.109,110). Modulation
of cathepsin D by BACE1 inhibitors was subsequently
confirmed experimentally111. We suggest that the genet-
ics of proteins related to the drug target be examined
closely and, if any are implicated in phenotypes of seri-
ous safety concern, selectivity of molecules over these
proteins should be assessed using relevant assays.
Besides its utility for designing prospective off-target
screens, genetics can be used retrospectively to identify
off-target interactions mediating drug-related AEs100.
For example, patients taking the Janus kinase 2 (JAK2)
inhibitor fedratinib exhibited neurological symptoms
similar to those of patients with the genetic disor-
der Wernicke’s encephalopathy; a similar syndrome,
thiamine-responsive encephalopathy type 2, is caused by
variants in the thiamine transporter 2 gene solute carrier
family 19 member 3 (SLC19A3)112,113. This led to the dis-
covery that fedratinib inadvertently inhibited thiamine
transporter 2 (ref.114).
Another potential example is that of thalidomide,
which has well-documented teratogenic effects. Recent
studies have linked thalidomide-induced developmen-
tal toxicity to degradation of the transcription factor
spalt-like transcription factor 4 (SALL4)115–117. LOF
variants in SALL4 cause limb shortening, congenital
heart disease and ear and eye abnormalities118,119 pheno­
copying birth defects caused by thalidomide120. When
a drug-associated AE is thought to be mediated by
off-target activities, proteins with genetic evidence of
involvement in that phenotype could be identified and
prioritized for experimental follow-up.
Human genetics resources can also be used to sup-
port evaluation of potential or observed off-target
effects associated with drug modalities other than
small molecules, such as therapeutic nucleic acids.
For example, RNAi-mediated seed hybridization-based
off-target effects were found to be a major driver of
hepatotoxicity observed with N-acetylgalactosamine
(GalNAc)-conjugated siRNAs in rodent toxicity
screens121. Similarly, hybridization-driven off-target
effects of antisense oligonucleotides have been shown
to contribute to hepatotoxicity of these drugs122. These
sequence-based off-target effects were subsequently
designed out during lead optimization. If off-target
sequences are bioinformatically predicted in silico or
experimentally observed for clinical candidate thera-
peutic nucleic acids, the human genetic variation and
phenotypic association of these off-targets should be
further evaluated. In support of this approach, draft
regulatory guidance on developing antisense oligo­
nucleotide and siRNA therapeutics to treat hepatitis B123
states that information on human genetic variation be
considered for each off-target match that is predicted
in silico. Lastly, for genome editing modalities, genetic
background can alter off-target risk124 and its impact
should be evaluated according to the latest FDA draft
guidelines125.
As novel modalities (beyond small molecules with
protein targets) are increasingly employed in drug
development, a genetics-driven approach to off-target
screening allows companies to go beyond traditional
‘druggable’ proteins, but this will require considera-
tion of secondary pharmacology that may be unique to
the modality.
Evaluating carcinogenicity risk
The carcinogenicity risk of novel therapeutics is espe-
cially difficult to evaluate experimentally because tum-
origenesis can be a lengthy process and may be evident
only after chronic exposure. Valuable insights can often
be gained through assessment of potential cancer pheno­
types associated with genetic variation (for example, rare
or common germline variation or somatic mutations) or
genetic modification of drug targets in animal models.
However, molecular, cellular and pathophysio­logical
differences between animal models and humans con-
found the extrapolation of preclinical carcinogenic-
ity findings to humans. Thus, there is an increasing
interest in systematically evaluating the potential for
human genetic associations of drug target genes with
tumorigenicity phenotypes using GWAS and Mendelian
inherited cancer syndrome data, and by lever­aging
databases of recurrent somatic mutations and epige-
netic and expression profiles in human tumour tissues
and cell lines126,127.

152 | February 2023 | volume 22 www.nature.com/nrd

0123456789();:

Cancer driver genes
For example, human genetic data have provided an
The genes in a tumour in which opportunity to assess the potential carcinogenesis risks
somatic mutations have been associated with therapeutic modulation of transform-
positively selected, facilitating ing growth factor-β (TGFβ) signalling. TGFβ cytokines
tumour growth.
likely play a dual role in tumorigenesis depending on
the context, which may be tumour-suppressive or
pro-tumorigenic depending on the phase of tumour
growth128,129. Recent in vitro data in cancer cell lines
shows that perturbation of TGFβ through inactiva-
tion or inhibition of its downstream effectors includ-
ing TGFβ receptor type 2 (TGFBR2) could help to
overcome a strongly inhibitory tumour microenviron-
ment and improve antitumour T cell responses130,131.
However, several studies highlight an association
between mutations in TGFBR1 and TGFBR2 and car-
cinogenic outcomes. Rare, heterozygous LOF mutations
in TGFBR1 are reported in Ferguson–Smith disease, an
autosomal dominant disorder characterized by devel-
opment of multiple self-healing squamous epitheli-
oma, an invasive skin tumour132. In line with this, an
increased incidence of keratoacanthoma (a low-grade
epithelial skin tumour) and skin squamous cell car-
cinoma (SCC) have been reported in recent clinical
trials of oncology drugs that inhibit TGFβ signalling
such as the multi-kinase inhibitor sorafenib 133–135.
Sequencing of two of the sorafenib-induced neo-
plasms identified two somatic missense mutations in
TGFBR1 (ref.136). These results demonstrate the value
of human genetics to inform potential carcinogenic
risks when scrutinizing data provided by in  vitro
cell models130,131.
Genomic features of cancer (for example, frequent
somatic mutations, copy number changes, gene fusions
or expression changes in tumour versus normal) can
provide a causal link between target genes and car-
cinogenicity. Some well-known examples include
somatic activating mutations in BRAF observed in
up to 60% of melanomas134,137, internal tandem dupli-
cation of FLT3 (FLT3-ITD; approximately 25% of all
AML cases)138, BCR–ABL gene fusion in CML139 and
TP53 LOF mutations in numerous cancers140. In addi-
tion, AE outcome data from patients treated with drugs
that target these frequent somatic genetic changes
have provided evidence for additional mechanisms
for carcinogenicity; for example, the paradoxical acti-
vation of CRAF gives rise to skin SCC and other skin
malignancies in melanoma patients treated with BRAF
inhibitors141.
Table 3 | Selected off-target effects for which human genetics has been
informative retrospectively
Drug Primary Off target Adverse event Genetic disease
target associated with
off target
LY2811376, BACE1 CTSD Retinal degradation Neuronal ceroid
AMG-8718 lipofuscinosis
Fedratinib JAK2 SLC19A3 Encephalopathy Encephalopathy
Thalidomide Many SALL4 Limb deformities Duane-radial ray
syndrome
BACE1, β-secretase 1; CTSD, cathepsin D; JAK2, Janus kinase 2; SALL4, spalt-like transcription
factor 4; SLC19A3, solute carrier family 19 member 3.
NaTure RevIeWS | DRuG DISCOveRy
0123456789();:
Reviews
For the genes not yet established as cancer driver genes
or with no obvious link to cancer through Mendelian
cancer syndromes and GWAS, the carcinogenicity
risk could be assessed by correlating their functional
disruption with cancer longitudinal outcome data or
with cell growth or survival synthetic lethal screens, for
example, using RNAi or gene editing approaches142–144.
For example, a somatic mutation in a putative cancer
driver gene would be expected to be associated with
poor survival outcome or more aggressive disease and
would respond to a drug targeting the driver change,
resulting in tumour shrinkage and improvement in
survival145. Similarly, synthetic lethality as demon-
strated by cell death or a dramatic decrease in cell via-
bility can be expected between a driver somatic change
(hotspot somatic mutation, copy number change or over­
expression) in a potential oncogene and its knockout or
knock-down in synthetic lethal screens. Human genetic
association studies for the activation of oncogenes or
inhibition of tumour suppressors are also anticipated to
be leveraged for the evaluation of oncogenic potential
associated with genome engineering-based therapeutics
(for example, CRISPR–Cas9) in which unintended and
nonrandom off-target editing is observed146,147.
The COSMIC database represents a valuable resource
to identify cancer driver genes from a wealth of tumour
sequencing data 148. However, the inherent genetic
instability of tumours can make it difficult to identify
causative as opposed to correlative mutations. Recent
advances in human genome sequencing technologies
such as error-corrected next generation sequencing
and access to large population-based comprehensive
phenotype–genotype databases such as UK Biobank,
23andMe and deCODE genetics149 may provide addi-
tional opportunities to identify associations between
germline or somatic mutations and carcinogenic risk
in non-tumorigenic samples. Such studies could serve
to identify novel associations between GOF or LOF
mutations (either in the germline or somatically) and
tumorigenicity.
Complexities and challenges
The various examples discussed above highlight the
potential power of human genetics to inform drug
safety. Despite these utilities, there are significant chal-
lenges that limit translation of genetic insights into drug
safety risk assessment. First, making causal inferences
from genetic data is challenging and requires a deep
appreciation of statistical caveats. Second, once a causal
association has been strongly implicated between a geno­
type and a phenotype, there are additional limitations in
translating a genetic association to drug development
insights. These limitations fall into three broad areas:
pharmacological translation, gene isoform expression
patterns and disease context.
Pharmacological translation
A key limitation of translating genetic information to
safety assessment is the fact that genetic variation may
or may not accurately mimic the effects of a drug. The
type of genetic variation (GOF versus LOF, heterozygous
versus homozygous, coding versus regulatory, germline
volume 22 | February 2023 | 153
Reviews
versus somatic) will govern the connection between the
gene and phenotype studied. Likewise, the modality, tis-
sue targeting and/or distribution and dosing of a drug
(small molecule versus antibody versus therapeutic pro-
tein) will influence the extent and duration of on- and
off-target modulations.
Thus, there are several major limitations with respect
to interpretation of genetic associations with AEs. In the
first instance, pharmaceuticals might not result in com-
plete inhibition or antagonism of their target over the
entire course of treatment or lifetime, which potentially
would be required to ‘phenocopy’, or mirror the effect of,
LOF genotypes. Therefore, it may be possible to safely
drug a target where the target is partially modulated
(for example, 50–75%) without mimicking the genetic
phenotype. For example, serum lipoprotein a (Lp(a))
level is associated positively with cardiovascular risk and
inversely with risk of type 2 diabetes150,151. Molar con-
centration was identified as the attribute of Lp(a) that
affects risk of cardiovascular diseases but T2D risk was
associated only with low molar concentration of Lp(a)
(<3.5 nM). Thus, pharmacological reduction of serum
Lp(a) concentration in the individuals with the greatest
concentration down to the population median of 14 nM
is predicted to decrease coronary artery disease risk
without increasing T2D risk152. Accordingly, heterozy-
gous LOF variants may represent an appropriate model
to better inform the drug risk–benefit ratio. Although
genetic effects on gene dosage can be helpful for risk
assessment of therapeutics, there is a lack of precision as
well as dynamic variation that hampers interpretation.
Similarly, genetic effects may take months or
years to manifest (for example, RORγ candidia-
sis and mycobacteriosis 153 or MERTK and retinal
phenotypes154,155); translating this lifelong genotype–
phenotype association into drug modulation of a target
gene into a therapeutic window represents a significant
challenge.
It should also be recognized that proteins are fre-
quently modular in nature, composed of multiple
domains that enable broad functionality of the protein,
such as kinase domains, scaffold functions, co-activators,
co-repressor binding sites, subcellular and extracellular
localization motifs, and post-translational modification
sites156,157. Genetic variants that result in complete loss
of expression or early truncation, or splice site changes,
mimic RNA downregulation or protein degradation,
whereas damaging missense variants may mimic func-
tional inhibition or antagonized protein while keep-
ing other functions intact (for example, scaffolding).
Therefore, the phenotypes associated with these different
variants might be significantly different.
For instance, high expression and amplification of the
TMEM16A gene (encodes anoctamin 1 chloride chan-
nel) has been demonstrated in several cancers, reach-
ing 36% in oesophageal cancer158,159. Its blockade with
specific functional inhibitors of TMEM16A-mediated
chloride currents did not alter cell proliferation in
tumour cell lines, whereas PROTACs that decreased
TMEM16A protein levels showed dose-dependent inhi-
bition of cell proliferation160. This difference in efficacy
between PROTACs and a functional inhibitor has been
154 | February 2023 | volume 22 www.nature.com/nrd
0123456789();:
suggested to be due to cell proliferation being driven
by a scaffolding behaviour of TMEM16A (via EGFR
complex formation) rather than its role as a calcium
channel161,162. A similar discrepancy between PROTACs
and other modalities has been shown for inhibition of
SMARCA2/4 (ref.163).
Systematic drug target gene homology assessments
during early development that take into account human
genetic variation of a drug target within the intended
patient population and genetic variation in animal
breeds or strains support drug design and optimal spe-
cies selection for both pharmacology and toxicology
assessments. When the mechanism of a drug is not mod-
ulating the activity of a specific protein, but instead tar-
geted killing of a cell type, silencing of a mRNA molecule
or editing of genomic DNA itself, different principles are
likely to apply.
For example, in the case of a drug such as a bispecific
T cell engager (BiTE) that targets a certain type of cell
for killing based on surface marker expression, genetic
associations with that cell surface marker are likely not
as relevant to safety evaluation as the expression pat-
tern of the marker across cell types, which will be much
more informative. The impact of human genetic varia-
tion also needs to be carefully considered for the safety
assessment of cell and gene therapies, including genome
editing specificity (for example, variants creating novel
or higher potency off-target loci). Genetic variation (for
example, in DNA repair pathways) might influence viral
vector integration site patterns, although this has yet to
be reported.
Gene expression patterns
Cross-tissue expression patterns are an important medi-
ator for translating genetic associations to drug safety.
For example, if the genetic variant affects an isoform that
is not expressed in a given tissue, it cannot be expected to
be predictive of pharmacological effects on that tissue. In
addition, the characteristics of the therapeutic agent
in question such as mechanism of action, dose, chem-
ical properties and route of administration will affect
whether a genetic association is relevant to drug safety.
For example, if a drug is targeted specifically to tissue
in the liver and cannot cross the blood–brain barrier,
a neurological genetic association that is known to be
mediated solely by behaviour of the target protein in the
brain will not be relevant to evaluation of the drug.
Disease context
Some disease states may be promoted as a consequence
of excessive and/or sustained expression of genetic net-
works (for example, sustained cytokine expression in
autoimmune diseases such as rheumatoid arthritis164).
In such a setting pharmacological regulation results in
normalization, that is, a disease-relevant pathway may
be operating at multiple folds of normal levels, and
drug treatment inhibits this to the basal level, which
would not necessarily equate to the effects of the drug
or genetic variant in a normal non-pathogenic context.
One of the major advantages of Mendelian genetics
in translational sciences is the established link between
the variants and the gene function (with the caveat that

the level of confidence in this link relies on the quality
of the supporting literature). However, many Mendelian
disorders are characterized by complex and sometimes
variable presentations with many clinical features, which
makes it challenging to translate into predictable AEs.
Despite these limitations, a recent retrospective study
demonstrates that, across drug targets with reported
disease-causing variants, a decreased risk for AEs that
manifest in a given organ system is observed if the organ
system is not affected in the Mendelian disorder13.
Despite the increasing role of human genetics in
target validation, choosing a target free of genetically
predicted liabilities is still not a guarantee that AEs will
not be observed clinically, owing to the aforementioned
caveats. For this reason, human genetics should be
thought of as a tool for hazard identification, but can
never be expected to provide a complete picture of the
potential risks of a novel therapeutic agent.
Diversity and ethical considerations
Most human genetic research has been performed in
populations of European ancestry, which raises con-
cerns for the use of genetics for drug development165–167.
This bias in the data makes GWAS most powered to
enumerate risk alleles in populations of European
ancestry, resulting in comparatively poorer AE risk
prediction in other populations. As a matter of equity
but also to increase power for target validation to ben-
efit drug development in general, it is imperative for
genetic research to prioritize expanding data collection
in populations of non-European ancestry168.
In the USA, diversifying participation in genetic studies
and expanding the health benefits of genetic informa-
tion has been a stated goal of the All Of Us biobanking
project led by the NIH169. Expanding study popula-
tions will also involve performing research in low- and
middle-income countries, which prompts ethical con-
cerns around protecting the interests of research parti­
cipants and local scientists and benefit sharing170,171.
In Africa, both the publicly funded initiative H3Africa172
and the private venture 54Gene173 have centred local
researchers, institutions and economic development in
their genetic studies.
Resources and suggested best practices
A wide range of publicly available databases collecting
human genetics data can be used in evaluation of the
safety of intended targets or off-targets. In this section,
we aim to provide a list of these resources, highlight
common pitfalls to be avoided in the interpretation of
the data, and suggest a framework of best practices to
follow when using human genetics and genomics data
in early safety assessment of a target gene (Fig. 3).
A large array of human genomics databases are
available with information to aid variant interpretation,
including population sequencing results, variation tol-
erance scores, gene-level information including gene–
disease associations, and databases relating to GWAS
and PheWAS results and their interpretation. Selected
examples of each of these are described in Table 4.
Given the breadth and depth of human genetics data
available, statistical rigour is required during interrogation
NaTure RevIeWS | DRuG DISCOveRy
0123456789();:
Reviews
and interpretation, to avoid pitfalls. Critically evaluat-
ing the validity of these data is essential; for example,
candidate-gene association studies in the literature have
a high false-positive rate174. An understanding of sta-
tistical pitfalls is important, especially when interpret-
ing the results of GWAS or PheWAS. For example, the
P value threshold for a significant result should consider
the number of both independent variants and pheno-
types tested. Prior hypotheses about variants believed to
functionally mimic a therapeutic intervention and the
phenotypes likely to be affected can be used to limit
the search space and increase power, although caution
is recommended here. Below is a workflow suggesting
how genetic data could be considered in a typical early
target evaluation.
Determine the breadth of the search
If specific variants functionally linked to the gene are
associated with a phenotype relevant to the therapeutic
indication, this strengthens the plausibility of the rele-
vance of other phenotype associations to drug safety. If
there is no association with the therapeutic indication,
the reasons for a lack of association should be critically
considered with respect to the therapeutic hypothesis.
Are there no well-powered genetic data for the indi-
cation of interest? Does the available variation affect
the gene only in tissue contexts that are not relevant
to the indication?
Investigate germline genetic evidence for other
phenotypes
Mendelian genetics. Search the Mendelian gene litera-
ture, for example, Online Mendelian Inheritance in Man
(OMIM)175. The highest-confidence gene–phenotype
relationships are listed in OMIM as relationships where
“the molecular basis of the disorder is known.” Critically
evaluate the validity of these relationships using mod-
ern best practices176 and resources such as ClinGen177
and ClinVar178. Be especially critical of ‘candidate-gene’
or ‘candidate-SNP’ publications. Advantages of using
Mendelian diseases are that the causal gene is usually
clear and effects are usually strong; disadvantages are
that sample sizes of affected individuals are usually low.
Some integrative databases such as DisGeNet (https://
www.disgenet.org/dbinfo) combine Mendelian and
complex associations. Scrutinize the Mendelian litera-
ture for the phenotypic spectrum involved in the syn-
drome. If phenotypes are secondary to developmental
alterations, they may be less informative to drug varia-
tion. Estimate the power to detect or exclude additional
effects by looking at the sample size, penetrance and
variable expressivity of the syndrome.
GWAS and rare-variant association studies. Search the
GWAS literature, including the NHGRI-EBI GWAS
catalogue in the ‘Reported Genes’ and ‘Mapped Genes’
columns179. Be especially critical of associations that do
not reach genome-wide significance (P < 5 × 10−8) or that
did not replicate in a larger study in the catalogue of the
same phenotype. Advantages of using common SNP
association study results include the large sample size of
individuals carrying the variant, allowing high sensitivity
volume 22 | February 2023 | 155
Reviews

a Mendelian genetics
• Direction of effect: LOF or GOF? Relevance to therapeutic hypothesis
Disease-causing (inhibition vs agonism/activation)?
variants implicated in • Clinical features and organ systems affected?
Mendelian disorders • Dominant or recessive inheritance? If recessive, medical information
available for heterozygous carriers?
• Functional assessment (molecular and cellular effects) of variants?
• Disorder due to developmental malformations: if, yes relevance?

Human biallelic LOF data

Individuals with rare
biallelic LOF variants
• Medical history: comprehensive longitudinal clinical information
available?
• Clinical assessment: can medical observations be attributed to the
variants?
• Functional assessment (molecular and cellular effects) of variants?

Genetic association studies

Target Association in region GWAS
gene in or near gene • Statistical significance/replication?
• Direction of association (increased/decreased risk)?
• Phenotypes and affected organ systems?
• Causality assessment: association driven by variant affecting gene
function/expression?

PheWAS
• Association with other phenotypes?
• Statistical significance/replication?
Variant associated with • Direction of association (increased/decreased risk)?
indication • Phenotypes and affected organ system?
• Causality assessment/regional association context: associations driven
by variant?
Also consider • Association independent of indication association?
• Direct interacting
proteins Rare variant association studies
• Relevant proteins in • Statistical significance? Number of carriers? Replication?
pathway Association driven by • Phenotypes and affected organ systems?
• Protein family members rare variants • Effect of associated variants (partial LOF, GOF)?
• Direction of association (increased/decreased risk)?
• Causality assessment: association driven by rare variants affecting gene?

b
PheWAS of variant V: association with
phenotype P Identification of most likely causal variants
15
GWAS Yes Bayesian Set of most likely
−log10 P value

PheWAS variant summary fine-mapping causal variants

10 statistics No LD-based proxy SNP1 SNP2 SNP3 SNP4 SNP5
available Lead variant set definition Lead
5

Annotation of most likely causal variants

0
74.55 74.60 74.65 74.70 74.75 Disease-relevant pLOF/non-synonymous Non-coding/epigenetic
Chromosome 5 (Mb) functional assay variant annotations
Tissue 1 Tissue 2
GWAS regional association with phenotype P Gene 1 Gene 1 Gene 1
Gene 2 x Gene 2 Gene 2 x
15 PheWAS variant 100 Gene 3 Gene 3 Gene 3
Lead
variant 80
−log10 P value

10 60 Tissues relevance?
40 Interrogation of eQTL and pQTL studies
5
20
GWAS
0 0 summary Conditional Supporting eQTL/
Yes analysis and Posterior
74.55 74.60 74.65 74.70 74.75 pQTL evidence
statistics probability
Chromosome 5 (Mb) Significant available colocalization
eQTL/pQTL Yes anlysis
Gene 1 Gene 3 in tissue/cell Tissues
Gene 2 eQTL/pQTL 1 2 3 4
type in region summary GWAS
No LD Gene 1
statistics lead
available variant Gene 2 x
Gene 3
eQTL
lead
variant Tissue relevance?

156 | February 2023 | volume 22 www.nature.com/nrd

0123456789();:

◀ Fig. 3 | Proposed framework for incorporating human germline genetics in target
safety reviews. a | Human genetics provides a powerful toolkit to complement
target safety reviews and enables investigators to explore the role of target and
off-target genes in a large spectrum of drug safety-relevant phenotypes in humans,
using hypothesis-free or hypothesis-driven approaches. Comprehensive germline
genetic assessment includes interrogation of genome-wide association studies (GWAS),
phenome-wide association studies (PheWAS), rare variant associations, Mendelian
genetics and exploration of individuals with very rare high-impact variants. Important
considerations are required to establish the relevance of the genetic findings, including
robustness (quality of the data, report or analysis, statistical significance, replication,
statistical power), direction of effect and gene causal assessment. b | To interpret
association results from GWAS or PheWAS, a set of downstream analyses are warranted
to establish the causal link between the target or off-target gene and the phenotype
identified in the association analysis. Commonly used methods include identification
and annotation of the most likely causal variant for the association, and interrogation of
expression quantitative trait loci (eQTL) or protein quantitative trait loci (pQTL) to
investigate whether the phenotype association can be attributed to a role of the GWAS
causal variant on the target or off-target gene expression or protein levels in relevant
tissue or cell type. LD, linkage disequilibrium; LOF, loss of function; SNP, single nucleotide
polymorphism.
to modest effects on phenotype; and the sharing of
common variation worldwide, allowing the possibility
of replication in multiple populations. Disadvantages
are the challenges in causal gene assignment owing to
the regulatory rather than coding mechanisms usually
responsible and confounding owing to linkage disequi-
librium (LD). Determine level of confidence that the
target gene, rather than another nearby gene, is mecha-
nistically responsible for the association signal from the
common variant, as detailed in the following list.
• Inspect the regional pattern of LD by looking at nearby
GWAS results in the browser and in LD databases,
for example, LDLink180 to ensure that the association
is not a ‘shadow’ — for example, a signal driven by
LD with a much stronger causal association nearby.
Determine whether there are multiple independent
signals. Look for genes on the haplotype block.
• Look for candidate causal variants: identify the credi-
ble set of potential causal variants, using fine-mapping
approaches if GWAS summary data are available, or
by searching for variants in high LD if no summary
statistics data are available. Search for annotations that
link variants in credible set to genes (protein-coding
variant, or variants affecting regulatory elements)
using browsers, for example, OpenTargets Genetics,
RegulomeDB181 and HaploReg182.
• Look for molecular QTLs: protein QTL (pQTL) evi-
dence, available in the GWAS browser, and eQTL
evidence in browsers such as the GTEx Portal183 and
eQTLGen184. If an additional association signal has
been found at the locus from these sources, perform
colocalization analysis to confirm a shared mecha-
nism between the disease association and QTL signal,
using a method such as summary-based Mendelian
randomization (SMR)185 or coloc48. Consider the rel-
evance of the QTL tissue or cell type to the GWAS
phenotype.
• Formally test MR assumptions using an algorithm
such as MR-Egger74.
• Look for experimental validation in the literature,
paying close attention to whether alternative causal
genes identified above were tested.
NaTure RevIeWS | DRuG DISCOveRy
0123456789();:
Reviews
Search for other population-scale association studies
not included in the GWAS catalogue, such as reports of
individual results from GWAS where the genome-wide
hits were not reported, exome-sequencing studies,
rare-variant association study databases such as the
AstraZeneca PheWAS Portal 53 and GeneBass, and
custom array designs. Be mindful of the statistical
thresholds used.
If a putative functional variant is available (based on
association with the indication phenotype, or based
on other strong functional evidence), search for it in GWAS
and PheWAS databases, for example, PhenoScanner186,
GWAS ATLAS187, BIG188, the PheWAS Catalogue189 and
GRASP190. Apply a P value threshold (that is, multiple
testing correction) appropriate to the number of pheno-
types being tested. If there is a prior concern from pre-
clinical or clinical evidence, this can narrow the search
space and lower the significance threshold, relative to
searching phenome-wide. If a pleiotropic association
is identified, determine the likelihood that the associ-
ation is driven by the same causal variant or gene (true
pleiotropy) using the molecular QTL and LD databases
described above. If the variant is also associated with a
phenotype related to the indication, use colocalization
methods to confirm a shared mechanism between the
indication and the pleiotropic association.
Evaluate carcinogenicity risk
To understand whether the gene may be an oncogene
or tumour suppressor, search for the gene in databases
of somatic variation and expression in cancer, such as
the COSMIC Cancer Gene Census148, cBioPortal/The
Cancer Genome Atlas (TCGA)191 or IntOGen192.
Evaluate LOF data
If the drug mechanism involves decreasing the function
or expression of a protein, inspect patterns of LOF var-
iation in population cohorts such as the gnomAD por-
tal and the Bravo portal to TOPMED data193, FinnGen,
Genes and Health and Pakistan Genomic Resource,
applying the principles outlined by Minikel et al.95 to
carefully inspect the pattern of LOFs across exons and
their quality scores. If phenotype data are available for
heterozygous or complete knockouts identified through
biobank studies, examine these individuals’ phenotype
information. If enough carriers of a specific variant are
available, perform a candidate-variant PheWAS using an
algorithm optimized for rare variants such as SAIGE194. If
multiple rare LOF variants are identified throughout the
gene, perform a rare-variant association study using an
aggregation method such as SKAT195, SAIGE-GENE196,
or collapsing analysis197.
Future perspectives and conclusions
Human genetics studies along with clinical sample bio-
banks have opened many new avenues of research in
drug discovery and development. Indeed, in the com-
ing years vast amounts of genotype–phenotype data are
expected to be generated from biobank repositories as a
result of sequencing millions of exomes and genomes.
These data will undoubtedly reveal novel human genetic
variants. In addition, populations of non-European
volume 22 | February 2023 | 157
Reviews

Table 4 | Selected human genomics resources that might inform target safety assessment
Resource Description Website Refs.
1000 Genomes Population sequencing data (n = 2,504, WES and WGS) https://www.genome. 205

gov/27528684/1000-genomes-project
100,000 Genomes Consortium of genotype–phenotype interactions from UK https://www.genomicsengland.co.uk/ 206

Project population about-genomics-england/the-100000-

genomes-project/
AZ PheWAS Portal Repository of gene–phenotype associations generated using https://azphewas.com/ 53

exome-sequencing and phenotype data from the UK Biobank

BIG Database of GWAS results (brain imaging and other traits from https://big.stats.ox.ac.uk 188

UK Biobank)
BioGRID Database of protein, chemical and genetic interactions https://thebiogrid.org/ 207

Bravo TOPMed Population sequencing data https://bravo.sph.umich.edu/freeze5/hg38/ 193

ClinGen Curated database of clinically relevant variants https://www.clinicalgenome.org/ 177

ClinVar Database of clinically relevant variants https://www.ncbi.nlm.nih.gov/clinvar/ 178

COSMIC Cancer Gene Database of genes implicated in cancer https://cancer.sanger.ac.uk/census 148

Census
Decipher Database of clinically relevant variants https://decipher.sanger.ac.uk/ 208

DisGeNet Database of genes and variants associated with human https://www.disgenet.org/ 209

diseases
East London Genes & Population sequencing data (only list of pLOF variants, not http://www.genesandhealth.org/ 210

Health phenotypes, are freely available)

eQTLGen Database of blood eQTLs https://www.eqtlgen.org/ 184

FinnGen Consortium of genotype–phenotype interactions from Finnish https://www.finngen.fi/en/access_results 211

population
GeneATLAS Database of genotype–phenotype associations from UK http://geneatlas.roslin.ed.ac.uk/ 212

Biobank
GeneBass Repository of gene–phenotype associations generated using https://genebass.org/
exome-sequencing and phenotype data from the UK Biobank
gnomAD Population sequencing data and LOF intolerance scores http://gnomad.broadinstitute.org/ 92

GRASP Curated database of GWAS results https://grasp.nhlbi.nih.gov/Overview.aspx 190

GTEx Portal Database of gene expression and eQTLs across many tissues https://www.gtexportal.org/home/ 213

GWAS ATLAS Database of GWAS summary statistics https://atlas.ctglab.nl 187

GWAS Catalog Curated database of GWAS results https://www.ebi.ac.uk/gwas/ 214

HaploReg Tool to find noncoding regulatory features that overlap given http://compbio.mit.edu/HaploReg/ 182

SNP and LD SNPs

HGMD Database of clinically relevant variants (not freely available https://www.hgmd.cf.ac.uk/ac/index.php 215

to commercial organizations)
LDlink Calculations of linkage disequilibrium between https://ldlink.nci.nih.gov/ 180

1000 Genomes SNPs

MGI PheWeb Database of GWAS results from the Michigan Genomics https://pheweb.sph.umich.edu/
Initiative and UK Biobank
Missense Tolerance Ratio Regional variation tolerance scores http://mtr-viewer.mdhs.unimelb.edu.au 216

OMIM Curated database of genetic disease https://omim.org/ 217

Open Targets Database of genetic information in the context of drug https://www.opentargets.org/ 218

target validation
Orion Noncoding variation tolerance scores http://www.genomic-orion.org/ 219

PhenoScanner Database of genotype–phenotype associations http://www.phenoscanner.medschl.cam.ac.uk/ 220

PheWAS Catalogue Database of PheWAS results from BioVU https://phewascatalog.org/ 189

RegulomeDB Tool to find noncoding regulatory features that overlap https://www.regulomedb.org/ 181

given SNP
RVIS Gene-level variation tolerance score https://genic-intolerance.org/ 91

The cBioPortal for Curated somatic mutation database https://www.cbioportal.org/ 191,221

Cancer Genomics
UniProt Database of protein information https://www.uniprot.org/ 222

eQTL, expression quantitative trait loci; GWAS, genome-wide association studies; LD, linkage disequilibrium; LOF, loss of function; OMIM, Online Mendelian
Inheritance in Man; PheWAS, phenome-wide association studies; SNP, single nucleotide polymorphism; WES, whole exome sequencing; WGS, whole genome
sequencing.

158 | February 2023 | volume 22 www.nature.com/nrd

0123456789();:
 Reviews

ancestry are currently very under-represented in most
human genetics studies, resulting in reduced under-
standing of the range and functional consequences of
genetic variation within these populations198. Efforts by
the human genetics community will continue to redress
this population disparity. Furthermore, the expansion
of rare variant discovery into the noncoding portions
of the genome will provide an even greater opportunity
for the identification of candidate protective alleles for
various disease states.
Large-scale deeply phenotyped biobanks are being
created that contain arrays of biomarkers and catalogue
various disease states and subtypes. The development of
new genotyping technologies and computational meth-
odologies such as deep mutational scanning (DMS) and
the priority index (Pi) pipeline199 may provide mecha-
nistic insights, potentially predict the functional or clin-
ical impact of these variants of uncertain significance
and prioritize the genes and drug targets responsible for
GWAS signals200.
Recent advances in human epigenome mapping have
provided insights into the importance of human non-
coding DNA regulatory elements such as enhancers for
controlling cell-type-specific genome functions201,202.
Most human disease-associated genetic variation
maps to the regulatory genome, and many GWAS var-
iants directly affect transcription factor occupancy201.
Thus, increased attention is being given to integrating
cell- and tissue-specific epigenomic profiles, including
large-scale epigenome-wide association studies (EWAS),
for the characterization of genotype–phenotype relation­
ships203,204. The unprecedented access to genome-centric
phenotypic data, biobanks and computational tools is
enabling the identification of new drug target candi-
dates and providing deeper insights into disease aetiol-
ogy, including multifactorial or complex diseases with
multiple common genetic variants.
Human genetics is becoming an increasingly impor-
tant approach for early target de-risking during drug
development. How these genetically defined safety
concerns should be considered in terms of informing
preclinical and clinical development, patient risk–
benefit and patient management strategies remains to be
further defined. Drug targets and off-targets for which
predicted safety signals are identified by human genetic
data could be included in secondary pharmacology
screening or inform the strategy and design of early pre-
clinical safety screens or clinical monitoring strategies.
They may also provide a strategy of generating testable
hypotheses that can inform a proactive approach to fur-
ther assessing drug safety at early stages of development.
Furthermore, human genetic data from biobanks can
provide biological plausibility or implausibility for the
relevance of drug AEs reported from clinical develop-
ment and pharmacovigilance, and provide evidence for
post-marketing regulatory review. Unanticipated safety
signals that occur during later phase drug development,
or that are reported from pharmacovigilance, may be
further investigated for possible genetic associations to
support patient stratification and selection.
The use of human genetics in support of drug safety
assessment is gaining significant momentum and is com-
plementary to informatics evaluations of known and
putative target safety concerns9. However, genetics-based
evidence from emerging MR and biobank studies will
need to be better integrated with various preclinical and
clinical safety data resources. Pre-competitive sharing
of genetic and drug safety data, national initiatives for
biobanks, and industry and academic partnerships
(for example, OpenTargets) will be important stimuli for
further evaluation of broader drug safety assessment appli-
cations. We outline here some suggested best practices for
the rigorous interpretation of potential drug target-related
genotype–phenotype associations that should help guide
future discussions on the regulatory value of human
genetics-based safety assessment data, potentially facili-
tated by a cross-sector working group including industry,
academic and regulatory stakeholders.
Published online 19 October 2022

1. Visscher, P. M. et al. 10 years of GWAS discovery:
biology, function, and translation. Am. J. Hum. Genet.
101, 5–22 (2017).
2. Chong, J. X. et al. The genetic basis of Mendelian
phenotypes: discoveries, challenges, and
opportunities. Am. J. Hum. Genet. 97, 199–215
(2015).
3. Plenge, R. M., Scolnick, E. M. & Altshuler, D.
Validating therapeutic targets through human
genetics. Nat. Rev. Drug Discov. 12, 581–594 (2013).
4. Kamb, A., Harper, S. & Stefansson, K. Human genetics
as a foundation for innovative drug development.
Nat. Biotechnol. 31, 975–978 (2013).
5. Nelson, M. R. et al. The support of human genetic
evidence for approved drug indications. Nat. Genet.
47, 856–860 (2015).
6. King, E. A., Davis, J. W. & Degner, J. F. Are drug
targets with genetic support twice as likely to be
approved? Revised estimates of the impact of genetic
support for drug mechanisms on the probability of
drug approval. PLoS Genet. 15, e1008489 (2019).
7. Monticello, T. M. et al. Current nonclinical testing
paradigm enables safe entry to first-in-human
clinical trials: the IQ consortium nonclinical to clinical
translational database. Toxicol. Appl. Pharmacol.
334, 100–109 (2017).
8. Waring, M. J. et al. An analysis of the attrition
of drug candidates from four major pharmaceutical
companies. Nat. Rev. Drug Discov. 14, 475–486
(2015).
9. Roberts, R. A. Understanding drug targets: no
such thing as bad news. Drug Discov. Today 23,
1925–1928 (2018).
10. Okada, Y. et al. Genetics of rheumatoid arthritis
contributes to biology and drug discovery. Nature
506, 376–381 (2014).
11. Diogo, D. et al. Phenome-wide association studies
across large population cohorts support drug
target validation. Nat. Commun. 9, 4285
(2018).
12. Jerome, R. N. et al. Leveraging human genetics
to identify safety signals prior to drug marketing
approval and clinical use. Drug Saf. 43, 567–582
(2020).
13. Nguyen, P. A., Born, D. A., Deaton, A. M., Nioi, P.
& Ward, L. D. Phenotypes associated with genes
encoding drug targets are predictive of clinical trial
side effects. Nat. Commun. 10, 1579 (2019).
14. Cao, J. et al. Targeting acyl-CoA:diacylglycerol
acyltransferase 1 (DGAT1) with small molecule
inhibitors for the treatment of metabolic diseases.
J. Biol. Chem. 286, 41838–41851 (2011).
15. Denison, H. et al. Diacylglycerol acyltransferase 1
inhibition with AZD7687 alters lipid handling and
hormone secretion in the gut with intolerable side
effects: a randomized clinical trial. Diabetes Obes.
Metab. 16, 334–343 (2014).
16. Haas, J. T. et al. DGAT1 mutation is linked to a
congenital diarrheal disorder. J. Clin. Invest. 122,
4680–4684 (2012).
17. Tegeder, I. et al. GTP cyclohydrolase and
tetrahydrobiopterin regulate pain sensitivity and
persistence. Nat. Med. 12, 1269–1277 (2006).
18. Bonafe, L., Thony, B., Penzien, J. M., Czarnecki, B.
& Blau, N. Mutations in the sepiapterin reductase
gene cause a novel tetrahydrobiopterin-dependent
monoamine-neurotransmitter deficiency without
hyperphenylalaninemia. Am. J. Hum. Genet. 69,
269–277 (2001).
19. Thony, B. & Blau, N. Mutations in the
BH4-metabolizing genes GTP cyclohydrolase I,
6-pyruvoyl-tetrahydropterin synthase, sepiapterin
reductase, carbinolamine-4a-dehydratase, and
dihydropteridine reductase. Hum. Mutat. 27,
870–878 (2006).
20. Booth, B. Painful Truth: The Successful Failure Of
A Biotech Startup. Forbes (17 November 2017);
https://www.forbes.com/sites/brucebooth/2017/
11/17/painful-truth-successful-failure-of-a-biotech-
startup
21. Sharfe, N., Dadi, H. K., Shahar, M. & Roifman, C. M.
Human immune disorder arising from mutation
of the alpha chain of the interleukin-2 receptor.
Proc. Natl Acad. Sci. USA 94, 3168–3171 (1997).
22. Caudy, A. A., Reddy, S. T., Chatila, T., Atkinson, J. P.
& Verbsky, J. W. CD25 deficiency causes an immune
dysregulation, polyendocrinopathy, enteropathy,
X-linked-like syndrome, and defective IL-10 expression
from CD4 lymphocytes. J. Allergy Clin. Immunol. 119,
482–487 (2007).

NaTure RevIeWS | DRuG DISCOveRy volume 22 | February 2023 | 159

0123456789();:
Reviews
23. Goudy, K. et al. Human IL2RA null mutation mediates
immunodeficiency with lymphoproliferation and
autoimmunity. Clin. Immunol. 146, 248–261 (2013).
24. Tang, Q. et al. Central role of defective interleukin-2
production in the triggering of islet autoimmune
destruction. Immunity 28, 687–697 (2008).
25. Prasad, N. et al. Is basiliximab induction, a novel risk
factor for new onset diabetes after transplantation for
living donor renal allograft recipients? Nephrology 19,
244–250 (2014).
26. Lo, B. et al. Patients with LRBA deficiency show
CTLA4 loss and immune dysregulation responsive to
abatacept therapy. Science 349, 436–440 (2015).
27. Lopez-Herrera, G. et al. Deleterious mutations in
LRBA are associated with a syndrome of immune
deficiency and autoimmunity. Am. J. Hum. Genet. 90,
986–1001 (2012).
28. Alangari, A. et al. LPS-responsive beige-like anchor
(LRBA) gene mutation in a family with inflammatory
bowel disease and combined immunodeficiency.
J. Allergy Clin. Immunol. 130, 481–488.e482 (2012).
29. Charbonnier, L. M. et al. Regulatory T-cell deficiency
and immune dysregulation, polyendocrinopathy,
enteropathy, X-linked-like disorder caused by loss-of-
function mutations in LRBA. J. Allergy Clin. Immunol.
135, 217–227 (2015).
30. Bertrand, A., Kostine, M., Barnetche, T., Truchetet, M. E.
& Schaeverbeke, T. Immune related adverse events
associated with anti-CTLA-4 antibodies: systematic
review and meta-analysis. BMC Med. 13, 211 (2015).
31. Bouhassira, E. E. et al. An alanine-to-threonine
substitution in protein 4.2 cDNA is associated
with a Japanese form of hereditary hemolytic anemia
(protein 4.2NIPPON). Blood 79, 1846–1854 (1992).
32. Bruce, L. J. et al. Absence of CD47 in protein
4.2-deficient hereditary spherocytosis in man: an
interaction between the Rh complex and the band 3
complex. Blood 100, 1878–1885 (2002).
33. Jiang, Z., Sun, H., Yu, J., Tian, W. & Song, Y. Targeting
CD47 for cancer immunotherapy. J. Hematol. Oncol.
14, 180 (2021).
34. Jin, Y. et al. Genome-wide association studies of
autoimmune vitiligo identify 23 new risk loci and
highlight key pathways and regulatory variants.
Nat. Genet. 48, 1418–1424 (2016).
35. Petukhova, L. et al. Genome-wide association study
in alopecia areata implicates both innate and adaptive
immunity. Nature 466, 113–117 (2010).
36. Betz, R. C. et al. Genome-wide meta-analysis in
alopecia areata resolves HLA associations and reveals
two new susceptibility loci. Nat. Commun. 6, 5966
(2015).
37. Choi, L. et al. Evaluating statistical approaches to
leverage large clinical datasets for uncovering
therapeutic and adverse medication effects.
Bioinformatics 34, 2988–2996 (2018).
38. Rao, A. S. et al. Large-scale phenome-wide association
study of PCSK9 variants demonstrates protection
against ischemic stroke. Circ. Genom. Precis. Med. 11,
e002162 (2018).
39. Ference, B. A. et al. Mendelian randomization study
of ACLY and cardiovascular disease. N. Engl. J. Med.
380, 1033–1042 (2019).
40. Interleukin-6 Receptor Mendelian Randomisation
Analysis (IL6R MR) Consortium The interleukin-6
receptor as a target for prevention of coronary heart
disease: a mendelian randomisation analysis. Lancet
379, 1214–1224 (2012).
41. Walker, V. M., Davey Smith, G., Davies, N. M. &
Martin, R. M. Mendelian randomization: a novel
approach for the prediction of adverse drug events
and drug repurposing opportunities. Int. J. Epidemiol.
46, 2078–2089 (2017).
42. Swerdlow, D. I. et al. HMG-coenzyme A reductase
inhibition, type 2 diabetes, and bodyweight: evidence
from genetic analysis and randomised trials. Lancet
385, 351–361 (2015).
43. Alghamdi, J. et al. Risk of neuropsychiatric adverse
effects of lipid-lowering drugs: a Mendelian
randomization study. Int. J. Neuropsychopharmacol.
21, 1067–1075 (2018).
44. Interleukin 1 Genetics Consortium. Cardiometabolic
effects of genetic upregulation of the interleukin 1
receptor antagonist: a Mendelian randomisation
analysis. Lancet Diabetes Endocrinol. 3, 243–253
(2015).
45. Bush, W. S., Oetjens, M. T. & Crawford, D. C.
Unravelling the human genome-phenome
relationship using phenome-wide association studies.
Nat. Rev. Genet. 17, 129–145 (2016).
46. Denny, J. C., Bastarache, L. & Roden, D. M.
Phenome-wide association studies as a tool to
160 | February 2023 | volume 22 www.nature.com/nrd
advance precision medicine. Annu. Rev. Genomics
Hum. Genet. 17, 353–373 (2016).
47. Duffy, A. et al. Tissue-specific genetic features inform
prediction of drug side effects in clinical trials. Sci. Adv.
6, eabb6242 (2020).
48. Giambartolomei, C. et al. Bayesian test for
colocalisation between pairs of genetic association
studies using summary statistics. PLoS Genet. 10,
e1004383 (2014).
49. Evans, D. M. & Davey Smith, G. Mendelian
randomization: new applications in the coming age
of hypothesis-free causality. Annu. Rev. Genomics
Hum. Genet. 16, 327–350 (2015).
50. Szustakowski, J. D. et al. Advancing human genetics
research and drug discovery through exome
sequencing of the UK Biobank. Nat. Genet. 53,
942–948 (2021).
51. Bycroft, C. et al. The UK Biobank resource with
deep phenotyping and genomic data. Nature 562,
203–209 (2018).
52. Backman, J. D. et al. Exome sequencing and analysis
of 454,787 UK Biobank participants. Nature 599,
628–634 (2021).
53. Wang, Q. et al. Rare variant contribution to human
disease in 281,104 UK Biobank exomes. Nature 597,
527–532 (2021).
54. Millard, L. A. et al. MR-PheWAS: hypothesis
prioritization among potential causal effects of body
mass index on many outcomes, using Mendelian
randomization. Sci. Rep. 5, 16645 (2015).
55. Gill, D. et al. Associations of genetically determined
iron status across the phenome: a Mendelian
randomization study. PLoS Med. 16, e1002833
(2019).
56. Zheng, J. et al. Phenome-wide Mendelian randomization
mapping the influence of the plasma proteome on
complex diseases. Nat. Genet. 52, 1122–1131 (2020).
57. Smith, G. D. & Ebrahim, S. ‘Mendelian
randomization’: can genetic epidemiology contribute
to understanding environmental determinants of
disease? Int. J. Epidemiol. 32, 1–22 (2003).
58. Abifadel, M. et al. Mutations in PCSK9 cause
autosomal dominant hypercholesterolemia.
Nat. Genet. 34, 154–156 (2003).
59. Timms, K. M. et al. A mutation in PCSK9 causing
autosomal-dominant hypercholesterolemia in a Utah
pedigree. Hum. Genet. 114, 349–353 (2004).
60. Di Taranto, M. D. et al. Identification and in vitro
characterization of two new PCSK9 gain of
function variants found in patients with familial
hypercholesterolemia. Sci. Rep. 7, 15282 (2017).
61. Cohen, J. et al. Low LDL cholesterol in individuals
of African descent resulting from frequent nonsense
mutations in PCSK9. Nat. Genet. 37, 161–165
(2005).
62. Cohen, J. C., Boerwinkle, E., Mosley, T. H. Jr. &
Hobbs, H. H. Sequence variations in PCSK9, low LDL,
and protection against coronary heart disease.
N. Engl. J. Med. 354, 1264–1272 (2006).
63. Kathiresan, S. et al. Six new loci associated with
blood low-density lipoprotein cholesterol, high-density
lipoprotein cholesterol or triglycerides in humans.
Nat. Genet. 40, 189–197 (2008).
64. Willer, C. J. et al. Newly identified loci that influence
lipid concentrations and risk of coronary artery
disease. Nat. Genet. 40, 161–169 (2008).
65. Hooper, A. J., Marais, A. D., Tanyanyiwa, D. M.
& Burnett, J. R. The C679X mutation in PCSK9 is
present and lowers blood cholesterol in a Southern
African population. Atherosclerosis 193, 445–448
(2007).
66. Schmidt, A. F. et al. PCSK9 genetic variants and risk
of type 2 diabetes: a Mendelian randomisation study.
Lancet Diabetes Endocrinol. 5, 97–105 (2017).
67. Williams, D. M., Finan, C., Schmidt, A. F., Burgess, S.
& Hingorani, A. D. Lipid lowering and Alzheimer
disease risk: a Mendelian randomization study.
Ann. Neurol. 87, 30–39 (2020).
68. Benn, M., Nordestgaard, B. G., Frikke-Schmidt, R.
& Tybjaerg-Hansen, A. Low LDL cholesterol, PCSK9
and HMGCR genetic variation, and risk of Alzheimer’s
disease and Parkinson’s disease: Mendelian
randomisation study. BMJ 357, j1648 (2017).
69. Schmidt, A. F. et al. Phenome-wide association
analysis of LDL-cholesterol lowering genetic
variants in PCSK9. BMC Cardiovasc. Disord. 19,
240 (2019).
70. Sabatine, M. S. et al. Efficacy and safety of
evolocumab in reducing lipids and cardiovascular
events. N. Engl. J. Med. 372, 1500–1509 (2015).
71. Wright, R. S. et al. Pooled patient-level analysis
of inclisiran trials in patients with familial
0123456789();:
hypercholesterolemia or atherosclerosis. J. Am. Coll.
Cardiol. 77, 1182–1193 (2021).
72. Leiter, L. A. et al. Alirocumab safety in people with
and without diabetes mellitus: pooled data from
14 ODYSSEY trials. Diabet. Med. 35, 1742–1751
(2018).
73. Da Dalt, L. et al. PCSK9 deficiency reduces insulin
secretion and promotes glucose intolerance: the role
of the low-density lipoprotein receptor. Eur. Heart J.
40, 357–368 (2019).
74. Bowden, J., Davey Smith, G. & Burgess, S.
Mendelian randomization with invalid instruments:
effect estimation and bias detection through Egger
regression. Int. J. Epidemiol. 44, 512–525 (2015).
75. Garrelfs, S. F. et al. Lumasiran, an RNAi therapeutic
for primary hyperoxaluria type 1. N. Engl. J. Med.
384, 1216–1226 (2021).
76. McGregor, T. L. et al. Characterising a healthy adult
with a rare HAO1 knockout to support a therapeutic
strategy for primary hyperoxaluria. eLife 9, e54363
(2020).
77. Paisan-Ruiz, C. et al. Cloning of the gene containing
mutations that cause PARK8-linked Parkinson’s
disease. Neuron 44, 595–600 (2004).
78. Zimprich, A. et al. Mutations in LRRK2 cause
autosomal-dominant parkinsonism with pleomorphic
pathology. Neuron 44, 601–607 (2004).
79. Simon-Sanchez, J. et al. Genome-wide association
study reveals genetic risk underlying Parkinson’s
disease. Nat. Genet. 41, 1308–1312 (2009).
80. Whiffin, N. et al. The effect of LRRK2 loss-of-function
variants in humans. Nat. Med. 26, 869–877 (2020).
81. Samson, M. et al. Resistance to HIV-1 infection
in caucasian individuals bearing mutant alleles of
the CCR-5 chemokine receptor gene. Nature 382,
722–725 (1996).
82. Liu, R. et al. Homozygous defect in HIV-1 coreceptor
accounts for resistance of some multiply-exposed
individuals to HIV-1 infection. Cell 86, 367–377
(1996).
83. Emmelkamp, J. M. & Rockstroh, J. K. CCR5
antagonists: comparison of efficacy, side effects,
pharmacokinetics and interactions–review of the
literature. Eur. J. Med. Res. 12, 409–417 (2007).
84. Tebas, P. et al. Gene editing of CCR5 in autologous
CD4 T cells of persons infected with HIV. N. Engl.
J. Med. 370, 901–910 (2014).
85. Nag, A. et al. Human genetic evidence supports
MAP3K15 inhibition as a therapeutic strategy
for diabetes. medRxiv https://doi.org/10.1101/
2021.11.14.21266328 (2021).
86. Narasimhan, V. M. et al. Health and population
effects of rare gene knockouts in adult humans
with related parents. Science 352, 474–477
(2016).
87. Van Hout, C. V. et al. Exome sequencing and
characterization of 49,960 individuals in the UK
Biobank. Nature 586, 749–756 (2020).
88. Lim, E. T. et al. Distribution and medical impact
of loss-of-function variants in the Finnish founder
population. PLoS Genet. 10, e1004494 (2014).
89. Tanigawa, Y. et al. Rare protein-altering variants in
ANGPTL7 lower intraocular pressure and protect
against glaucoma. PLoS Genet. 16, e1008682
(2020).
90. Saleheen, D. et al. Human knockouts and phenotypic
analysis in a cohort with a high rate of consanguinity.
Nature 544, 235–239 (2017).
91. Petrovski, S., Wang, Q., Heinzen, E. L., Allen, A. S.
& Goldstein, D. B. Genic intolerance to functional
variation and the interpretation of personal genomes.
PLoS Genet. 9, e1003709 (2013).
92. Karczewski, K. J. et al. The mutational constraint
spectrum quantified from variation in 141,456
humans. Nature 581, 434–443 (2020).
93. Petrovski, S. et al. The intolerance of regulatory
sequence to genetic variation predicts gene dosage
sensitivity. PLoS Genet. 11, e1005492 (2015).
94. Begum, T., Ghosh, T. C. & Basak, S. Systematic
analyses and prediction of human drug side effect
associated proteins from the perspective of protein
evolution. Genome Biol. Evol. 9, 337–350 (2017).
95. Minikel, E. V. et al. Evaluating drug targets through
human loss-of-function genetic variation. Nature 581,
459–464 (2020).
96. Trochet, D., Prudhon, B., Vassilopoulos, S. &
Bitoun, M. Therapy for dominant inherited diseases by
allele-specific RNA interference: successes and pitfalls.
Curr. Gene Ther. 15, 503–510 (2015).
97. Rook, M. E. & Southwell, A. L. Antisense
oligonucleotide therapy: from design to the huntington
disease clinic. BioDrugs 36, 105–119 (2022).

98. Nagasaka, M. et al. Beyond osimertinib: the
development of third-generation EGFR tyrosine
kinase inhibitors for advanced EGFR+ NSCLC.
J. Thorac. Oncol. 16, 740–763 (2021).
99. Bowes, J. et al. Reducing safety-related drug
attrition: the use of in vitro pharmacological profiling.
Nat. Rev. Drug Discov. 11, 909–922 (2012).
100. Whitebread, S. et al. Secondary pharmacology:
screening and interpretation of off-target activities -
focus on translation. Drug Discov. Today 21,
1232–1242 (2016).
101. ICH. Guidance for Industry: S7A Safety Pharmacology
Studies for Human Pharmaceuticals (2001).
102. Hamon, J. et al. In vitro safety pharmacology profiling:
what else beyond hERG? Future Med. Chem. 1,
645–665 (2009).
103. Curran, M. E. et al. A molecular basis for cardiac
arrhythmia: HERG mutations cause long QT syndrome.
Cell 80, 795–803 (1995).
104. Kannankeril, P., Roden, D. M. & Darbar, D.
Drug-induced long QT syndrome. Pharmacol. Rev.
62, 760–781 (2010).
105. Paulussen, A. D. et al. Genetic variations of KCNQ1,
KCNH2, SCN5A, KCNE1, and KCNE2 in drug-induced
long QT syndrome patients. J. Mol. Med. 82,
182–188 (2004).
106. Chen, Q. et al. Genetic basis and molecular
mechanism for idiopathic ventricular fibrillation.
Nature 392, 293–296 (1998).
107. Deaton, A. M. et al. Rationalizing secondary
pharmacology screening using human genetic
and pharmacological evidence. Toxicol. Sci. 167,
593–603 (2019).
108. Liu, X. et al. A proteomic platform to identify off-target
proteins associated with therapeutic modalities that
induce protein degradation or gene silencing. Sci. Rep.
11, 15856 (2021).
109. Siintola, E. et al. Cathepsin D deficiency underlies
congenital human neuronal ceroid-lipofuscinosis.
Brain 129, 1438–1445 (2006).
110. Gisolfi, C. V., Summers, R. W., Schedl, H. P. &
Bleiler, T. L. Intestinal water absorption from select
carbohydrate solutions in humans. J. Appl. Physiol.
73, 2142–2150 (1992).
111. Zuhl, A. M. et al. Chemoproteomic profiling reveals
that cathepsin D off-target activity drives ocular
toxicity of beta-secretase inhibitors. Nat. Commun.
7, 13042 (2016).
112. Debs, R. et al. Biotin-responsive basal ganglia disease
in ethnic Europeans with novel SLC19A3 mutations.
Arch. Neurol. 67, 126–130 (2010).
113. Kono, S. et al. Mutations in a thiamine-transporter
gene and Wernicke’s-like encephalopathy. N. Engl.
J. Med. 360, 1792–1794 (2009).
114. Zhang, Q. et al. The Janus kinase 2 inhibitor fedratinib
inhibits thiamine uptake: a putative mechanism for
the onset of Wernicke’s encephalopathy. Drug. Metab.
Dispos. 42, 1656–1662 (2014).
115. Donovan, K. A. et al. Thalidomide promotes
degradation of SALL4, a transcription factor
implicated in Duane radial ray syndrome. eLife 7,
e38430 (2018).
116. Matyskiela, M. E. et al. SALL4 mediates teratogenicity
as a thalidomide-dependent cereblon substrate.
Nat. Chem. Biol. 14, 981–987 (2018).
117. Belair, D. G. et al. Thalidomide inhibits human
iPSC mesendoderm differentiation by modulating
CRBN-dependent degradation of SALL4. Sci. Rep. 10,
2864 (2020).
118. Kohlhase, J. et al. Okihiro syndrome is caused by
SALL4 mutations. Hum. Mol. Genet. 11, 2979–2987
(2002).
119. Kohlhase, J. et al. Mutations at the SALL4 locus
on chromosome 20 result in a range of clinically
overlapping phenotypes, including Okihiro syndrome,
Holt-Oram syndrome, acro-renal-ocular syndrome,
and patients previously reported to represent
thalidomide embryopathy. J. Med. Genet. 40,
473–478 (2003).
120. Vargesson, N. Thalidomide-induced teratogenesis:
history and mechanisms. Birth Defects Res. C.
Embryo Today 105, 140–156 (2015).
121. Janas, M. M. et al. Selection of GalNAc-conjugated
siRNAs with limited off-target-driven rat
hepatotoxicity. Nat. Commun. 9, 723 (2018).
122. Burel, S. A. et al. Hepatotoxicity of high affinity
gapmer antisense oligonucleotides is mediated by
RNase H1 dependent promiscuous reduction of very
long pre-mRNA transcripts. Nucleic Acids Res. 44,
2093–2109 (2016).
123. US Department of Health and Human Services.
Chronic Hepatitis B Virus Infection: Developing
NaTure RevIeWS | DRuG DISCOveRy
Drugs for Treatment: Guidance for Industry, https://
www.fda.gov/media/117977/download (2022).
124. Scott, D. A. & Zhang, F. Implications of human genetic
variation in CRISPR-based therapeutic genome
editing. Nat. Med. 23, 1095–1101 (2017).
125. Guidance Document: Human Gene Therapy Products
Incorporating Human Genome Editing (US Food and
Drug Administration, 2022); https://www.fda.gov/
regulatory-information/search-fda-guidance-documents/
human-gene-therapy-products-incorporating-human-
genome-editing
126. Moggs, J. G., MacLachlan, T., Martus, H. J. &
Bentley, P. Derisking drug-induced carcinogenicity
for novel therapeutics. Trends Cancer 2, 398–408
(2016).
127. Fielden, M. R. et al. Modernizing human cancer risk
assessment of therapeutics. Trends Pharmacol. Sci.
39, 232–247 (2018).
128. Dumont, N. & Arteaga, C. L. The tumor
microenvironment: a potential arbitrator of the
tumor suppressive and promoting actions of TGFbeta.
Differentiation 70, 574–582 (2002).
129. Caja, F. & Vannucci, L. TGFbeta: a player on multiple
fronts in the tumor microenvironment. J. Immunotoxicol.
12, 300–307 (2015).
130. Qin, T. et al. A novel highly potent trivalent TGF-beta
receptor trap inhibits early-stage tumorigenesis
and tumor cell invasion in murine Pten-deficient
prostate glands. Oncotarget 7, 86087–86102
(2016).
131. Grenga, I. et al. Anti-PD-L1/TGFbetaR2 (M7824)
fusion protein induces immunogenic modulation of
human urothelial carcinoma cell lines, rendering them
more susceptible to immune-mediated recognition
and lysis. Urol. Oncol. 36, 93.e1–93.e11 (2018).
132. Goudie, D. R. et al. Multiple self-healing squamous
epithelioma is caused by a disease-specific spectrum
of mutations in TGFBR1. Nat. Genet. 43, 365–369
(2011).
133. Lacouture, M. E. et al. Cutaneous keratoacanthomas/
squamous cell carcinomas associated with
neutralization of transforming growth factor beta by
the monoclonal antibody fresolimumab (GC1008).
Cancer Immunol. Immunother. 64, 437–446 (2015).
134. Strauss, J. et al. Phase I trial of M7824
(MSB0011359C), a bifunctional fusion protein
targeting PD-L1 and TGFbeta, in advanced solid
tumors. Clin. Cancer Res. 24, 1287–1295 (2018).
135. Arnault, J. P. et al. Keratoacanthomas and squamous
cell carcinomas in patients receiving sorafenib.
J. Clin. Oncol. 27, e59–e61 (2009).
136. Arnault, J. P. et al. Skin tumors induced by
sorafenib; paradoxic RAS-RAF pathway activation
and oncogenic mutations of HRAS, TP53, and
TGFBR1. Clin. Cancer Res. 18, 263–272 (2012).
137. Carlino, M. S. et al. Correlation of BRAF and
NRAS mutation status with outcome, site of
distant metastasis and response to chemotherapy in
metastatic melanoma. Br. J. Cancer 111, 292–299
(2014).
138. Daver, N., Schlenk, R. F., Russell, N. H. & Levis, M. J.
Targeting FLT3 mutations in AML: review of current
knowledge and evidence. Leukemia 33, 299–312
(2019).
139. Goldman, J. M. Chronic myeloid leukemia: a historical
perspective. Semin. Hematol. 47, 302–311 (2010).
140. Muller, P. A. & Vousden, K. H. p53 mutations in
cancer. Nat. Cell Biol. 15, 2–8 (2013).
141. Cox, A. D. & Der, C. J. The raf inhibitor paradox:
unexpected consequences of targeted drugs.
Cancer Cell 17, 221–223 (2010).
142. McDonald, E. R. III et al. Project DRIVE: a
compendium of cancer dependencies and synthetic
lethal relationships uncovered by large-scale, deep
RNAi screening. Cell 170, 577–592.e10 (2017).
143. de Weck, A. et al. Correction of copy number induced
false positives in CRISPR screens. PLoS Comput. Biol.
14, e1006279 (2018).
144. Rauscher, B., Heigwer, F., Breinig, M., Winter, J.
& Boutros, M. GenomeCRISPR – a database
for high-throughput CRISPR/Cas9 screens.
Nucleic Acids Res. 45, D679–D686 (2017).
145. Flaherty, K. T. et al. Inhibition of mutated, activated
BRAF in metastatic melanoma. N. Engl. J. Med. 363,
809–819 (2010).
146. Fu, Y. et al. High-frequency off-target mutagenesis
induced by CRISPR-Cas nucleases in human cells.
Nat. Biotechnol. 31, 822–826 (2013).
147. Pattanayak, V. et al. High-throughput profiling of
off-target DNA cleavage reveals RNA-programmed
Cas9 nuclease specificity. Nat. Biotechnol. 31,
839–843 (2013).
0123456789();:
Reviews
148. Forbes, S. A. et al. COSMIC: somatic cancer genetics
at high-resolution. Nucleic Acids Res. 45, D777–D783
(2017).
149. Abul-Husn, N. S. & Kenny, E. E. Personalized medicine
and the power of electronic health records. Cell 177,
58–69 (2019).
150. Expert Panel on Detection, Evaluation, and Treatment
of High Blood Cholesterol in Adults. Executive
summary of the third report of the national cholesterol
education program (NCEP) Expert panel on detection,
evaluation, and treatment of high blood cholesterol
in adults (Adult Treatment Panel III). JAMA 285,
2486–2497 (2001).
151. Mora, S. et al. Lipoprotein(a) and risk of type 2
diabetes. Clin. Chem. 56, 1252–1260 (2010).
152. Gudbjartsson, D. F. et al. Lipoprotein(a) concentration
and risks of cardiovascular disease and diabetes.
J. Am. Coll. Cardiol. 74, 2982–2994 (2019).
153. Okada, S. et al. Impairment of immunity to Candida
and Mycobacterium in humans with bi-allelic
RORC mutations. Science 349, 606–613 (2015).
154. Gal, A. et al. Mutations in MERTK, the human
orthologue of the RCS rat retinal dystrophy gene,
cause retinitis pigmentosa. Nat. Genet. 26, 270–271
(2000).
155. Sayama, A. et al. UNC569-induced morphological
changes in pigment epithelia and photoreceptor
cells in the retina through MerTK inhibition in mice.
Toxicol. Pathol. 46, 193–201 (2018).
156. Koonin, E. V., Wolf, Y. I. & Karev, G. P. The structure
of the protein universe and genome evolution. Nature
420, 218–223 (2002).
157. Ekman, D., Bjorklund, A. K., Frey-Skott, J. &
Elofsson, A. Multi-domain proteins in the three
kingdoms of life: orphan domains and other unassigned
regions. J. Mol. Biol. 348, 231–243 (2005).
158. Wang, H. et al. Cell-specific mechanisms of TMEM16A
Ca2+-activated chloride channel in cancer. Mol. Cancer
16, 152 (2017).
159. Crottes, D. & Jan, L. Y. The multifaceted role of
TMEM16A in cancer. Cell Calcium 82, 102050 (2019).
160. Bill, A. et al. Small molecule-facilitated degradation of
ANO1 protein: a new targeting approach for anticancer
therapeutics. J. Biol. Chem. 289, 11029–11041 (2014).
161. Bill, A. et al. ANO1/TMEM16A interacts with EGFR
and correlates with sensitivity to EGFR-targeting
therapy in head and neck cancer. Oncotarget 6,
9173–9188 (2015).
162. Bill, A. & Alex Gaither, L. The mechanistic role of the
calcium-activated chloride channel ANO1 in tumor
growth and signaling. Adv. Exp. Med. Biol. 966,
1–14 (2017).
163. Farnaby, W. et al. BAF complex vulnerabilities in
cancer demonstrated via structure-based PROTAC
design. Nat. Chem. Biol. 15, 672–680 (2019).
164. Burska, A., Boissinot, M. & Ponchel, F. Cytokines as
biomarkers in rheumatoid arthritis. Mediators Inflamm.
2014, 545493 (2014).
165. Sirugo, G., Williams, S. M. & Tishkoff, S. A. The
missing diversity in human genetic studies. Cell 177,
26–31 (2019).
166. Popejoy, A. B. & Fullerton, S. M. Genomics is failing
on diversity. Nature 538, 161–164 (2016).
167. Fatumo, S. et al. A roadmap to increase diversity in
genomic studies. Nat. Med. 28, 243–250 (2022).
168. Hindorff, L. A. et al. Prioritizing diversity in human
genomics research. Nat. Rev. Genet. 19, 175–185
(2018).
169. All of Us Research Program Investigators. et al. The
“All of Us” Research Program. N. Engl. J. Med. 381,
668–676 (2019).
170. de Vries, J. et al. Ethical issues in human genomics
research in developing countries. BMC Med. Ethics
12, 5 (2011).
171. Munung, N. S. & de Vries, J. Benefit sharing
for human genomics research: awareness and
expectations of genomics researchers in Sub-Saharan
Africa. Ethics Hum. Res. 42, 14–20 (2020).
172. Pennisi, E. Genomes arising. Science 371, 556–559
(2021).
173. Maxmen, A. The next chapter for African genomics.
Nature 578, 350–354 (2020).
174. Munafo, M. R. & Gage, S. H. Improving the reliability
and reporting of genetic association studies.
Drug Alcohol Depend. 132, 411–413 (2013).
175. Amberger, J. S. & Hamosh, A. Searching online
Mendelian inheritance in man (OMIM):
a knowledgebase of human genes and genetic
phenotypes. Curr. Protoc. Bioinformatics 58,
1.2.1–1.2.12 (2017).
176. Richards, S. et al. Standards and guidelines for the
interpretation of sequence variants: a joint consensus
volume 22 | February 2023 | 161
Reviews
recommendation of the American College of Medical
Genetics and Genomics and the Association for
Molecular Pathology. Genet. Med. 17, 405–424 (2015).
177. Rehm, H. L. et al. ClinGen – the clinical genome
resource. N. Engl. J. Med. 372, 2235–2242 (2015).
178. Landrum, M. J. et al. ClinVar: improving access to
variant interpretations and supporting evidence.
Nucleic Acids Res. 46, D1062–D1067 (2018).
179. Buniello, A. et al. The NHGRI-EBI GWAS catalog
of published genome-wide association studies,
targeted arrays and summary statistics 2019.
Nucleic Acids Res. 47, D1005–D1012 (2019).
180. Machiela, M. J. & Chanock, S. J. LDassoc: an online
tool for interactively exploring genome-wide association
study results and prioritizing variants for functional
investigation. Bioinformatics 34, 887–889 (2018).
181. Boyle, A. P. et al. Annotation of functional variation in
personal genomes using RegulomeDB. Genome Res.
22, 1790–1797 (2012).
182. Ward, L. D. & Kellis, M. HaploReg: a resource for
exploring chromatin states, conservation, and
regulatory motif alterations within sets of genetically
linked variants. Nucleic Acids Res. 40, D930–D934
(2012).
183. GTEx Consortium. The GTEx Consortium atlas of
genetic regulatory effects across human tissues.
Science 369, 1318–1330 (2020).
184. Võsa, U. et al. Unraveling the polygenic architecture
of complex traits using blood eQTL metaanalysis.
bioRxiv https://doi.org/10.1101/447367 (2018).
185. Zhu, Z. et al. Integration of summary data from GWAS
and eQTL studies predicts complex trait gene targets.
Nat. Genet. 48, 481–487 (2016).
186. Staley, J. R. et al. PhenoScanner: a database of human
genotype-phenotype associations. Bioinformatics 32,
3207–3209 (2016).
187. Watanabe, K. et al. A global overview of pleiotropy
and genetic architecture in complex traits. Nat. Genet.
51, 1339–1348 (2019).
188. Elliott, L. T. et al. Genome-wide association studies
of brain imaging phenotypes in UK Biobank. Nature
562, 210–216 (2018).
189. Denny, J. C. et al. Systematic comparison of
phenome-wide association study of electronic medical
record data and genome-wide association study data.
Nat. Biotechnol. 31, 1102–1110 (2013).
190. Leslie, R., O’Donnell, C. J. & Johnson, A. D. GRASP:
analysis of genotype-phenotype results from 1390
genome-wide association studies and corresponding
open access database. Bioinformatics 30, i185–i194
(2014).
191. Cerami, E. et al. The cBio cancer genomics portal: an
open platform for exploring multidimensional cancer
genomics data. Cancer Discov. 2, 401–404 (2012).
192. Gonzalez-Perez, A. et al. IntOGen-mutations identifies
cancer drivers across tumor types. Nat. Methods 10,
1081–1082 (2013).
193. Taliun, D. et al. Sequencing of 53,831 diverse
genomes from the NHLBI TOPMed Program. Nature
590, 290–299 (2021).
194. Zhou, W. et al. Efficiently controlling for case-control
imbalance and sample relatedness in large-scale genetic
association studies. Nat. Genet. 50, 1335–1341 (2018).
195. Wu, M. C. et al. Rare-variant association testing for
sequencing data with the sequence kernel association
test. Am. J. Hum. Genet. 89, 82–93 (2011).
196. Zhou, W. et al. Scalable generalized linear mixed model
for region-based association tests in large biobanks
and cohorts. Nat. Genet. 52, 634–639 (2020).
197. Povysil, G. et al. Rare-variant collapsing analyses
for complex traits: guidelines and applications.
Nat. Rev. Genet. 20, 747–759 (2019).
198. Martin, A. R. et al. Clinical use of current polygenic
risk scores may exacerbate health disparities.
Nat. Genet. 51, 584–591 (2019).
199. Fang, H. et al. A genetics-led approach defines the
drug target landscape of 30 immune-related traits.
Nat. Genet. 51, 1082–1091 (2019).
200. Newberry, R. W., Leong, J. T., Chow, E. D.,
Kampmann, M. & DeGrado, W. F. Deep mutational
scanning reveals the structural basis for alpha-synuclein
activity. Nat. Chem. Biol. 16, 653–659 (2020).
201. Maurano, M. T. et al. Systematic localization of
common disease-associated variation in regulatory
DNA. Science 337, 1190–1195 (2012).
202. Gupta, R. M. et al. A genetic variant associated
with five vascular diseases is a distal regulator
of endothelin-1 gene expression. Cell 170,
522–533.e515 (2017).
203. Flanagan, J. M. Epigenome-wide association studies
(EWAS): past, present, and future. Methods Mol. Biol.
1238, 51–63 (2015).
162 | February 2023 | volume 22 www.nature.com/nrd
204. Lappalainen, T. & Greally, J. M. Associating cellular
epigenetic models with human phenotypes.
Nat. Rev. Genet. 18, 441–451 (2017).
205. 1000 Genomes Project Consortium. A global
reference for human genetic variation. Nature 526,
68–74 (2015).
206. Turro, E. et al. Whole-genome sequencing of patients
with rare diseases in a national health system. Nature
583, 96–102 (2020).
207. Chatr-Aryamontri, A. et al. The BioGRID interaction
database: 2017 update. Nucleic Acids Res. 45,
D369–D379 (2017).
208. Firth, H. V. et al. DECIPHER: database of chromosomal
imbalance and phenotype in humans using ensembl
resources. Am. J. Hum. Genet. 84, 524–533 (2009).
209. Pinero, J. et al. The DisGeNET knowledge platform for
disease genomics: 2019 update. Nucleic Acids Res.
48, D845–D855 (2020).
210. Finer, S. et al. Cohort profile: East London Genes
& Health (ELGH), a community-based population
genomics and health study in British Bangladeshi and
British Pakistani people. Int. J. Epidemiol. 49, 20–21i
(2020).
211. Locke, A. E. et al. Exome sequencing of Finnish
isolates enhances rare-variant association power.
Nature 572, 323–328 (2019).
212. Canela-Xandri, O., Rawlik, K. & Tenesa, A. An atlas
of genetic associations in UK Biobank. Nat. Genet. 50,
1593–1599 (2018).
213. Consortium, G. T. et al. Genetic effects on gene
expression across human tissues. Nature 550,
204–213 (2017).
214. MacArthur, J. et al. The new NHGRI-EBI Catalog of
published genome-wide association studies (GWAS
Catalog). Nucleic Acids Res. 45, D896–D901 (2017).
215. Stenson, P. D. et al. The human gene mutation
database: towards a comprehensive repository of
inherited mutation data for medical research, genetic
diagnosis and next-generation sequencing studies.
Hum. Genet. 136, 665–677 (2017).
216. Traynelis, J. et al. Optimizing genomic medicine in
epilepsy through a gene-customized approach to
missense variant interpretation. Genome Res. 27,
1715–1729 (2017).
217. Amberger, J. S., Bocchini, C. A., Schiettecatte, F.,
Scott, A. F. & Hamosh, A. OMIM.org: online
Mendelian inheritance in man (OMIM(R)), an online
catalog of human genes and genetic disorders.
Nucleic Acids Res. 43, D789–D798 (2015).
218. Koscielny, G. et al. Open Targets: a platform for
therapeutic target identification and validation.
Nucleic Acids Res. 45, D985–D994 (2017).
219. Gussow, A. B. et al. Orion: detecting regions of the
human non-coding genome that are intolerant to
variation using population genetics. PLoS ONE 12,
e0181604 (2017).
220. Kamat, M. A. et al. PhenoScanner V2: an expanded
tool for searching human genotype-phenotype
associations. Bioinformatics 35, 4851–4853 (2019).
221. Gao, J. et al. Integrative analysis of complex cancer
genomics and clinical profiles using the cBioPortal.
Sci. Signal. 6, pl1 (2013).
222. The UniProt Consortium. UniProt: the universal protein
knowledgebase. Nucleic Acids Res. 46, 2699 (2018).
223. Scott, S. A. et al. Clinical pharmacogenetics
implementation consortium guidelines for CYP2C19
genotype and clopidogrel therapy: 2013 update.
Clin. Pharmacol. Ther. 94, 317–323 (2013).
224. Weinshilboum, R. M. & Sladek, S. L. Mercaptopurine
pharmacogenetics: monogenic inheritance of
erythrocyte thiopurine methyltransferase activity.
Am. J. Hum. Genet. 32, 651–662 (1980).
225. CPIC® Guideline for Thiopurines and TPMT and
NUDT15 (Clinical Pharmacogenetics Implementation
Consortium, 2018); https://cpicpgx.org/guidelines/
guideline-for-thiopurines-and-tpmt/
226. Chung, W. H. et al. Medical genetics: a marker for
Stevens–Johnson syndrome. Nature 428, 486 (2004).
227. Ferrell, P. B. Jr & McLeod, H. L. Carbamazepine,
HLA-B*1502 and risk of Stevens–Johnson
syndrome and toxic epidermal necrolysis: US
FDA recommendations. Pharmacogenomics 9,
1543–1546 (2008).
228. Chen, P. et al. Carbamazepine-induced toxic effects
and HLA-B*1502 screening in Taiwan. N. Engl. J. Med.
364, 1126–1133 (2011).
229. McCormack, M. et al. HLA-A*3101 and
carbamazepine-induced hypersensitivity reactions in
Europeans. N. Engl. J. Med. 364, 1134–1143 (2011).
230. Lindpaintner, K. The impact of pharmacogenetics
and pharmacogenomics on drug discovery. Nat. Rev.
Drug Discov. 1, 463–469 (2002).
0123456789();:
231. Roses, A. D. Pharmacogenetics and drug
development: the path to safer and more effective
drugs. Nat. Rev. Genet. 5, 645–656 (2004).
232. Roses, A. D. Pharmacogenetics in drug discovery
and development: a translational perspective.
Nat. Rev. Drug Discov. 7, 807–817 (2008).
233. Nelson, M. R. et al. The genetics of drug efficacy:
opportunities and challenges. Nat. Rev. Genet. 17,
197–206 (2016).
234. Wei, C. Y., Lee, M. T. & Chen, Y. T. Pharmacogenomics
of adverse drug reactions: implementing personalized
medicine. Hum. Mol. Genet. 21, R58–R65 (2012).
235. Alfirevic, A. & Pirmohamed, M. Adverse drug
reactions and pharmacogenomics: recent advances.
Per. Med. 5, 11–23 (2008).
236. Collins, S. L., Carr, D. F. & Pirmohamed, M. Advances
in the pharmacogenomics of adverse drug reactions.
Drug. Saf. 39, 15–27 (2016).
237. Cook, J. C., Wu, H., Aleo, M. D. & Adkins, K. Principles
of precision medicine and its application in toxicology.
J. Toxicol. Sci. 43, 565–577 (2018).
238. Cacabelos, R., Cacabelos, N. & Carril, J. C. The role
of pharmacogenomics in adverse drug reactions.
Expert. Rev. Clin. Pharmacol. 12, 407–442 (2019).
239. Lesko, L. J. & Woodcock, J. Translation of
pharmacogenomics and pharmacogenetics:
a regulatory perspective. Nat. Rev. Drug Discov. 3,
763–769 (2004).
240. Maliepaard, M. et al. Pharmacogenetics in the evaluation
of new drugs: a multiregional regulatory perspective.
Nat. Rev. Drug Discov. 12, 103–115 (2013).
241. Ehmann, F. et al. Pharmacogenomic information in
drug labels: European Medicines Agency perspective.
Pharmacogenomics J. 15, 201–210 (2015).
242. Relling, M. V. & Evans, W. E. Pharmacogenomics
in the clinic. Nature 526, 343–350 (2015).
243. Cecchin, E., Roncato, R., Guchelaar, H. J., Toffoli, G.
& Ubiquitous Pharmacogenomics, C. Ubiquitous
pharmacogenomics (U-PGx): the time for
implementation is now. An Horizon2020 program
to drive pharmacogenomics into clinical practice.
Curr. Pharm. Biotechnol. 18, 204–209 (2017).
244. van der Wouden, C. H. et al. Development of the
PGx-passport: a panel of actionable germline genetic
variants for pre-emptive pharmacogenetic testing.
Clin. Pharmacol. Ther. 106, 866–873 (2019).
245. Yang, T. et al. Genotype-guided dosing versus
conventional dosing of warfarin: a meta-analysis of 15
randomized controlled trials. J. Clin. Pharm. Ther. 44,
197–208 (2019).
Author contributions
The authors contributed equally to all aspects of the article.
Competing interests
A.M.D., L.D.W., M.F. and J.Y. are former employees of
Amgen. K.J.C. and P.N. are employees of AstraZeneca.
A.D.R.E. is a former employee of Novartis Institutes for
BioMedical Research and is currently employed by GentiBio,
Inc. J.M. is an employee of Novartis Institutes for BioMedical
Research. D.D. is an employee of Takeda Development Center
America. D.A.K. is an employee of GlaxoSmithKline. M.R.N. is
an employee of Deerfield Management Company, L.P. F.D.S.
is a former employee and currently a part-time contractor of
Merck Sharp & Dohme Corp., a subsidiary of Merck & Co.,
Inc., Kenliworth, NJ, USA. A.M.D. and L.D.W. are employees
and stockholders of Alnylam Pharmaceuticals. J.Y. is an
employee of Pfizer.
Peer review information
Nature Reviews Drug Discovery thanks Kathleen Meyer,
Munir Pirmohamed and the other, anonymous, reviewer for
their contribution to the peer review of this work.
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.
Springer Nature or its licensor holds exclusive rights to this
article under a publishing agreement with the author(s) or
other rightsholder(s); author self-archiving of the accepted
manuscript version of this article is solely governed by the
terms of such publishing agreement and applicable law.
Related links
Genes and Health: http://www.genesandhealth.org/research
Pakistan Genomic Resource: https://www.cncdpk.com/page/
Pakistan-Genomic-Resource-(PGR)
© Springer Nature Limited 2022