You are on page 1of 14

REVIEWS

Mechanisms of tissue and cell-type


specificity in heritable traits
and diseases
Idan Hekselman1 and Esti Yeger-Lotem   1,2*
Abstract | Hundreds of heritable traits and diseases that are caused by germline aberrations in
ubiquitously expressed genes manifest in a remarkably limited number of cell types and tissues
across the body. Unravelling mechanisms that govern their tissue-specific manifestations is
critical for our understanding of disease aetiologies and may direct efforts to develop treatments.
Owing to recent advances in high-throughput technologies and open resources, data and tools
are now available to approach this enigmatic phenomenon at large scales, both computationally
and experimentally. Here, we discuss the large prevalence of tissue-selective traits and diseases,
describe common molecular mechanisms underlying their tissue-selective manifestation and
present computational strategies and publicly available resources for elucidating the molecular
basis of their genotype–phenotype relationships.

Heritable traits and


Heritable traits and diseases are commonly considered heritable diseases that were genetically solved decades
diseases in terms of their affected organs, tissues or cell types. ago, including Gaucher disease5, Huntington disease6,
Phenotypes with a heritable For example, cystic fibrosis (CF) brings to mind the PD7–9 and cardiomyopathies10. Identifying the mecha­
monogenic or polygenic respiratory system, whereas Parkinson disease (PD) nisms that underlie the tissue-selective pathogenesis
component due to inherited or
de novo germline aberrations
is linked with dopaminergic neurons. More generally, could provide an important first step towards elucidating
present throughout the body. analysis of over 1,200 Mendelian diseases shows that disease aetiologies. Since these mechanisms could high-
most Mendelian diseases manifest clinically in a highly light components that are critical for the normal physiol-
Germline aberrations tissue-specific manner (Fig. 1a). Tissue-specific manifes- ogy of the disease-related tissue11 or, alternatively, disease
Aberrations (inherited or
tation is expected when caused by somatic aberrations modifiers that contribute to the resilience of unaffected
de novo) that are common to
all cells harbouring the
localized to particular tissues (Box 1); yet, it is intrigu- tissues, they could open new therapeutic directions.
individual’s genome. ing in heritable traits and diseases as these are caused The large number of tissue-selective heritable diseases
by  germline aberrations present throughout the body1. that are yet to be molecularly deciphered, as well as the
Causal genes An intuitive explanation for the tissue specificity of important implications of their elucidation for basic sci-
Genes containing a variant that
was found to lead to disease.
heritable traits and diseases is that their causal genes are ence and therapeutic strategies, call for new approaches
expressed physiologically in a limited number of tissues. for their investigation. A stepping-stone in the devel-
However, genes causal for these traits and diseases are opment of such approaches is the unprecedented
often expressed widely throughout the human body1–3 scale of available molecular profiles of human tissues,
(Fig.  1b) without tissue-specific preference (Fig.  1c) . organs and cell types (henceforth collectively denoted
Do we understand what, then, leads to the tissue-specific as tissues for brevity). Since the beginning of the new
manifestation of various heritable traits and diseases? millennium, diverse human tissues have been interro-
1
Department of Clinical
The genetic diagnosis of heritable diseases, namely gated via multiple omics profiling technologies (Table 1).
Biochemistry & Pharmacology, identifying the causal gene or variant, has been greatly Transcriptomic profiling was first applied to a few sam-
Faculty of Health Sciences, facilitated by clinical usage of next-generation sequenc- ples per tissue12,13. With the advent of next-generation
Ben-Gurion University of the ing, which is executed routinely and at large scales4. sequencing, RNA sequencing (RNA-seq) of physiolo­
Negev, Be’er-Sheva, Israel.
By contrast, the molecular understanding of disease gical tissues from hundreds of human donors was car-
2
National Institute for aetiologies, namely identifying how the causal aberra- ried out by the Genotype–Tissue Expression (GTEx)
Biotechnology in the Negev,
Ben-Gurion University,
tion leads to disease symptoms, currently lags behind. project14 and others15. Most recently, single-cell RNA-seq
Be’er-Sheva, Israel. The elucidation of disease aetiologies typically requires (scRNA-seq) datasets of human tissues are becoming
*e-mail: estiyl@bgu.ac.il focused, low-throughput experiments that were tailored available through, for example, the Human Cell Atlas16.
https://doi.org/10.1038/ to the disease in question. Consequently, pathological Several ensuing projects arose to aggregate available data
s41576-019-0200-9 mechanisms have remained elusive even for well-studied from consortia-based projects and others. For example,

NaTure RevIewS | GeNeTICs volume 21 | March 2020 | 137


Reviews

a 1,200 Fig. 1 | The tissue selectivity of heritable diseases and


traits and their associated genes. a | Germline aberrations
that underlie diseases are present throughout the body
(red diamonds) yet often manifest clinically in few tissues

Number of Mendelian diseases


(red star, pointed by an arrow), as shown for 1,252 Mendelian
800 diseases2,44. b | In contrast to the tissue specificity of
diseases, disease-associated genes are typically expressed
widely across tissues. The number of tissues expressing a
gene is shown for 15,515 protein-coding genes expressed
in any tissue (blue), for 1,044 genes causal for the 1,252
400
Mendelian diseases shown in part a (red) and for 402 genes
significantly associated with 20 complex traits47 (yellow).
Expression values were downloaded from the Genotype–
Tissue Expression (GTEx) portal and refer to GTEx v.7.
0 Only genes whose median expression level in a tissue
1 2 3 4 5 6–30 exceeded 10 transcripts per million (TPM) were considered
Number of disease-manifesting tissues to be expressed in that tissue. Subregions of the same tissue
Unaffected tissues were collated and appear as one (relevant for brain, heart,
Disease-manifesting tissue artery, cervix, colon, skin and oesophagus tissues), and
the maximal expression level was taken. Although the
distributions partially depend on the expression level
threshold chosen154, the conclusion of most genes being
b Expressed genes
expressed across multiple tissues is robust for alternative
30
Protein-coding expression level cut-offs of 5 and 15 TPM (see Supplementary
Mendelian disease genes information for further details on the data analysis, including
Expressed genes (%)

Trait-associated the effects of different expression level cut-offs). c | Most


20 disease-associated genes are not preferentially expressed
in any tissue. The distribution of genes preferentially
expressed in 0, 1, 2 or at least 3 tissues is shown for protein-
10 coding genes (left), genes causal for Mendelian diseases
(middle) and genes significantly associated with complex
traits (right). Preferential expression was derived from ref.51,
and similarly defined as genes with a preferential score
0 >2 in at least one tissue according to data from GTEx v.6
2

10

0
1–

3–

5–

7–

–1

–1

–1

–1

–2

–2

–2

–2

–2

–3

(see Supplementary information for further details on the


9–

11

13

15

17

19

21

23

25

27

29

Number of expressing tissues data analysis).

These collaborative efforts have enabled a change


c Preferentially expressed genes
in the way disease aetiologies can be approached.
Protein-coding Mendelian disease genes Trait-associated In addition to studying them in the context of the dis-
eased tissue, comparative analyses of disease-related
versus unaffected tissues can be performed. Indeed,
emerging experimental and computational methods
are starting to provide explanatory insights for several
diseases1,2,11,24–31. Owing to open data resources14,18,24,25,32,33
and to the enthusiasm with which they have been met by
the computational community, data and tools are now
available to apply these methods and others to many
1 2 >3 0 1 2 >3 0 1 2 >3 0
diseases. In this Review, we first discuss current knowl-
edge of the prevalence of tissue-selective manifestation
recount2 reprocessed ~70,000 human RNA-seq sam- of hereditary diseases and its limitations, followed by
ples17, and the volume of sequencing and array data in a description of the known mechanisms by which tis-
ArrayExpress and Gene Expression Omnibus continues sue selectivity is mediated. Finally, we discuss the open
to grow. Tissue profiling has been extended to additional resources, designed for non-computational investigators,
omics assays. Proteomic profiling of tissues was carried to facilitate the exploration of the molecular mechanisms
out by the Human Protein Atlas (HPA) consortium18 underlying the numerous traits and diseases that await
and others19. Large-scale profiling of DNA regulatory resolution.
elements and chromatin modifications mostly from
cell lines was carried out by the Encyclopedia of DNA Tissue selectivity is prevalent among diseases
Elements (ENCODE)20 and Functional Annotation Extensive RNA-seq and proteomic analyses across
of the Mammalian Genome (FANTOM5)21, whereas human tissues have repeatedly revealed that, whereas tis-
Roadmap Epigenomics performed similar chroma- sues differ greatly in their morphology and functions, the
tin profiling but with a more overt focus on primary majority of the genes and proteins are expressed widely
tissues rather than cell lines22, with other efforts aimed at across the human body and only a minority are tissue
creating tissue chromatin contact maps23. specific14,18,34. In particular, the subset of protein-coding

138 | March 2020 | volume 21 www.nature.com/nrg


Reviews

Box 1 | The tissue specificity of somatic mutations In endocrine disorders, damage in hormone-producing
cells, such as the insulin-producing pancreatic β-cells
Somatic mutations tend to vary in numbers and patterns between tissues. in the case of type 1 diabetes mellitus, manifests clini-
These variations were attributed to differences in the exposure of each tissue to cally in remote tissues harbouring hormone receptors42.
mutation-inducing substances or microenvironmental factors. For example, lungs are Likewise, in Bartter syndrome, decreased absorption
exposed to tobacco smoke, which elicits lung cancers, whereas skin is exposed to
of ions in the middle part of the nephron overloads
ultraviolet light, which elicits melanoma. More recently, the difference in cell division
rates, and consequently in the rates of replication errors, was revealed as a major source α-intercalated cells of the distal collecting duct 43.
for somatic variation149, as a strong correlation between stem cell division rate and Altogether, since pathologies are not typically studied
cancer incidence was shown across cancers and countries149. holistically across a range of tissues and developmen-
The question of whether the impact of specific somatic mutations tends to be tissue tal states, current knowledge of the tissue selectivity of
specific has been thoroughly studied in the context of cancer150. It was shown that, genetic diseases might be noisy and partial due to clinical
whereas mutation of certain cancer-driving genes, such as TP53 (which encodes p53) and/or technical complexities.
and KRAS, lead to cancers across several tissues, many more cancer drivers lead to Notwithstanding these caveats, multiple large-scale
cancers in specific tissues151. As has been observed for heritable disorders, despite analyses of associations between diseases and affected tis-
the pathological effects being frequently tissue specific, the cancer driver genes
sues were conducted across various subsets of Mendelian
affected by the mutations are rarely expressed in a tissue-specific manner1,152,153.
diseases. These analyses, which were based on a combi-
Multiple mechanistic explanations have been suggested for this phenomenon150 such
as differences in the epigenetic landscape across tissues140. Nevertheless, how somatic nation of expert curation and text mining of PubMed1,2
mutations affect certain tissues but not others is still considered an important open and OMIM44, have repeatedly shown that Mendelian
question in cancer research140. diseases have a clear tendency for tissue-specific mani­
festation (Fig. 1a). Putting this phenomenon in an evo-
lutionary context, tissue-specific diseases might be less
genes that are causal for hereditary diseases is not more detrimental than diseases that broadly affect the body,
tissue specific than protein-coding genes in general and thus more likely to be heritable. Tissue specificity
(Fig. 1b). Given their widespread expression, it is intrigu- is also common among complex diseases and traits, of
ing to investigate the extent to which genetic diseases which neurodegenerative disorders, cardiovascular dis-
themselves are tissue selective. eases, autoimmune diseases and systolic blood pressure
According to the well-established database of Online are just a few examples45–47. In fact, tissue specificity
Mendelian Inheritance in Man (OMIM)35, over 7,000 has been leveraged repeatedly to illuminate the aetiol-
Mendelian disorders have been identified so far. More ogy of  complex traits and to interpret genetic findings
than half (3,879 disorders) were fully diagnosed geneti- obtained via genome-wide association studies (GWAS)45.
cally and map to variation in 3,910 protein-coding genes. In summary, although our knowledge of the full list of
The tissues that are clinically affected by each disease are affected tissues per disease is partial, affected tissues
generally recognized by signs and symptoms presented remain a rather small subset compared to the broad
by patients. Until the last decade, knowledge of tissue expression of the underlying causal genes (Fig. 1b). Why,
associations was not organized in a structured manner then, are only certain tissues affected by an inherent
in publicly available databases. This gap has been aberration, while other tissues that express the aberrant
bridged by computational techniques, for example, by gene remain robust? In the next section, we discuss sev-
Pathogenic tissues applying text-mining techniques to abstracts in PubMed eral molecular mechanisms that elicit or contribute to the
The tissues that elicit the to identify significant co-mentioning of disorders and tissue-specific manifestation of monogenic and complex
disease. tissues1. The association between diseases and clinically diseases.
affected tissues became more readily accessible with the
Complex traits
Traits caused by variations in
creation of disease-oriented public databases, such as Molecular mechanisms of tissue specificity
multiple genes or non-coding the Human Phenotype Ontology (HPO)36 and Disease Why does a disease that stems from a germline aberra-
genomic regions potentially in Ontology37, which record phenotypic abnormalities for tion manifest in a tissue-specific manner? A main claim
combination with other factors thousands of diseases (Table 2). Efforts to further expand is that the susceptible tissue, which clinically manifests
such as environmental
and refine the knowledge of tissues affected by diseases the disease, has a signifying characteristic function or
exposure or lifestyle.
are still ongoing38. feature that is disease related. For example, liver and
Genome-wide association The association between diseases and affected tissues muscle, where glycogen metabolism is carried out, are
studies is not always straightforward. Diseases may have sub- affected in glycogen storage diseases; long-lived neu-
(GWAS). Studies that scan tle phenotypes in tissues other than the main affected rons are sensitive to the cumulative effect of protein
genetic variants across
individuals to identify variants
tissues, which might escape notice. For example, one misfolding; and skin is exposed to ultraviolet radiation
that are significantly associated of the earliest symptoms of several neurodegenerative and therefore prone to DNA damage and melanoma.
with a trait (the variants are disorders that can be easily missed is the loss of smell39. However, such clear associations between tissue features
known as risk alleles when they Likewise, certain tissues might be affected in only a sub- and pathology are often missing48. For example, it is still
are associated with disease
set of the patients. A well-known example is CF, where debatable why familial mutations in the widely expres­
occurrence).
patients present large variability in organ involvement, sed cell-cycle negative regulator gene RB1 cause mostly
Monogenic including variable lung dysfunction, male infertility and retinoblastoma, without affecting other tissues49.
A trait caused by variation in a pancreatic insufficiency40. Similarly, certain patients With the accumulation of tissue-wide molecular pro-
single gene. with familial mutations in the BRCA1 or BRCA2 genes files, several features of causal genes and susceptible tis-
Susceptible tissue
develop breast cancer, whereas others develop ovarian sues became apparent (Fig. 2). Below we present cellular
The tissue that manifests a trait cancer41. Additionally, affected tissues might differ from expression-based mechanisms that are common among
or disease. the actual  pathogenic tissues where damage initiated. monogenic diseases, regulatory and network-based

NaTure RevIewS | GeNeTICs volume 21 | March 2020 | 139


Reviews

Table 1 | Representative human tissue profiling resources


Database No. of tissues/ Profiling method Clinical Refs
samples/donors state
mRNA
GTEx 51 (+ 2 CL)/~11,500/~700 RNA-seq Normal 14

HPA  44/122/122 RNA-seq Normal 18

Protein
HPA  32/122/122 IHC Normal and 18

diseased
Human Proteome Map 30/85/3 Mass spectrometry Normal 155

Non-coding RNAs
TissueAtlas (microRNAs) 61/61/2 Microarray Normal 156

DASHR 86 (+ 51 PC, 48 CL)/> 800/NR Data integration Normal 157

FANTOM5 (microRNAs, lncRNAs, 400/150 (+ 570 PC, 250 CL)/3 Various Normal 33,158

promoters, enhancers)
Regulatory elements
GTEx (eQTLs) 48/~ 10,000/~ 600 eQTLs Normal 14

ENCODE and Roadmap >120/NR/hundreds ChIP-seq, DNase-seq, Normal 22,32

Epigenomics ATAC-seq, FAIRE-seq


3D Genome 109/113/NR Hi-C, ChIA-PET, Normal 122

Capture Hi-C, PL AC-seq


TiGER 30 tissues Data integration Normal 159

Single cells expression profiles


Human Cell Atlas 3/17/17 scRNA-seq Normal and 16

diseased
Single Cell Portal 68 studies scRNA-seq Normal and
diseased
ATAC-seq, assay for transposase-accessible chromatin using sequencing; ChIA-PET, chromatin interaction analysis by paired-end tag;
ChIP-seq, chromatin immunoprecipitation coupled to sequencing; CL , cell lines; DASHR , Database of Small Human non-coding
RNAs; DNase-seq, DNase I hypersensitive site sequencing; ENCODE, Encyclopedia of DNA Elements; eQTLs, expression quantitative
trait loci; FAIRE-seq, formaldehyde-assisted identification of regulatory elements; FANTOM5, Functional Annotation of the Mammalian
Genome; GTEx, Genotype–Tissue Expression; HPA , Human Protein Atlas; IHC, immunohistochemistry ; lncRNA , long non-coding
RNA ; NR , not reported; PC, primary cells; PL AC-seq, proximity ligation-assisted ChIP-seq; RNA-seq, RNA sequencing; scRNA-seq,
single-cell RNA sequencing; TiGER , Tissue-specific Gene Expression and Regulation.

mechanisms that are common among monogenic, poly- Tissue susceptibility could also result from preferential
genic and complex diseases, and non-cell-autonomous expression of the causal gene, that is, its over­expression
mechanisms. These mechanisms, while presented sepa- relative to its levels in unaffected tissues51. Large-scale
rately, are often interdependent and can serve as a starting analyses of hundreds of monogenic disease genes1,3
point for elucidating the tissue-selective pathogenesis. revealed that preferential expression of causal genes in
susceptible tissues was a significant trend that was robust
Expression-based mechanisms to expression normalization procedures, was not domi-
This class of mechanisms is typical of monogenic diseases nated by a specific disease or tissue1, and was common in
caused by aberrations in protein-coding genes (Fig. 2a). about a third to half of the diseases explored2,3. For exam-
ple, in Duchenne and Becker muscular dystrophy, the
Distinct expression in disease-related tissues. A simple causal gene encodes dystrophin (DMD), which anchors
mechanism for tissue-specific susceptibility is the exclu- the extracellular matrix to the cytoskeleton, and the
Tissue-exclusive expression
sive expression of the causal gene in susceptible tissues. preferential expression of DMD in muscle is consistent
When expression of a gene For example, the caveolin 3 (CAV3) gene encoding the with the importance of its function to muscle physiology.
exceeds a predefined cut-off in muscle-specific form of the caveolin protein family is Another example, revealed by scRNA-seq of the mouse
a single tissue. This is in expressed primarily in heart and skeletal muscle, and respiratory system, is of a rare cell type expressing high
contrast to tissue-selective
consequently, when mutated, causes cardiomyopathies levels of the CFTR gene that is causal for CF; depletion
expression, which refers to
expression in a subset of and skeletal muscle disorders. Likewise, certain genes of that cell type led to pulmonary CF phenotypes11,26.
tissues (>1 tissue). causal for kidney diseases were shown, via scRNA-seq Tissue-selective expression is also observed at the
of mouse kidney, to be expressed exclusively in the isoform level. Tissue-specific transcript isoforms were
Preferential expression disease-related region of the nephron50. However, most observed for about half of the expressed human genes,
When expression of a gene is
elevated in a certain tissue
causal genes do not show a tissue-exclusive expression with isoforms resulting mostly from alternative tran-
relative to its expression in (Fig. 1b), making this compelling mechanism pertinent scription start and termination sites rather than alter-
other tissues. to a small subset of diseases. native splicing52. Preferential expression of isoforms53

140 | March 2020 | volume 21 www.nature.com/nrg


Reviews

Table 2 | Databases associating between traits, genes and tissues


Database Traits Association No. of associations Ref.
Online Mendelian Mendelian diseases Disease, gene >7,000 diseases, 35

Inheritance in Man >3,900 genes


NHGRI–EBI GWAS Complex diseases and traits Trait, variant ~90,000 SNP–trait 160

Catalogue associations
Human Phenotype Ontology Diseases Phenotype, disease, >156,000 terms 36

tissue and gene annotated to diseases


DisGeNET Diseases and traits Disease, gene >24,000 diseases, 161

>17,000 genes
Disease Ontology Diseases Disease, tissue >9,000 terms associated 37

to clinical vocabularies
Orphanet Rare diseases Disease, gene >6,000 diseases
NHGRI–EBI GWAS, US National Human Genome Research Institute–European Bioinformatics Institute Catalogue of published
genome-wide association studies; SNP, single-nucleotide polymorphism.

and tissue-specific splicing patterns of causal genes and degraded in cartilage but not in non-cartilage cells, lead-
risk alleles were also demonstrated, although their ing to cartilage-specific collagen X haploinsufficiency58.
physiological roles often remained unclear54,55. A large-scale study of protein-truncating mutants
Could tissue susceptibility also occur due to under- estimated that 17% of the rare nonsense mutants pre-
expression of the causal gene? A study that focused dicted to trigger NMD have heterogeneous effects
on the tumour suppressor breast cancer gene BRCA1, across tissues59. Another example of a tissue-specific
which maintains genome stability, showed that haplo­ post-transcriptional process is the bystander effect, a
insufficiency for BRCA1 led to cell type-specific phenomenon whereby a broadly expressed misfolded
genomic instability and premature senescence, parti­ protein overloads the proteostasis machinery, and thus
cularly in human mammary epithelial cells56. On a larger compromises the correct folding of unrelated meta­
scale, analysis of 51 cancer-initiating genes that are stable proteins60. In Caenorhabditis elegans, for exam-
causal for heritable cancer syndromes revealed that ple, the broadly expressed misfolded DAF-28 protein
loss-of-function cancer genes were under-expressed led to neuron-specific dysfunction due to disrupting the
in susceptible tissues, whereas gain-of-function cancer normal biogenesis of the unrelated endogenous DAF-7
genes were overexpressed (although some challenges protein; DAF-7 is expressed specifically in neurons and
with the robustness of the computational approach its secretion is critical for proper developmental signal-
were indicated)1. This analysis suggests that, in loss-of- ling60. Due to their indirect nature, these downstream
function cancer aberrations, the protective function processes are challenging to decipher.
of the compromised under-expressed gene becomes
insufficient. In gain-of-function cancer aberrations, Tissue-restricted functional redundancy. In the
the enhanced malfunction of the aberrant gene product aforementioned mechanisms, the causal genes had
is toxic. The latter effect was also observed in multiple tissue-specific properties, functions or impacts. Here,
protein-misfolding diseases, such as synucleinopathies, in contrast, pathogenesis might emerge because of
where aggregates containing the α-synuclein protein limited expression of a different gene or entity, which
accumulate in neurons or glial cells8. is functionally redundant with the causal gene and acts
as its backup. Accordingly, the backup masks the causal
Tissue-specific post-transcriptional processes. Gene aberration across unaffected tissues, but is insufficient
products typically undergo various types of post- or absent specifically in susceptible tissues. One type of
transcriptional processing that can affect their half- backup is a paralogue (a homologue) of the causal gene.
life and function, making the relationship between Paralogous genes have been repeatedly shown to have
transcript levels, protein levels and protein activity redundant functionality and to compensate for the loss
quite complicated and gene specific57. By affecting of each other in multiple species61–64 (reviewed in ref.65),
the final gene products, post-transcriptional process- and dosage sharing between paralogues is suggested to
ing that occurs in a tissue-specific manner can lead be one of the first steps following gene duplication66.
to tissue-specific impact. One example is the post- In several cases, backup by paralogues has been shown
transcriptional process of nonsense-mediated decay to be dosage dependent67,68. For example, the essenti-
(NMD), which is responsible for eliminating mRNA ality of the paralogous human helicases DDX3Y and
transcripts that contain premature stop codons. DDX3X increased when the expression level of the other
NMD was shown to exhibit some tissue specificity, paralogue decreased, and vice versa64.
which could in turn affect clinical manifestation in a The role of paralogues of causal genes was recently
tissue-specific manner. For example, Schmid meta- tested in a systematic manner across 112 tissue-specific
physeal chondrodysplasia is caused by heterozygous hereditary diseases2. In 43% of the cases, paralogues
Paralogue
A homologous gene present in
mutations in the COL10A1 gene encoding collagen were shown to be under-expressed relative to the causal
the same organism, typically X expressed in cartilage. In patients with premature gene specifically in the susceptible tissue. Their lower
having redundant functionality. stop-codon mutations, mutant mRNAs were completely expression suggested that their backup capability was

NaTure RevIewS | GeNeTICs volume 21 | March 2020 | 141


Reviews

Mechanism Affected tissue Unaffected tissue Fig. 2 | Illustration of the different mechanisms under­
lying tissue-selective manifestation of heritable traits
a Expression-based mechanisms and diseases. Each row of the figure corresponds to a
mechanism and portrays its unique impact in the affected
1 Exclusive expression
tissue versus unaffected tissues. Although each mechanism
is portrayed separately for clarity , they are often inter­
dependent. Below we use ‘trait genes’ (marked red) to
2 Preferential expression refer collectively to disease-causing genes and risk alleles.
a | Mechanisms pertaining to the relative abundance of
trait genes. Exclusive or preferential expression of the trait
gene in the susceptible tissue (rows 1,2). Distinct post-
3 Post-transcriptional transcriptional processing of trait genes in the susceptible
processes tissue, such as enhanced nonsense-mediated decay (scis-
sors) of the trait gene’s transcript (row 3). A compensatory
factor (pentagon) that is functionally redundant with the
4 Reduced functional trait gene, is downregulated particularly in the susceptible
redundancy
tissue (dashed lines), and thus unable to provide sufficient
F1 F2 F3 F4 F5 F1 F2 F3 F4 F5 compensation. F1 to F5 correspond to the functions of the
trait gene and its compensatory factor, with functions F2 to
b Regulatory mechanisms F4 being common to both (row 4). b | Mechanisms pertaining
1 Disrupted regulatory to regulatory elements. Disruption of regulatory sites
elements accessible specifically in susceptible tissues, such as a
genomic region (red line) free of nucleosomes (row 1).
Disruption of expression quantitative trait loci (eQTLs) that
2 Effects of eQTLs act specifically in the susceptible tissue. This subclass of
genomic regulatory elements (red rectangle) affects the
expression levels of target genes (red arrow), located in cis
or in trans to the eQTL, in a tissue-dependent manner (row 2).
3 Disrupted chromatin Disruption of chromatin contacts (red region) alters the
contacts
spatial organization of genomic regions, thereby affecting
tissue-specific enhancer–promoter interactions (row 3).
c | Tissue-disrupted networks. Disruption of molecular
c Tissue-disrupted networks interactions of trait genes that occur exclusively in the
susceptible tissue will impact that tissue alone (row 1).
1 Disrupted molecular
interactions Disruption of a tissue-specific pathway , denoted as series
of arrows, that involves the trait gene. The pathway con-
verges specifically in the susceptible tissue and is not
2 Disrupted pathway active in unaffected tissues (row 2). Disruption of gene
modules (red shaded regions). The module that contains
known trait genes and suspected trait genes (pale-coloured)
is formed specifically in the susceptible tissue and not
3 Disrupted gene- in unaffected tissues (row 3). d | Non-cell-autonomous
module integrity mechanisms. The susceptible tissue exclusively responds
to a ubiquitous signal (grey circles) via a tissue-specific
receptor, and thus disruption of this response will impact
d Non-cell-autonomous mechanisms this tissue alone (row 1). The susceptible tissue responds via
a ubiquitous receptor to a signal that occurs exclusively in
1 Responsiveness to its microenvironment (row 2).
signals

treatment works by manipulating the splicing of its


2 Microenvironment paralogous gene, SMN2, to encode an otherwise lowly
expressed SMN2 isoform that is similar to the wild-type
SMN1 (refs77,78).

Regulatory mechanisms
compromised in the susceptible tissue, thereby making This class of mechanisms pertains to aberrations in
that tissue more sensitive to the effects of the aberration2. genomic regions containing regulatory elements.
Additional functional redundancies, such as alternative Such aberrations are typical of complex diseases and
metabolic pathways69, common regulation by distinct traits mapped via GWAS, and often correspond to
transcription factors (TFs) 70, higher-order genetic non-coding, small-effect risk alleles79. Although the
interactions71,72 and nonsense-induced transcriptional functional consequence of many risk alleles remains
compensation73,74, are yet to be examined as disease obscure, several tissue-selectivity mechanisms were
modifiers in a large scale. Notably, backup mecha­ demonstrated across traits (Fig.  2b) . Below, we dis-
nisms are already of therapeutic interest75,76. For exam- cuss regulatory mechanisms ranging from local and
ple, spinal mus­cular atrophy is caused by aberration small-scale effects on proximal genes to remote effects
in the SMN1 gene. Its antisense oligonucleotide-based across larger chromosomal scales.

142 | March 2020 | volume 21 www.nature.com/nrg


Reviews

Disruption of gene-specific regulatory elements. Disruptions of chromatin contacts. Application of chromo­


Gene-specific regulatory elements have been repeat- some conformation capture techniques revealed that
edly shown to function in a tissue-selective manner. For the human genome is partitioned into megabase-scale
example, transcriptional regulatory interactions were topologically associated domains (TADs)86,87 that can
shown to be more tissue specific than expected by the bridge the distances between enhancers and promot-
expression of the TFs and target genes themselves51. The ers and thus affect gene regulation. Genetic aberrations
disruption of gene-specific regulatory elements could that disrupt TADs were shown to rewire enhancer–
therefore elicit tissue-specific phenotypes, as thoroughly promoter interactions and lead to the misexpression of
demonstrated by the case of the obesity-associated locus genes, resulting in pathogenic phenotypes such as limb
FTO. Analysis of the chromatin state of FTO across 127 malformations88. Whereas many TADs were shown to
human cell types revealed that it harbours an enhancer be highly conserved across tissues, a subclass of fre-
that is specific to pre-adipocyte cells80. Further analysis quently interacting regions showed strong tissue speci-
revealed that the FTO risk allele disrupts a conserved ficity in local chromatin interactions and was found near
repressor motif, thereby leading to de-repression of its tissue-specific genes23. The tissue-aware consideration of
target genes, and consequently to a developmental shift TAD structures, their boundaries and their position rela­
towards pro-obesity80. tive to genes and enhancers could therefore illuminate
Systemic cross-trait studies of the relationships the tissue-specific pathogenicity of genetic aberrations89.
between gene-specific regulatory elements and non-
coding risk alleles showed that non-coding risk alleles Tissue-disrupted networks
tend to be enriched in regulatory regions that were The mechanisms below refer to alterations in the molecu­
active in trait-susceptible tissues81,82. For example, single- lar interaction patterns of single genes or gene subsets,
nucleotide polymorphisms (SNPs) associated with which are relevant to both monogenic and polygenic
plasma low-density lipoprotein concentration over- diseases (Fig. 2c).
lapped significantly with peaks of the activation marker
H3K4me3 in liver83, and SNPs associated with type 2 Disrupted tissue-specific molecular interactions. Widely
diabetes mellitus and related traits were enriched speci­ expressed proteins can carry different functions across
fically in enhancers detected in pancreatic islets84. tissues by interacting with distinct molecules per tis-
Subsequently, enrichment of trait-associated SNPs sue90, allowing a ubiquitous aberration to manifest as a
within active regulatory regions became a hallmark of tissue-specific phenotype. For instance, the gene dystro-
pathogenic tissues and was utilized to illuminate patho- glycan 1 (DAG1), which is causal for muscular dystrophy–
genic cell types81. Yet, the relationship between risk alleles dystroglycanopathy, encodes a globally expressed cell
and active regulatory elements is not straightforward. adhesion receptor. Specifically in muscle, DAG1 forms
A study of tissue-specific regulation inferred by inte- the dystrophin–glycoprotein complex by interacting
grating data of active chromatin noted only modest with DMD91 and CAV3 (ref.92) proteins, which are expres­
enrichment of SNPs associated with Crohn’s disease, sed primarily in muscle, thereby explaining its muscle-
rheumatoid arthritis and schizophrenia within tissue- specific pathology. A systematic analysis of over 300
specific versus broadly active regulatory elements, sugges­ diseases showed that, like DAG1, causal genes tend to
ting that additional genes expressed in disease-relevant form tissue-specific protein–protein interactions (PPIs)
cells might impact pathogenecity82. preferentially in susceptible versus unaffected tissues3.
Do disease-causing mutations in such genes indeed
Effects of eQTLs. Expression quantitative trait loci (eQTLs) affect PPIs? A large-scale yeast two-hybrid analysis of
are a class of non-coding regulatory elements whose poly­ 220 causal genes revealed that 61% of the disease-causing
morphism correlates with variation in the expression alleles exhibited partial loss of their wild-type PPIs93
levels of one or more genes85. eQTLs can reside in cis or in (denoted edgetic perturbation) 94. Taken together,
trans with their target genes, and lead to their upregulation these studies suggest that disease-causing mutations
or downregulation in a tissue-dependent manner. Owing could perturb PPIs that are specific to the susceptible
to the massive genome and transcriptome sequencing tissue, thereby contributing to tissue-specific disease
carried by the GTEx consortium, thousands of eQTLs manifestation3.
were identified across tissues14. Tissue eQTLs were tested
for association with complex traits by calculating their Disrupted tissue-specific pathways and gene modules.
colocalization with risk alleles for traits that manifest in Pathways are biological processes performed by sub-
the same tissues24,45. For example, a systematic study of sets of gene products and other molecules, and may
complex traits and diseases, including metabolic, cardio- involve metabolic reactions, as in glycolysis, or regula-
vascular, anthropometric, autoimmune and neurodegen- tory and protein interactions, as in signalling cascades.
erative traits, showed that eQTLs identified in susceptible The activity of pathways varies across tissues, with some
tissues were significantly enriched for trait-associated pathways being highly tissue specific and thus causing
Expression quantitative SNPs45. This suggested that trait-associated SNPs act tissue-specific pathologies when disrupted. For instance,
trait loci by altering gene regulation in trait-related tissues. the glycogenolysis pathway, whereby the polysaccha-
(eQTLs). Genomic regions However, as mentioned for the preceding mechanism, ride glycogen is degraded, takes place mainly in liver
containing DNA sequence
variants that influence the
in most traits, enriched eQTLs were generally tissue and muscle, and thus glycogen storage diseases95, which
mRNA expression level shared and not tissue specific45, again suggesting the are caused by aberrations in glycogenolysis enzymes,
of a gene. involvement of additional genes and/or mechanisms. manifest primarily in those tissues.

NaTure RevIewS | GeNeTICs volume 21 | March 2020 | 143


Reviews

Whereas pathways can be delineated, gene modules across tissues; however, certain tissues maintain their
are defined more loosely as subsets of functionally own unique environment via a functional barrier that
related genes that tend to be co-expressed, encode separates them from the blood, such as the blood–brain
products that physically interact with each other, or lead and blood–testis barriers. The importance of the tissue
to similar phenotypes when individually perturbed96. microenvironment in general, and of tissue barriers in
The concept of gene modules can be extended to that particular, is revealed by diseases that are associated with
of disease modules, which comprise genes associated barrier leakage. For example, blood–brain barrier leak-
with a certain disease28,97. The incorporation of addi- age was observed in patients with Alzheimer disease101
tional functionally related genes into disease modules and disruption of the blood–testis barrier was associated
could illuminate disease mechanisms. For example, with male infertility102.
a disease module of asthma that incorporated the Notably, the tissue microenvironment is also critical
local interactome neighbourhood of known asthma for therapy. For example, the Gaucher lysosomal storage
genes was enriched for genes with modest P values disorder results from mutations in the glucocerebrosi­
in an asthma GWAS, and for genes that were differ- dase enzyme that catalyses the breakdown of gluco­
entially expressed in response to treatment with an cerebroside, which consequently builds up in lysosomes.
asthma-specific drug27. Gaucher types differ in the tissues they affect: Gaucher
Similarly to pathways, modules may be estab- type 1 (where brain is spared) and the non-neurological
lished in a tissue-dependent manner47,98. A study of symptoms of Gaucher type 3 are treated with enzyme
37 GWAS traits showed that risk alleles often perturb replacement therapy; this treatment cannot be applied
regulatory gene modules that were highly specific to to the neurological symptoms of Gaucher type 3 due to
disease-relevant tissues47. Another study of disease the blood–brain barrier.
modules for 70 tissue-specific polygenic diseases, The fate and function of immune cells are parti­
which determined module connectivity per tissue by cularly affected by the tissue microenvironment103.
the interactome connectivity between the genes pref- Memory T cells, for example, which are central players
erentially expressed in that tissue, showed that disease in the adaptive immune response, were shown to
module connectivity was significantly high particularly acquire distinct phenotypes, depending on tissue-
in disease-related tissues98. These studies suggest that specific environmental cues104. Likewise, tissue-resident
the integrity of a disease module is important for its macro­phages present tissue-specific functional charac­
tissue-specific functionality and that aberrations that teristics105, such as their capability for clearing mutant
compromise this integrity could lead to tissue-specific cells 106. The impact of the microenvironment on
pathologies. the emergence of autoimmune disorders is demon-
strated by coeliac disease, which manifests in the
Non-cell-autonomous mechanisms small intestine upon exposure to gluten protein pre­
The mechanisms below involve inter-cellular inter­ sent in grains. Coeliac disease tends to be inherited in
actions and pertain to monogenic and polygenic diseases families harbouring a specific isoform of the HLA-DQ
(Fig. 2d). protein that is part of a major histocompati­bility
complex (MHC) class II antigen-presenting cell
Tissue-specific responsiveness to signals. Tissue-specific receptor. The isoform-bearing receptor has higher
proteins tend to be enriched for proteins function- affinity to gluten-related peptides, the exposure to
ing as receptors99, indicating that one determinant of which is limited to the gastrointestinal tract, thereby
tissue-specific functions is the variance in the respon- contributing to the activation of a tissue-specific
siveness of tissues to signals. The impact of such signals autoimmune process.
on disease predisposition was recently demonstrated for
breast and ovarian cancers caused by familial mutations Further understanding tissue-specific mechanisms
in BRCA1 (ref.100). Breast and ovaries are the primary Many more mechanisms that underlie tissue selec-
tissues that express the oestrogen receptor and thus tivity await discovery. Mechanisms involving post-
respond to oestrogen. A study showed that oestrogen transcriptional modifications as well as mechanisms that
signalling induces DNA double-strand breaks in breast pertain to the cellular composition and spatial organiza-
cancer cells and epithelial cells of mouse mammary tion of each tissue have typically remained hidden since
glands, leading to pathological topoisomerase II–DNA molecular profiling has been mostly applied to trans­
complexes; these pathological complexes are eliminated criptomes, and particularly to transcriptomes of bulk tis-
by the normal function of BRCA1, yet accumulate when sues or cell populations. Additionally, broad surveys of
BRCA1 is depleted100. This mechanism demonstrates tissue expression across many tissues have largely been
how tissue-specific pathogenesis can emerge, despite done on unaffected individuals, most of which were
both the signal and the disease-causing gene being adults, and thus do not reflect the impact of germline
widely expressed. aberrations on unaffected tissues nor reveal expression
patterns in early developmental stages and childhood.
Tissue-specific microenvironment. Cells continuously Inclusion of emerging types of data, such as single-cell
interact with their microenvironment by cell–cell phys- measurements, proteomics and spatiotemporal profiles,
ical contacts and by processing chemical signals in the as well as broad surveys of developmental stages107 and of
form of diffusible molecules. Blood-transferred mole- disease states across affected and unaffected tissues108,109,
cules and infiltrating immune cells tend to be shared will refine current observations and unravel additional

144 | March 2020 | volume 21 www.nature.com/nrg


Reviews

molecular mechanisms underlying tissue-selective and strategies that we describe are exploratory; they can
susceptibility. serve to refute candidate mechanisms and to support
or speed the detection of potentially relevant mecha-
Exploring tissue-selective mechanisms nisms. As with other exploratory investigations, poten-
The identification of the molecular mechanism under­ tial mecha­nisms should be scrutinized via literature
lying disease manifestation is a huge challenge that often mining and verified experimentally. We open this sec-
remains unmet even for well-studied diseases. Yet, by tion with tools that can be used to associate a disease
focusing on the narrower question of illuminating or trait of interest with its causal genes, risk alleles and
the molecular basis of their tissue selectivity, we may affected tissues. We then describe tools for the explor-
obtain important insight that could enhance the search atory analysis of potential mechanisms pertaining to
for treatment80,110. This narrower question can now be protein-coding and non-coding disease genes and
approached owing to the tremendous amounts of molec- risk alleles (Fig. 3). We end with a short description of
ular data of diseases and tissues that have become avail­ experimental strategies.
able for exploration. Below, we point to publicly avail­able
large-scale databases and web tools that allow non- Connecting traits and diseases to genetic factors and
computational investigators to harness this extensive susceptible tissues
information towards a better under­standing of the Given a heritable disease or trait, this step aims to gather
mechanisms underlying the tissue selectivity of their available information on their known causal genes and
disease of interest. These resources have also been risk alleles as well as the tissues and cell types in which
of great importance to computational researchers they manifest, thereby providing the foundation for
and enabled many of the studies cited herein. The tools studying their tissue selectivity (Table 2). Investigations of

Tissue-selective Experimental models


trait/disease
Cell line
TissueNet (cellular focus)
GIANT
GTEx DifferentialNet Reactome
OMIM HumanBase
NHGRI–EBI GWAS HPA
HPO/DO Network
Orphanet Expression Pathways and
modules Organoid
(tissue focus)
Coding

Non-coding
Animal model
Regulatory Chromatin (in vivo/inter-tissue focus)
elements contact maps
eQTLs
ENCODE 3D Genome
Roadmap Epigenomics GTEx
FANTOM5
IHEC

Fig. 3 | A schematic flow chart illuminating tissue-selectivity mechanisms. The analysis starts with a tissue-selective
trait or disease (left). The underlying genetic factors can be extracted from Online Mendelian Inheritance in Man (OMIM)
and the US National Human Genome Research Institute–European Bioinformatics Institute Catalogue of published
genome-wide association studies (NHGRI–EBI GWAS). Additionally , affected tissues can be deduced from Human
Phenotype Ontology (HPO), Disease Ontology (DO) and Orphanet. The analysis continues with exploration of potential
mechanisms that might explain the tissue-selective disease manifestation, by using tissue-aware resources and tools
(middle). Mechanisms pertaining to protein-coding genes (upper part) include expression-based mechanisms, which can
be explored via resources such as the Genotype–Tissue Expression (GTEx) and the Human Protein Atlas (HPA) portals;
interaction-based mechanisms, which can be evaluated via tools such as TissueNet, Genome-wide Integrated Analysis of
gene Networks in Tissues (GIANT), and DifferentialNet; and pathway-based and module-based mechanisms, which can be
inferred from resources such as Reactome and HumanBase. Mechanisms pertaining to non-coding variations (lower part)
include aberrant tissue-specific regulatory elements, which can be deduced from resources such as the Encyclopedia of
DNA Elements (ENCODE); aberrant tissue expression quantitative trait loci (eQTLs), available through GTEx; and aberrant
chromatin contacts, available via, for example, 3D Genome. Additional mechanisms, tools and resources are mentioned
in the main text. The analysis concludes with experimental testing of putative mechanisms in tissue-aware modelling
systems (right). These include modelling systems aimed at testing cellular mechanisms such as tissue culture and cell lines,
modelling systems aimed at testing intra-tissue mechanisms such as organoids, or modelling systems aimed at testing
in vivo, inter-tissue mechanisms such as model organisms. FANTOM5, Functional Annotation of the Mammalian Genome;
IHEC, International Human Epigenome Consortium.

NaTure RevIewS | GeNeTICs volume 21 | March 2020 | 145


Reviews

Mendelian diseases typically start with querying the estab- pathway or module genes, or upstream receptors.
lished and manually curated OMIM database35. OMIM Additional tissue expression resources appear in Table 1.
points to the genetic factors that are known or suspected
to cause Mendelian diseases, along with richly curated Exploring tissue-specific regulatory mechanisms
phenotypic, molecular and clinical data. Investigations of Multiple non-coding variants and risk alleles, which
risk alleles for complex diseases and traits can addition- are often identified in GWAS, lead to phenotypes by
ally query the US National Human Genome Research affecting gene regulation. However, the interpretation
Institute–European Bioinformatics Institute (NHGRI– of these variants is typically more complex than that of
EBI) GWAS Catalogue111. To associate a disease or trait protein-coding genes and multiple approaches with vary-
of interest with its clinical manifestation, including its ing levels of complexity have been devised to address this
susceptible tissues, pathology-oriented ontology data- challenge24,116–119. Here, we focus on web tools and portals
bases come to aid. HPO36 and Disease Ontology37 cover that are amenable to exploration by non-computational
hundreds of diseases. HPO can be queried by dis- researchers. In general, these resources offer detailed
ease and/or phenotype to obtain its associated genes, views into the epigenome of multiple tissues and cell
affected anatomical systems and related diseases, which types, and include multiple histone marks (methylation
can be explored in a hierarchical manner (for exam- and acetylation at various sites), DNase I hypersensitiv-
ple, moving from kidney abnormality to renal mor- ity sites, TF binding sites, DNA methylation, predicted
phology). Orphanet, an online database dedicated to microRNA sites, enhancers, chromatin contacts, and
rare diseases, also presents phenotypic abnormalities more, which were mapped by ENCODE, Roadmap
by the frequency of their occurrence in patients, and Epigenomics32, FANTOM5 and other projects23 (Table 1).
thus could also illuminate causative versus secondary Besides portals dedicated to these projects, data can
phenotypes and tissues112. be accessed via epigenome-specific data portals such
as the International Human Epigenome Consortium
Exploring expression-based mechanisms (IHEC) portal120.
Although the fraction of tissue-exclusive and pref- A typical workflow with these tools is to select the
erentially expressed genes is not high (Fig. 1b,c), it is epigenetic marks and the samples for analysis and view
worthwhile to check the expression patterns of causal them while focusing on the query region. Viewing is
genes in susceptible versus unaffected tissues, as a enabled via dedicated regulation tracks in browsers
clear differential expression could immediately indi- such as the UCSC genome browser and the WashU epi­
cate a causal mechanism1,3. Several resources enable genome browser. The genetic variation track that con-
the rigorous examination of the expression patterns tains common and pathological variations could further
of genes across tissues (Table 1). GTEx and HPA offer illuminate the likely impact of variants of interest in the
user-friendly views of the expression of wild-type genes query region. The involvement of a query variant in
across tissues and are based on analysis of hundreds long-range chromatin interactions within tissues can
and tens of donors, respectively. GTEx additionally be observed via additional browsers121, such as the 3D
supports gender-based views and enables analysis of Genome Browser122. The tools described above are not
tissue-specific splicing by presenting the expression lev- quantitative or probabilistic and thus are more suitable
els of exons, exon junctions and isoforms per gene across for qualitative analysis.
tissues (note that only reads that mapped to known Examples of web tools offering probabilistic analy-
exons are considered, leaving new exons undetected). ses include FORGE, which identifies cell type-specific
HPA additionally annotates genes as preferentially enrichments of query variants in DNase I hypersen-
expressed and portrays protein expression levels across sitivity sites123, and eFORGE, which identifies cell
tissues. Despite GTEx and HPA datasets not contain- type-specific regulatory signals in a set of differentially
ing overt information on post-transcriptional changes methylated positions identified in epigenome-wide asso-
and focusing on samples from adults, their ease of use ciation studies117. The involvement of a query variant in
makes them a focal point for querying the expression tissue-specific eQTLs and splice QTLs can be assessed
patterns of causal genes, risk alleles and other factors. via the GTEx portal, which calculates the likelihood
Data of gene expression patterns at the cell-type level per query.
are also emerging via single-cell resources, such as the Although each specialized tool may seem detached,
Human Cell Atlas16, and are already viewable via mouse the combination of epigenome datasets that can identify
cell atlases113–115. regulatory elements active in given tissues, expression
Notably, expression resources might also illumi- datasets that can show which transcripts are expressed
nate pathogenic versus affected tissues and cell subsets, in a given tissue, and chromatin contact maps that link
such as when the causative cells or tissues express the potential enhancers and promoters, could provide a big
causal gene but the affected cells or tissues do not, as leap forward in interpreting tissue genotype–phenotype
demonstrated for kidney diseases43. Likewise, alongside relationships80,124.
exploring the expression patterns of causative genes and
risk alleles, these resources can be used to explore the Exploring network-based mechanisms
expression patterns of functionally related genes that Functional understanding of the tissue-specific impact
might impact disease manifestation in a tissue-specific of variation on protein functions can be gained by
manner, such as molecular interactors (for example, exploring the physical interactions and functional rela-
kinases), modifier genes (such as a paralogue), relevant tionships among proteins in tissue contexts and by using

146 | March 2020 | volume 21 www.nature.com/nrg


Reviews

tissue-aware tools for the exploration of pathways and susceptible tissues. The association of genes to known
network modules. pathways can be explored via the Reactome pathway
database131. Reactome records hundreds of curated
Tools for exploring molecular relationships in tissue pathways and presents their activity across tissues
contexts. The tissue-specific impact of a ubiquitously according to the tissue expression levels of pathway
expressed disease gene could be mediated by the tissue genes (similar approaches to estimate the tissue activi-
specificity of its molecular interactions. How extensive ties of known cellular and metabolic pathways have been
is our knowledge of such interactions? Transcriptional used in multiple studies)132–134. Apart from exploring
regulatory interactions have been mapped across tissues tissue-specific pathways, the set of disease-associated
and cell lines for several TFs32,33. PPIs, by contrast, despite genes can also be compared against ubiquitous path-
them being extensively mapped (over 380,000 PPIs ways that might be upregulated in the susceptible tis-
involving ~20,000 proteins have been detected experi- sue. For example, a recent analysis of PD-associated
mentally), have often been measured in model systems genes revealed that they were significantly enriched
and thus lack tissue contexts. To partially close this gap, in a lysosomal gene set that was highly expressed in
tissue contexts for PPIs have been inferred from tissue astrocytic, microglial and oligodendrocytic subtypes,
expression profiles. Accordingly, PPIs involving proteins suggesting that PD risk is associated with a perturba-
that were absent or lowly expressed in a given tissue were tion of a ubiquitous cellular process affecting a range of
disregarded or downplayed in that tissue125,126. cellular subtypes9.
As tissue proteomic profiles were limited, the infer- In addition to exploring known pathways, one can
ence of tissue contexts for PPIs relied on tissue trans­ explore whether disease-associated genes constitute a
criptomes, under the assumption that transcript levels gene module within the interactome of the suscepti-
correlate with protein abundance. This assumption has ble tissue. This analysis is available via HumanBase9,
been much debated, yet multiple studies found mode­ which also offers tools that use tissue interactomes for
rate correlations at steady state19,127, especially when interpreting GWAS31. Given that a recent study found
gene-specific conversions from transcript to protein that common genetic variation affects both overall
levels were applied57. Attesting to the validity of this risk and clinical presentation of severe neurodevelop-
gap-closing procedure, tissue molecular interaction mental brain disorders that were likely to be mono-
networks (interactomes) comprising transcription regu­ genic135, network-based approaches could be harnessed
lation47,51, PPIs3,128 or functional relationships (also to illuminate the contribution of common alleles in
including co-expression and other types of non-physical a tissue.
relationships)31 were shown to be advantageous over
unfiltered interactomes in uncovering protein functions, Notes on experimental model systems
prioritizing disease genes or interpreting GWAS47,128. Experimental assessment of candidate tissue-specific
Several tools enable the exploration of physical inter- mechanisms is a complex task. Part of the challenge is
actions and other functional relationships for genes of specific to the traits and mechanisms in question such
interest and their products (Table 3). Some tools addition- as defining an indicative phenotypic readout and assess-
ally highlight tissue-specific inter­actions, which could ing the potentially minor effect of a single mechanism.
be extremely useful for illuminating tissue-selectivity A more general aspect of the complexity relates to the
mechanisms. These include TissueNet for PPIs129, experimental system representing the different types
Genome-wide Integrated Analysis of gene Networks of tissues. Cell lines in culture are the simplest models;
in Tissues (GIANT) for functional relationships31, and however, they do not fully recapitulate the physiology of
DifferentialNet for upregulated or downregulated PPIs130. intact tissues in vivo and certainly not inter-tissue com-
munication. By contrast, in vivo models are typically
Tools for exploring the tissue activity of pathways and challenging, expensive and time consuming.
gene modules. Tissue-specific pathogenesis might stem A model system that has arisen in recent years is
from the involvement of causal genes and risk alleles induced pluripotent stem cells that are subsequently dif-
in pathways and gene modules that are specific to ferentiated into cell types of interest136. Patient-derived
cells can be used to model tissues that might be inacces-
Table 3 | Representative tissue-aware databases of molecular relationships sible or challenging to grow such as neurons. Moreover,
Database Details Ref.
they can be genetically manipulated, such that isogenic
pairs with and without a relevant mutation can be created
TissueNet PPIs, highlights tissue-specific versus ubiquitous proteins 129
and differentiated into distinct cell types. The cell types
as well as differentially expressed proteins
that are functionally affected by the mutation relative to
DifferentialNet Differential PPIs, highlights causal genes 130
the non-mutant cells can then be revealed. The mode
HIPPIE PPIs, confidence scored 162
of action of the mutation could be tested, for example,
HumanBase Functional interactions, confidence scored 31
by comparing the expression profiles of these otherwise
SPECTRA PPIs, tissue and tumour networks 163 genetically identical cells or by comparing the interaction
GNAT Co-expression relationships 164 profiles of the wild-type and mutant gene products.
As an alternative approach, organoids provide a
Reactome Expression of pathway-associated proteins 131
system that encompasses multiple cell types and their
GNAT, Genetic Network Analysis Tool; HIPPIE; Human Integrated Protein–Protein Interaction
Reference; PPI, protein–protein interaction; SPECTRA , SPECific Tissue/Tumour Related PPI interactions but without much of the organismal com-
networks Analyser. plexity or time delays of full in vivo models137. Similarly

NaTure RevIewS | GeNeTICs volume 21 | March 2020 | 147


Reviews

to haematopoietic transplants, organoids can be crea­ mechanisms await characterization, such as develop-
ted with mixed genotypes and utilized to assess the mental mechanisms124, secondary modifier genes144,
effect of mutant cells on other, non-mutant cell types. and the impact of common variants135 and environmen-
Additionally, the emerging technique of cell-lineage tal factors145. Whereas some mechanisms were exposed
tracking can be used to track developmental disorders, owing to disease-directed efforts, as in the case of
thereby unravelling the stages and cell types in which BRCA1 (ref.100), other mechanisms were revealed owing
the mutation has a functional effect 138. Combined to recent advancements in high-throughput omics tech-
with the huge progress in genetic editing of model sys- niques, as in the case of CF11,26, and the emergence of
tems and organisms, recently enhanced with tools for publicly available online tools and resources presenting
efficient polygene editing139, these technologies pave heterogeneous omics datasets, as in the cases of FTO80
the way for easier modelling and, ultimately, a better and ICR124 (Fig. 3). Continued technological progress and
understanding of disease aetiologies. reduction in costs are likely to expand the array of omics
techniques towards interrogating the meta­bolome, pro-
Conclusions and future perspectives teome146, microbiome, inter-cellular communication
Tissue selectivity of traits and diseases is a widespread networks29,147,148 and other ‘–omes’ across a variety of
phenomenon that encompasses hundreds of heritable physiological systems and diversified groups of cohorts.
monogenic and polygenic diseases and traits (Fig. 1). Largely owing to the huge amount of data they hold,
It is also relevant for somatic genetic diseases, including omics resources are sometimes perceived as overwhelm-
cancers, as many cancer-driver genes are ubiquitously ing and are consequently used only for preliminary
expressed and yet the cancers they elicit are enigmat- qualitative analyses. Sophisticated analytical tools and
ically tissue selective140 (Box 1). The understanding of statistical methods continuously improve; however,
the pathophysiological mechanisms that render certain many of them are employed only by computational
tissues susceptible to genetic perturbation while experts. Given the large numbers of traits and diseases
other tissues remain robust is important, not only for awaiting resolution and the critical needs of patients,
genetically solved diseases, but also for unsolved dis- more efforts should be invested in making these power­
eases. The tendency of disease genes to be preferen- ful tools and resources accessible and comprehensible
tially expressed and functionally related to each other to a wider community of non-computational researchers
in affected tissues, or to colocalize with tissue eQTLs, and clinicians. Together with investing in holistic studies
was harnessed for the identification of causal fac- of ubiquitously expressed disease genes that may lead to
tors31,116,118,141, and simi­lar approaches helped illuminate tissue-specific gene-dependent and variant-dependent
affected tissues and cell types38,142. These pathophysio- phenotypes, these efforts will facilitate the discovery of
logical mechanisms could also be employed for develop- pathophysiological mechanisms and consequently con-
ing tissue-selective treatments as demonstrated for the tribute to better diagnosis, understanding and treatment
calcium-sensing receptor143. of traits and diseases.
We described multiple types of mechanisms that
underlie tissue selectivity (Fig. 2), yet many important Published online 8 January 2020

1. Lage, K. et al. A large-scale analysis of tissue-specific 9. Reynolds, R. H. et al. Moving beyond neurons: the role 20. ENCODE Project Consortium. An integrated
pathology and gene expression of human disease of cell type-specific gene regulation in Parkinson’s encyclopedia of DNA elements in the human genome.
genes and complexes. Proc. Natl Acad. Sci. USA 105, disease heritability. NPJ Parkinsons Dis. 5, 6 (2019). Nature 489, 57–74 (2012).
20870–20875 (2008). 10. Li, X. & Zhang, P. Genetic determinants of myocardial An expansive resource of epigenetic and regulatory
A pioneering large-scale analysis of expression dysfunction. J. Med. Genet. 54, 1–10 (2017). signals in human cells.
mechanisms in the context of tissue-selective 11. Montoro, D. T. et al. A revised airway epithelial 21. Andersson, R. et al. An atlas of active enhancers
hereditary diseases. hierarchy includes CFTR-expressing ionocytes. Nature across human cell types and tissues. Nature 507,
2. Barshir, R. et al. Role of duplicate genes in 560, 319–324 (2018). 455–461 (2014).
determining the tissue-selectivity of hereditary Example of a cell type-specific mechanism observed 22. Roadmap Epigenomics Consortium et al. Integrative
diseases. PLOS Genet. 14, e1007327 (2018). via single-cell mapping of a healthy tissue. analysis of 111 reference human epigenomes. Nature
3. Barshir, R., Shwartz, O., Smoly, I. Y. & Yeger-Lotem, E. 12. Su, A. I. et al. A gene atlas of the mouse and human 518, 317–330 (2015).
Comparative analysis of human tissue interactomes protein-encoding transcriptomes. Proc. Natl Acad. An expansive resource of epigenetic signals in
reveals factors leading to tissue-specific manifestation of Sci. USA 101, 6062–6067 (2004). human cells.
hereditary diseases. PLOS Comput. Biol. 10, e1003632 13. Jongeneel, C. V. et al. An atlas of human gene 23. Schmitt, A. D. et al. A compendium of chromatin
(2014). expression from massively parallel signature sequencing contact maps reveals spatially active regions in
Tissue-specific protein interaction maps and their (MPSS). Genome Res. 15, 1007–1014 (2005). the human genome. Cell Rep. 17, 2042–2059
relationship to the tissue selectivity of hereditary 14. GTEx Consortium. et al. Genetic effects on gene (2016).
diseases. expression across human tissues. Nature 550, 24. Barbeira, A. N. et al. Exploring the phenotypic
4. Bamshad, M. J. et al. Exome sequencing as a tool for 204–213 (2017). consequences of tissue specific gene expression
Mendelian disease gene discovery. Nat. Rev. Genet. An unprecedented resource of transcriptomes and variation inferred from GWAS summary statistics.
12, 745–755 (2011). eQTLs across physiological human tissues. Nat. Commun. 9, 1825 (2018).
5. Moaven, N., Tayebi, N., Goldin, E. & Sidransky, E. 15. Uhlen, M. et al. Transcriptomics resources of human 25. Yao, V. et al. An integrative tissue-network approach
Rare Diseases Advances in Predictive, Preventive and tissues and organs. Mol. Syst. Biol. 12, 862 (2016). to identify and test human disease genes.
Personalised Medicine 69-90 (Springer Netherlands, 16. Rozenblatt-Rosen, O., Stubbington, M. J. T., Regev, A. Nat. Biotechnol. 36, 1091–1099 (2018).
2015). & Teichmann, S. A. The human cell atlas: from vision 26. Plasschaert, L. W. et al. A single-cell atlas of the
6. Holmans, P. A., Massey, T. H. & Jones, L. Genetic to reality. Nature 550, 451–453 (2017). airway epithelium reveals the CFTR-rich pulmonary
modifiers of Mendelian disease: Huntington’s disease 17. Collado-Torres, L. et al. Reproducible RNA-seq ionocyte. Nature 560, 377–381 (2018).
and the trinucleotide repeat disorders. Hum. Mol. analysis using recount2. Nat. Biotechnol. 35, Example of a cell type-specific mechanism observed
Genet. 26, R83–R90 (2017). 319–321 (2017). via single-cell mapping of a healthy tissue.
7. Hernandez, D. G., Reed, X. & Singleton, A. B. Genetics 18. Uhlen, M. et al. Proteomics. Tissue-based map of the 27. Sharma, A. et al. A disease module in the interactome
in Parkinson disease: Mendelian versus non-Mendelian human proteome. Science 347, 1260419 (2015). explains disease heterogeneity, drug response
inheritance. J. Neurochem. 139 (Suppl. 1), 59–74 An unprecedented proteomic resource across and captures novel pathways and genes in asthma.
(2016). physiological human tissues. Hum. Mol. Genet. 24, 3005–3020 (2015).
8. Goedert, M., Jakes, R. & Spillantini, M. G. The 19. Wilhelm, M. et al. Mass-spectrometry-based draft 28. Huttlin, E. L. et al. Architecture of the human
synucleinopathies: twenty years on. J. Parkinsons Dis. of the human proteome. Nature 509, 582–587 interactome defines protein communities and disease
7 (Suppl. 1), S51–S69 (2017). (2014). networks. Nature 545, 505–509 (2017).

148 | March 2020 | volume 21 www.nature.com/nrg


Reviews

29. Kveler, K. et al. Immune-centric network of cytokines 56. Sedic, M. et al. Haploinsufficiency for BRCA1 leads to 83. Trynka, G. et al. Chromatin marks identify critical cell
and cells in disease context identified by computational cell-type-specific genomic instability and premature types for fine mapping complex trait variants.
mining of PubMed. Nat. Biotechnol. 36, 651–659 senescence. Nat. Commun. 6, 7505 (2015). Nat. Genet. 45, 124–130 (2013).
(2018). 57. Edfors, F. et al. Gene-specific correlation of RNA and 84. Parker, S. C. J. et al. Chromatin stretch enhancer
30. Gandal, M. J. et al. Shared molecular neuropathology protein levels in human cells and tissues. Mol. Syst. Biol. states drive cell-specific gene regulation and harbor
across major psychiatric disorders parallels polygenic 12, 883 (2016). human disease risk variants. Proc. Natl Acad. Sci. USA
overlap. Science 359, 693–697 (2018). 58. Bateman, J. F., Freddi, S., Nattrass, G. & Savarirayan, R. 110, 17921–17926 (2013).
31. Greene, C. S. et al. Understanding multicellular Tissue-specific RNA surveillance? Nonsense-mediated 85. Albert, F. W. & Kruglyak, L. The role of regulatory
function and disease with human tissue-specific mRNA decay causes collagen X haploinsufficiency in variation in complex traits and disease. Nat. Rev.
networks. Nat. Genet. 47, 569–576 (2015). Schmid metaphyseal chondrodysplasia cartilage. Genet. 16, 197–212 (2015).
Tissue-specific networks of functional molecular Hum. Mol. Genet. 12, 217–225 (2003). 86. Dixon, J. R. et al. Topological domains in mammalian
relationships and an online tool. 59. Rivas, M. A. et al. Human genomics. Effect of predicted genomes identified by analysis of chromatin
32. Davis, C. A. et al. The Encyclopedia of DNA Elements protein-truncating genetic variants on the human interactions. Nature 485, 376–380 (2012).
(ENCODE): data portal update. Nucleic Acids Res. 46, transcriptome. Science 348, 666–669 (2015). 87. Nora, E. P. et al. Spatial partitioning of the regulatory
D794–D801 (2018). 60. Klabonski, L., Zha, J., Senthilkumar, L. & Gidalevitz, T. landscape of the X-inactivation centre. Nature 485,
33. Lizio, M. et al. Update of the FANTOM web resource: A bystander mechanism explains the specific 381–385 (2012).
high resolution transcriptome of diverse cell types phenotype of a broadly expressed misfolded protein. 88. Lupianez, D. G. et al. Disruptions of topological
in mammals. Nucleic Acids Res. 45, D737–D743 PLOS Genet. 12, e1006450 (2016). chromatin domains cause pathogenic rewiring of
(2017). 61. DeLuna, A. et al. Exposing the fitness contribution of gene-enhancer interactions. Cell 161, 1012–1025
34. Ramskold, D., Wang, E. T., Burge, C. B. & Sandberg, R. duplicated genes. Nat. Genet. 40, 676–681 (2008). (2015).
An abundance of ubiquitously expressed genes 62. Conant, G. C. & Wagner, A. Duplicate genes and 89. Kaiser, V. B. & Semple, C. A. When TADs go bad:
revealed by tissue transcriptome sequence data. robustness to transient gene knock-downs in chromatin structure and nuclear organisation in
PLOS Comput. Biol. 5, e1000598 (2009). Caenorhabditis elegans. Proc. Biol. Sci. 271, 89–96 human disease. F1000Res 6, 314 (2017).
35. Amberger, J., Bocchini, C. A., Scott, A. F. & Hamosh, A. (2004). 90. Bossi, A. & Lehner, B. Tissue specificity and the human
McKusick’s online Mendelian Inheritance in Man 63. White, J. K. et al. Genome-wide generation and protein interaction network. Mol. Syst. Biol. 5, 260
(OMIM). Nucleic Acids Res. 37, D793–D796 (2009). systematic phenotyping of knockout mice reveals (2009).
A comprehensive resource for Mendelian disorders. new roles for many genes. Cell 154, 452–464 91. Ilsley, J. L., Sudol, M. & Winder, S. J. The interaction
36. Kohler, S. et al. The human phenotype ontology in (2013). of dystrophin with β-dystroglycan is regulated by
2017. Nucleic Acids Res. 45, D865–D876 (2017). 64. Wang, T. et al. Identification and characterization of tyrosine phosphorylation. Cell Signal. 13, 625–632
37. Kibbe, W. A. et al. Disease ontology 2015 update: essential genes in the human genome. Science 350, (2001).
an expanded and updated database of human diseases 1096–1101 (2015). 92. Sotgia, F. et al. Caveolin-3 directly interacts with the
for linking biomedical knowledge through disease data. 65. Diss, G., Ascencio, D., DeLuna, A. & Landry, C. R. C-terminal tail of β-dystroglycan. Identification
Nucleic Acids Res. 43, D1071–D1078 (2015). Molecular mechanisms of paralogous compensation of a central WW-like domain within caveolin family
38. Finucane, H. K. et al. Heritability enrichment of and the robustness of cellular networks. J. Exp. Zool. members. J. Biol. Chem. 275, 38048–38058
specifically expressed genes identifies disease-relevant B Mol. Dev. Evol. 322, 488–499 (2014). (2000).
tissues and cell types. Nat. Genet. 50, 621–629 (2018). 66. Lan, X. & Pritchard, J. K. Coregulation of tandem 93. Sahni, N. et al. Widespread macromolecular
39. Zou, Y. M., Lu, D., Liu, L. P., Zhang, H. H. & Zhou, Y. Y. duplicate genes slows evolution of subfunctionalization interaction perturbations in human genetic disorders.
Olfactory dysfunction in Alzheimer’s disease. in mammals. Science 352, 1009–1013 (2016). Cell 161, 647–660 (2015).
Neuropsychiatr. Dis. Treat. 12, 869–875 (2016). 67. Aoidi, R., Maltais, A. & Charron, J. Functional 94. Zhong, Q. et al. Edgetic perturbation models of human
40. Rave-Harel, N. et al. The molecular basis of partial redundancy of the kinases MEK1 and MEK2: rescue inherited disorders. Mol. Syst. Biol. 5, 321 (2009).
penetrance of splicing mutations in cystic fibrosis. of the Mek1 mutant phenotype by Mek2 knock-in 95. Ozen, H. Glycogen storage diseases: new perspectives.
Am. J. Hum. Genet. 60, 87–94 (1997). reveals a protein threshold effect. Sci. Signal. 9, ra9 World J. Gastroenterol. 13, 2541–2553 (2007).
41. Antoniou, A. et al. Average risks of breast and ovarian (2016). 96. Hartwell, L. H., Hopfield, J. J., Leibler, S. &
cancer associated with BRCA1 or BRCA2 mutations 68. Yamauchi, Y. et al. Two genes substitute for the mouse Murray, A. W. From molecular to modular cell biology.
detected in case series unselected for family history: Y chromosome for spermatogenesis and reproduction. Nature 402, C47–C52 (1999).
a combined analysis of 22 studies. Am. J. Hum. Genet. Science 351, 514–516 (2016). 97. Feldman, I., Rzhetsky, A. & Vitkup, D. Network
72, 1117–1130 (2003). 69. Sambamoorthy, G. & Raman, K. Understanding the properties of genes harboring inherited disease
42. Concannon, P., Rich, S. S. & Nepom, G. T. Genetics of evolution of functional redundancy in metabolic mutations. Proc. Natl Acad. Sci. USA 105, 4323–4328
type 1A diabetes. N. Engl. J. Med. 360, 1646–1654 networks. Bioinformatics 34, i981–i987 (2018). (2008).
(2009). 70. Sameith, K. et al. A high-resolution gene expression 98. Kitsak, M. et al. Tissue specificity of human disease
43. Bokhari, S. R. A., Zulfiqar, H. & Mansur, A. Bartter atlas of epistasis between gene-specific transcription module. Sci. Rep. 6, 35241 (2016).
Syndrome (StatPearls Publishing, 2019). factors exposes potential mechanisms for genetic 99. Emig, D. & Albrecht, M. Tissue-specific proteins
44. Basha, O. et al. Differential network analysis of human interactions. BMC Biol. 13, 112 (2015). and functional implications. J. Proteome Res. 10,
tissue interactomes highlights tissue-selective 71. Taylor, M. B. & Ehrenreich, I. M. Higher-order genetic 1893–1903 (2011).
processes and genetic disorder genes. Preprint at interactions and their contribution to complex traits. 100. Sasanuma, H. et al. BRCA1 ensures genome integrity
bioRxiv https://doi.org/10.1101/612143 (2019). Trends Genet. 31, 34–40 (2015). by eliminating estrogen-induced pathological
45. Gamazon, E. R. et al. Using an atlas of gene regulation 72. Kuzmin, E. et al. Systematic analysis of complex topoisomerase II–DNA complexes. Proc. Natl Acad.
across 44 human tissues to inform complex disease- genetic interactions. Science 360, eaao1729 (2018). Sci. USA 115, E10642–E10651 (2018).
and trait-associated variation. Nat. Genet. 50, 73. El-Brolosy, M. A. et al. Genetic compensation 101. Da Mesquita, S. et al. Insights on the pathophysiology
956–967 (2018). triggered by mutant mRNA degradation. Nature 568, of Alzheimer’s disease: the crosstalk between amyloid
46. Menche, J. et al. Disease networks. Uncovering 193–197 (2019). pathology, neuroinflammation and the peripheral
disease-disease relationships through the incomplete 74. Ma, Z. et al. PTC-bearing mRNA elicits a genetic immune system. Neurosci. Biobehav. Rev. 68,
interactome. Science 347, 1257601 (2015). compensation response via Upf3a and COMPASS 547–562 (2016).
47. Marbach, D. et al. Tissue-specific regulatory circuits components. Nature 568, 259–263 (2019). 102. Pelz, L., Purfurst, B. & Rathjen, F. G. The cell adhesion
reveal variable modular perturbations across complex 75. Jdey, W. et al. Drug-driven synthetic lethality: molecule BT-IgSF is essential for a functional blood-
diseases. Nat. Methods 13, 366–370 (2016). bypassing tumor cell genetics with a combination of testis barrier and male fertility in mice. J. Biol. Chem.
48. Gracanin, A., Dreijerink, K. M., van der Luijt, R. B., AsiDNA and PARP inhibitors. Clin. Cancer Res. 23, 292, 21490–21503 (2017).
Lips, C. J. & Hoppener, J. W. Tissue selectivity in 1001–1011 (2017). 103. Faria, A. M. C., Reis, B. S. & Mucida, D. Tissue
multiple endocrine neoplasia type 1-associated 76. Lee, J. S. et al. Harnessing synthetic lethality to adaptation: implications for gut immunity and
tumorigenesis. Cancer Res. 69, 6371–6374 (2009). predict the response to cancer treatment. Nat. tolerance. J. Exp. Med. 214, 1211–1226 (2017).
49. Dyson, N. J. RB1: a prototype tumor suppressor and Commun. 9, 2546 (2018). 104. Liu, Y., Ma, C. & Zhang, N. Tissue-specific control of
an enigma. Genes. Dev. 30, 1492–1502 (2016). 77. Finkel, R. S. et al. Nusinersen versus sham control tissue-resident memory T cells. Crit. Rev. Immunol. 38,
50. Park, J. et al. Single-cell transcriptomics of the mouse in infantile-onset spinal muscular atrophy. N. Engl. 79–103 (2018).
kidney reveals potential cellular targets of kidney J. Med. 377, 1723–1732 (2017). 105. Okabe, Y. & Medzhitov, R. Tissue biology perspective
disease. Science 360, 758–763 (2018). 78. Mercuri, E. et al. Nusinersen versus sham control in on macrophages. Nat. Immunol. 17, 9–17 (2016).
51. Sonawane, A. R. et al. Understanding tissue-specific later-onset spinal muscular atrophy. N. Engl. J. Med. 106. Nakayama, T. et al. Tissue-specific and
gene regulation. Cell Rep. 21, 1077–1088 (2017). 378, 625–635 (2018). time-dependent clonal expansion of ENU-induced
52. Reyes, A. & Huber, W. Alternative start and 79. Visscher, P. M. et al. 10 years of GWAS discovery: mutant cells in gpt delta mice. Environ. Mol. Mutagen.
termination sites of transcription drive most transcript biology, function, and translation. Am. J. Hum. Genet. 58, 592–606 (2017).
isoform differences across human tissues. Nucleic 101, 5–22 (2017). 107. Cardoso-Moreira, M. et al. Gene expression across
Acids Res. 46, 582–592 (2018). 80. Claussnitzer, M. et al. FTO obesity variant circuitry mammalian organ development. Nature 571,
53. Ungewitter, E. & Scrable, H. 40p53 controls the and adipocyte browning in humans. N. Engl. J. Med. 505–509 (2019).
switch from pluripotency to differentiation by 373, 895–907 (2015). 108. Newman, J. R. B. et al. Disease-specific biases in
regulating IGF signaling in ESCs. Genes. Dev. 24, Deciphering the tissue-specific regulatory alternative splicing and tissue-specific dysregulation
2408–2419 (2010). mechanism underlying a disease variant via revealed by multitissue profiling of lymphocyte gene
54. Kim, H. K., Pham, M. H. C., Ko, K. S., Rhee, B. D. cross-tissue exploration. expression in type 1 diabetes. Genome Res. 27,
& Han, J. Alternative splicing isoforms in health and 81. Maurano, M. T. et al. Systematic localization of 1807–1815 (2017).
disease. Pflugers Arch. 470, 995–1016 (2018). common disease-associated variation in regulatory 109. Puram, S. V. et al. Single-cell transcriptomic analysis
55. Raj, T. et al. Integrative transcriptome analyses of the DNA. Science 337, 1190–1195 (2012). of primary and metastatic tumor ecosystems in head
aging brain implicate altered splicing in Alzheimer’s 82. Boyle, E. A., Li, Y. I. & Pritchard, J. K. An expanded and neck cancer. Cell 171, 1611–1624. e24 (2017).
disease susceptibility. Nat. Genet. 50, 1584–1592 view of complex traits: from polygenic to omnigenic. 110. Sumner, C. J. & Crawford, T. O. Two breakthrough
(2018). Cell 169, 1177–1186 (2017). gene-targeted treatments for spinal muscular atrophy:

NaTure RevIewS | GeNeTICs volume 21 | March 2020 | 149


Reviews

challenges remain. J. Clin. Invest. 128, 3219–3227 135. Niemi, M. E. K. et al. Common genetic variants 161. Pinero, J. et al. DisGeNET: a comprehensive platform
(2018). contribute to risk of rare severe neurodevelopmental integrating information on human disease-associated
111. MacArthur, J. et al. The new NHGRI-EBI catalog of disorders. Nature 562, 268–271 (2018). genes and variants. Nucleic Acids Res. 45,
published genome-wide association studies (GWAS 136. Hamazaki, T., El Rouby, N., Fredette, N. C., D833–D839 (2017).
Catalog). Nucleic Acids Res. 45, D896–D901 (2017). Santostefano, K. E. & Terada, N. Concise review: 162. Alanis-Lobato, G., Andrade-Navarro, M. A. &
112. Rath, A. et al. Representation of rare diseases in induced pluripotent stem cell research in the era of Schaefer, M. H. HIPPIE v2.0: enhancing meaningfulness
health information systems: the Orphanet approach precision medicine. Stem Cells 35, 545–550 (2017). and reliability of protein-protein interaction networks.
to serve a wide range of end users. Hum. Mutat. 33, 137. Clevers, H. Modeling development and disease with Nucleic Acids Res. 45, D408–D414 (2017).
803–808 (2012). organoids. Cell 165, 1586–1597 (2016). 163. Micale, G., Ferro, A., Pulvirenti, A. & Giugno, R.
113. Han, X. et al. Mapping the mouse cell atlas by 138. Pennisi, E. Development cell by cell. Science 362, SPECTRA: an integrated knowledge base for
microwell-seq. Cell 172, 1091–1107 e17 (2018). 1344–1345 (2018). comparing tissue and tumor-specific PPI networks
114. Tabula Muris, C. et al. Single-cell transcriptomics of 139. Grunwald, H. A. et al. Super-Mendelian inheritance in human. Front. Bioeng. Biotechnol. 3, 58 (2015).
20 mouse organs creates a Tabula Muris. Nature mediated by CRISPR-Cas9 in the female mouse 164. Pierson, E. et al. Sharing and specificity of co-expression
562, 367–372 (2018). germline. Nature 566, 105–109 (2019). networks across 35 human tissues. PLOS Comput. Biol.
115. Saunders, A. et al. Molecular diversity and 140. Haigis, K. M., Cichowski, K. & Elledge, S. J. Tissue- 11, e1004220 (2015).
specializations among the cells of the adult mouse specificity in cancer: the rule, not the exception.
brain. Cell 174, 1015–1030. e16 (2018). Science 363, 1150–1151 (2019). Acknowledgements
116. Gamazon, E. R. et al. A gene-based association method 141. Bastarache, L. et al. Phenotype risk scores identify We thank A. Rudich, V. Chalifa-Caspi, A. Monsonego and
for mapping traits using reference transcriptome data. patients with unrecognized Mendelian disease members of the Yeger-Lotem lab for their helpful comments.
Nat. Genet. 47, 1091–1098 (2015). patterns. Science 359, 1233–1239 (2018).
117. Breeze, C. E. et al. eFORGE: a tool for identifying cell 142. Ongen, H. et al. Estimating the causal tissues for Author contributions
type-specific signal in epigenomic data. Cell Rep. 17, complex traits and diseases. Nat. Genet. 49, Both authors contributed to all aspects of the article.
2137–2150 (2016). 1676–1683 (2017).
118. Hormozdiari, F. et al. Colocalization of GWAS and 143. Leach, K., Conigrave, A. D., Sexton, P. M. & Competing interests
eQTL signals detects target genes. Am. J. Hum. Genet. Christopoulos, A. Towards tissue-specific The authors declare no competing interests.
99, 1245–1260 (2016). pharmacology: insights from the calcium-sensing
119. Lappalainen, T. & Greally, J. M. Associating cellular receptor as a paradigm for GPCR (patho)physiological Peer review information
epigenetic models with human phenotypes. Nat. Rev. bias. Trends Pharmacol. Sci. 36, 215–225 (2015). Nature Reviews Genetics thanks M. Kuijjer and the other,
Genet. 18, 441–451 (2017). 144. Khanna, H. et al. A common allele in RPGRIP1L anonymous, reviewer(s) for their contribution to the peer
120. Bujold, D. et al. The International Human Epigenome is a modifier of retinal degeneration in ciliopathies. review of this work.
Consortium Data Portal. Cell Syst. 3, 496–499. e2 Nat. Genet. 41, 739–745 (2009).
(2016). 145. Lakhani, C. M. et al. Repurposing large health Publisher’s note
121. Yardimci, G. G. & Noble, W. S. Software tools for insurance claims data to estimate genetic and Springer Nature remains neutral with regard to jurisdictional
visualizing Hi-C data. Genome Biol. 18, 26 (2017). environmental contributions in 560 phenotypes. claims in published maps and institutional affiliations.
122. Wang, Y. et al. The 3D Genome Browser: a web-based Nat. Genet. 51, 327–334 (2019).
Supplementary information
browser for visualizing 3D genome organization and 146. eGTEx Project. Enhancing GTEx by bridging the gaps
Supplementary information is available for this paper at
long-range chromatin interactions. Genome Biol. 19, between genotype, gene expression, and disease.
https://doi.org/10.1038/s41576-019-0200-9.
151 (2018). Nat. Genet. 49, 1664–1670 (2017).
123. Dunham, I., Kulesha, E., Iotchkova, V., Morganella, S. 147. Boisset, J. C. et al. Mapping the physical network
& E, B. FORGE: a tool to discover cell specific of cellular interactions. Nat. Methods 15, 547–553 Related links
enrichments of GWAS associated SNPs in regulatory (2018). 3D Genome: http://promoter.bx.psu.edu/hi-c/index.html
regions. F1000 Res. 4, 18 (2015). 148. Zhou, X. et al. Circuit design features of a stable Database of small Human Non-coding RNAs (DAsHR):
124. Oz-Levi, D. et al. Noncoding deletions reveal a gene two-cell system. Cell 172, 744–757. e17 (2018). http://dashr2.lisanwanglab.org/index.php
that is critical for intestinal function. Nature 571, 149. Tomasetti, C., Li, L. & Vogelstein, B. Stem cell DifferentialNet: http://netbio.bgu.ac.il/diffnet/
107–111 (2019). divisions, somatic mutations, cancer etiology, and Disease Ontology: http://disease-ontology.org/
125. Yeger-Lotem, E. & Sharan, R. Human protein interaction cancer prevention. Science 355, 1330–1334 (2017). DisGeNeT: http://www.disgenet.org/
networks across tissues and diseases. Front. Genet. 6, 150. Schneider, G., Schmidt-Supprian, M., Rad, R. & encyclopedia of DNA elements (eNCODe):
257 (2015). Saur, D. Tissue-specific tumorigenesis: context https://www.encodeproject.org/
126. Hekselman, I., Sharon, M., Basha, O. & Yeger-Lotem, E. matters. Nat. Rev. Cancer 17, 239–253 (2017). Functional Annotation of the Mammalian Genome
Analyzing Network Data in Biology and Medicine 151. Sack, L. M. et al. Profound tissue specificity in (FANTOM5): http://fantom.gsc.riken.jp/5/
(ed Pržulj, N.) 459–489 (Cambridge Univ. Press, proliferation control underlies cancer drivers and Genetic Network Analysis Tool (GNAT):
2019). aneuploidy patterns. Cell 173, 499–514.e23 (2018). http://mostafavilab.stat.ubc.ca/gnat/#
127. Schwanhausser, B. et al. Global quantification of 152. Ostrow, S. L., Barshir, R., DeGregori, J., Yeger-Lotem, E. Genotype-Tissue expression (GTex):
mammalian gene expression control. Nature 473, & Hershberg, R. Cancer evolution is associated with https://gtexportal.org/home/
337–342 (2011). pervasive positive selection on globally expressed HiPPie: http://cbdm-01.zdv.uni-mainz.de/~mschaefer/hippie/
128. Magger, O., Waldman, Y. Y., Ruppin, E. & Sharan, R. genes. PLOS Genet. 10, e1004239 (2014). Human Cell Atlas: https://www.humancellatlas.org/
Enhancing the prioritization of disease-causing genes 153. Schaefer, M. H. & Serrano, L. Cell type-specific Human Phenotype Ontology (HPO): https://hpo.jax.org/app/
through tissue specific protein interaction networks. properties and environment shape tissue specificity Human Protein Atlas (HPA): https://www.proteinatlas.org/
PLOS Comput. Biol. 8, e1002690 (2012). of cancer genes. Sci. Rep. 6, 20707 (2016). Human Proteome Map:
129. Basha, O. et al. The tissueNet v.2 database: 154. Kryuchkova-Mostacci, N. & Robinson-Rechavi, M. http://www.humanproteomemap.org/index.php
a quantitative view of protein-protein interactions A benchmark of gene expression tissue-specificity HumanBase: https://hb.flatironinstitute.org/
across human tissues. Nucleic Acids Res. 45, metrics. Brief. Bioinform. 18, 205–214 (2017). Us National Human Genome Research institute–european
D427–D431 (2017). 155. Kim, M. S. et al. A draft map of the human proteome. Bioinformatics institute (NHGRi–eBi) GwAs Catalogue:
130. Basha, O., Shpringer, R., Argov, C. M. & Yeger-Lotem, E. Nature 509, 575–581 (2014). https://www.ebi.ac.uk/gwas/
The DifferentialNet database of differential protein- 156. Ludwig, N. et al. Distribution of miRNA expression Online Mendelian inheritance in Man (OMiM):
protein interactions in human tissues. Nucleic Acids across human tissues. Nucleic Acids Res. 44, https://www.omim.org/
Res. 46, D522–D526 (2018). 3865–3877 (2016). Orphanet: https://www.orpha.net/
131. Fabregat, A. et al. The reactome pathway 157. Leung, Y. Y. et al. DASHR: database of small human Reactome: https://reactome.org/
knowledgebase. Nucleic Acids Res. 46, D649–D655 noncoding RNAs. Nucleic Acids Res. 44, D216–D222 Roadmap epigenomics Project:
(2018). (2016). http://www.roadmapepigenomics.org/
132. Jerby, L., Shlomi, T. & Ruppin, E. Computational 158. de Rie, D. et al. An integrated expression atlas of single Cell portal: https://portals.broadinstitute.org/single_cell
reconstruction of tissue-specific metabolic models: miRNAs and their promoters in human and mouse. sPeCific Tissue/Tumour Related PPi networks Analyser
application to human liver metabolism. Mol. Syst. Nat. Biotechnol. 35, 872–878 (2017). (sPeCTRA): https://alpha.dmi.unict.it/spectra/
Biol. 6, 401 (2010). 159. Liu, X., Yu, X., Zack, D. J., Zhu, H. & Qian, J. TiGER: TissueAtlas: https://ccb-web.cs.uni-saarland.de/tissueatlas/
133. Ji, X. et al. Identification of susceptibility pathways a database for tissue-specific gene expression and TissueNet: http://netbio.bgu.ac.il/tissuenet/
for the role of chromosome 15q25.1 in modifying lung regulation. BMC Bioinformatics 9, 271 (2008). Tissue-specific Gene expression and Regulation (TiGeR):
cancer risk. Nat. Commun. 9, 3221 (2018). 160. Buniello, A. et al. The NHGRI-EBI GWAS Catalog of http://bioinfo.wilmer.jhu.edu/tiger/
134. Schultz, A. & Qutub, A. A. Reconstruction of published genome-wide association studies, targeted
tissue-specific metabolic networks using CORDA. arrays and summary statistics 2019. Nucleic Acids
PLOS Comput. Biol. 12, e1004808 (2016). Res. 47, D1005–D1012 (2019). © Springer Nature Limited 2020

150 | March 2020 | volume 21 www.nature.com/nrg

You might also like