Modulating gene regulation to treat genetic disorders

Abstract

Over a thousand diseases are caused by mutations that alter gene expression levels. The potential of nuclease-deficient zinc fingers, TALEs or CRISPR fusion systems to treat these diseases by modulating gene expression has recently emerged. These systems can be applied to modify the activity of gene-regulatory elements — promoters, enhancers, silencers and insulators, subsequently changing their target gene expression levels to achieve therapeutic benefits — an approach termed cis-regulation therapy (CRT). Here, we review emerging CRT technologies and assess their therapeutic potential for treating a wide range of diseases caused by abnormal gene dosage. The challenges facing the translation of CRT into the clinic are discussed.

Alterations in the levels of gene expression above or below a certain threshold can have a significant impact on the phenotype and lead to a wide variety of human diseases. Mutations in the coding region of a gene can lead to lower or higher gene expression levels, resulting in loss or gain of protein function, whereas mutations in gene-regulatory elements — also known as cis-regulatory elements (CREs) — can affect the magnitude, timing or location (cell/tissue type) of gene expression (BOX 1). According to the ClinVar database¹, it is estimated that 516 genes lead to human disease in a recessive manner due to loss-of-function mutations in both copies, whereas an estimated 660 genes are known to cause disease when haploinsufficient². In addition, the ClinVar¹ and ClinGen³ databases indicate that mutations in 65 genes are known to be pathogenic due to gain of function.

Box 1 |. Causes of abnormal gene dosage.

Abnormal gene dosage underlies numerous diseases and can be due to various causes. For example, mutations in cis-regulatory elements such as promoters or enhancers could lead to their target gene being upregulated or downregulated, whereas mutations in silencers could cause aberrant expression levels of their target gene. These and other causes of abnormal gene dosage and their definitions are detailed below.

Haploinsufficiency

Originating from the Greek word ‘haplo’, which means ‘single’, and the word ‘insufficiency’, one of the two gene copies has a loss-of-function mutation that leads to having only a single functional copy. This produces less than normal gene expression that is insufficient for proper gene function, subsequently causing a disease phenotype.

Recessive loss of function

The two copies of a gene are non-functional leading to no apparent gene expression.

Microdeletions/microduplications

Chromosomal deletions/duplications that are too small to be detected by standard cytogenetic methods, typically being 1–3 Mb long.

Chromosomal abnormalities

Part of or a whole chromosome is absent (deletion), duplicated (duplication) or moved to another location (translocation), or part of a chromosome is flipped end to end (inversion).

Promoter mutations

Promoters are gene-regulatory elements that control the initiation of transcription and reside upstream of the gene’s transcription start site. Mutations in promoters could lead to a decrease or increase in gene expression.

Enhancer mutations

Enhancers are gene-regulatory elements that are distant to the gene’s transcription start site and turn on and control gene expression levels by interacting with the promoter. Mutations in enhancers could lead to a decrease or increase in gene expression.

Silencer mutations

Silencers are gene-regulatory elements that are distant to the gene’s transcription start site and silence promoter activity. Mutations in these sequences could lead to altered gene expression of their target gene.

Parent-of-origin genomic imprinting

Alleles of certain genes are imprinted or DNA methylated and inactivated based on their parent of origin, whereas the other allele remains active and expressed. Mutations in the expressed allele can lead to disease whereas the other normal allele remains imprinted.

Random X-inactivation

The process by which, in mammalian XX females, one of the X chromosomes is transcriptionally inactivated or silenced into heterochromatin. Mutations in one gene copy or both gene copies on the X chromosome can lead to disease.

CREs include promoters that reside right next to the gene’s transcription start site and turn on its expression at specific times, locations and levels. Promoters are regulated by other distal CREs, such as enhancers and silencers, which act in a temporal and tissue-specific manner. Enhancers turn on the promoter and can be considered the ‘promoters of the promoter’, whereas silencers shut down promoters. In addition, insulators prevent promiscuous interactions between enhancers and the gene’s promoter. Systematic genomic approaches can identify and functionally characterize these CREs in a high-throughput manner and help understand the cis-regulatory circuitry of gene expression^4,5. This includes the recent use of clustered regularly interspaced short palindromic repeats (CRISPR)–CRISPR-associated protein 9 (CRISPR–Cas9) and other editing systems to carry out large-scale genomic screens to identify CREs^6,7.

Mutations in CREs can cause a wide variety of human diseases⁵. For example, mutations in the promoters of globin genes cause various haemoglobinopathies and thalassaemias⁸, mutations in the limb enhancer of the sonic hedgehog (SHH) gene cause various limb malformations⁹ and mutations in a silencer of the γ-globin gene leads to hereditary persistence of fetal haemoglobin¹⁰. In addition, developmental defects such as limb malformations, sex reversal and neurodevelopment disorders can be caused by genetic defects resulting from loss of insulator function that can alter 3D DNA topology and looping^11,12.

Although various technologies to therapeutically modulate gene expression have shown promise, they are limited in their application, especially for dosage-sensitive diseases, raising the need for alternative therapeutic approaches. For example, gene replacement therapy can restore normal gene function by delivering a fully functional copy of the mutant gene to the patient, representing a promising approach to cure recessive conditions where no functional gene copy exists¹³. However, the limited 4,700-bp packaging capacity of recombinant adeno-associated viruses (rAAVs; the delivery vehicle of choice for gene replacement therapy) precludes their use for genes that have long coding sequences (CDSs). From the 516 recessive disease-causing and 660 haploinsufficient disease-causing genes, 121 and 135 genes, respectively, cannot be packaged into an rAAV efficiently, due to their CDS being longer than 3,000 bp in length (rAAV needs additional sequences ~1,700 bp in length to drive CDS expression). Some of the diseases caused by specific mutations could potentially be treated by oligonucleotide-based therapeutic modalities¹⁴, and antisense oligonucleotide (ASO)-based drugs have been approved for several autosomal recessive diseases^15,16. However, as ASO-based drugs are largely dependent on the premise that enough transcript from the disease-causing gene will be present in order to intervene, alternative approaches for targeting loss-of-function mutations that result in complete lack of transcript from the mutant alleles are needed. In addition, gene editing using zinc fingers, transcription activator-like effector nucleases (TALENs) and CRISPR–Cas9 systems have the potential to correct the disease-causing mutations¹⁷, but these editing tools could leave ‘DNA scars’ due to off-target binding and would need to be custom-tailored for each individual mutation.

cis-Regulation therapy (CRT) represents an alternative therapeutic approach that precisely targets CREs in the genome and alters the expression levels of their target genes to rescue disease phenotypes. This approach takes advantage of nuclease-deficient gene-editing systems, such as zinc fingers, transcription activator-like effectors (TALEs) or CRISPR–dead-Cas9 (dCas9) (FIG. 1a; TABLE 1) that are fused to proteins or effector domains, such as transcriptional activators or repressors, epigenetic modifiers, chromatin remodellers or DNA looping factors, which can modulate gene expression when targeted to CREs¹⁸ (FIG. 1b–d; TABLE 2). A major advantage of this approach is that it controls the endogenous gene-regulation machinery in a way that can modulate gene expression to physiologically relevant levels.

Fig. 1 |. DNA targeting modules and effector domains used in CRT.

a | The programmable DNA targeting modules (DTMs; zinc fingers, transcription activator-like effectors (TALEs), clustered regularly interspaced short palindromic repeats (CRISPR)/single guide RNA (sgRNA)) can be designed to target and recognize a specific DNA sequence. These modules can be fused with protein/effector domains that can affect the transcription of the targeted gene. b | Effector domains can activate or repress transcription of the target gene. c | Effector domains that modify the epigenetic state of the targeted element, either by marking DNA CpG methylation, as depicted by the DNA methylase enzyme DNA methyltransferase 3a (Dnmt3a), or removing the DNA CpG methylation mediated by ten–eleven translocation methylcytosine dioxygenase 1 (Tet1) enzyme. Effector domains can also modify the histones of the target DNA; for example, the G9a enzyme can write histone H3 methylation of lysine 9 methylation (H3K9me), LSD1 demethylase can remove H3K4me and p300 can write H3 acetylation of lysine 27 (H3K27ac) on the target DNA. These effector modules can modulate transcription of the nearby gene as a consequence of the modification. d | Effector domains can also modify chromatin looping by facilitating the DNA–DNA interaction of the pair of target DNA sequences, based on the interaction affinity of the looping domain fused with the DNA targeting module. CRT, cis-regulation therapy; dCas9, dead-Cas9; LSD1, lysine demethylase 1A; PAM, protospacer adjacent motif.

Table 1 |.

Properties of programmable DNA targeting systems

DNA targeting system	DNA recognition/length of DNA target	Engineering	Specificity/off-target potential	Immunogenicity	Refs
Zinc fingers	Each Cys2-His2 zinc finger can recognize 3–4 bp Customizable with tandem cassettes of zinc finger modules Targeting more than 9–12 bp lowers specificity	Each Cys2-His2 zinc finger is composed of 28–30 amino acids and a linker of 6–8 amino acids (TGEK(R)P) Small and easy to fuse with larger effector domains to clone into rAAV vectors Time-consuming and expensive	Specific up to 9 bp Designed to recognize 5’-(GNN)N-3’ Depends on specificity of the zinc finger concatemer More than 3 zinc fingers are less specific and, hence, could have high off-target effects	Low Can be humanized to be less immunogenic	168–170
TALE	Each TALE recognizes a single base pair Comprehensive but needs 5’T and 3’A (5’-T… A-3’) 10–20 bp but also customizable for longer lengths	33–35 amino acid repeat arrays with amino acids 12–13 (RVD) providing DNA base specificity Small and easy to fuse with larger effector domains to express in AAV vectors Time-consuming and expensive	Highly specific based on unique target sequence and its recognition length	Not studied	171,172
CRISPR	Each sgRNA scaffold can recognize 14–23 nucleotides next to a PAM site in the genome Orthologous CRISPR systems have different PAM recognition sites Multiple sgRNAs spanning longer lengths can be designed	Size of dCas9 varies from different orthologous CRISPR systems, from around 3,000–4,200 bp dCas9 fusions with larger effector domains are difficult to package in rAAV vectors Easy to engineer and cost-effective	Depends on the target sequence in the genome and Cas protein Variable specificity as can tolerate mismatches towards 3’ of sgRNA target sequence	Pre-existing immunity against Cas9 protein derived from pathogenic bacteria has been found in human populations	173,174

CRISPR, clustered regularly interspaced short palindromic repeats; dCas9, dead-Cas9; PAM, protospacer adjacent motif; rAAV, recombinant adeno-associated virus; RVD, repeat variable di-residue; sgRNA, single guide RNA; TALE, transcription activator-like effector.

Table 2 |.

Effector domain fusion systems

Effector domain	Description	Length (bp)	Refs
Transcriptional activators
VP64	4 tandem copies of VP16, a herpes simplex virus type 1 transcription factor	150	117,175
VP160	10 tandem copies of VP16, a herpes simplex virus type 1 transcription factor	400	176
VPR	Tandem fusion of VP64, p65 (RELA; NF-κB transcription factor P65) and Rta (human herpesvirus 8 regulator of transcription activation)	1,568	177
Transcriptional repressors
KRAB	Kruppel-associated box (KRAB) is a protein domain found in many transcriptional repressors	185	178
KRAB-MeCP2	KRAB connected to the transcription repression domain of methyl DNA binding protein 2 (MeCP2)	194	179
Histone modifiers
LSD1	Lysine-specific demethylase 1 (LSD1) that removes histone H3 methylation at lysine 4 (H3K4me)	2,611	85
p300	Histone acetyltransferase domain of the gene EP300; marks H3K27 acetylation	1,850	86
EHMT2/G9a	Histone 3 lysine 9 methyltransferase	1,584	84
SUV39H1	Histone 3 lysine 9 methyltransferase	1,584	84
DNA methylation editors
Dnmt3a	DNA methyltransferase 3a; de novo DNA methylation	2,738	74
Dnmt3b	DNA methyltransferase 3b; de novo DNA methylation	950	180
Dnmt3a-3l	Fusion of DNA methyltransferase 3a and 3l for de novo DNA methylation	1,680	181
Tet1	Tet methylcytosine dioxygenase 1; removes DNA methylation	2,170	74
Tet2	Tet methylcytosine dioxygenase 2; removes DNA methylation	3,018	182
Tet3	Tet methylcytosine dioxygenase 3; removes DNA methylation	2,916	182
Scaffold-assembly systems
SAM	Synergistic activation mediator (SAM); a 3-component system that uses MS2 (bacteriophage coat protein) RNA aptamers on the sgRNA (first component) to assemble effector domains (transcriptional activators or repressors, for example) to dCas9 or dCas9 fusion (third component)	MS2 510 dCas9 3–4.2 kb Effector size variable	183
SunTag-scFv	dCas9 fused with a SunTag peptide scaffolding array (GCN4) that can be detected by scFv (single-chain variable fragment) protein linked with any effector domain; for example, VP64 or p65 for transcriptional activation or Dnmt3a for DNA methylation	GCN4 array 1,000–2,000, scFv 75 dCas9 3–4.2 kb Effector size variable	184
DNA looping modulators
LDB1-SA	LIM domain binding 1 self-association domain	600	91,92
CLOuD9	Chromatin Loop reOrganization using nuclease Deficient Cas9; phytohormone inducible DNA looping mediated by dimerization of locus-bound dCas9 fusion of ABI1 and PYL1 domains	ABI1 840 PYL1 540	93
dCas9-Zip	dCas9 fused to bivalent leucine zipper domains	630	95

dCas9, dead-Cas9; sgRNA, single guide RNA.

In this Review, we provide a brief overview of the fundamental principles for these programmable platforms and focus on proof-of-concept studies in various genetic conditions where CRT has been shown to be effective in cell and animal models. In addition, we assess delivery options, primarily viral vehicles and lipid nanoparticles, that can be used to deliver CRT modalities to different tissues. Strategies to overcome obstacles that need to be addressed before CRT can enter the clinic are also discussed.

Programmable platforms for CRT

There are three current platforms that can be considered for CRT: the DNA targeting modules zinc fingers, TALENs and CRISPR–Cas9, which can be designed to bind to a specific DNA sequence of a CRE (FIG. 1a; TABLE 1). Zinc finger nucleases and TALENs are first and second-generation ‘gene editors’, respectively, that have a separate programmable DNA binding module, which can be fused with Fok1 nuclease to generate DNA cuts. CRISPR–Cas systems are third-generation ‘gene editors’ that work as a ribo-protein complex, with a guide RNA and a Cas nuclease.

Zinc fingers are derived from Cys₂-His₂ domains of zinc finger proteins and can be programmed to bind to specific DNA sequences depending on the amino acid residues at positions −1, 2, 3 and 6 of their α-helix. A tandem array of four zinc fingers linked together can specifically recognize ~12 bp. The recognition specificity of tandem zinc fingers decreases with length of the array due to the context of multiple zinc fingers and the surrounding DNA sequence. Conversely, each TALE domain can recognize a single base pair with high specificity, depending on the amino acids at positions 12 and 13 termed the repeat variable di-residue. Repetitive sequences found in TALEs makes it difficult to engineer long arrays, and therefore they can efficiently be designed to target around 10–20 bp. Designing the DNA targeting module of CRISPR–single guide RNA (sgRNA) is the easiest among the three approaches. The unique targeting sequence in sgRNA can be between 14 and 24 bp, with 20–21 bp being most efficient for this system. The type II CRISPR–sgRNA system generally requires a specific sequence that is 2–6 bp in length, called the protospacer adjacent motif (PAM) sequence, to be present in the genomic target.

All three platforms have different advantages and disadvantages that can be considered before selecting them for a particular study (TABLE 1). Zinc fingers and TALEs can be fused with an effector domain (FIG. 1) that can carry out various site-specific modifications without cutting the DNA. For CRISPR–Cas systems, a mutation in the nuclease domain of the Cas enzyme can be used to generate a dCas that will not cut the DNA and can be fused with an effector domain. There are four major classes of effector domains that are fused with DNA targeting modules that are routinely used: effectors that can directly modulate transcription of the target gene; effectors that modify DNA CpG methylation; effectors that are either writers or erasers of particular histone modifications; and effectors that facilitate chromatin looping interactions (FIG. 1b–d; TABLE 2). CRE activity can also be modulated by editing the CRE itself using zinc finger nucleases, TALENs and traditional CRISPR–Cas9 (REFS^19–21), as well as by using CRISPR–Cas9-mediated base editing²² or prime editing technologies²³. However, due to the scope of this Review, we will focus on applications of non-editing DNA targeting modules for CRT.

Applications of CRT

Gene upregulation for haploinsufficiency

There are an estimated 660 genes that when haploinsufficient lead to human diseases, ranging from epilepsy to cancer, polycystic kidney disease, blindness and many others². Analyses of large-scale sequencing studies^24,25 estimate this number to be much higher, finding 3,230 loss-of-function intolerant genes, 88% of which are predicted to be due to severe haploinsufficiency. A CRT approach to upregulate the expression of the existing normal gene copy could be a potential treatment for these diseases. Upregulating the mRNA output of the normal functional allele could increase its protein levels above the expression threshold that leads to the disease. This approach can also be designed for genes irrespective of their CDS length, which is a major advantage over rAAV-based gene replacement therapy.

The feasibility of using a CRT approach to upregulate gene expression for an haploinsufficient disease was initially demonstrated by the rescue of haploinsufficiency-caused obesity in mice². Heterozygous loss-of-function mutations (haploinsufficiency) in the genes single-minded family basic helix–loop–helix (bHLH) transcription factor 1 (SIM1) or melanocortin 4 receptor (MC4R) are a major cause of human obesity²⁶. Heterozygous knockout of either of these genes in mice robustly recapitulates human hyperphagia and obesity^27,28. A CRISPR activation (CRISPRa) approach was used to target the CREs of either gene to rescue the obesity phenotype (FIG. 2a). To upregulate the existing functional gene copy of Sim1, the promoter and hypothalamic enhancer of this gene were targeted using a nuclease-deficient Cas9 (dCas9) fused to the transcriptional activator VP64 (4 tandem copies of viral protein 16)² (TABLES 1,2). Transgenic mice were generated that can ubiquitously express a CRISPRa system designed to target the Sim1 promoter or enhancer. Interestingly, Sim1 upregulation was only observed in the tissues where the targeted regulatory element is active. In the promoter-targeted CRISPRa mouse, Sim1 was upregulated in the hypothalamus and kidney, whereas in the Sim1 hypothalamus enhancer-targeted mice the upregulation was only observed in the hypothalamus. Sim1 is expressed robustly in the kidney and hypothalamus, whereas in several other tissues the Sim1 locus is thought to be under repressed chromatin. As open chromatin regions in the genome have been shown to be more accessible for CRISPR binding^29,30, the tissue-specific upregulation of Sim1 in the kidney and hypothalamus is speculated to be due to the open chromatin state of its promoter and its enhancer. This CRT approach may therefore be advantageous over gene replacement therapy, where it is often difficult to achieve tissue-specific expression. rAAV-mediated delivery of the CRISPRa system into the hypothalamus of either Sim1 or Mc4r haploinsufficient mice rescued the weight gain phenotype in both mouse models. Furthermore, Sim1 haploinsufficient mice that were injected with CRISPRa maintained a body weight similar to wild-type mice with sustained levels of Sim1 upregulation during a 9-month post-injection follow-up.

Fig. 2 |. Using CRT to upregulate gene transcription.

a | Rescue of haploinsufficiency in mice, using obesity as an example, via a clustered regularly interspaced short palindromic repeats activation (CRISPRa) approach targeting either the single-minded family basic helix–loop–helix (bHLH) transcription factor 1 (Sim1) gene promoter or a hypothalamus enhancer (hEn). Heterozygous knockout mice of the Sim1 gene are missing the transcription start site and exon 1, as depicted by the red dotted line. Promoter targeting upregulates Sim1 expression in the kidney and hypothalamus (Hypoth), whereas enhancer targeting upregulates its expression only in the hypothalamus. Pink bars represent low gene expression levels and violet bars represent upregulated levels that are similar to normal. b | Rescue of muscular dystrophy-associated phenotypes in mice using a CRISPRa approach that targets the promoter of an alternative gene with similar function. An autosomal recessive mutation in laminin subunit-α2 (Lama2) leads to complete loss of function causing muscle degeneration and loss of the myelin sheath on peripheral nerves. CRISPRa upregulation of laminin subunit-α1 (Lama1) results in its expression in muscles and Schwann cells that can rescue the Lama2 loss of function-associated phenotypes. Red rectangles represent low gene expression levels of Lama2 and green rectangles represent CRISPRa upregulated levels of Lama1 in tissues where it is not usually expressed (muscle and Schwann cells). CRT, cis-regulation therapy; dCas9, dead-Cas9; pr, promoter.

A similar CRISPRa approach ameliorated epilepsy in a mouse model of sodium voltage-gated channel α-subunit 1 (Scn1a) haploinsufficiency³¹. Heterozygous loss-of-function mutations in the SCN1A gene cause Dravet syndrome, which is characterized by severe epilepsy that begins early in life, followed by behaviour and psychomotor abnormalities along with cognitive impairment in later years. Standard rAAV-based gene therapy is not feasible for Dravet syndrome due to the ~6-kb CDS length of SCN1A. A CRT approach was developed to upregulate Scn1a by targeting its promoter using a lentivirus-based CRISPRa system (dCas9 fused to 10 copies of VP16 (V160)). Primary GABAergic interneurons, which are the target cell type for Dravet syndrome, were derived from Scn1a haploinsufficient mice that have a loss-of-function nonsense mutation that leads to epileptic seizures detected from 1 month of age³² and were infected with lentivirus packaged with CRISPRa. These infected GABAergic interneurons showed upregulation of Scn1a along with an increase in the frequency of action potentials and the firing rate³¹. Intracerebroventricular injection of CRISPRa packaged into rAAV in neonatal pups upregulated Scn1a in heterozygous knockout mice, resulting in improvements in action potential frequency in parvalbumin neurons and a reduction in hyperthermia-induced seizures. Interestingly, upregulation of Scn1a was observed equally from both mutant and wild-type alleles, but the mutant receptor was unable to dock to the cell membrane.

For gene therapy, stringent dosing is extremely important in order to limit ectopic or non-physiological expression levels, as these could have a harmful effect. The CRISPRa studies discussed above indicate that tight control of transcription could potentially be maintained using CRT. In the Sim1–CRISPRa study, increasing the CRISPRa–rAAV dose in wild-type mice could not proportionally increase the levels of Sim1 (REF.²). In the Scn1a study, a significant upregulation, both at the Scn1a mRNA and protein levels, was not detected in mature wild-type neurons, suggesting that the transcription of Scn1a could be under negative-feedback regulation. Targeting the endogenous CREs could thus have an added advantage to limit overexpression due to the inherent transcriptional feedback mechanisms. However, it is worth noting that CRISPRa upregulation of Mc4r in the hypothalamus or Sim1 in the kidney led to about threefold higher expression than wild-type levels², suggesting that gene expression thresholds could be tissue-dependent. More in-depth in vivo work needs to be done in order to determine these tissue-specific feedback mechanisms and how fine-tuning gene expression can help to design better therapeutic strategies. Approaches to express CRISPR conditionally or changing the activity of CRISPR by incorporating nucleotide variations in the targeting sgRNAs could be adopted in designing CRT to fine-tune gene expression^33–35.

Upregulating an alternative gene

Diseases caused by complete loss of function of both gene copies could potentially be rescued by upregulating the expression level of another gene with a similar function in the disease-associated tissues. This approach was recently shown to ameliorate disease phenotypes in two different mouse models of muscular dystrophy^36,37.

Congenital muscular dystrophy type 1A (MDC1A; OMIM 607855) is an autosomal recessive muscular dystrophy with variability in onset and severity depending on the type of mutation in the laminin subunit-α2 (LAMA2) gene. LAMA2 is a large gene (CDS 9,369 bp) that encodes a subunit of the extracellular protein laminin 211, which is important for myotube stability and survival, Schwann cell migration, neurite growth and myelination of axons. rAAV-based gene replacement therapies are not feasible for LAMA2 due to its CDS length. Given the finding that transgenic overexpression of another member of the laminin gene family, laminin subunit-α1 (LAMA1; CDS 9,228 bp), which is not expressed in skeletal muscle or Schwann cells, rescues myopathy and peripheral neuropathy phenotypes in a MDC1A mouse model³⁸, CRISPRa upregulation of LAMA1 in skeletal muscle and Schwann cells could be a potential treatment strategy for MDC1A. Lama2 homozygous mutant mice (dy^2j/dy^2j) that have a splicing mutation³⁹ leading to a muscular dystrophy phenotype were injected in the temporal vein at the pre-symptomatic stage (2 days of age) with an rAAV–CRISPRa system (VP64 fused to dCas9 on both amino and C termini) that targets the Lama1 promoter. This treatment led to similar upregulation levels (around twofold) of Lama1 using either one or three different sgRNAs, with mice exhibiting a reduction in skeletal muscle fibrosis (FIG. 2b), larger fibre size and an absence of hindlimb contractures 7 weeks post injection³⁶. Phenotypic improvement (fibres and muscle function and mobility) was also observed when mice were treated at a post-symptomatic stage (3 weeks), suggesting a broad therapeutic window.

Duchenne muscular dystrophy (DMD; OMIM 310200) is an X-linked disorder caused by mutations in the dystrophin gene, primarily affecting boys, leading to a progressive muscular dystrophy that causes mortality around the age of 20 years. Although rAAV-based gene replacement therapy for the full CDS of dystrophin is not possible (CDS 14 kb), truncated versions of dystrophin, mini-dystrophin and micro-dystrophin can be packaged into rAAVs and are in clinical trials^40–43 despite some concerns for immune response against synthesized versions of dystrophin. In addition, rAAV-mediated delivery of alternative candidates that could ameliorate DMD-associated phenotypes, such as follistatin (FST; CDS 954 bp) and B4GALNT2 (β−1, 4-N-acetylgalactosaminyltransferase 2; CDS 1443 bp), has been under clinical trials for the past few years^44,45, but with limited efficacy. FST is a myostatin inhibitor, and its overexpression increases muscle mass and strength in the DMD mdx mouse model⁴⁶. Of note, the phenotype of this mouse model differs from human DMD, having reduced endomysial fibrosis and successful fibre regeneration⁴⁷. Another candidate gene, utrophin (with a CDS length of 10,302 bp), can substitute for the function of dystrophin, and its transgenic overexpression in skeletal muscle ameliorates disease phenotypes in mdx mice⁴⁸. Micro-utrophin, a truncated version of full-length utrophin, packaged into an rAAV improved DMD pathology without any observed immunogenicity^49,50. CRISPRa using dCas9–VP160 (10 copies of VP16; TABLE 2) targeting utrophin in myoblasts from a patient with DMD restored β-dystroglycan, a transmembrane cytoskeletal protein absent in patients with DMD⁵¹. Another candidate for DMD therapy is klotho (cDNA 3,039 bp), which encodes a transmembrane protein that is involved in muscle growth and is silenced in mdx mice⁵². Klotho is progressively silenced during ageing and is associated with morbidity in many age-related diseases. Thus, reactivating klotho is considered to have anti-ageing benefits^53,54. Using a CRISPRa-modified approach of the synergistic activation mediator (SAM) system (TABLE 2), termed CRISPR–Cas9 target gene activation, upregulation of Fst, utrophin or klotho increased muscle mass and strength in the mdx mouse model³⁷. It would be interesting to see whether simultaneous upregulation of these alternative candidate genes in DMD using a CRISPRa approach may have cumulative benefits. Together, these studies demonstrate the therapeutic potential of modulating the expression of an alternative gene to rescue the complete loss of function of the causative gene.

Gene downregulaiton to treat disease

Chromosomal alterations leading to a gain in the copy number of genes via gene duplication, segmental duplication or trisomy, such as Down syndrome, often result in a gain of gene expression. In addition, one of the two gene copies could have a mutation leading to a gain of function contributing towards disease pathology. Analysis of orphan drugs listed in the US Food and Drug Administration (FDA) suggested that there is a huge lack of drug development for diseases caused by gain-of-function mutations⁵⁵. In these cases, downregulating the mutant allele could potentially rescue the disease.

Huntington disease (OMIM 143100) is caused by a CAG expansion in exon 1 in one copy of the huntingtin (HTT) gene, causing progressive cognitive, motor and behavioural decline starting at age 30–40 years⁵⁶. Downregulation of the mutated gene copy in Huntington disease could lessen its dominant negative effect and alleviate Huntington disease-associated pathology. Various modalities such as small interfering RNA or ASO have shown some promise for Huntington disease treatment^57,58. Downregulation of the mutant HTT allele using a TALE-based (TABLE 1) CRT approach has also been investigated. A transcriptional repressor domain of the Kox1-repressor (KRAB; TABLE 2) was coupled to TALEs that were designed to specifically target the mutant allele of HTT. Three SNPs of the mutant allele, rs762855 (upstream of the HTT transcription start site), rs3856973 (in the first intron of HTT) and rs2024155 (in the third intron of HTT), were chosen for allele-specific targeting using individual TALE–KRAB fusions. Introduction of this TALE system into fibroblasts derived from a patient with Huntington disease significantly reduced the expression of the mutant gene copy by one of the three TALE–KRAB fusions⁵⁹ (FIG. 3a). In another study, the KRAB domain fused to zinc finger proteins (ZF-KRAB; TABLE 1) was designed to bind to the longer CAG repeats in the mutant allele with higher affinity than the wild-type allele⁶⁰. rAAV-mediated delivery of ZF-KRAB into mice brains of Huntington disease mouse model R6/2 (which has ~150 CAG repeats in the Htt gene) showed 60% repression of the mutant allele (FIG. 3a), which improved Huntington disease-associated motor coordination function phenotypes, such as hindlimb clasping and rotarod test scores⁶⁰. However, although the R6/2 mouse model mimics many of the progressive neurological phenotypes observed in patients with Huntington disease, its relevance is limited for rescue studies as it does not recapitulate the full spectrum of clinically significant phenotypes⁶¹. Another Huntington disease transgenic mouse model, YAC128, which has a yeast artificial chromosome coding the human HTT gene with 125 N-terminal CAG repeats, expresses the mutant HTT at levels equivalent to 75% of the endogenous wild-type mouse Htt and robustly recapitulates Huntington disease abnormalities, such as age-dependent cortical neurodegeneration and cognitive and motor dysfunction. This suggests that a 25% reduction of the mutant allele may not be beneficial in alleviating Huntington disease-associated phenotypes, and approaches that downregulate HTT below this threshold may have therapeutic benefits^58,62.

Fig. 3 |. Using CRT to downregulate gene transcription.

a | Downregulation of the mutant huntingtin (HTT) gene copy using a cis-regulation therapy (CRT) module fused to a transcriptional repressor domain of the Kox1-repressor (Kruppel-associated box (KRAB)). Transcription activator-like effectors (TALEs) fused with KRAB designed to specifically target the mutant gene copy in fibroblasts derived from a patient with Huntington disease via an allele-differentiating SNP in the promoter region of the human HTT (middle panel) or multiple KRAB domains fused to zinc finger proteins (ZF-KRABs) that recognize the CAG repeats and tile the entire CAG repeat region of the mutant allele with higher affinity in the mouse Htt locus (bottom panel). Blue rectangles represent expression of wild-type HTT allele, and red rectangles represent expression levels of the mutant allele containing more than 40 copies of the CAG repeat (CAG^>40). b | Fatty acid binding protein 4 (Fabp4), a lipid chaperon, shows elevated levels in adipocytes and positively correlates with the severity of the obesity phenotype. Downregulation of Fabp4 via dead-Cas9 (dCas9)–KRAB fusion that is delivered to adipocytes using an adipose targeting peptide leads to amelioration of obesity-associated phenotypes in mice exposed to a high-fat diet. Pr, promoter; sgRNA, single guide RNA.

Facioscapulohumeral muscular dystrophy (OMIM 158900) is caused by the deletion of D4Z4 repeats in the proximal sub-telomeric region of 4q35, leading to derepression of the double homeobox protein4 (DUX4) gene, primarily in muscle. Patients with facioscapulohumeral muscular dystrophy show variable penetrance, some being non-symptomatic with moderate expression of DUX4, and in some patients even the slightest increase in DUX4 expression can induce progressive muscular degeneration^63,64. Modifier genes identified as the bromodomain adjacent to zinc finger domain protein 1A (BAZ1A), lysine-specific demethylase 4C (KDM4C), bromodomain containing 2 (BRD2) and SWI/SNF-related matrix-associated actin-dependent regulator of chromatin subfamily A member 5 (SMARCA5) are known to play a role in determining the dosage sensitivity of DUX4. CRISPR interference (CRISPRi) using dCas9–KRAB targeting the promoters of DUX4 or its modifier genes BAZ1A, KDM4C, BRD2 and SMARCA5 reduced DUX4 levels in primary myoblasts derived from patients with facioscapulohumeral muscular dystrophy⁶⁵. These results show that CRISPRi-based CRT can be used to downregulate a modifier gene to rescue disease phenotypes.

Downregulating an elevated biomarker associated with a certain disease could also be an effective therapeutic strategy. One such example is fatty acid binding protein 4 (Fabp4), a lipid chaperon that is specifically expressed in adipocytes and is elevated in obesity, cardiovascular disease, fatty liver disease and diabetes⁶⁶. Downregulating Fabp4 gene expression in adipocytes could have beneficial effects for treating these conditions (FIG. 3b). A DNA plasmid coding a dCas9–KRAB fusion and sgRNA targeting the second exon of Fabp4 was delivered to adipocytes using a non-viral delivery approach⁶⁷. This approach used an adipose tissue targeting peptide sequence, ATS (CKGGRAKDC), that specifically binds to the membrane protein prohibitin that is expressed in white adipocytes⁶⁸. A stretch of nine positively-charged arginines were added to this peptide sequence (ATS-9R) that can bind to the negatively charged DNA cargo coding the CRISPRi components. Delivery of ATS-9R–CRISPRi both to cultured adipocytes and to adipocytes of mice exposed to a high-fat diet, reduced Fabp4 gene expression and lipogenesis in mice. In the mice fed a high-fat diet, obesity, inflammation, steatosis and serum levels of free fatty acids, triglycerides and markers of liver function were reduced.

The feasibility of developing CRISPRi-based CRT using in vitro models is tenable for many diseases. An important issue is to ascertain the threshold for dose reduction of the mutant allele in vivo such that this can lead to therapeutic benefits, in particular for mutations that have a dominant negative effect. CRISPRi-based CRT can be used to investigate this dose–effect relationship, and the design of variable sgRNAs can be used to titrate gene expression levels³³; however, further research is required to develop CRT-based modalities that can target specific alleles. Nevertheless, CRT could be a promising alternative for conditions where other gene silencing modalities, such as ASOs, RNAi, small interfering RNA or short hairpin RNA, are difficult to design or are ineffective.

Modulating DNA methylation

DNA methylation is an epigenetic mechanism that can affect gene expression by the addition of methyl groups primarily to cytosine nucleotides. DNA methylation in promoters or enhancers is generally associated with their repression, subsequently leading to suppression of gene transcription. Aberrant levels of DNA methylation can lead to abnormal gene dosage and are associated with many neurological diseases⁶⁹. For example, Fragile X syndrome (FXS; OMIM 300624) is caused by a CGG trinucleotide repeat expansion of over 200 copies in the 5′UTR of the FMRP translational regulator 1 (FMR1) gene promoter, which leads to hypermethylation of the FMR1 promoter, subsequently silencing gene expression⁷⁰. Demethylation of the FMR1 promoter can reactivate gene function and rescue FXS-associated phenotypes^71–73. Modulating the levels of DNA methylation could be possible using CRT tools that alter the disease-associated gene expression levels.

Altering DNA methylation at specific genomic locations is possible using DNA targeting molecules, such as zinc fingers, TALEs or CRISPR–dCas9 fused to DNA methyltransferase 3a (Dnmt3a), or DNA demethylase, such as ten–eleven translocation methylcytosine dioxygenase 1 (Tet1)⁷⁴ (FIG. 1d; TABLE 2). Demethylating the hypermethylated 5′UTR region containing trinucleotide repeat CGG in cell models of FXS using the dCas9–Tet1 fusion could activate FMR1 gene expression (FIG. 4a). In induced pluripotent stem cells and differentiated neurons from male-derived fibroblasts that contain around 450 CGG repeats, dCas9–Tet1 targeted to the repeat region using a sequence-specific guide RNA led to approximately 96% reduction in DNA methylation levels in the FMR1 5′UTR and restored FMR1 mRNA levels up to 90% in the induced pluripotent stem cells and 80% in neurons⁷⁰. Electrophysiological properties in the methylation-edited neurons were also restored. However, in induced pluripotent stem cells derived from patients with FXS differentiated into postmitotic neurons, the direct DNA methylation editing only led to 20% reduction in methylation levels without any significant reactivation of the FMR1 gene. Despite this, restoration of electrophysiological spontaneous hyperactivity in these neurons was observed. These neurons maintained their edited methylation status of the FMR1 promoter even after they were xenografted into the mouse brain cortex. It is important to note that the target sequence of sgRNA used here was highly repetitive and, consequently, resulted in off-target demethylation of approximately 200 genomic regions.

Fig. 4 |. Using CRT to modulate DNA methylation or DNA looping.

a | The DNA demethylase enzyme ten–eleven translocation methylcytosine dioxygenase 1 (Tet1) fused with dead-Cas9 (dCas9) targeting the CGG trinucleotide repeat region of the FMRP translational regulator 1 (FMR1) gene can demethylate the locus and reactivate FMR1 gene expression in Fragile X syndrome. This epigenetic modification can result in improved neuronal function in induced pluripotent stem cell (iPSC)-derived neurons. b | Programmed DNA looping mediated by zinc finger protein fusion to the LIM domain binding 1 (Ldb1) self-association domain (Lbd1-SA) to rescue a promoter mutation in the adult globin gene (mHBB) that results in thalassaemia due to loss of function of HBB. This zinc finger protein–Ldb1-SA module brings the promoter of the fetal globin genes (HBGII and HBGI) physically together with the locus control region (LCR) forcing fetal globin gene expression and is expected to improve red blood cell morphology in the thalassaemia condition. CRT, cis-regulation therapy.

The FMR1 study provides an important proof-of-principle example that DNA demethylation or epigenome editing could potentially be used for therapeutic purposes. However, a major issue that still remains is how long DNA demethylation editing can be maintained, as none of the cell or animal-based studies has shown maintenance for more than a few weeks. DNA demethylation editing is based on the premise that it leads to a change in the epigenomic signatures that ultimately leads to transcriptional changes. This is quite different from the aforementioned gene modulation using promoter/enhancer CRISPRa or CRISPRi that directly brings transcriptional activators or repressors to specific CREs. To overcome tight regulation of epigenetic marks in a sustainable manner could be a potential challenge in achieving long-lasting DNA demethylation.

Conversely, several studies have shown that de novo DNA methylation editing/marking can persist as an epigenetic memory during cell division^75–77. This strategy, where transient expression of fusion modules can lead to permanent epigenetic changes over cell divisions, can also be referred to as a ‘hit-and-run’ strategy. The triple fusion of DNMT3A–DNMT3L and KRAB with a DNA targeting module can lead to long-term methylation and persistent gene silencing over several cell divisions even when transiently expressed. For example, this approach specifically reduced the expression of the β₂-microglobulin (B₂M) gene by 78% in an erythroleukaemia cell line using dCas9 fused to KRAB, DNMT3A and DNMT3L (REF.⁷⁵). The DNA methylation mark at the B₂M locus propagated for up to 180 days and 36 cell divisions after removal of the fusion construct. The level of B₂M protein in blood and urine is a crucial prognostic marker for many pathological conditions, such as amyloidosis, multiple myeloma and lymphoma, and reducing it can provide therapeutic benefits⁷⁸. Interestingly, efforts to reactivate the epigenetically silenced B₂M locus by targeting dCas9–VP160 or dCas9–p300 were unsuccessful, whereas dCas–Tet1 fusion could demethylate and reactivate the locus up to 45% (REF.⁷⁵). This suggests that changes in the local epigenetic state are a prerequisite before transcriptional upregulation and that the epigenetic state of the locus plays a crucial role in facilitating accessibility for CRISPR–dCas9 binding^29,30. In another study, a dCas9–DNMT3A-3L fusion targeted to the cyclin-dependent kinase inhibitor 2A (CDKN2A) gene methylated CpG nucleotides ‘permanently’ and silenced this gene in primary breast cells from healthy donors⁷⁶. More research is needed to determine whether long-lasting effects of this approach can be replicated in vivo. The aforementioned ‘hit-and-run’ strategy to methylate DNA and permanently silence the locus, even in the absence of fusion constructs, is of particular interest where constant expression of CRT tools can lead to side effects.

Modifying chromatin to alter gene expression

Histone modifications.

Chromatin structure and organization can discretely modulate the activity of CREs that regulate gene expression. Several studies have implicated aberrant chromatin structure and histone modifications in organism development, cell differentiation and human disease^12,79,80. Post-translational modifications of histones at specific amino acid residues assign particular structure and function to the chromatin. There are several histone post-translational modifier proteins that either add or remove various chemical groups — such as phosphorylation, methylation, acetylation, ubiquitylation and others — at specific amino acid residues of histones that determine the transcriptional and functional state of the locus⁸¹. For example, histone H3 dimethylated or trimethylated at lysine 9 (H3K9me2, H3K9me3) makes DNA transcriptionally repressed, and acetylation of the same lysine residue (H3K9ac) marks transcriptionally active DNA. Similarly, H3K27ac is predominantly found in active promoters and enhancers. Targeting histone-modifying enzymes to specific DNA locations can add or remove chemical groups from the histones and could modulate gene expression.

Customizable DNA binding modules — zinc fingers, TALEs and CRISPR–dCas9 — fused to histone-modifying enzymes (TABLE 2) have been successfully used to alter gene expression (FIG. 1d). Silencing vascular endothelial growth factor A (VEGFA) is of therapeutic significance in cancer, as VEGFA expression has been associated with pro-angiogenic effects^82,83. To show whether VEGFA can be silenced, zinc fingers fused to histone 3 lysine 9 methyltransferase (EHMT2/G9a or SUV39H1; TABLE 2) was targeted to its promoter, enriching methylation of H3K9 and repressing its expression in HEK293T cells⁸⁴. Similarly, in cells, TALEs fused to the lysine demethylase 1A (LSD1) have been used to deactivate enhancers and reduce their target gene expression by modifying the histone signatures (H3K4me2 and H3K27ac) around the locus⁸⁵. Alternatively, customizable DNA binding modules fused with histone acetyltransferase, such as the E1A binding protein p300 (EP300), targeted to enhancers can be used to enrich H3K27ac and increase their target gene expression (FIG. 1d; TABLE 2). Histone acetyltransferase EP300 fusion with dCas9, TALEs or zinc fingers was shown to efficiently enrich for H3K27ac in promoters or enhancers and robustly upregulate target gene expression in HEK293T cells⁸⁶.

The aforementioned studies suggest that it might be more straightforward to ‘write’ histone marks, as observed with fusions of H3K9 methyltransferase (G9a, SUV39H1) or H3K27 acetyltransferase (EP300), than to ‘erase’ existing histone marks such as H3K4me2 and H3K27ac (LSD1). However, more testing is needed to provide sufficient support for this hypothesis. Previous studies show that nucleosome occupancy can affect CRISPR targeting, with nucleosome free regions being more accessible for sgRNA–Cas9 or dCas9 binding²⁹. The knowledge of the tissue-specific chromatin accessibility of a genomic region could help develop algorithms for better targeting of histone modifier fusion systems and may explain why some loci are more resilient to histone modification editing than others.

Several major concerns need to be addressed for developing epigenome editing for CRT applications. For example, reversal of a compact chromatin structure could be a daunting task in postmitotic cells, for example neurons, whereas it could be easier to achieve in dividing tissues. In addition, maintaining a stable epigenome mark in a long-lasting manner in both dividing and terminally differentiated tissue may be challenging. Some histone marks are shown to contribute towards the epigenetic memory of the locus by inheritance of a heterochromatin state during somatic cell division; for example, X-inactivation and centromeric heterochromatin. Recently, direct evidence for this hypothesis was shown by a study that marked nucleosomes of specific loci in parental cells by dCas9–biotinylation fusion and followed their reintegration into the chromatin of daughter cells²⁹. This system allowed tracking of local redeposition of biotinylated nucleosomes for repressed or active chromatin domains in mouse embryonic stem cells. Repressed chromatin domains that were marked with H3K27me3 in daughter cells directly inherited the biotinylated nucleosomes. Another study reported that propagation of H3K27me3 could be possible, similar to the aforementioned ‘hit-and-run’ strategy for DNA methylation, by targeting dCas9 fused with Dnmt3a–Dnmt3l and the enhancer of zeste 2 polycomb repressive complex 2 subunit (Ezh2) gene, an H3K27 histone methyltransferase⁸⁷. This fusion system was able to silence the erb-b2 receptor tyrosine kinase 2 (ERBB2; also known as HER2) gene for up to 57 cell divisions in human colon carcinoma HCT116 cells⁸⁷. It would be interesting to see whether H3K27me3 editing can be maintained during cell division in vivo so that it could be developed into a ‘hit-and-run’ CRT strategy.

The programmable histone modifiers generated so far are undoubtedly promising tools in the epigenome editing toolkit. However, it remains to be seen how effective these tools will be for reversing disease phenotypes.

3D looping.

Gene-regulatory elements physically interact with one another in the nucleus to control gene expression¹². For example, an enhancer that is distant to its target promoter can contact and activate its promoter by forming loops. There are many emerging examples of diseases that are caused by abnormal DNA looping. One of the most-studied regions for 3D regulatory interactions is the human β-globin locus. During embryonic development, fetal globin genes that have high oxygen affinity are expressed in erythroid cells and are later switched at birth to other globin genes with lower affinity⁸⁸. The temporal expression of these globin genes is controlled by the locus control region that switches the looping interaction between the fetal globin genes and the adult globin genes, mediated by the self-association domain of the looping factor LIM domain binding 1 (LDB1; TABLE 2). LDB1 can bind to distant DNA elements and bring them together via dimerization. Patients with mutations in the adult β-globin gene, haemoglobin subunit-β (HBB), develop sickle cell anaemia (OMIM 603903) or β-thalassaemia (OMIM 613985). The finding that some patients who present a less severe form of the disease have elevated levels of fetal globin genes during adulthood^89,90, led to investigations of the effects of derepressing or reactivating the fetal globin genes for developing treatments for sickle cell anaemia or β-thalassaemia.

As DNA looping is responsible for the switch from fetal globin gene expression to adult globin, forced looping of the locus control region with fetal globin in adulthood could potentially reverse gene expression. To switch back the looping interaction from the adult globin gene promoter to the fetal, a zinc finger protein fused with the self-association domain of Ldb1 was designed to bind to the fetal globin gene promoter (FIG. 4b). In mouse adult erythroblasts and human CD34⁺⁺ cells (derived from healthy human bone marrow), this approach robustly reactivated fetal globin gene expression⁹¹. In a follow-up study, transduction of haematopoietic cells isolated from patients with sickle cell disease with lentiviral-mediated zinc finger–Ldb1 fusion targeting the promoter of the fetal globin gene, resulted in robust reactivation of the fetal γ-globin compared with the reactivation seen with the approved drug, hydroxyurea (HU)⁹². Consequently, mutant β-globin or sickle globin levels decreased along with the sickling phenotype in erythroid sickle cell disease cells under hypoxia⁹². It is also worth noting that this forced looping strategy can be particularly advantageous in patients with sickle cell anaemia or β-thalassaemia as it can reduce the expression of the mutant adult β-globin gene and induce fetal globin expression at the same time.

To develop a customizable forced looping method, an intriguing strategy has been adopted from the plant phytohormone S-(+)-abscisic acid (ABA) system, originally shown to induce the co-localization of two protein molecules upon ABA addition⁹³. Upon ABA binding, the ABA receptor allosterically interacts with ABA insensitive 1 (ABI1) protein via its PYL1 domain. Borrowed from plant phytohormone signalling and repurposed with a CRISPR component, this system was named the CLOuD9 (Chromatin Loop reOrganization using nuclease Deficient Cas9) targeting system (TABLE 2). The fusion of Staphylococcus aureus dCas9 with the PYL1 domain and Streptococcus pyogenes dCas9 with the ABI1 domain induced dimerization of the two different dCas9s upon induction with ABA in K562 and HEK293T cells⁹⁴. In this study, the reversible looping was induced between the adult β-globin promoter and locus control region using the CLOuD9 system in K562 human erythroleukaemia cells, which are known to express high levels of the fetal globin gene. Interestingly, the bioavailability of this ABA-induced proximity system was detected in serum up to 4 hours, even following oral administration in the mouse⁹³. Although considered non-toxic by Environmental Protection Agency, repeated dosing of ABA using the CLOuD9 system has yet to be investigated. Another CRISPR-based bivalent system with orthologous dCas9 fused to leucine zipper domains (dCas9-Zip; TABLE 2), which has high binding affinities, was developed to induce looping based on varying concentrations of dCas9-Zip (REF.⁹⁵). The inducible systems for customizable DNA looping have not been thoroughly investigated in in vivo models and neither have the off-targeting effects of their non-specific looping interactions. However, the ability of these systems to modulate gene expression by modifying DNA ‘looping’ hold significant potential for CRT applications.

CRT treatment for additional diseases

CRT has the potential to treat many other diseases in addition to those discussed above. For example, CRT tools could be designed to target multiple genes simultaneously in heterozygous micro-deletions or duplications (BOX 1). Certain autosomal recessive mutations, for example, a missense mutation that leads to lower protein levels of a cell-membrane transporter, can be rescued by upregulating expression such that a physiologically relevant amount of transporter is trafficked to the cell membrane. Mutations in CREs, such as promoters, enhancers or silencers, can be rescued by upregulation or downregulation of their target genes, if the mutations result in loss of function or gain of function, respectively (BOX 1). For example, a mutation in the HBB promoter that leads to lower HBB expression, resulting in haemoglobinopathy or thalassaemia, can be rescued by upregulation. In addition, there are several diseases that are caused by mutations in one gene copy whereas the other copy is inactive. For example, in parent-of-origin genomic imprinting or random X-inactivation⁹⁶ (BOX 1), CRT could be a potential strategy to reactivate the silenced allele and rescue the disease phenotype.

Delivery methods for CRT

Successful CRT requires safe and efficient delivery of the various modalities in vivo. Large biomolecules, primarily nucleic acids, are delivered in vivo broadly by viral or non-viral-based methods. Similar to gene therapy, CRT modules need to be constantly expressed in order to achieve the desired expression levels of the targeted gene in a long-lasting manner. For transient expression of CRT tools, mRNA or protein-based deliveries can be applied as mentioned above, using a ‘hit and run’ strategy. Each method has its own advantages and disadvantages and could be optimized for CRT depending on the specific biomedical applications and delivery routes. Below, we discuss both viral and non-viral delivery methods that can be applied for DNA, mRNA or protein-based delivery of CRT tools.

Viral delivery

Viruses hijack the genetic machinery in the host genome to replicate their own genome. They contain genetic material in the form of DNA or RNA and have an infectious proteinaceous coat, a ‘capsid’, that can attach to cellular membranes and assist cell entry. As viruses are extremely effective in entering cells, they have been extensively manipulated to deliver genetic payloads without causing side effects.

Gene therapy applications have extensively explored how to tame these viruses for effective delivery of a recombinant gene. Several criteria are taken into consideration when designing viral vectors, including safety, toxicity and pathogenicity, cell type specificity, infection efficiency, payload capacity of the viral genome, stability of their genomes within the infected cells and pre-existing immunity in the patient population⁹⁷. Compromising any of these criteria can lead to failure of gene therapy in the clinic.

Many different viruses, including vaccinia, vesicular stomatitis virus, polio, lentivirus, γ-retrovirus, adenovirus, adeno-associated virus (AAV) and herpes simplex virus, have been used for vaccines, onco-therapies and gene therapy⁹⁸. Retroviruses and lentiviruses are mainly used for ex vivo therapies, whereas AAV is favoured for direct gene delivery in vivo and is the most commonly used delivery vehicle for most clinical gene therapy applications¹³. There is growing interest in developing AAV-based zinc finger, TALE or CRISPR therapies to achieve efficient delivery. Packaging of nuclease-deficient zinc fingers, TALENs and CRISPR fusion systems into AAV vectors is possible when the fusion proteins are small. For gene activation, VP64, which is 180 bp in length, can easily fit into an AAV when fused with dCas9, but an EP300 (1,850 bp) fusion with dCas9 is longer than the packaging capacity for AAV vectors (TABLE 2). Zinc fingers and TALEs are smaller in size and their fusions with large proteins could fit into AAV vectors.

Other CRISPR–dCas9-based fusion systems, such as DNA methylation editing, histone modification editing or the zinc finger system for forced looping, have all been delivered by lentiviruses that have a much higher payload capacity than AAVs. However, lentiviruses are associated with higher pathogenicity and immune response and integrate into the genome, thus limiting their therapeutic application.

In general, there are several safety concerns associated with recombinant viral vectors. Viral capsids can be immunogenic to some extent and elicit a humoral response. Strict manufacturing practices have to be followed in order to achieve the purity and titres necessary for therapeutically effective viral vectors. In addition, individuals may have pre-existing neutralizing antibodies, which can make gene therapy ineffective and induce various side effects. Repeat dosing could also be a treatment challenge as patients can develop an immune response from the initial dose. However, the field of gene therapy has significantly matured over the past few decades, and strategies to mitigate the risks for viral vectors are emerging, transforming the treatment options for genetic diseases^99–102.

Non-viral delivery

Nanoparticles made of ‘synthetic material’ have enormous potential as an alternative non-viral delivery mechanism for nucleotides or proteins. These synthetic particles can package larger genetic payloads and exhibit low immunogenicity compared with viral delivery platforms, providing a safer delivery option. Synthetic particle-mediated delivery could be an approach for patients with existing immunity for viral vectors or for patients in whom repeated viral dosing would be ineffective.

Nanoparticles made from lipids, polymers and gold have been extensively studied as potential delivery modules for gene therapy. Cationic (positively charged) lipid or polymer complexes, also known as liposomes, are commonly used to deliver anionic (negatively charged) nucleic acids in cells. These are stable nanocomplexes/particles that enter the cell through endocytosis. Some of the commonly used synthetic cationic lipids include DOSPA (2,3-dioleyloxy-N-[2(sperminecarboxamido)ethyl]-N,N-dimethyl-1-propanaminium trifluoroacetate), DOTMA (N-[1-(2,3-dioleyloxy) propyl]-N,N,N-trimethyl-ammonium chloride), DOGS (dioctadecylamido-glycylspermine) and DOTAP (1,2-dioleoyl-3-trimethylammonium propane). They all have both hydrophilic and lipophilic properties and are often combined with neutral lipid such as cholesterol or DOPE (1,2-dioleoyl-sn-glycero-3-phosphoethanolamine) to enhance cell membrane fusion and improve gene expression and stability. Lipid-based particles can be covalently conjugated with cell type targeting moieties such as antibodies, proteins, receptor ligands and vitamins to enhance the targeting efficiency in desired cell types. These combination liposomes have been shown to efficiently deliver DNA cargo in vivo into tissues such as the spleen, kidney, muscle, liver, heart or lungs^103–106. Polymer-based particles are mainly polyethyleneimine (PEI) or poly-l-lysine (PLL) based and have high efficiency for cell delivery, but due to their toxicity and non-biodegradable properties are less preferred for gene therapy¹⁰⁷. Other polymers such as amino esters have lesser toxicity, but have sub-par delivery efficiency. Notably, a study that screened over 56 non-viral delivery agents comprising different lipids, polymers or magnetic/gold/carbon nanoparticles for efficient delivery of CRISPR in a human melanoma cell line (A375) found cationic liposomes to be the most efficient¹⁰⁸. In vivo delivery of CRISPR liposomes in mouse liver to edit the transthyretin (Ttr) gene could successfully reduce the serum Ttr levels¹⁰⁹. Lipid nanoparticles have also been used to deliver a CRISPR–dCas9 fused with an adenine base editor system, which has a long CDS (5,200 bp) that cannot be packaged into AAV vectors¹¹⁰. Targeted delivery of CRISPRi in adipocytes via ATS⁶⁷, as discussed above, is another potential non-viral delivery approach that could be used for CRT (FIG. 3b).

In summary, synthetic nanoparticles can be an appealing option for CRT delivery, in particular for large payloads. In addition, these non-viral-based methods are applicable for mRNA and protein delivery of CRT tools in cases where one-time expression may be enough to achieve therapeutic benefits. This option is of particular interest for therapeutic editing using a cutting Cas9 where the persistent expression of Cas9 is undesirable. However, most of the CRT tools may need constant expression to maintain the modified expression of the target gene.

Challenges and road to the clinic

Despite the proof-of-concept studies described in this Review, there are many challenges that need to be addressed in order to develop CRT as a viable approach to treat patients. For example, studies are needed to characterize the immunogenicity of DNA targeting fusion modules or to determine at what levels their expression could be physiologically safe. The following are some of the most crucial factors that must be considered in order to bring CRT to the clinic.

Expression level thresholds

CRT is based on modulating gene expression levels to treat disease. Each gene might require different dosage levels in order to provide physiological benefits. For CRISPRa, upregulating a gene beyond a certain threshold level could have detrimental effects. Similarly, determining how much downregulation would be beneficial to correct a certain disease is also necessary to develop effective CRISPRi applications. Fusion domains that can modulate gene expression optimally according to the physiological needs would be extremely effective. In both the Sim1 (REF.²) and Scn1a (REF.³¹) studies, there is evidence to suggest that targeting CREs may preserve the aspects of a negative-feedback mechanism that can control physiological levels of mRNA. Drug-inducible approaches that can turn gene expression of a DNA targeting molecule on/off could potentially be used to maintain the expression level thresholds of the target gene^111,112. However, drug-inducible systems must have FDA approval to be used along with CRT. Other systems such as anti-CRISPR proteins that inhibit the activity of CRISPR–Cas9 can be investigated to provide stringent control for CRT^113–115. Another important concern is how accurately the animal model recapitulates the human phenotype, as the expression level thresholds between species could differ.

CRE target selection

An understanding of the CRE location and function of the ‘disease causing genes’ is necessary for CRT design. Genome annotation databases, such as the UCSC Genome Browser¹¹⁶, are useful in identifying sequences in the promoter or other CREs that can be targeted. Large-scale screens in cells have shown that most promoters are amenable to CRISPRa/CRISPRi and sgRNAs can be selected based on the required expression levels^35,117. However, as many genes have alternative promoters that regulate different tissue-specific transcripts¹¹⁸, knowledge about the activity of the promoters in the disease-associated cell/tissue type is imperative to their selection as targets for CRT. Targeting enhancers has an added advantage in defining CRT tissue specificity². Large-scale functional genomic studies in numerous cell types and tissues have identified and characterized distal enhancers^{4,5,119–122}, and 3D chromatin interaction analyses have provided valuable insights into linking these enhancers with their target promoters¹²³. High-throughput studies targeting enhancers using CRISPRi demonstrate that many enhancers could be usable targets for CRT^121,124,125. As most of these high-throughput studies are done in cell lines, the regulatory circuitry of many of these enhancers in vivo is still unknown, and a systematic understanding of their activity is necessary to design the optimal targeting strategies. Characterization of other CREs, such as insulators and silencers, has been lagging due to the deficiency of high-throughput functional assays. Recent development of these assays has expanded the capability to detect silencers in mammalian genomes^126,127. The current understanding about mammalian insulators is largely driven by characterizing CTCF binding and there is a paucity of systematic assays that can identify insulators in an unbiased manner^128,129. CTCF-bound chromatin insulators are known to facilitate chromatin looping and could thus be amenable for modulating chromatin loop domains¹³⁰. In summary, detailed functional characterization, including in vivo studies, will be needed to select CREs that will be effective targets for CRT.

Tissue specificity

Similar to gene replacement therapy, achieving tissue specificity will be a major safety concern in CRT. Studies have shown that tissue/cell type-specific modulation of gene expression by CRT could be dependent on the targeted genomic locus. Large-scale Cas9-based cell culture screens have indicated a targeting preference for nucleosome free regions^30,131. Future work that will screen for tissue-specific CREs can facilitate in finding the appropriate targets for CRT. The aforementioned transgenic Sim1 CRISPRa targeting showed tissue-specific upregulation based on the targeted regulatory element² (FIG. 2a). However, this specificity will be dependent on the context of the locus or sequence to be targeted (TABLE 1). For example, klotho, a gene that is silenced in muscle, was activated using the strong CRISPR-based SAM system in the muscle of mdx mice (DMD model)^52,37. To this end, knowledge about the status of chromatin accessibility can help in designing efficient targeting modules for CRT. Tissue specificity can also be achieved by designing the CRT constructs to be expressed under tissue-specific promoters or enhancers, as used to drive transgene expression for conventional gene therapy. For example, a muscle-specific promoter of the human ‘muscle creatine kinase’ (MCK) gene was used to drive micro-dystrophin delivered via intramuscular injections^41,132. In addition, AAV serotypes that have tissue type specificity can be used for packaging CRT modules to broadly define tissue specificity¹³. Similarly, nanoparticle-mediated delivery of CRT reagents can be targeted to certain cell types by conjugating lipids with tissue-specific peptides^133–135. Another key consideration in achieving long-lasting effects of CRT is the proliferative capacity of the treated cell type. For example, CRT modules delivered to the central nervous system may remain active for a very long time as the central nervous system consists of predominantly non-dividing cell populations. On the other hand, when targeting the liver, which is a highly proliferative tissue, repeated dosing may be required.

Off-target effects

The DNA targeting modules used in CRT need to be highly specific without any off-targeting. As CRT technologies include a nuclease-deficient DNA targeting module that does not edit the DNA, the off-targeting effects of these tools would not cause nucleotide changes termed ‘DNA scars’. However, CRT tools could bind to non-specific regions in the genome in a concentration-dependent manner that may affect the expression levels of nearby genes. The off-targets of CRT tools can be assessed by assays that examine, in a genome-wide manner, DNA binding, activity of their fused effector domain (for example, H3K27ac for EP300 fusion) and transcriptional changes. Chromosome conformation capture (3C)-based genomic-wide assays should be carried out to rule out the promiscuous effect of expressing a DNA looping domain. Improvements in the binding specificity of DNA targeting modules (zinc fingers, TALENs and CRISPR) could lower the off-target effects of CRT tools^136–139. For example, the secondary structure or target sequence of the guide RNA can determine the binding efficiency of Cas9/dCas9 systems^35,138. As the target efficiency of the CRISPR–dCas9 system often depends on the effector domain fused to it, sgRNA designing tools that take into account the effector fusion for the CRISPR–dCas9 system could help to improve the on-target efficiency of these fusion modules¹⁴⁰. Improved in silico tools are required to predict off-targets that also take into account known human variation (using currently available databases such as dbSNP¹⁴¹, ExAC²⁴, gnomAD²⁵ and others) when designing the DNA targeting modules and could also exploit allele-specific polymorphisms to improve targeting, if allele-specific expression would be beneficial. Finally, the model system used for off-target evaluation also plays an important role in translating this information to patients. Evaluating the off-targets in primary cells or cells derived from patients will be more appropriate than using established cell lines that can have genomic aberrations.

Cytotoxicity

Most of the CRT tools discussed in this Review consist of a fusion with an effector domain of known proteins. These proteins include transcriptional activators, repressors, chromatin-modifying enzymes or other catalytic domains adopted from different biological systems. Expressing these recombinant proteins in vivo could have cytotoxic effects. For example, a broad level of DNA demethylation was observed for the CRISPR–dCas9–TET system that did not correspond to its off-target binding⁷⁴. Thus, it would be necessary to ascertain the activity of fused catalytic domains on a genome-wide level. Standard cytotoxicity evaluations, such as metabolic activity, cell growth, cell division, protease activity and others, need to be carried out for CRT tools at different concentrations^142–144. Strict manufacturing practices would ensure the purity and effective concentrations of the CRT tools packaged into viral or non-viral vehicles. For example, the amount of AAV titres, the size of nanoparticles and their ratio of cargo versus empty particles are required to show efficacy without inducing any cytotoxicity.

Immunogenicity

The immunogenicity of CRT tools, which include fusion domains with nuclease-deficient zinc fingers, TALEs and CRISPR, has not been assessed. However, the evaluation of immunogenicity for orthologous CRISPR–Cas systems suggests that pre-existing innate and adaptive immune responses against Cas9 protein may hinder CRISPR-based therapies^145,146. Modifying the Cas9 protein to avoid existing antibodies can overcome this hurdle¹⁴⁵, at least for the first dose. Adapting the immune response by activating Cas9-specific regulatory T cells can also mitigate the effects of effector T cells against the Cas9 protein^147,148. Orthologous CRISPR–Cas9 systems from non-pathogenic bacteria that have no prior exposure to human populations can also be explored for immune response assessments^149–153. In an effort to lower the risk of immunogenicity for the ZF-KRAB fusion that was designed to lower Htt expression in a Huntington disease mouse model, a host-matched ZF-KRAB construct was designed containing crucial modifications that resulted in less microglial proliferation and reduced cytotoxic responses when injected into the mouse brain¹⁵⁴. Similarly, human-adapted fusion domains can be generated, which can be less immunogenic. Long-term evaluation of CRISPR–Cas9-treated DMD mouse models has shown that immune responses may vary depending on the age of the mice at the time of treatment¹⁵⁵. This study suggests the importance of age-dependent evaluation of the immune responses for CRT. Existing data and knowledge on overcoming immunogenicity of the viral and non-viral delivery modules can be applied for CRT applications^156–158. The engineering of rAAV capsids to evade immune responses has also shown promise for safer gene therapies^156,159,160 and could be adopted for CRT.

Delivery

The optimal delivery route for CRT tools would largely depend on which tissues need to be treated. Choosing the best delivery route for CRT modules along with extensive safety and toxicity studies in rodent model systems and non-human primate models is crucial in order for CRT to be successful in the clinic. The same delivery tools as those approved for conventional gene therapy could be used for CRT. For example, for central nervous system disorders, AAV vectors injected intravenously that can cross the blood–brain barrier or intrathecal injections can be used for spinal cord delivery¹⁶¹. Intravitreal delivery is a standard route for retinal disorders¹⁶². It is also crucial to ascertain the number of cells to be treated, which may vary depending on the target tissue, delivery route, bioavailability and stability of the CRT module. The vector construction of CRT modules and the delivery vehicle for packaging are important determinants in addressing DNA stability and integration concerns. Recombinant AAVs are replication deficient, persist as episomes and are preferred for their stable gene expression with a low potential for integration¹³. These properties make AAV an ideal vector for non-dividing cells, but in dividing tissues AAVs are less stable and are gradually lost during cell divisions. Increasing the titres of AAVs may not circumvent this problem, as it can lead to spurious genomic integrations and genotoxicity^163,164. Non-viral delivery platforms, such as liposomes, could overcome genome integration concerns, but they are prone to degradation and may need frequent dosing¹³⁵. Liposomes would be preferred vehicles to deliver CRT modules as mRNA or proteins where transient expression is required. Some of these challenges can be addressed by improving delivery vehicles, such as engineering stable and degradation-resistant lipids for liposome delivery¹⁶⁵. Advances in AAV vector generation and capsid engineering can also address some of the aforementioned concerns for viral delivery platforms^159,166,167. The choice of delivery platform would ultimately need to take into account several criteria, including the stability of the vector, consistent expression of CRT modules and low genotoxicity due to potential integration.

Genotyping and phenotyping

As CRT applications are dependent on the genetic nature of the mutation, it will be necessary to reliably genotype patients. Genetic testing for patients should be done according to Clinical Laboratory Improvement Amendments (CLIA) certification to validate their mutation. Mutations in the coding sequence of a gene should be characterized thoroughly for their functional consequence before deciding which CRT strategy can be applied for rescue. To this end, genotype–phenotype correlations and family history should be considered whenever possible, but this could be challenging especially for de novo mutations that are found only in the patient. For many mutations, allele-specific targeting would be appropriate to avoid any deleterious effects arising due to targeting of both of the alleles. For example, in a gene downregulation strategy for Huntington disease, only the mutant allele needs to be targeted and not the wild-type allele^59,60 (FIG. 3a). Methods to improve allele-specific targeting should be further investigated. To target a specific allele, phasing would be necessary, matching the region that is targeted to the allele-carrying mutation by matching SNPs that are on the same chromosome. DNA sequencing approaches that take advantage of long-read sequencing, genotyping or sequencing of the patient’s parents or siblings, if available, and/or various human allele annotation data sets can be applied to match/phase the allele with the mutation and the targeted region.

Outlook

CRT has enormous potential for treating human disease. CRT can be applied to upregulate or downregulate genes that cause disease due to lower or higher expression levels, respectively, and can also be used to specifically alter the expression of genes that can assist in ameliorating disease phenotypes. Proteins that can induce epigenetic modifications, such as methylation or histone modification, or DNA looping can be harnessed in CRT to modulate gene expression. It is crucial to examine the benefits and risks of using CRT over other therapeutic modalities. Through a collaborative effort among researchers, patient communities, clinicians, regulatory agencies and industry, CRT platforms may become a viable approach to treat many genetic diseases currently lacking therapeutic options.

Glossary

Cis-Regulatory elements

(CREs). DNA sequences that regulate the transcription of a neighbouring gene.

Packaging

Assembly of the nucleic acids and capsid during virus generation.

DNA scars

Irreversible and unintended DNA changes caused mainly due to off-targeting by DNA targeting modules with functional nucleases.

DNA looping

Physical DNA–DNA interaction in the genome within 3D nuclear space.

Nanoparticles

Particles that are between 1 and 100 nm in diameter.

Intracerebroventricular

A route of delivery via injection into the cerebrospinal fluid in cerebral ventricles.

Trinucleotide repeat expansion

A specific 3-bp DNA sequence that has more copies than normal in the genome.

Bioavailability

The proportion of the therapeutic agent upon administration that has an active effect.

Off-targeting

The effects arising due to non-specific and unintended targeting of DNA targeting modules such as zinc fingers, transcription activator-like effector (TALE) and CRISPR in the genome.

Delivery routes

The methods of administration of a therapeutic agent based on the site of action.

Capsid

(Also known as a viral envelope). The proteinaceous shell that packages the genetic material of the virus. its structure is important in determining viral stability, delivery and host interactions.

Pre-existing immunity

The adaptive immune response of the body due to pre-exposure to an antigen.

AAV serotypes

(Adeno-associated virus serotypes). The variations in the capsid surface proteins of an adeno-associated virus that can define its transduction efficiency in different tissue or cell types.

Blood–brain barrier

The blood–brain barrier is the membrane made from endothelial cells surrounding the blood vessels that selectively allows solutes to transfer from the blood to the central nervous system.

Intrathecal

A route of delivery via injection into the spinal canal in order to avoid the blood–brain barrier selective permeability.

Intravitreal

A route of delivery into the vitreous humour of the eye.

Episomes

Circular DNA that is not integrated in the genome.