Odd genetic observations in the 1980s and 1990s indicated that RNA may play a general role in regulating gene expression, when it was reported that both sense and antisense RNAs could modulate gene expression transcriptionally and post-transcriptionally, referred to as ‘co-suppression’, ‘gene silencing’ or ‘RNA interference’ (RNAi). The finding that introducing sense and antisense RNAs together resulted in systemic silencing of target genes led to the dissection of the RNAi pathways, showing that dsRNA precursors are processed to form small RNAs that guide DNA methylation and cleavage of orthologous sequences in mRNAs. It was soon discovered that these pathways are used extensively to control gene expression during differentiation and development via ‘microRNAs’ (miRNAs) derived from introns and other non-protein-coding transcripts, the numbers of which increased during animal evolution. Similar regulatory RNA pathways were characterized in plants. Related small RNAs, ‘piRNAs’, were found to be required for germ and stem cell development in animals, some of which are produced from transposon-derived sequences. Other classes of small regulatory RNAs were found to be generated from tRNAs, rRNAs, snoRNAs, snRNAs, gene promoters and splice junctions, and small RNAs were shown to mediate intergenerational and interspecies communication. A related pathway, termed ‘CRISPR’, was found to target cleavage of bacteriophage genomes, manipulation of which has revolutionized genetic analysis and genetic engineering. Both CRISPR and RNAi use RNAs to guide generic effector proteins sequence-specifically to target RNAs and DNAs, a highly efficient and flexible system of gene control.

Unusual Genetic Phenomena Involving RNA

The demonstration that RNAs can use base pairing to form double-stranded structures dates back to 1956. ³ In the 1960s and 1970s, numerous reports showed the existence of double-stranded RNAs (dsRNAs) derived from viral RNA genomes replicating in eukaryotic cells. It had also been known since the 1960s that synthetic and naturally occurring viral, bacterial and fungal dsRNAs can induce antiviral cytokines called ‘interferons’ ^a and inhibit protein synthesis in animal cells at low concentrations, an important feature of innate immunity. ^8–15 The first candidate RNA drug (reported in 1967) was a fungal extract that contained “an anti-viral agent … [which] is an RNA of viral origin”. ^{10 , 15 , 16}

The discovery of interferons and their induction by dsRNAs explained, at least in part, a phenomenon known since the 1920s as ‘virus interference’, ⁴ whereby viral infection results in resistance to subsequent infections by the same or related viruses. ^{17 , 18} The mechanism was subsequently shown to involve two interferon-induced, dsRNA-dependent enzymes that inhibit translation and activate an endoribonuclease. ^{19 , 20} In 1985, Masayori Inouye and colleagues developed an antisense platform (‘micRNA’) to regulate host gene expression, which has “great potential as a novel cellular immune system for blocking bacteriophage or virus infection”, ²¹ with Sergio Giunta and Giuseppe Groppa suggesting that

Upon virus infection and exposure to viral nucleic acids, the cell synthesizes micRNAs that … constitute some of the dsRNAs that are involved both in interferon induction and in the cellular metabolic pathways activated by interferon … and may also be preferentially cleaved by cellular enzymes which would account for the still unexplained discrimination between cleavage of cellular and viral mRNAs in interferon treated cells. ²²

presaging both the mechanism and the sensing of foreign nucleic acids by the Toll pathway. ^b

In the late 1980s, Andrew Fire, Craig Mello and others showed that the introduction of DNA constructs containing fragments of genes in C. elegans caused suppression of the endogenous homologous gene. ³² As expected, this silencing was observed when constructs were designed to drive antisense RNA transcription. ³³ Surprisingly, however, silencing was also observed when designed to drive ‘sense’ gene expression, which was expected to do the opposite, i.e., increase the mRNA levels. ³³ Adding to this puzzle, in 1995, Su Guo and Ken Kemphues obtained the same result by directly introducing in vitro transcribed RNAs in either the sense or antisense orientation, which caused ‘silencing’ of the corresponding gene in C. elegans. ³⁴

Similar silencing effects in plants and fungi were also reported in the 1990s, ³⁵ when it was established that the presence of multiple copies of introduced transgenes resulted in repression not only of the transgenes, but also of the homologous endogenous genes. ^c

The iconic example is flower color in petunia, where the effect of silencing of endogenous loci by antisense RNAs was already well established, as famously illustrated a decade earlier by the changes of pigmentation in transgenic petunia flowers expressing antisense RNAs targeting genes specifying the anthocyanin biogenesis pathway, ^39–42 which appeared to result from “increased RNA turnover”. ⁴³

In 1990, two groups did the opposite experiment but unexpectedly obtained the same result – they overexpressed a pigment synthesis gene hoping to produce deep purple petunia flowers, but instead obtained predominantly white flowers, with remarkable mosaic patterns, including one dubbed “The Cossack Dancer” (Figure 12.1). ^{41 , 44 , 45}

Figure 12.1

Homology-dependent gene silencing. The ‘Cossack Dancer' variegated pattern in transgenic petunia expressing a chimeric chalcone synthase pigmentation gene resulting in co-suppression of the endogenous genes. (Reproduced from Napoli et al. with (more...)

These phenomena were variously termed ‘co-suppression’, ‘homology-dependent gene silencing’, ‘transgene silencing’ (‘quelling’ in fungi) ^{35 , 46–49} or ‘RNA-mediated interference’ (RNAi), ⁵⁰ and later shown also to occur in Drosophila. ^{51 , 52}

Despite being initially regarded as a peculiar phenomenon, as early as in 1986 ‘co-suppression’ began to be used as a biotechnological tool to modulate plant phenotypes, ^{36 , 53–55} only a few years after the first transgenic plants had been constructed by the introduction of foreign genes. ⁵⁶ Similar to the puzzling RNA interference observed in worms, it was suggested that the silencing caused by introduced copies of ‘sense’ transgenes in plants ^{44 , 57 , 58} might be triggered by “unexpected” promoter activity in the antisense orientation. ^{59 , 60} Nevertheless, this remained a matter of discussion, as the mechanisms behind these phenomena were unknown and appeared to be heterogeneous. ⁶⁰

Several models involving RNA-RNA, RNA-DNA and DNA-DNA pairing interactions (or RNA-DNA triplexes ⁴⁵ ) were proposed and investigated by the pioneering groups, those of Michael Wassenegger, Marjori and Antonius Matzke, David Baulcombe, Jan Kooter and others. ^{48 , 61–65} In some cases, silencing appeared to be caused by epigenetic mechanisms; in particular sequence-specific RNA-directed DNA methylation ^{48 , 65–69} (see below and Chapter 14), which was termed ‘transcriptional gene silencing’ (TGS).

Other evidence suggested that antisense RNAs interfered with translation or directed the degradation of the target mRNA via dsRNA formation, ^{43 , 54 , 58 , 70} termed ‘post-transcriptional gene silencing’ (PTGS), without a supported idea of how ‘sense’ RNAs might result in the same outcome, or whether there may be more than one RNA-mediated pathway, although there was evidence that TGS and PTGS are mechanistically related. ^{71 , 72}

The finding that the silencing of specific genes was able to spread systemically in plants indicated the participation of a diffusible molecule, ^73–76 likely RNA itself, ^{47 , 54 , 64 , 77 , 78} with the prediction that there existed a “sequence-specific signaling mechanism in plants that may have roles in developmental control as well as in protection against transposons and viruses”. ⁷⁶

In 1999, Baulcombe and Andrew Hamilton, considering the lack of evidence for long antisense RNAs, hypothesized that low molecular weight RNAs might be involved but had escaped detection due to their small size. They showed that short antisense RNAs (estimated to be ∼25nt) are produced in transgenic or virus-infected plants undergoing PTGS when both antisense and target mRNAs (cellular or viral) are expressed. They proposed that these RNAs acted as guides for PTGS, stressing that small RNAs “are long enough to convey sequence specificity yet small enough to move through plasmodesmata”, and are thus able to act as “the systemic signal and specificity determinants of PTGS” (Figure 12.2). ⁷⁹

Figure 12.2

The identification of small sense and antisense RNAs in post-transcriptional gene silencing in plants. (Reproduced from Hamilton and Baulcombe with permission of the American Association for the Advancement of Science.) The left panel shows that transgenic (more...)

The RNA Interference Pathway

In 1998, Andrew Fire, Craig Mello and co-workers resolved the sense-antisense conundrum by showing (inadvertently, as part of an experimental control) that addition of sense and antisense RNAs together potently triggered systemic and heritable gene silencing in C. elegans, which “was at least two orders of magnitude more effective than either single strand alone”. ⁸¹ Although the mechanism of silencing was still unknown, these observations explained the RNAi phenomenon, suggesting that the previous silencing observed with ‘sense’ RNAs was due to contaminating dsRNAs formed during in vitro RNA synthesis and antisense transcription.

Independently, also in 1998, Peter Waterhouse and colleagues showed that gene silencing and transgene-mediated virus resistance in plants were induced and targeted by dsRNAs, ⁸² and Elisabetta Ullu and colleagues showed that dsRNA induced mRNA degradation in trypanosomes. ⁸³ It was soon also demonstrated that dsRNAs trigger gene silencing in Drosophila ^84–87 and mammalian cells. ^88–90

These findings were followed by genetic dissection of the biochemical pathways and proteins involved. ^{91 , 92} The first identified was an RNA-dependent RNA polymerase (RdRP) named qde-1 (for quelling defective mutant 1), which had been predicted to participate in RNA silencing (as, by definition, they would produce dsRNAs). ^{93 , 94} Although RdRPs had been detected in prokaryotes and eukaryotes decades before, mostly associated with RNA virus replication, RdRP activity had also been described in tissues of “apparently healthy plants” whose product seemed to be “double-helical RNA”, ⁹⁵ with other reports showing the existence of non-viral RdRPs. ^{96 , 97} In 1999, Carlo Cogoni and Giuseppe Macino showed that RdRP was involved in gene silencing, ⁹⁴ with others subsequently showing that homologs are present not only in fungi, but also in many plants and nematodes, ^d indicating a common pathway. ^101–104

Many other components of the ‘RNAi machinery’ were soon identified and shown to occur in plants, invertebrates and vertebrates. ^{92 , 105} To cut another long story short, work from the laboratories of David Bartel, Greg Hannon, Thomas Tuschl, Phil Zamore and others showed that long dsRNA is recognized by dsRNA-binding proteins (of which there are many versions) that interact with and activate an enzyme named Dicer ^e to cleave the bound RNA into short dsRNA fragments of ~20–23nt in length (which can be amplified by RdRP, hence making RNAi auto-catalytic), ^{91 , 101 , 104 , 107–109} as had been proposed previously, ¹¹⁰ explaining the powerful and relatively stable systemic and transgenerational silencing obtained with small initial amounts of RNAs.

The short dsRNA fragments, called siRNAs (small interfering RNAs), ⁸⁹ are then unwound and one of the two strands (the ‘guide strand’, determined by the strength of the hydrogen bond interaction at the 5′-ends of the RNA duplex, known as the ‘asymmetry rule’) is loaded into an effector complex, the ‘RNA-induced silencing complex’ (RISC). The guide strand pairs with a complementary sequence in an mRNA (or other target RNA molecule) and induces its cleavage by an ‘Argonaute’ ^f protein, the catalytic component of the RISC. ^{89 , 107–109 , 113}

There is a large family of Argonaute proteins ^{114 , 115} and a range of small RNA signaling pathways, especially in plants. ^116–118 In eukaryotes, Argonaute proteins are found both in the cytoplasm, where they are involved in target mRNA degradation, whereas others are nuclear. ¹¹⁹ Argonaute orthologs also occur in prokaryotes, ¹¹⁴ where they chop foreign plasmids or phage DNA (or transcribed sequences from foreign DNA) ^g into small (13–25 nt) guide sequences, which then target and cleave complementary DNA, ^120–124 indicating that RNA-guided targeting of genes for viral defense and gene regulation dates back to the dawn of biology. ^h This system is similar to but distinct from bacterial and archaeal CRISPR-Cas systems that also derive guides from foreign DNA (see below), although some CRISPR-Cas loci encode Argonaute proteins that associate with RNA guides, suggesting that the two mechanisms work synergistically in some organisms. ^{120 , 124}

Moreover, the identification of highly conserved protein complexes that mediate small regulatory RNA functions is consistent with principles found in RNA control of other processes, such as the actions of snRNAs and snoRNAs in eukaryotes (Chapter 8), and sRNAs in bacteria, where Hfq acts as a general cofactor for antisense RNAs that rely on short stretches of base pairing ^127–132 (Chapter 9).

Genetic studies showed that membrane-bound dsRNA-binding proteins are required for systemic transmission of RNAi. ¹³³ There are two such proteins in C. elegans, ^{134 , 135} and two orthologs in vertebrates, the latter of which transport internalized extracellular dsRNAs from endocytic compartments into the cytoplasm for immune activation. ^{136 , 137} However, while systemic RNAi transmission is well understood in plants, ¹³⁸ the extent of and mechanisms for exocrine RNA transport and signaling in mammals ⁱ are unclear; in part it appears to involve packaging of small RNAs (including microRNAs – see below) into exosomes and other circulating vesicles, ^141–144 reported to alter gene expression in immune and other cells, ^145–150 and to influence aging, ¹⁵¹ as well as tumor invasion and metastasis in cancer, ^152–154 although there are conflicting reports. ^{155 , 156}

Transcriptional Gene Silencing: RNA-Directed DNA Methylation

Consistent with a nuclear role, there is now considerable evidence that TGS (as distinct from PTGS) imposed by dsRNAs involves methylation of cognate sequences in the genome, ^{69 , 157–160} which, although at that time best documented in plants, led Wassenegger to predict in 2000 that “RNA-directed DNA methylation is involved in epigenetic gene regulation throughout eukaryotes”, including mammalian X-chromosome inactivation and parental imprinting. ¹⁶¹ This prediction was confirmed in 2004 when Manika Pal-Bhadra, Tatsuo Fukagawa and colleagues showed that RNAi is required for heterochromatin formation in Drosophila and vertebrate cells, ^{162 , 163} and Kevin Morris and colleagues showed that siRNAs can induce methylation and TGS of homologous loci in mammalian cells, ¹⁶⁴ since confirmed by many others ¹⁶⁵ and shown to also involve Argonaute proteins. ^166–168

The groups of Rob Martienssen, Shiv Grewal, Danesh Moazed, Robin Allshire, Caroline Dean and others showed that RNAi pathways also control many aspects of chromosome structure and dynamics. ^169–179 They demonstrated that transposon-dense heterochromatic regions are not transcriptionally inert, but express RNAs with key functions, many of which involve targeting of nascent transcripts and chromosome-associated RNAs by small RNA species. ^{180 , 181} These include regulation of centromere function, ^182–185 meiotic and mitotic chromosome segregation, ¹⁶⁹ repression of meiotic genes by targeting nearby retrotransposon-derived repeats, ¹⁸⁶ other aspects of meiotic progression including chromosome condensation, ¹⁶⁹ rectification of copy number imbalances, ¹⁶⁹ genome rearrangements ^{187 , 188} and chromosome reassembly, ^{189 , 190} plant flowering time ¹⁹¹ and DNA damage responses from fission yeasts to mammals, ^{169 , 192 , 193} with interconnections to transcription and splicing. ^{180 , 181} RNAi pathways are required for transposon silencing, thought to protect the genome against uncontrolled transposition of mobile elements, ^194–198 but have been adapted to regulate developmental processes. ^{186 , 199 , 200}

In plants, RNA-directed DNA methylation induces TGS through the production by specialized plant RNA polymerases of 24nt RNAs that are loaded into AGO4, ^201–203 which then recruits DNA methyltransferase and directs DNA methylation to regulate plant-pathogen interactions, stress responses and development. ^{159 , 204–207} Similar pathways operate in animals, although less is known about them. ^{167 , 169 , 196–198 , 205} The notable common feature is the use of embedded transposon-derived sequences as the targets of locus-specific regulation by RNAs, which may explain, in large part, the widespread presence of such sequences in genomes of higher organisms (Chapters 10 and 16).

The observations in C. elegans and other organisms showed that RNA signals also mediate transgenerational epigenetic inheritance, ^{81 , 208–211} requiring, inter alia, a nuclear Argonaute protein ²¹² and RNA modification ²¹³ (Chapter 17).

Research and Biotechnology Applications

The ability of natural or modified small siRNAs, delivered directly or via a vector that produces a ‘small/short hairpin RNA’ (shRNA), to inhibit gene expression specifically and potently in plants and animals has been widely adopted to investigate and modulate gene function in vivo and in vitro.

In mammals, siRNA was first used to ‘knock down’ protein-coding genes in oocytes and newly fertilized zygotes by Florence Wianny and Magdalena Zernicka-Goetz, where the phenotypes observed were similar to those observed with null mutants of the endogenous genes. ⁸⁸ siRNA-directed gene knockdown was later shown to be generally applicable to mammalian cells ⁸⁹ and applied to systematically target protein-coding genes in culture to identify those involved in various aspects of cell biology (e.g., ²¹⁴ ), viral replication (e.g., ²¹⁵ ) and drug responses or sensitivity (e.g., ²¹⁶ ).

The attraction of RNAi (or any nucleic acid-based system) for human therapy is substantial, as it flexibly employs a relatively common chemistry to inhibit the expression of target genes, notwithstanding the possibility of off-target effects, which are progressively being addressed. ²¹⁷ The longstanding problem is in vivo delivery, especially tissue specificity and transport across the blood-brain barrier to treat neurological disorders, ²¹⁸ as dsRNAs are not easily absorbed except by the liver. ²¹⁹ The delivery problem has been solved in part by utilizing viruses, notably AAV (adenovirus-associated virus, because of its low immunogenicity, many tissue-specific serotypes, and low rate of chromosomal integration), as well as other approaches and vehicles such as artificial exosomes, liposomes, nanoparticles and peptide-conjugates that have been engineered to deliver stably expressed shRNAs and small and large RNAs in vivo. ^220–225

There are less technical problems in plants, and shRNA-based transgenes have been widely used to engineer or to exogenously deliver, for instance, viral and fungal resistance in horticultural crops and ornamental plants. ^{226 , 227}

Viral vectors may also be used to deliver normal copies of defective genes, although their payload is limited by the amount of DNA that can be packaged in the virus, and there may be residual immunogenicity and genotoxicity problems. ^{218 , 228 , 229} Nonetheless, AAV vectors carrying replacement genes or antisense oligonucleotides to rescue or bypass splicing defects have been used successfully to treat different conditions, such as spinal muscular atrophy, ^{230 , 231} muscular dystrophy ²³² and retinal dystrophy. ²³³

However, the use of siRNA and shRNA in research and gene therapy is being rapidly supplanted by a far more efficient and precise genome manipulation system, based on RNA targeting of engineered nucleases, called CRISPR, discussed later in this chapter.

MicroRNAs

Closely related endogenous RNA regulatory pathways that regulate differentiation and development were waiting in the wings to be discovered. As described in Chapter 9, a 22nt small RNA, lin-4, had been described by Victor Ambros and colleagues in 1993 to control developmental timing in C. elegans. ²³⁴ In 2000, a second small (~21-nt) regulatory RNA species, let-7, was identified by Gary Ruvkun and colleagues, produced from a locus that, like lin-4, controls developmental timing by negatively regulating the lin-41 (protein-coding) gene through partial complementarity to target sequences in its 3′UTR. ²³⁵ The similarity to the lin-4 system led the authors to conclude that

“the mechanism by which they [lin-4 and let-7] regulate the expression of their target could be related” and that they “may constitute elements of a cascade of stage-specific regulatory RNAs that control the temporal sequence of events in C. elegans development”. ²³⁵

Shortly thereafter, Amy Pasquinelli, Ruvkun and colleagues found that the let-7 sequence, size, and ‘hairpin’ structure of its precursor are highly conserved in invertebrates and vertebrates, ²³⁶ with the latter having multiple paralogs. ²³⁷ The timing of let-7 expression during development and its target sites in the lin-41 mRNA are also conserved. ^{236 , 238} These “small temporal RNAs” ²³⁶ became known as ‘microRNAs’ (miRNAs).

The similarity of size of miRNAs and siRNAs (and the small RNAs described in plant PTGS), suggested a relationship between them, ²³⁶ confirmed when Alla Grishok and colleagues found that inactivation of Dicer and other components of the RNAi pathway impaired miRNA processing and activity. ²³⁹

After the realization of the possible existence of other small RNA “classes”, and despite “lingering skepticism” and wariness “of what we imagined was a vast detritus of partially degraded RNAs in cells”, ²⁴⁰ the development of strategies for large-scale isolation and sequencing of small RNAs with “probable regulatory roles”, ^{237 , 241 , 242} as well as bioinformatic searches using the available genome sequences, ²⁴³ led to the cloning and characterization of large numbers of small ~21nt RNAs from a variety of organisms, including mammals, ^{237 , 241 , 242 , 244–249} “deviants no longer”. ²⁵⁰

As Ruvkun observed at the time:

why have more of the miRNAs not been revealed by genetic analysis? …the number of genes in the tiny RNA world may turn out to be very large, numbering in the hundreds or even thousands in each genome. Tiny RNA genes may be the biological equivalent of dark matter — all around us but almost escaping detection, until first revealed by C. elegans genetics. ²⁵¹

Subsequent studies showed that miRNAs are also produced from processed introns and other non-coding transcripts, ^j including RNAs derived from repetitive sequences, that form an internally folded dsRNA stem-loop structure with an imperfectly paired double-stranded stem. ^{254 , 259–265} This structure is recognized and cleaved from precursor host RNAs by a ‘Microprocessor’ complex comprised of the RNaseIII endoribonuclease Drosha and its partner dsRNA-binding protein Pasha/DGCR8, ^266–270 or generated directly from the debranching of lariat intermediates formed during intron excision (‘mirtrons’) ^271–273 to yield a hairpin-shaped RNA of ~65–70 nucleotides in length. This ‘pre-miRNA’ is then cut by Dicer in the cytoplasm, loaded into the RISC complex, which (after ejection of the ‘passenger’ strand) targets cognate mRNAs for deadenylation of their polyA tails and degradation and/or translational inhibition by binding to complementary sequences primarily in 3′UTRs (Figure 12.3). ^{91 , 239 , 274–276}

Figure 12.3

Biogenesis and mechanism of action of the main classes of small regulatory RNA. (Reproduced from Jinek and Doudna with permission of Springer Nature.)

Thousands of tissue-specific miRNAs ^{244 , 277–280} – some of which are conserved over long evolutionary distances, ²⁸¹ whereas others are more lineage-restricted ^{247–249 , 282} – have been identified and shown to regulate a wide range of differentiation and cellular processes. ^{274 , 277 , 283 , 284} These include maintenance of pluripotency and stem cell states, ^{277 , 285} cell proliferation, ²⁸⁶ epithelial to mesenchymal transition, ²⁸⁷ brain morphogenesis, ²⁸⁸ hematopoietic differentiation, ²⁸⁹ neuronal asymmetry, ²⁹⁰ dendritic spine development, ²⁹¹ muscle development, ²⁹² programmed cell death, ²⁹³ innate immune responses, ²⁹⁴ metabolism, ²⁹⁵ leaf development ^{296 , 297} and flowering time, ²⁹⁸ among many others. ^k MicroRNAs also undergo developmentally regulated post-transcriptional modifications, ^{299 , 300} and exhibit altered expression in developmental disorders and cancers ^301–303 (dubbed ‘oncoMiRs’ ³⁰⁴ ).

A feature of miRNAs, which generally distinguishes them from siRNAs, is that the precursor dsRNA structure that is recognized by Drosha does not have perfect base complementarity, and either strand may be utilized. Importantly, both miRNAs and siRNAs can repress gene expression without activating the interferon pathway, ⁸⁹ which is triggered by longer dsRNAs. ³⁰⁵ MicroRNAs appear to have evolved before multicellularity, as they are also found in unicellular algae, ³⁰⁶ although their repertoire expanded greatly during metazoan, vertebrate and mammalian evolution. ^{307 , 308}

Curiously, it appears that an individual miRNA may recognize sites in many mRNAs, ^309–311 and that many, if not most, mRNAs, also contain bona fide recognition sites for multiple miRNAs. ^{312 , 313} The regulatory logic for this reciprocal multiplexing is unclear and, while miRNAs appear to locate their targets through an 8-nucleotide ‘seed’ sequence, ^{297 , 310} the rules for target recognition are still being teased out, ^314–316 as also is the place of miRNAs in decisional hierarchies (Chapter 15). ³¹⁷

Although miRNAs are thought to operate primarily on mRNAs in the cytoplasm, they also operate in the nucleus to target pre-mRNAs, ³¹⁸ as well as enhancer RNAs ³¹⁹ (to modulate chromatin architecture; Chapters 14 and 16) and nuclear RNAs such as 7SK. ³²⁰ They also bind to long non-coding RNAs, sometimes thought to function as miRNA ‘sponges’ or ‘decoys’, ^321–323 although they are likely legitimate and common targets in their own right. Some miRNAs also trigger the production of 21–22nt siRNAs from transposons, ³²⁴ referred to as ‘easiRNAs’ (epigenetically activated small interfering RNAs), to control genome dosage response and hybridization barriers in plants. ³²⁵ MicroRNAs also influence nuclear DNA methylation, ^{326 , 327} but the intersecting small RNA pathways and networks have yet to be understood.

Nonetheless, a sort of consensus has emerged. In general, siRNAs are perfectly complementary to their targets and promote their degradation, and in plants are used for defense against viruses and transposons, as well as adaptive responses to environmental stress. ³²⁸ siRNAs are also produced in animals - in oocytes, embryonic and somatic cells, from pseudogenes ^l and TEs and other naturally formed dsRNAs ^{331 , 332} - where they regulate transposons, genome dosage response and developmental processes, likely interrelated. ^{94 , 195 , 198 , 265 , 329 , 330 , 333–337} On the other hand, miRNAs are imperfectly complementary to mRNA (usually) 3′UTR sequences and often direct translation inhibition, as well as less well characterized targets, for the regulation of developmental processes as described above.

Piwi-Associated RNAs

In 2006, a related general class of small RNAs was described, ^338–342 although they had been detected a few years earlier by Alexei Aravin and colleagues in analyses of tandem repeat and retrotransposon silencing ³³³ and small RNA sequencing datasets in the testis. ³⁴³ A subgroup of Argonaute proteins, known as the ‘Piwi family’, ^m was known to be required for germ- and stem-cell development in animals, three homologs (Mili, Miwi and Miwi2) being essential for spermatogenesis in mammals ^{345 , 346} (Figure 12.4). These proteins bind ‘Piwi-interacting RNAs’, or piRNAs, which are longer than miRNAs, generally ranging from 25 to 33nt, and have a characteristic 2′O-methylation ⁿ of the 3′ terminal ribose, ³⁴⁸ which is recognized by the PAZ domain of Piwi proteins, ^{349 , 350} with those interacting with different Miwi/Mili proteins having a different mean length. ^{346 , 351 , 352}

Figure 12.4

Biphasic production of piRNAs and Miwi expression during mouse sperm development, (Adapted from Zheng and Wang, under Creative Commons Attribution License.)

Thousands of genomic loci have been found to produce piRNAs. ^o These RNAs are present in millions of copies in testes and are so abundant that, like the discovery of small nuclear RNAs in the 1960s, they were discovered by visualization in gel electrophoretic displays. PiRNAs are also expressed in ovaries ^353–358 and somatic cells, ^{354 , 359} where they regulate gene expression in stem cells and embryonic development, ^{357 , 358} and in the brain, where they are required for neurogenesis and memory formation ^360–363 (Chapter 17). piRNA expression is also dysregulated in cancer. ³⁶⁴ They are not produced by Drosha processing of dsRNA intermediates but the best understood mechanism for piRNAs derived from repeat sequences and present in diverse organisms is through a ‘ping-pong’ amplification system from single-stranded RNA templates. ^98–100

piRNAs are thought to be (mainly) involved in the repression of transposons, ^p likely their ancestral function. ^{196 , 346 , 352 , 367–371} They are also involved in parental imprinting. ³⁷² However, their range is much greater. For instance, male-specific chromosome Y-derived long non-coding transcripts, named Pirmy and Pirmy-like RNAs, which exhibit a large number of splice variants in testis, act as templates for piRNAs that regulate autosomal genes. ³⁷³ piRNAs are also derived from other sequences, including intergenic transcripts, pseudogenes and the 3′UTRs of mRNAs, ^{112 , 374 , 375} the latter targeting the introns of other genes. ³⁷⁶ Along with their roles in development and brain function, these observations lead to the conclusion that genomes have co‐opted the TE‐derived piRNA system for regulating transposon activity and other dimensions of genome regulation. ^375–379 These include the modulation of nuclear architecture at target loci containing TEs, shown in a Drosophila ovarian somatic cell line and involving directing these genomic regions to the nuclear periphery, heterochromatin formation and chromatin conformational changes to promote transcriptional silencing; thus providing a possible mechanism for the proposed role of TEs as “functional motifs” for the dynamic regulation of chromatin state and nuclear architecture ^{380 , 381} (see Chapters 14 and 16).

In sperm development in Drosophila and mammals, there are two bursts and classes of piRNAs and two corresponding rounds of Piwi protein expression. The first class is expressed before the meiotic ‘pachytene’ stage of spermatogenesis, is derived from transposon‐rich clusters, and associates with Mili and Miwi2. The second class is expressed during the pachytene stage from thousands of genomic loci, is extremely abundant and associates with Mili and Miwi1. ^382–384 Their function is unknown, ³⁸⁵ although an intriguing possibility is that this dual system evolved to enhance evolvability (Chapter 18), with quality control, when progeny numbers are small. ³⁸⁶ There is widespread formation of dsRNAs in the testis, different in profile from that observed in somatic cells. ³⁸⁷ Moreover, the loss of fertility in Piwi mutants lacking piRNAs is not due to de-repression of repetitive elements: pachytene piRNAs repress gene expression by directing the cleavage of meiotic transcripts required for sperm function. ^{388 , 389}

It may also be that the phenotypes of the loss of piRNA function are mediated by epigenetic silencing of histone genes, ³⁹⁰ implicating small RNAs in the transmission of epigenetic information across generations (Chapter 17). With many more piRNA loci yet to be studied, there is much to be learned about this large class of small RNAs in animals.

Other Classes of Small RNAs

The landscape is even more complicated, as most RNAs form double-stranded structures that can be and frequently are processed into small RNAs.

A variety of small RNA species are produced from rRNAs (‘rRFs’) in a developmental stage-specific manner in fungi, plants and animals. ^391–393 Most derive from 28S rRNA, and have various sizes, some shown to be siRNAs (‘phasiRNAs’), miRNAs and piRNAs, as well as ‘qiRNAs’ ³⁹⁴ that function in DNA damage response. ^q The 18S rRNA is also the source of smaller RNAs, including miRNAs, as well as a longer species of 130nt that may be an miRNA precursor. miRNAs are also processed from the 5.8S rRNA and the 45S rRNA precursors. ³⁹¹ The piRNA pathway is required to control the abundance of rRNA-derived siRNAs and the ribosomal RNA pool in C. elegans. ³⁹⁸

In all branches of life, tRNAs are processed into highly abundant smaller species (‘tRFs’ for tRNA-derived fragments, or ‘tsRNAs’ for tRNA-derived small RNAs), ^399–403 initially thought to be degradation products, ⁴⁰⁴ despite their production being cell-state and tissue-specific ^{399 , 405 , 406} and (at least in some cases) Dicer-dependent, ^{400 , 407} possibly playing a role in the global regulation of RNA silencing. ⁴⁰⁸ Some tRNA fragments function as miRNAs. ^409–411 The production of tRFs is influenced by methylation of the tRNA by the RNA methyltransferases Dnmt2 ⁴¹² and Nsun2, which may be involved in transgenerational inheritance (see below) and loss of which causes neurodevelopmental disorders. ⁴¹³ tRNAs and tRNA fragments are also modified and regulated by TET2, which oxidizes 5-methylcytosine in DNA ⁴¹⁴ (Chapter 14). tRFs and modifications thereof are stress-induced and inhibit translation, ^415–419 control retrotransposons, ^{420 , 421} promote cell proliferation, ⁴⁰¹ and are depleted in immortalized cell lines. ⁴⁰⁶

tRFs also convey long distance regulatory signals. Their levels are increased in Arabidopsis roots upon phosphate starvation, ⁴²² and rhizobial (symbiotic bacterial) tRFs modulate host nodulation in legumes via the RNAi pathway. ⁴²³ They are present, along with miRNAs, in secreted exosomes ¹⁴⁴ from stem cells ⁴²⁴ and activated T-cells. ⁴²⁵ They are also abundant in semen and sperm, ^{420 , 426 , 427} and have been found to contribute to intergenerational inheritance of acquired metabolic disorders, with evidence that this involves their methylation ^428–430 (see Chapter 17). tRFs have also been shown to affect the biogenesis of snoRNAs, scaRNAs and snRNAs, as well as histone levels and chromatin organization ⁴³¹ (Chapter 14).

tRNA-like small RNAs are also produced from the processing of long non-protein-coding RNAs, such as MALAT1, ^{432 , 433} one of the most highly expressed genes in the vertebrate genome, which also produces piRNAs. ⁴³⁴

All known snoRNAs (in fungi, plants and animals, from yeast to human) are processed to produce smaller species: those derived from C/D snoRNAs show a bimodal size distribution of 17–19nt and >27nt, the latter similar in size to piRNAs. ⁴³⁶ Those derived from H/ACA snoRNAs are predominantly 20–24nt in length ⁴³⁶ (Figure 12.5), at least some of which have been shown to be Dicer-dependent, bind to Argonaute, function as miRNAs and have developmental effects. ^{265 , 435–439}

Figure 12.5

Processing of an miRNA from snoRNA ACA45. The blue bar shows the sequences of the putative miRNA products and the number of reads in Argonaute-associated small RNA sequencing data, while the red bar shows putative ‘star' product from the stem-loop (more...)

Likewise, spliceosomal snRNAs are processed into smaller fragments, ⁴⁰⁹ although their significance is unknown. Vault RNAs (Chapter 8) are processed as well, dependent on cytosine methylation, into small RNAs that bind to Argonaute proteins and exhibit miRNA-like functions. ⁴⁴⁰ Other short (80–100nt) intronically derived RNAs, some of which are highly conserved, also associate with Argonaute proteins and are capable of repressing mRNAs via target sequences in the 3′UTRs. ⁴⁴¹ All of these observations indicate deep evolutionary and complex functional connections between regulatory RNA pathways.

Finally, another class of small RNAs with a modal length of 17–18nt is associated with transcription start sites (‘tiRNAs’) and exonic splice sites (‘spliRNAs’) in animals, but not in plants, with evidence that they are associated with binding sites for epigenetic regulators and mark nucleosome positions, ^442–444 which may be subject to fine control in animals because of the required precision of ontogeny (Chapter 15).

RNA Communication between Species

Some of the early experimental RNAi methods in C. elegans indicated the potential of interspecies communication by small RNAs: “dsRNAs expressed in Escherichia coli could induce gene silencing in nematode larvae that feed on them … raising the possibility that dsRNAs may be transferred from prokaryotic pathogens to their hosts”. ⁴⁴⁵

Small RNAs have subsequently been shown to mediate communication between species, for pathogenicity, defense or symbiosis. For example, the small RNA SsrA produced by the bacterium Vibrio fischeri is loaded into outer membrane vesicles that are transported into the epithelial cells in the light organ of the squid and suppress antimicrobial responses and permit light-organ symbiosis between the bacterium and its host. ⁴⁴⁶

The soil bacterium Pseudomonas aeruginosa, which is pathogenic in many species, releases a small RNA termed P11 into C. elegans that is processed by the RNAi and piRNA machinery, and specifically silences the expression of the protein maco-1, which functions in chemotaxis and neuronal excitability, leading to avoidance behavior that is heritable for four generations, of value to both. ⁴⁴⁷ C. elegans transmits this memory of learned avoidance to naive animals and four generations of progeny through virus-like particles encoded by the Cer1 retrotransposon. ⁴⁴⁸

Similarly, the bacterial pathogen Listeria monocytogenes secretes a small RNA-binding protein named Zea that binds a subset of L. monocytogenes RNAs and RIG-I, the non-self-RNA innate immunity sensor in mammalian cells. In mice, this was shown to dampen the immune response and suggested to allow sustainable host-bacterium symbiosis. ⁴⁴⁹ Zea orthologs occur in other bacteria that are rarely associated with disease, suggesting that this may be a general mechanism to assist co-existence between bacteria and multicellular hosts, ⁴⁴⁹ especially in view of the growing awareness of the importance of the enteric microbiome in ruminant biology and mammalian physiology.

These may occur between eukaryotic kingdoms, as shown between the mycorrhizal fungus Pisolithus microcarpus and its host Eucalyptus grandis, in which the fungus miRNA Pmic_miR-8 is transported into the plant roots, targeting host genes to promote symbiosis. ⁴⁵⁰ Such interactions may influence pathogenicity and aggressiveness of parasites, as exemplified by the interaction between the fungal pathogen Botrytis cinerea and host plants, proposed to involve retrotransposon-derived small RNAs derived from the fungus and delivered into plant cells and cross-kingdom RNA interference (ckRNAi) to undermine host defense. ⁴⁵¹

Even complex ecological interactions may be affected by regulatory RNAs. This is suggested by the reported effect of short satellite RNAs (satRNAs, a type of subviral agent associated with some viruses, Chapter 8) that accompanies cucumber mosaic virus (CMV) and are processed into small RNA (sRNA) species in infected tobacco leaves. The production of these Y-sat sRNAs affects host gene expression and leads to a change of the color of the leaves from green to bright yellow, which preferentially attracts aphid vectors. Surprisingly, the Y-sat RNAs are also processed in the aphids feeding on the leaves, leading to the production of small RNAs that affect aphid physiology, turning them red and subsequently promoting wing formation. This in turn facilitates CMV and Y-sat RNA spread without apparent detrimental effects on the plant, thus suggesting a potential intricate ‘survival strategy’ by a subviral non-coding RNA, which depends on interaction with (and affects the interactions between) the CMV helper virus, the host plant and the insect vector. ⁴⁵²

It is early days in the exploration and understanding of the extent of communication by “social RNA” (a term coined by Eric Miska) ^{453 , 454} in the competition and cooperation within and between species in complex environments, but there are further documented examples of RNA exchange, including between parasitic nematodes and mammalian cells, ⁴⁵⁵ plants and fungal pathogens, ^{456 , 457} parasitic and host plants, ⁴⁵⁸ and even between queen and worker bees via royal and worker jellies, ^{459 , 460} a pathway which has been exploited to artificially introduce virus resistance to commercial hives. ⁴⁶¹

CRISPR

The exponential growth and analysis of DNA and RNA sequence data transformed molecular biology, no better demonstration being the serendipitous discovery ^r of the CRISPR RNA-based immune system in prokaryotes. ^466–470

In 1987, Yoshizumi Ishino and colleagues reported the existence of mysterious, partly palindromic ~30nt tandemly repeated sequences, separated by ~30 nt ‘spacers’ in E. coli. ⁴⁷¹ A similar peculiar array of interspersed repeats was described in the salt-tolerant archaeon Haloferax mediterranei by Francisco Mojica and colleagues in 1993, who went on to collate instances of such sequences in many different species, which they called ‘Short Regularly Spaced Repeats’. ^{472 , 473} Others doing related bioinformatic analyses showed that a common set of protein-coding ‘cas’ ^s genes lay adjacent to these repeats, which they termed ‘Clustered Regularly Interspaced Short Palindromic Repeats’, ⁴⁷⁴ whose acronym, CRISPR, being more euphonic, became adopted.

The function of CRISPR was revealed in 2005, when three groups, including Mojica’s, ^t discovered that the spacers separating the repeats corresponded to fragments of bacteriophages and plasmids, indicating that these sequences recorded past invasions and constituted a memory-defense system, ^477–479 reinforced by the bioinformatically inferred nuclease function of the CRISPR-associated cas genes and analogies with RNAi. ^{479 , 480}

This conclusion was confirmed when Philippe Horvath, Rodolphe Barrangou, Sylvain Moineau and colleagues working at the French company Danisco ^u showed in 2007 that recently selected phage-resistant mutants of the lactic acid-producing bacterium Streptococcus thermophilus, used in yogurt and cheese production, had acquired phage-derived sequences at their CRISPR loci and that phage isolates that overcame the immunity had single nucleotide changes that altered the sequences corresponding to the spacers. ⁴⁸¹ These and other investigators, notably Luciano Marraffini, John van der Oost and colleagues, also showed that the CRISPR array is transcribed and cleaved within the repeats to produce 61 nt ‘crRNAs’ wherein the spacer (target) sequence is flanked by the palindromic sequences of the repeats. The crRNAs are then loaded into a cas-encoded nuclease, ^481–484 which uses the spacer guide sequence to target and introduce a double strand break in cognate DNA via two distinct nuclease domains. ^v The Cas proteins are essentially RNA-programmable restriction enzymes ^{487 , 488} (Chapter 6), some of which are also used for transposon homing. ^{489 , 490}

A vital missing link was discovered when Emmanuelle Charpentier and colleagues, who had earlier shown that streptococcal virulence factors are controlled by regulatory RNAs, ⁴⁹¹ found that the third most highly expressed sequence (after rRNA and tRNA) in Streptococcus pyogenes is a small RNA that is transcribed from a sequence adjacent to the CRISPR locus and has 25 bases of near-perfect complementary to the repeats. This RNA, termed ‘tracrRNA’ (trans-activating CRISPR RNA), hybridizes to the transcribed CRISPR array to guide cleavage by host-encoded RNaseIII, ⁴⁹² is loaded into the cas nuclease along with a crRNA, and is essential for its function ⁴⁸⁵ (Figure 12.6). An endogenous guide RNA autoregulates CRISPR-Cas expression to mitigate autoimmune toxicity. ⁴⁹³

Figure 12.6

The type II CRISPR-Cas9 System from Streptococcus thermophilus. (Reproduced from Lander with permission of Elsevier.) Like RNAi, CRISPR-Cas systems rely on RNA guidance for target specificity. (a) The locus contains a CRISPR array, four protein-coding (more...)

There are a number of variants of CRISPR systems in bacterial and archaeal genomes. ^{496 , 497} Type I and type III systems employ a large multi-Cas protein complex for crRNA binding and target cleavage. ^{466 , 467} The simplest, Type II, consists of the CRISPR array, the gene encoding the tracrRNA, and four protein-coding cas genes, three of which are involved in generating new spacers and repeats and the other, cas9, encodes the targeting nuclease. ⁴⁸⁶

Virginijus Siksnys, Horvath, Barrangou and colleagues showed that the CRISPR system from S. thermophilus could be functionally transferred into E. coli and that all that was required to target and cleave a specific DNA sequence is the tracrRNA, the Cas9 nuclease and the corresponding crRNA, ⁴⁹⁸ which could be as short as 20 nt as the flanking repeat sequences are not required, paving “the way for engineering of universal programmable RNA-guided DNA endonucleases”. ⁴⁷⁵

Shortly thereafter, Charpentier, Martin Jinek, Jennifer Doudna and colleagues showed that the normally base-paired cRNA and tracrRNAs could be covalently linked and produced as a single chimeric guide RNA (sgRNA), simplifying delivery of the system. ^{485 , 486}

CRISPR systems have also been coopted naturally for various structural and regulatory functions by repurposing of diverged repeats encoded outside of CRISPR arrays, including by transposons for RNA-guided transposition, by plasmids for interplasmid competition, and by viruses for antidefense and interviral conflicts. There are also multiple highly derived CRISPR variants of yet unknown functions. ⁴⁹⁹

RNA-Directed Genome Editing

In 2013, the groups of Feng Zhang ⁵⁰⁰ and George Church ⁵⁰¹ demonstrated that the CRISPR system (Cas9 plus sgRNAs) could be used in human and mouse cells to introduce sequence-specific double strand breaks, ^w which, when repaired in vivo, lead to incorporation of guided mutations, the latest in a long line of ‘homing endonucleases’ with wide practical applications. ^{504 , 505} Zhang and colleagues showed that they could engineer the Cas9 sequence to disable one of its two nuclease domains, converting it into a ‘nickase’ enzyme that only introduces a single-strand break, facilitating the insertion of new sequences into genomes by homology-directed repair. They also showed that multiple guide sequences can be encoded into a single CRISPR array to enable the simultaneous alteration of several sites. ⁵⁰⁰

CRISPR technology has proven far more versatile and efficient than the previous methods developed for targeted genome modification, which employed nucleases linked to the DNA-binding domains of transcription factors whose sequence specificity could be engineered by altering amino acids in these domains (called ZFNs ^x and TALENs), an effective but relatively cumbersome methodology. ^506–508

Moreover, CRISPR technology has been adapted by engineering the Cas9 and other Cas nucleases to inactivate, manipulate, image and annotate specific DNA and RNA sequences in a wide variety of organisms, ^{495 , 509–514} including high-throughput mutagenesis to identify the roles of new genes and genetic pathways in cell and developmental biology. ⁵¹⁵ CRISPR has also been widely used in industrial biotechnology ^{516 , 517} and in agriculture to construct transgenic plants and animals with desirable properties. ^{518 , 519}

The toolkit and applications have been advancing at breathtaking pace. ⁵²⁰ For example, nuclease-inactive and RNA-targeting Cas proteins have been fused to a plethora of different effector proteins to alter gene expression, epigenetic modifications and chromatin interactions. ^{495 , 521–526} These include the fusion of transcriptional repressors or activators to catalytically inactive Cas9 to repress (‘CRISPRi’) or enhance (‘CRISPRa’) expression of specific protein-coding and non-coding genes (Figure 12.7). ^527–531

Figure 12.7

CRISPR applications. Top panel, genome editing; bottom panel, RNA-targeting, tagging and editing. (Reproduced from Pickar-Oliver and Gersbach with permission of Springer Nature.)

They also include the fusion of cytidine deaminase or reverse transcriptase to Cas9 by to enable single base replacement or the insertion of any desired sequence (programmed into the guide RNA) at any specific position in the genome, termed ‘base editing’ and ‘prime editing’, respectively. These have been demonstrated to reverse damaging mutations at high efficiency in cell culture ^{494 , 532–535} and (to a limited extent, but with increasing success) in vivo including amelioration of deafness and amyloidosis, ^536–538 which “in principle could correct up to 89% of known genetic variants associated with human diseases”. ⁵³⁴ CRISPR systems have also been adapted to edit RNA with high specificity ⁵³⁹ to correct a range of genetic defects, ^540–543 and to develop multiplexed assays for high sensitivity detection of viral RNAs. ^544–546

CRISPR is also being deployed to propagate the ingression of desirable genes into natural populations, first mooted by CF Curtis in 1968 (using translocations) ^{547 , 548} and again by Austin Burt in 2003 (using “site-specific selfish genes” such as homing endonucleases, group II introns and some types of transposable elements). ⁵⁰⁵ To date, effective introduction of ‘gene-drive’ systems based on CRISPR-Cas9 has been achieved in yeast, ^{549 , 550} fruitfly ⁵⁵¹ and mosquito, ^552–554 among others, with applications including eliminating vector-borne and parasitic diseases, promoting sustainable agriculture and curtailing invasive species, albeit with biosafety and ‘ethical’ concerns, ^{555 , 556} which in turn have motivated the design of tunable and reversible systems. ^557–560

Advanced nanoparticle delivery systems are being developed in parallel to enable high-efficiency tissue-specific delivery in vivo for gene therapy, vaccination, cancer treatment and investigative genetic manipulation. ^561–563 The efficiency and specificity of genome editing continues to improve, with constant factor being the flexibility of RNA guides to unlock the potential of targeting generic effector proteins to specific DNA or RNA locations.

A billion years ago, RNA guides also solved the challenges of regulating cell fate decisions and tissue architecture in multicellular organisms (Chapters 15 and 16).

Further Reading

1. Abbott A. (2016) The quiet revolutionary: How the co-discovery of CRISPR explosively changed Emmanuelle Charpentier’s life. Nature 532: 432–434. [PubMed: 27121823]
2. Castel S.E. and Martienssen R.A. (2013) RNA interference in the nucleus: Roles for small RNAs in transcription, epigenetics and beyond. Nature Reviews Genetics 14: 100–112. [PMC free article: PMC4205957] [PubMed: 23329111]
3. Gutbrod M.J. and Martienssen R.A. (2020) Conserved chromosomal functions of RNA interference. Nature Reviews Genetics 21: 311–331. [PMC free article: PMC9478574] [PubMed: 32051563]
4. Lander E.S. (2016) The heroes of CRISPR. Cell 164: 18–28. [PubMed: 26771483]
5. Ledford H. (2016) The unsung heroes of CRISPR. Nature 535: 342–344. [PubMed: 27443723]
6. Maraganore J. (2022) Reflections on Alnylam. Nature Biotechnology 40: 641–650. [PubMed: 35534556]
7. Özata D.M., Gainetdinov I., Zoch A., O’Carroll D. and Zamore P.D. (2019) PIWI-interacting RNAs: Small RNAs with big functions. Nature Reviews Genetics 20: 89–108. [PubMed: 30446728]
8. Pasquinelli A.E. (2015) MicroRNAs: Heralds of the noncoding RNA revolution. RNA 21: 709–710. [PMC free article: PMC4371344] [PubMed: 25780202]
9. Veigl S.J. (2021) Small RNA research and the scientific repertoire: A tale about biochemistry and genetics, crops and worms, development and disease. History and Philosophy of the Life Sciences 43: 30. [PubMed: 33624250]