A Guide to Guides: An Overview of SpCas9 sgRNA Scaffold Variants and Modifications

Abstract

The CRISPR/SpCas9 system has revolutionized biology by enabling precise and programmable genome modification. While substantial effort has focused on engineering the SpCas9 protein and spacer sequences, the single-guide RNA (sgRNA) scaffold is an equally critical determinant of activity. Since the canonical scaffold was introduced in 2012, numerous variants have been developed. Early designs sought to enhance editing efficiency; however, despite the first improved scaffold being reported in 2013, more than 80% of CRISPR plasmids deposited in the Addgene repository still use the original scaffold rather than an efficiency-optimized alternative, which may not provide optimal performance. Subsequent work has also addressed intra-sgRNA interactions that impair folding, as well as inter-sgRNA interactions that destabilize multiplexed arrays, yet these solutions remain largely overlooked. Beyond efficiency, scaffold engineering—and the inclusion of auxiliary RNA elements—has enabled new capabilities, including effector recruitment, conditional regulation, visualization, improved stability, and large-scale multiplexing. The main goal of this review is to (i) provide a structured overview of the diverse SpCas9 sgRNA scaffold variants and auxiliary RNA modifications developed to date, (ii) summarize their functional characteristics and contexts of use, thereby illustrating how scaffold engineering continues to expand the functional scope of CRISPR technologies, and (iii) present a curated sequence resource comprising more than 230 scaffold variants and 80 auxiliary modifications to support experimental design and benchmarking.

1. Introduction

Since its landmark demonstration for targeted genome modification in eukaryotes by several research groups in 2013 [1,2,3,4,5], CRISPR technology has profoundly reshaped biological research. One key reason for the rapid adoption of SpCas9 (Streptococcus pyogenes Cas9) is the simplicity of the synthetic system, which requires only two components: the Cas9 protein and an RNA molecule. In the native system, the RNA component consists of two separate molecules—the trans-activating CRISPR RNA (tracrRNA) and the CRISPR RNA (crRNA)—that hybridize to function together. In contrast, most researchers employ a single-guide RNA (sgRNA) design. Strikingly, this chimeric RNA molecule was already described in the very first paper demonstrating SpCas9 activity in vitro by the Doudna lab [6]. The history of CRISPR research has been reviewed extensively elsewhere, including by Gostimskaya [7], and will not be covered in detail here.

Since 2012, CRISPR technology has evolved far beyond its original use with SpCas9. The CRISPR toolbox now includes a wide range of enzymes from diverse families [8]. Its applications have expanded from simple gene knockouts to advanced modalities such as base editing, prime editing, epigenetic editing, RNA editing, as well as CRISPR interference (CRISPRi) and CRISPR activation (CRISPRa) [9]. In addition to genome and transcriptome modifications in laboratory contexts, CRISPR has also been adapted for other applications, including chromosome imaging [10], molecular recording [11], biocomputation [12], diagnostics [13], and therapeutics [14].

Several scaffold variants have been designed to improve editing efficiency, of which a few have seen practical use, namely the Flip + Extension (F + E), optimized sgRNA, and cr772 scaffolds. For further details on these and other designs, see Section 3.1. Despite these advances, an analysis of mammalian CRISPR plasmids deposited in the Addgene repository indicates that the original scaffold, first described by Jinek and colleagues [6], remains by far the most widely used (Figure 1A,B; Supplementary Tables S1–S6). A broader analysis of all plasmids in the Addgene repository, regardless of host species, yielded similar results: among 10,774 plasmids containing at least one sgRNA cassette with either the original scaffold or one of the 11 reported efficiency-enhanced variants, 84.07% include the original scaffold, followed by F + E at 13.45%, the optimized scaffold at 2.40%, and cr772 at 0.06%. To avoid bias from plasmids that contain multiple sgRNAs, a per-plasmid classification was also carried out, where a plasmid was considered to use an enhanced scaffold if any of its sgRNA cassettes contained one. Even under this permissive criterion, 83.87% of plasmids still relied exclusively on the original scaffold (Supplementary Table S7). However, this scaffold is not necessarily the optimal choice, as multiple other variants have been shown to improve editing efficiency.

Although gRNAs from other CRISPR systems, such as Cas12f [15,16,17] and Cas12j [18,19] enzymes, have been extensively engineered to enable efficient genome editing, the scope of this review is limited to SpCas9-associated sgRNAs. The focus is specifically on scaffold designs that can be synthesized in vivo using only canonical RNA bases. Readers interested in chemically modified sgRNAs or sgRNAs containing universal bases are referred to prior literature on this topic [20,21,22,23,24]. Most published work has focused on complete sgRNA scaffolds rather than on sequence modifications of the individual RNA components (tracrRNA and crRNA). A notable exception is the study by Scott and colleagues (2019) [25], which investigated tracrRNA sequence variants by substituting uridines with other nucleotides in an RNP (ribonucleoprotein) delivery format. Another example is the Genome-editing Optimized Locked Design (GOLD)-gRNA, which combined sequence alterations with chemical modifications and also introduced the sgRNA scaffold variant T-lock [22], discussed in Section 3.1.

The objective of this review is threefold: (i) to provide a structured overview of the diverse SpCas9 sgRNA scaffold variants and auxiliary RNA modifications developed to date, (ii) to summarize their functional characteristics and contexts of use, and (iii) to offer a curated sequence resource to support experimental design and benchmarking. The methodology used to compile and classify scaffold variants and modifications is described in Supplementary Information S2.

In the following sections, the structural organization of the SpCas9 sgRNA is described first (Section 2), followed by an overview of alternative scaffold variants (Section 3) that have been developed primarily to enhance editing efficiency and reduce homology-related difficulties. Section 4 presents additional sgRNA scaffold modifications that incorporate auxiliary RNA elements for expanded applications, including multiplexing, effector recruitment, and conditional control.

2. Structure of an sgRNA

The original (canonical) sgRNA design introduced by Jinek and colleagues is approximately 100 nucleotides in length, composed of the 20-nt spacer, the scaffold, and the uridine-tail (U-tail) at the 3′ end [6]. Figure 2 illustrates this structure and its key elements.

At the 5′ end lies the 20-nt spacer, which determines target specificity. Although the vast majority of studies employ 20-nt spacers—generally the most efficient configuration—spacers of 18–21 nt can outperform the canonical length at certain loci [29]. Because sgRNAs are commonly transcribed from polymerase III (Pol III) promoters, which require a guanine (U6) or adenine (U3) as the first transcribed base, many guides carry an additional G or A at their 5′ end. When this extra base does not correspond to the native first nucleotide of the spacer, the resulting guide becomes G/A + 20 nt in length, effectively yielding a 21-nt spacer. Beyond the canonical design, several spacer variants have been developed. Truncated gRNAs (tru-gRNAs) of 17–18 nt improve specificity by increasing sensitivity to mismatches, though spacers of 14–16 nt typically lose cleavage activity [30], but can still be directed to specific targets at the genome level [31]. At the opposite extreme, self-targeting sgRNAs (stgRNAs) have been engineered for cellular recording, carrying spacers as long as 70 nt [32]. While the focus of this paper is on scaffold modifications, it should be emphasized that the spacer is of paramount importance in CRISPR experimental design. Beyond machine learning approaches for optimal spacer selection [33,34,35], several spacer-engineering strategies have been explored: safeguard gRNAs introduce cytosine stretches at the 5′ end to modulate Cas9 activity [36]; extended gRNAs (x-gRNAs) incorporate additional spacer-dependent nucleotides to improve specificity [37]; and intentional mismatches within the spacer can be used to titrate activity [38].

Directly adjacent to the spacer lies the scaffold (positions 21–96, full transcript numbering), which in the canonical design is 76 nucleotides long. The first element is the repeat–anti-repeat (RAR) duplex, composed of a lower stem (also called the proximal duplex), a bulge, and an upper stem (also called the distal duplex) [39,40]. At the apex of the upper stem is a tetraloop of four unpaired nucleotides. The RAR duplex is followed by stem-loop 1 (SL1), a structural element that has also been referred to as the “nexus.” In these descriptions, however, the nexus is defined more broadly to include an additional nucleotide at position 63 of the full transcript [40]. A short linker then connects to two additional stem-loops, SL2 and SL3. It should be noted that the nomenclature differs across publications: in some, SL1 is referred to as the nexus, and SL2 and SL3 are renamed SL1 and SL2, respectively. In addition, the terms “stem-loop” and “hairpin” are often used interchangeably in this context [39,40]. In this review, the delineation of the elements follows the classification introduced by Nishimasu and colleagues [39].

At the 3′ end, sgRNAs transcribed from RNA polymerase III (Pol III) promoters naturally acquire a short U-tail due to termination at a stretch of thymidines in the DNA template, typically around seven Ts, generating a variably sized U-tail (most often 3–5 uridines) [41]. Instead of using a simple T-stretch, some vector designs incorporate the Saccharomyces cerevisiae SUP4 terminator, which features a (T)₇G(T)₆ motif derived from the tRNA^Tyr locus [42]. This terminator has been adopted not only in yeast vectors, but also in some mammalian [43] and plant plasmids [44]. The U-tail is not required for sgRNA function and can be removed in multiplexing systems (see Section 4.2) without any negative impact on activity. One study also used a Pol II promoter–Pol II terminator configuration without any processing elements to release the sgRNA, resulting in transcripts with long 3′ extensions and a total length of 259 nt (excluding the poly(A) tail), which were nevertheless functional in rice [45].

3. Alternative Scaffolds

In this section, several alternative scaffolds that have been developed will be discussed. Multiple studies have shown that the sgRNA scaffold can tolerate numerous mutations without loss of activity [39,40], enabling the development of a wide range of functional scaffold variants. In a high-throughput experiment, it was also shown that the repeat–anti-repeat (RAR) duplex is highly permissive for modifications, as ~1.2 × 10⁶ sgRNA variants with 25-nt random internal cassettes displayed no apparent sequence constraints on sgRNA–dCas9 complex formation [46]. In addition, large insertions of up to 324 nucleotides in internal stem-loops have been reported for dCas9 applications [46,47]. This permissiveness to sequence modifications has enabled the development of a wide range of functional scaffold variants. Currently, all improved-efficiency scaffolds are derived from the canonical scaffold through base substitutions and/or short stem-loop extensions (≤10 nt). To minimize sequence homology among multiple sgRNA units in multiplex constructs (see Section 3.2.2), researchers have also developed divergent scaffold variants that substantially deviate from the canonical design. Figure 3 summarizes the historical development of improved-efficiency scaffold variants and illustrates two strategies used to address sequence-homology issues occurring both within an sgRNA and between multiple sgRNAs.

Another class of scaffold variants incorporates well-defined RNA modules into the canonical scaffold, rather than relying solely on sequence modifications of its native structure. These modules are typically designed for specific functions, for instance, to stabilize RNA folding, recruit effector proteins, or enable multiplex processing. While such constructs also constitute novel scaffold types, they are discussed separately in Section 4.

3.1. Scaffolds for Improved Genome Editing Efficiency

The first scaffold modification aimed at enhancing sgRNA efficiency was introduced in 2013 in the context of CRISPR-based chromosome imaging, and was coined the Flip + Extension (F + E) sgRNA [26]. Two alterations were combined into a single scaffold variant. First, the upper stem of the RAR was extended, with the rationale that a longer duplex would improve Cas9–sgRNA binding. This extension restores the native base identities at the corresponding positions of the crRNA–tracrRNA duplex, bringing the design closer to the native Streptococcus pyogenes configuration. Second, the uridine at position 4 of the scaffold was replaced with adenine, and the corresponding nucleotide at position 26 was converted from adenine to uridine to maintain base pairing (Figure 2, inset). This adjustment was designed to overcome premature transcription termination, since poly-T stretches are known Pol III termination signals. Subsequent work confirmed that the T4 motif alone was shown to cause ~75% termination efficiency in HEK293T cells, while six or more consecutive thymidines resulted in full termination [41]. While this study found the termination effect to be independent of cell type, another report suggested that editing efficiency could still be maintained when higher sgRNA amounts were present, but not under conditions of limited availability [52]. Another study showed that the U4A mutation (with the corresponding A26U change) mitigates the inhibitory “TT-motif” when present within the last four bases of sgRNA spacers. This motif is defined as TT + Y or a 2T + 2Y pyrimidine tract and is disproportionately enriched in low-activity sgRNAs [53]. Together, the A-U inversion and the extension alterations in the FE sgRNA produced a ~5-fold increase in signal-to-background ratio in dCas9-based CRISPR imaging and improved gene regulation by CRISPR interference (CRISPRi) as compared to the original scaffold; the term “HEAT” (Hybridization-Extended A–T inversion) scaffold was introduced in a later publication [22]. In a pooled dropout screen in mammalian cells targeting essential genes, this scaffold outperformed the original, yielding a substantially higher fraction of guides with strong dropout and fewer inactive guides compared to the unmodified scaffold [54]. In contrast, transient transformation experiments in tomato (Solanum lycopersicum) protoplasts did not show an increase in mutation frequency with this scaffold [55].

A follow-up study published in 2015 systematically tested stem extensions (1–10 nt) and alternative substitutions at position 4. This work confirmed that the F + E scaffold with a 5 bp stem extension was optimal but found that using a U→G substitution at position 4 (rather than the original U→A), together with the corresponding complementary change at position 26, yielded the highest knockout efficiency in mammalian cells. This version of the scaffold was then termed optimized scaffold (Figure 2, inset). In 15 out of 16 sgRNAs tested in mammalian cells, the optimized scaffold structure significantly increased knockout efficiency compared to the original scaffold. When two sgRNAs were used in combination to generate deletions, efficiency improved dramatically—by approximately tenfold in all four pairs of sgRNAs examined. The new design also substantially enhanced performance in lentiviral delivery systems, with efficiency gains even greater than those observed in plasmid-based expression [27]. This scaffold was also tested in rice (Oryza sativa), and was found to be superior to the original design [56]. In mammalian genome-wide screens, direct comparisons with the original scaffold indicated that, in the context of large-scale pooled assays, the optimized scaffold did not significantly change overall screen performance. Nevertheless, the data showed higher on-target activity, accompanied by increased off-target effects [57]. This trade-off between activity and fidelity mirrors similar findings with SpCas9 variants engineered for higher activity or fidelity [58,59], suggesting that such an inverse relationship may be a general feature of CRISPR optimization.

In 2020, a study reported the generation of 995 scaffold variants based on the F + E sgRNA design, systematically covering all single-nucleotide substitutions, all base-pair substitutions, and a set of designed combinations of these changes. The starting sequence also included a U13A mutation for library construction purposes. Notably, several variants with substitutions in stem-loop 2 consistently showed increased activity, revealing a stable rank ordering of performance despite substantial variation in absolute values. Six cr-variants—cr748, cr289, cr622, cr772, cr532, and cr961—were highlighted, in that order, as outperforming the standard F + E scaffold [38]. A follow-up study applying these scaffolds in prime editing experiments reported overall efficiency improvements, with variability depending on the specific spacer, and identified cr772 as providing the most pronounced enhancement [60].

A more recent advance is the T-lock scaffold, reported in 2022. This design addresses potential sgRNA misfolding precipitated by the spacer sequence. Here, the SL2 was elongated with an additional extension of high thermodynamic stability (Tm = 71 °C), serving as a nucleation site to stabilize correct folding. This design also uses a 4-nucleotide RAR stem extension, rather than the 5-nucleotide extension of the F + E scaffold. While chemical modifications further enhanced editing efficiency, the unmodified T-lock scaffold also improved gene editing performance with both in vitro-transcribed and in vivo–transcribed gRNAs. Adding extra locked hairpins at the tetraloop or the 3′ end, on top of the T-lock design, did not provide further improvements. In vivo-transcribed gRNAs showed up to a ~2.5-fold increase in editing efficiency in the most favorable case. Notably, this scaffold was not directly benchmarked against the earlier optimized sgRNA designs [22].

The development of prime editing introduced a new genome editing strategy in which a nickase-Cas9 is fused to a reverse transcriptase, which uses information encoded in a reverse transcriptase template (RTT) to introduce novel sequences into the genome [61]. In this approach, the RTT and primer-binding site (PBS) are appended to the 3′ end of the sgRNA. However, this additional sequence can interfere with sgRNA folding. To address this, a 2022 study optimized the sgRNA scaffold for prime editing by introducing an A48G and a corresponding G61C substitution, thereby reinforcing SL2 and improving stability. Additionally, the scaffold contains C63G and G75C substitutions relative to the original sgRNA. This variant was termed the “altered prime editing guide RNA-2” (apegRNA-2) [48]. Whether this scaffold also improves knockout efficiency with standard Cas9 has not been tested. Notably, the key modification in apegRNA-2 is located within SL2, the same hairpin stabilized in the T-lock design, suggesting that it may similarly enhance editing efficiency in knockout contexts. Conversely, it remains to be evaluated whether the T-lock scaffold can also protect against misfolding in prime editing systems.

In 2024, researchers leveraged artificial intelligence (AI) to design a novel sgRNA scaffold variant in a preprint describing the RNAGenesis model. RNAGenesis is a large-scale AI model trained on diverse RNA sequences and structures, enabling both accurate structural prediction and de novo functional design of RNA molecules. Several scaffold variants were generated and experimentally tested, with the RGen-6 variant showing the best performance. This mutant scaffold carries nine point mutations compared with the original scaffold, affecting the tetraloop, SL1, the linker between the SL1 and SL2, and —with the majority of changes (5/9) — SL2 itself. The RGen-6 scaffold showed greater stability, with enhanced G–C base pairing, stronger predicted folding energy, and overall improved structural robustness and folding reliability. Functionally, RGen-6 increased editing outcomes by up to two-fold for knockout efficiency, 2.5-fold for base editing (BE), and 1.2-fold for prime editing (PE) [49]. Unfortunately, this scaffold was only benchmarked against the original sgRNA design by Jinek and colleagues, published more than a decade earlier [6], and not against any of the more recently developed scaffolds.

In 2025, researchers expanded prime-editing guide RNA (pegRNA) scaffold engineering by creating diversified scaffold libraries intentionally designed to avoid long internal repeats, as these disrupt yeast-based homologous-recombination DNA assembly. To ensure compatibility with this assembly method, the authors restricted all shared sequence overlaps to ≤40 nucleotides. As a consequence of this design constraint, the scaffolds identified in their screen inherently fall within the broader class of reduced-homology sgRNA scaffolds, as discussed in Section 3.2.2. A total of 413 scaffold variants were evaluated in mammalian cells; these will be referred to as diversified pegRNA scaffolds for consistency, given that the original authors did not introduce a formal designation. These included replacement-type designs, which introduced structure-preserving 5-nt substitutions in the RAR and 4-nt substitutions in SL2, as well as extension-type designs, which added 5-nt insertions at regions predicted to tolerate structural flexibility based on pegRNA secondary structure modeling. The two design philosophies performed differently: replacement designs consistently produced higher gene-editing efficiencies than extension designs. In total, 17 variants outperformed the standard scaffold across every tested genomic locus, and all of these belonged to the replacement class. Notably, each of these high-performing designs also lacked the canonical T₄ motif (positions 2–5), consistent with its previously discussed role as a Pol III transcription-termination signal [50]. As with apegRNA-2, these scaffolds have not been tested for improved knockout efficiency with standard Cas9.

Although considerable progress has been made in optimizing sgRNA scaffolds, most plasmids continue to rely on the original design. Notably, more recent scaffolds such as T-lock (2022) [22] and RGen-6 (2024) [49] have not yet been benchmarked against earlier optimized variants. Moreover, the literature indicates that optimized scaffolds can improve efficiency in certain contexts but not in others, suggesting that additional factors—such as sgRNA abundance and the host species—play an important role. A timeline of enhanced scaffold variants is presented in the top panel of Figure 3, and a detailed visual overview of their structural differences is provided in Supplementary Figure S1.

3.2. Scaffolds with Reduced Sequence Homology

3.2.1. Spacer–Scaffold Complementarity (Intra-sgRNA Stability)

While broadly optimized scaffolds can enhance editing efficiency, the wide diversity of spacer sequences also entails a risk of scaffold misfolding through unintended hybridization with the spacer. This issue was first reported by Thyme et al. (2016) [62], who showed that many inactive gRNAs form non-productive complexes with SpCas9 both in vitro and in vivo, thereby sequestering Cas9 away from functional sgRNAs. In cases where an inactive sgRNA could be matched with a highly similar but active one, the researchers constructed hybrid (“chimeric”) sgRNAs by combining parts of the two spacers. This revealed that single-nucleotide substitutions could convert an inactive sgRNA into an active one. Although most inactive sgRNAs were predicted to form hairpins within the spacer region, several were instead predicted to form interactions between the spacer and the scaffold.

More recently, Riesenberg and colleagues (2022) [22] also addressed this issue. By introducing base substitutions into SL1, SL2, and SL3 of the tracrRNA (rather than the sgRNA scaffold) to minimize hybridization, they found that these modifications only rarely enhanced editing efficiency. Among the variants tested, only the t2 and t5 tracrRNA variants exhibited activity superior to the original tracrRNA, and then only for certain spacers.

In contrast to Thyme and colleagues [62], Huszár and colleagues (2023) [63] provided evidence that interactions between the spacer and the scaffold are an important cause of position-dependent sequence preferences. Systematic experimental interrogation of spacer–scaffold complementarity showed that only motifs targeting key structural units of the sgRNA—the RAR duplex and SL1—can efficiently impair SpCas9 activity. In a complementary approach, the authors analyzed a dataset comprising one million sgRNA–target pairs, representing an extensive sequence–activity collection. This large-scale analysis confirmed that inhibitory motifs are enriched in the SL1 region, where even short motifs of 5–8 nucleotides were sufficient to exert repressive effects. While both the RAR duplex and SL1 are susceptible to disruptive pairing, their sensitivities differ: RAR inhibition requires motifs of at least ~11 nucleotides with Tm ≥ 37 °C, whereas SL1 can already be disrupted by shorter motifs of ~7 nucleotides, though with a slightly higher Tm threshold (≥40 °C), highlighting the greater intrinsic vulnerability of SL1. The authors further demonstrated that extension of the RAR stem, originally introduced by Chen et al. (2013) in the F + E sgRNA design [26], can enhance sgRNA robustness against inhibitory spacer sequences complementary to the RAR region. By contrast, extending SL1 did not provide similar protection against SL1-complementary motifs, indicating that stabilization strategies effective for the RAR do not translate to other structural elements of the scaffold [63]. The researchers also designed self-complementary spacer sequences, of which approximately half (37/88) exhibited reduced activity. Remarkably, the use of “rescue” scaffold variants was able to restore activity in nearly all cases (36/37). To minimize interactions between the spacer and scaffold, several designs were introduced, including the “RAR rescue-1”, “RAR rescue-2”, “linker rescue”, “SL2 rescue”, “SL1 rescue-1”, and “SL1 rescue-2” scaffolds, many of which were repurposed from an earlier study [39]. This rescue-scaffold strategy is illustrated in the bottom panel of Figure 3. Importantly, the authors also demonstrated that spacer–scaffold complementarity is not limited to standard CRISPR–Cas9 editing as prime editing was likewise affected. In this case, short scaffold-complementary motifs within the reverse-transcription template (RTT) of the pegRNA that paired with the SL2 unit were sufficient to inhibit prime-editing efficiency. The authors also developed the Scaffold-Complementary Motifs Finder (SCMF) web tool, which identifies spacer sequences likely to be impaired by scaffold complementarity; such spacers can still function when paired with a suitable rescue scaffold [63].

3.2.2. Recombination in Multiplex sgRNA Arrays (Inter-sgRNA Stability)

In multiplex applications, where several genes are targeted simultaneously for genome modification or expression control, a common workflow involves cloning multiple sgRNAs into a single plasmid. However, the presence of homologous sequences is well-known to cause plasmid instability through homologous recombination [64,65,66]. To mitigate this, different promoter strategies can be employed: multiple Pol III promoter variants can be used to drive expression of individual sgRNAs within the array, or alternatively, Pol II promoters can be combined with processing systems that liberate each sgRNA from a precursor transcript (see Section 4.2). Nevertheless, the canonical sgRNA scaffold (76 nt) is sufficiently long to trigger instability when repeated multiple times within the same construct. While reports do exist of multiplex assemblies containing up to 49 sgRNAs in plants [67] and 31 sgRNAs in human cells [68], alternative scaffold designs have also been explored to reduce recombination issues. This strategy is illustrated in the bottom panel of Figure 3.

In plants, the use of different scaffolds to support multiplex editing has been demonstrated; starting from the optimized sgRNA scaffold, Wang and colleagues (2024) [69] used structure-guided and random mutagenesis to yield 45 functional variants, termed Scfvnn (with nn serving as a numerical identifier). In combination with different tRNA sequences, these scaffolds enabled the development of a fusion PCR protocol to rapidly assemble sgRNA–tRNA subclusters containing up to eight or nine spacers. When coupled with Golden Gate cloning, entire vectors carrying as many as 24 target sites were generated in a single step, significantly accelerating the construction of multiplex constructs. Proof-of-concept experiments in Nicotiana benthamiana and tomato (Solanum lycopersicum) confirmed the efficiency of this approach, with a single vector carrying two guides per gene (24 guides in total) simultaneously editing 10 out of 12 target genes in transgenic tomato plants. Construction of this array required nine different scaffold variants, although the Scfv09 variant did not yield any detectable editing activity. To further support adoption, the authors introduced SupClusterWriter, a web server that automates sgRNA cluster design using scaffold variants and tRNA spacers, thereby streamlining the construction of large-scale multiplex editing vectors.

The use of multiplexing scaffolds has also been explored for other modalities. In 2019, a set of 28 non-repetitive sgRNA scaffolds was published to enable simultaneous repression of multiple genes in E. coli. Using these so-called extra-long sgRNA arrays (ELSAs), it was demonstrated that up to 22 sgRNAs could be stably co-expressed, allowing simultaneous repression of as many as 13 genes, with repression levels reaching up to 3500-fold [70]. This set of scaffolds is truncated compared to the original scaffold (61 vs. 76 nt) and lacks SL3 [71]. Similarly, in 2025, the Non-repetitive RiboJ-Aided Multiplexed Base Editing (NR-RAMBE) system was reported, consisting of 38 non-repetitive 76-nt scaffolds developed for base editing in E. coli. In this system, the ribozyme RiboJ was used to excise individual sgRNAs from the array [72]. In the same year, the previously mentioned diversified pegRNA scaffolds were also reported, with shared repeats intentionally limited to ≤40 nt to enable future compatibility with large-scale synthetic DNA assembly in yeast [50]. In addition to the 17 variants that improved prime-editing efficiencies at all tested loci, the library also contains other scaffolds that outperform the canonical pegRNA at a subset of loci. Although these do not exhibit consistent gains across all targets, they may still be valuable in multiplexed editing applications.

3.3. Alternative Scaffolds with Similar Efficiency

As already noted, the sgRNA scaffold can tolerate extensive mutations without negatively affecting editing efficiency. Studies on the tracrRNA further showed that 3′ truncations only minimally reduce cleavage efficiency in vitro; even truncating the last 10 nucleotides—thereby removing SL3—retains nearly full nuclease activity [71]. Likewise, another study showed that specific mutations in both the lower and upper stems of the repeat–anti-repeat duplex, truncation of the linker to a single uridine, and specific substitutions of up to eight nucleotides in SL2 and 3 do not substantially impair gene editing activity [39]. In this work, a lower stem variant “mutated proximal duplex” was also engineered by substituting uridines at positions 2–5 with cytosines, with compensatory mutations at positions 36–39 to preserve base pairing. Notably, another study introduced these mutations into the “optimized scaffold”, generating a new variant that we term the “GCCC-esgRNA”. This design effectively prevented self-editing of the sgRNA in rice (Oryza sativa) when using Cas9-NG-based base editors, in contrast to the standard optimized scaffold [51]. By contrast, in rice (Oryza sativa) the near-PAMless SpRY Cas9 derivative showed increased self-editing with the GCCC-sgRNA scaffold and reduced on-target activity compared with the original GTTT-based scaffold. It should be noted that in this study the modifications were introduced into the original scaffold, rather than into the optimized scaffold used in the Cas9-NG–based base editor experiments of the previous work [73].

Additionally, the compiled dataset (Supplementary Table S8, entry 14–43) includes sgRNA scaffolds that exhibit broadly similar efficiencies, including 17 scaffolds from one mutational analysis [40], 6 from a separate mutational analysis [39]—some of which were later repurposed as “rescue” scaffolds as previously mentioned [63]—and 4 scaffolds obtained through directed evolution [74]. Although validation of these scaffolds has so far been largely confined to the original studies, they nevertheless represent a valuable resource for potential future applications.

4. Other Modifications and Extensions

In this section, other types of scaffold modifications are discussed, in which additional sequences are appended or integrated into the scaffold to confer new functionalities. An overview of these systems is provided in

Table 1. Summary of experimentally validated SpCas9 sgRNA scaffold modifications categorized by functional type. Representative elements, integration sites, roles, typical applications, and biological kingdoms in which validation has been demonstrated are indicated for each modification class. Key references are also included, although the list should not be considered as exhaustive. Sequences of these elements can be retrieved from Supplementary Table S9.

Category	Representative Elements	Integration Site	Role	Typical Application	Validated In
Visualization and recruitment (Section 4.1)	Pepper, Broccoli MS2, PP7, Com	Internal 3′-end	Effector or fluorophore recruitment	Different modalities	Animals [47,75,76] Plants [44] Fungi [43] Bacteria [75]
Multiplexing (Section 4.2)	Ribozymes Csy4 tRNA	Inter-guide	Transcript processing	Multiplexing	Animals [77] Plants [78] Fungi [79] Bacteria [72]
Stability (Section 4.3)	G2U1 evopreQ1	Internal 3′-end	Lower degradation	Prime editing (other modalities possible)	Animals [80,81] Plants [82] Bacteria [83]
DNA donor fusion (Section 4.4)	DNA donor (variable)	5′-end	Colocalization of donor and SpCas9	HDR	Fungi [84]
Conditional control (Section 4.5)	Guanine aptamer Theophylline aptamer	Internal 3′-end	Conditional sgRNA control	Classical CRISPR (other modalities possible)	Animals [85] Bacteria [86]
Fusion with other RNA elements (Section 4.6)	shRNA miRNA binding site lcRNA	Inter-guide 3′-end	Regulatory control	Different modalities	Animals [46,87,88]
Barcoding (Section 4.7)	iBAR CR1cs1 CR1cs2	Internal 3′-end	sgRNA tracking	Pooled screens	Animals [89,90]
Mobile RNA motifs (Section 4.8)	FT tRNA TLS	3′-end	Systemic mobility	Classical CRISPR (other modalities possible)	Plants [91,92]

.

4.1. Systems for Visualization and Recruitment

For genomic DNA imaging, fluorogenic RNA aptamers have been embedded into the scaffold. The first sequence that was used for this purpose was the Broccoli aptamer. This aptamer becomes fluorescent upon binding the GFP-like chromophores DFHBI/DFHBI-1T (3,5-Difluoro-4-hydroxybenzylidene imidazolinone) [47,93]. Pepper is another fluorogenic aptamer that binds HBC (Hydroxybenzylidene Cyanine) dyes that fluoresce upon binding [94], and can likewise be incorporated into the sgRNA scaffold [95]. Interestingly, Pepper has also been adapted in a different role through the Tat degron (tDeg) system. In this context, Pepper is not used as a dye-binding aptamer but instead as a protein-binding aptamer: it binds a Tat peptide fused to a degron tag on a fluorescent protein. Without this interaction, the degron drives rapid protein degradation. When Pepper binds the Tat peptide, the degron is masked, stabilizing the fluorescent protein and generating a signal specifically at the targeted genomic locus [96].

More generally, the insertion of protein-binding RNA aptamers into internal stem-loops of the sgRNA scaffold has become a frequently used strategy. These aptamers recruit RNA-binding proteins (RBPs), which are typically fused to effector domains to perform specific functions such as base editing, prime editing, epigenetic editing, or gene expression regulation [97]. Most aptamer–RBP pairs originate from bacteriophages and have been engineered into sgRNA scaffolds, such as MS2 aptamer–MS2 coat protein (MCP) [43,98,99], PP7 aptamer–PP7 coat protein (PCP) [43,98,99], BoxB-λ N22 peptide [99], Com-Com [43], Qβ aptamer–Qβ coat protein (QCP) [100], and A9-QCP [97]. These aptamers are typically added internally to one or more of the stem-loops, but they have also been added to the 3′ end of the scaffold [101]. Moreover, several types of aptamers have also been combined into a single scaffold [43,47,99]. Importantly, aptamers can also be harnessed to combine multiple modalities with a single Cas protein. For example, dCas9 can be used together with two distinct sgRNA scaffolds: one engineered with aptamers that recruit transcriptional activators, and another carrying aptamers that recruit repressors [64,67]. This allows simultaneous activation and repression of different target genes within the same system [43,102]. It has also been used to recruit adenine base editors (ABE) and cytosine base editors (CBE) to distinct genomic loci, as well as to induce deletions through the use of paired nickases. In the same study, the authors further proposed that if an additional guide with a truncated protospacer (14–15 nt), which does not support DNA cleavage, is used to recruit an epigenetic effector, repressor, activator, or fluorescent marker, the system could be expanded into a quadruple-function CRISPR platform [44].

Another strategy for recruiting effectors to the sgRNA scaffold is the use of PUF-binding sites (PBSs), which are specifically recognized by Pumilio/FBF (PUF) proteins [103]. The recognition code of PUF proteins has been fully elucidated, enabling each of the eight tandem repeats in the PUF domain to be engineered to bind one of the four RNA bases [104]. As a result, PUF domains can be readily programmed to recognize virtually any 8-nt RNA motif. In the Casilio system, sgRNAs are extended at the 3′ end with one or more PBSs (sgRNA-PBS), while effector proteins are fused to engineered PUF domains [103]. This design allows programmable and multiplexed recruitment of diverse effectors, including transcriptional regulators, epigenetic modifiers, and fluorescent proteins. Notably, this approach has also been applied to visualize non-repetitive DNA sequences using a single sgRNA, thereby expanding the imaging capacity of CRISPR–dCas9 systems [105].

4.2. Systems for sgRNA Multiplexing

In contrast to other CRISPR families, such as the Cas12 family, where nucleases can autonomously process their own guide RNAs from arrays, SpCas9 lacks this self-processing ability and therefore generally requires external mechanisms [106]. While it is possible to express multiple sgRNAs from individual RNA polymerase III (Pol III) promoters, this approach can become limiting due to size constraints. Pol III promoters are typically a few hundred base pairs in length [107], although shorter synthetic variants as small as 70 bp have been developed [108]. Moreover, the repetition of promoter sequences in such arrays can make them prone to recombination, further limiting stability. Nevertheless, in the BREEDIT genome editing platform, up to 12 guides have been successfully expressed from a single construct, each driven by its own Pol III promoter [109], and another study even reported up to 49 guides in one construct [67], demonstrating that large-scale promoter-driven arrays are technically feasible in some contexts. Another limitation is that commonly used Pol III promoters require a specific nucleotide as the first transcribed base—for example, adenine for U3 and guanine for U6 [110]. Finally, Pol III promoters are usually constitutively active, which restricts their use in conditional gene expression systems. In contrast, RNA polymerase II (Pol II) promoters offer more flexibility, but their direct use for sgRNA expression is complicated by issues such as 5′ capping and 3′ polyadenylation [111] (see Section 4.3).

For multiplexing, a 2013 study described a simple system in which two sgRNAs were directly fused with a SpeI restriction site between them. The fusion array was transcribed from a Pol II promoter and tested in stably transformed wheat (Triticum aestivum) suspension cultures, where it successfully induced targeted mutations. The INOX gene was chosen as the target, and two sgRNAs were designed to recognize adjacent genomic sites separated by 26 bp. Whereas the individual sgRNAs produced mutation frequencies of 17.9% and 20.7%, the dual-sgRNA construct still yielded efficiencies of 12.7% and 18.4%, respectively. In addition, complete deletions spanning both sites occurred in 2.8% of cases [112]. A comparable system in rice (Oryza sativa) achieved up to 93.7% editing efficiency by co-expressing SpCas9 and two sgRNAs from a single Pol II promoter, matching the performance of ribozyme-based strategies (discussed below) [45]. This “direct-fusion” system, which does not rely on processing elements to release individual sgRNAs, has also been successfully applied in plant viral vectors [113,114,115]. Another study in E. coli also found that an array of 6 sgRNAs could mediate base editing at all 6 targeted sites without making use of elements to splice out the guides, although the efficiency was lower than when ribozymes were used [72].

Among processing-based strategies, the earliest was reported in 2013, where each sgRNA was flanked by a Hammerhead (HH) ribozyme at the 5′ end and a hepatitis delta virus (HDV) ribozyme at the 3′ end. This configuration, termed the RGR (Ribozyme–gRNA–Ribozyme) cassette, enabled the transcript to self-cleave both in vitro and in vivo, releasing functional sgRNAs that successfully induced mutations in yeast (Saccharomyces cerevisiae) [110]. This strategy was later applied in mammalian cells as well [116,117]. A variant was developed for plants, in which ribozyme cleavage sites were placed between the SpCas9 ORF and the sgRNA, with a full-length HH ribozyme and an additional cleavage site included at the 3′ end of the transcript. In this configuration, the ribozyme could act in cis to process both the junction between Cas9 and the sgRNA as well as the transcript terminus [118]. A drawback of many ribozymes is that their cleavage efficiency often requires sequence adaptation to the surrounding context, complicating construct design [72]. An exception is RiboJ, derived from the Satellite tobacco ringspot virus (sTRSV) ribozyme, which includes an additional 23-nt hairpin at its 3′ end [84]. In an E. coli study, 15 different ribozymes were tested, of which 12 showed functional activity. Truncated variants were also engineered to remove dispensable hairpin loops [72].

In a landmark paper by Nissim and colleagues in 2014, authors described the use of the type III CRISPR/Cas-associated Csy4 protein from Pseudomonas aeruginosa. This protein recognizes a 28 bp sequence, cleaves the RNA, and remains bound to the upstream RNA fragment. While this first report placed a single sgRNA between two recognition sites [117], later studies extended the system to multiplex as many as 10 sgRNAs within a single transcript [78]. In this study, they also found that a truncated 20 bp recognition sequence was more efficient than the full length 28 bp version. In another follow-up, addition of a Csy4 recognition site to the 3′ end of prime editing gRNAs (pegRNAs) improved editing efficiency, which was attributed to reduced self-circularization of the guide RNA due to the stabilizing Csy4 hairpin [119]. Unlike other systems mentioned in this section, Csy4 requires the co-expression of an additional ORF (624 bp) [117], which can be a drawback in size-limited delivery systems such as viral vectors.

In 2015, Xie and colleagues reported tRNA–sgRNA arrays that are processed by endogenous tRNA-processing enzymes into individual sgRNAs. This polycistronic tRNA–gRNA (PTG) system was tested in rice (Oryza sativa) with two guides, where it achieved higher mutagenesis efficiency at individual target sites compared to plants expressing a single sgRNA without the tRNA cassette [120]. This rice derived tRNA sequence has also been successfully used in the plant Arabidopsis thaliana [121], but also, for example, cross-kingdom in Drosophila melanogaster [122], and Danio rerio [123]. While the rice tRNA works well in several species, the resulting repetitiveness of arrays constructed with one repeating tRNA element has led to the excursion into new variants, such as in Danio rerio [123] and plants [69]. Because tRNA molecules contain internal promoter elements [124], sgRNAs can be expressed at appreciable levels even without an upstream dedicated promoter. This background activity is undesirable when researchers aim to spatiotemporally control sgRNA expression through Pol II promoters. To address this, engineered tRNAs have been developed that eliminate such background expression, including the ΔC55G tRNA^Pro variant [125].

When leveraging ribozyme-, Csy4-, and tRNA-based multiplexing systems, a single Pol II promoter can simultaneously drive the expression of an open reading frame (e.g., SpCas9) and sgRNA arrays from a single transcript [126]. Because transcript processing removes the poly(A) tail from any ORF that is not placed in the last position, which would otherwise destabilize the mRNA, the 3’ end can be protected using RNA triple helices [117,127]. An important consideration for multiplexing is the occurrence of position effects, where the position of a guide within an array influences its activity. While some studies have reported position effects in all systems described here (including arrays driven by individual Pol III cassettes), other studies found no such effect. This variability suggests that additional parameters—such as the host species, transformation method, or other context-specific factors—play a significant role [128].

In the context of multiplexing, fusion guide RNAs (fgRNAs) should also be mentioned. In this design, a conventional Cas9 sgRNA is extended at the 5′ end with a Cpf1 (Cas12a) crRNA sequence. This format enabled orthogonal genome manipulation, as Cas9 and Cpf1 can be directed simultaneously: for example, the authors demonstrated knockout of one gene by Cpf1 together with transcriptional activation of another gene by Cas9 in mammalian cells, all using a single fgRNA molecule [129]. Finally, sgRNA arrays can also be cleaved by introducing shRNAs and miRNA-complementary binding sites, this will be discussed in Section 4.6.

4.3. Systems to Enhance sgRNA Stability

One of the earliest motifs reported to enhance sgRNA stability was the G2U1 quadruplex, an 11-nt sequence element with a strong tertiary structure. In zebrafish (Danio rerio) embryos harboring an endogenous eGFP gene, sgRNAs carrying the 3′ G2U1 motif yielded indel frequencies approximately 13% higher than the unmodified sgRNA. Similar results were obtained at the endogenous tyrosinase (tyr) locus, where the 3′ G2U1 modification enhanced cleavage efficiency by up to 11 percentage points. Notably, these experiments relied on microinjection of in vitro–transcribed sgRNAs [80], although it should be mentioned that these motifs can theoretically also be transcribed in vivo.

Initially designed for prime editing systems, where long 3′ extensions are particularly sensitive to degradation, RNA degradation–resistant motifs were developed specifically to stabilize such vulnerable RNA elements. In 2021, the prequeosine1-1 riboswitch aptamer (evopreQ1) was identified as a particularly promising motif for the 3′ end of prime editing guide RNAs. A trimmed variant of this aptamer, tevopreQ1, retains functionality at only 37 nucleotides. Even in its full form, at 42 nucleotides, evopreQ1 remains comparatively compact. In the same study, the 48-nt Moloney murine leukemia virus (MMLV) mpknot frameshifting pseudoknot was also shown to enhance RNA stability. These sgRNAs, termed engineered pegRNAs (epegRNAs), enhanced prime editing efficiency by 3–4-fold across various human cell lines without increasing off-target activity [60]. A separate study in 2022 systematically screened five exoribonuclease-resistant RNA (xrRNA) motifs derived from different flaviviruses. Among these, the Zika virus xrRNA emerged as the most effective stabilizer, though xrRNAs from Murray Valley encephalitis (MVE) virus and Dengue virus also performed well. Using prime editing, the authors reported an average enhancement of up to 3.1-, 4.5-, and 2.5-fold for base conversions, small deletions, and small insertions, respectively [81]. Finally, in the moss Physcomitrium patens, a turnip yellow mosaic virus (TYMV) 3′ tRNA-like structure (TLS) was incorporated at the 3′ end of prime editing sgRNAs in prime-editing experiments. Both the full 86-nucleotide and the truncated 41-nucleotide versions were functional and improved editing efficiency [130]. It remains unclear whether these modifications improve editing efficiencies in other CRISPR modalities, but they are likely most beneficial in applications that require long 3′ sgRNA extensions.

Circular gRNAs (cgRNAs) were first reported in 2022, motivated by the inherent instability of conventional linear sgRNAs, which are highly susceptible to degradation by RNA exonucleases. To address this limitation, cgRNAs were designed using the autocatalytic splicing activity of an RNA cyclase ribozyme. Their formation, which has strict length requirements, was optimized in vivo in E. coli. A construct containing a 251 bp insert, termed 251cgRNA, proved to be functional. In vivo, 251cgRNA enhanced the activity of both CBEs and ABEs in E. coli, outperforming linear sgRNAs [83]. A similar strategy has also been applied in mammalian cells using the Tornado system (Twister-optimized RNA for durable overexpression). In this design, transcripts are expressed with the RNA of interest flanked by Twister ribozymes, which undergo spontaneous autocatalytic cleavage. The processed RNA ends are subsequently ligated by the ubiquitous endogenous RNA ligase RtcB [131], generating circular sgRNAs. Functional assays demonstrated that circular gRNAs were less effective in both CRISPR activation and genome editing, showing reduced activity for on-target as well as off-target activity compared with linear sgRNAs [132].

Although not transcribed in vivo, in vitro–transcribed sgRNAs carrying a 5′ cap structure and 3′ polyadenylation were reported to enhance gene-editing efficiency when co-electroporated with SpCas9 protein into mammalian cells [133]. However, subsequent studies showed that the effects of these features may be context-dependent. In Saccharomyces cerevisiae, a 5′ cap did not impair sgRNA function in CRISPR interference (CRISPRi), whereas the presence of a 3′ poly(A) tail inhibited its activity [134]. By contrast, in rice (Oryza sativa) polyadenylated sgRNAs still supported high editing efficiencies [45]. A separate study in mammalian cells reported that the 5′ cap structure impaired Cas9 activity, while polyadenylation had no detectable effect [111]. The mammalian MICR-ON, -i, and -BE platforms for example (see Section 4.6) are also predicated on the observation that transcripts generated by RNA polymerase II typically do not yield functional gRNAs without further processing [88]. For this reason, polymerase II promoters are typically not used without any intervening sequences to release the sgRNA from the transcript [126], although this can be feasible depending on the context (see Section 4.2).

Multiplexing-associated elements may also potentially increase sgRNA stability. For example, as discussed in Section 4.2, the recognition motif of Csy4 is frequently used for multiplexing purposes. However, this structured hairpin has also been reported to slow mRNA degradation [135] and to stabilize both intronic sgRNAs [117] and prime editing sgRNAs [136], although it has not been directly established that it can also stabilize conventional sgRNAs. Similarly, self-cleaving ribozymes have been employed for multiplexing, and their cleavage products can also enhance stability by limiting RNA degradation. Small ribozymes such as the hammerhead and hepatitis delta virus (HDV) ribozymes generate cleavage products with 2′,3′-cyclic phosphate (cP) and 5′-hydroxyl termini [137]. These termini may in some contexts stabilize RNA molecules, as many exonucleases require a 5′-phosphate for efficient binding [138,139]. However, this effect is not universal: in certain systems, cP ends can also promote more rapid degradation [140]. Another potential stabilizing factor comes from tRNA structures, which might confer resistance against degradation. In the original report by Xie et al. (2015) [120], where tRNAs were first introduced as processing elements for sgRNA arrays, transcript levels of two sgRNAs were ~3- and ~31-fold higher compared to their counterparts lacking tRNA spacers. Whether this reflects protection from degradation or transcriptional enhancement remains unclear. Current evidence strongly supports the latter explanation, as tRNA genes contain two internal promoters [124], but direct experimental evidence is still lacking. Likewise the protein-binding RNA aptamers MS2, PP7, and BoxB have also been reported to enhance prime editing efficiency—even in the absence of their protein binding partners—when appended to the 3′-end of prime editing sgRNAs [136]. Overall, while some evidence suggests that these elements can stabilize RNA molecules, their ability to directly enhance the stability of conventional sgRNAs has not yet been conclusively demonstrated.

4.4. DNA Donor Fusions

Another modification explored is the direct fusion of a homologous recombination (HR) template to the sgRNA. One of the earliest demonstrations was the HI-CRISPR (Homology-Integrated CRISPR) system in Saccharomyces cerevisiae, reported in 2014. In this approach, the 5′ end of the crRNA spacer was fused to a ∼100 bp donor sequence containing homology arms, thereby linking the guide and repair template in a single molecule. Despite other reports indicating that additions at the 5′ end of sgRNAs are detrimental to SpCas9 cleavage activity [141], this design achieved up to 100% efficiency in targeted gene disruption. It should be noted that this system was based on the dual crRNA:tracrRNA format rather than a single-guide RNA (sgRNA) [84]. A later adaptation using synthetic oligo–derived crRNA–donor fusions in Saccharomyces cerevisiae demonstrated that the approach can also be applied to high-throughput genome editing at single-nucleotide resolution [142]. Indeed, this concept has since been explored more extensively using synthesized RNAs or RNA:DNA hybrids [143,144,145], but these approaches fall outside the scope of this review.

It is worth noting that prime editing guide RNAs (pegRNAs) also carry additional sequence information, in this case at the 3′ end of the scaffold. These extensions include both a reverse-transcription template for incorporation into the genome and a primer-binding site to enable base pairing with the target DNA [146]. While not donors in the traditional sense, prime editing systems can nevertheless mediate sequence replacements of up to ~88 base pairs and insertions of up to ~100 base pairs [147].

4.5. Conditional Control Systems

Aptamers have also been used to engineer ligand-responsive sgRNAs. The earliest design made use of an aptamer is fused to a ribozyme, creating an aptazyme. Ligand binding activates the ribozyme, which then cleaves itself and thereby releases a functional sgRNA. This approach was then applied to develop theophylline-controlled genome and base editing systems, as well as guanine-responsive transcriptional regulation. However, these systems are not entirely leak-free and retain residual sgRNA activity even in the absence of their ligand [148]. Another limitation of this system is that it relies on a blocking sequence to inactivate the gRNA in the absence of a ligand. Because this blocking sequence is partially complementary to the spacer, it must be redesigned for each new target.

To overcome this limitation, Lin and colleagues developed a universal design in which the blocking sequence targets the conserved lower and upper stem regions of the sgRNA scaffold rather than the spacer, thereby avoiding the need to redesign for each new target. In this strategy, both the blocking motif and a theophylline-binding aptamer are appended to the 3′ end of the sgRNA, creating Small-molecule–Activated Allosteric Aptamer-Regulated (SMART) sgRNAs. While SMART-sgRNAs provide a more universal design by blocking conserved stem regions, their efficiency in living cells remains limited. For example, the optimized construct sgB18-A30-S7 retained ~52% residual activity in the OFF state and recovered only ~31% of activity upon theophylline induction in HEK293T cells [85].

In contrast to blocking-motif approaches such as SMART-sgRNAs, another study engineered ligand responsiveness by directly inserting aptamers into the sgRNA scaffold. By systematically testing different insertion sites, the authors identified optimized ligand-responsive sgRNAs, namely a theophylline-activated variant (ligRNA+) and a theophylline-inactivated variant (ligRNA−). Both designs were functional for most, but not all, spacers. For the few non-responsive spacers, base-pairing between the spacer and the aptamer was proposed as a possible explanation for the lack of theophylline sensitivity. Furthermore, incorporation of the 3-methylxanthine (3MX) aptamer yielded sgRNA constructs termed ligRNA^±3mx which specifically respond to this chemical. These variants allow CRISPRi to be either activated or deactivated in a dose-dependent manner, spanning a dynamic range of more than 10-fold. In parallel, thiamine-responsive ligRNAs were developed, although these exhibited a narrower dynamic range (~6-fold) [149]. Another study screened a library of theophylline-aptamer insertions to identify switchable aptamer–sgRNAs (agRNAs) and identified two variants, A9 and GU19, with good retargetability. Introduction of a C22A point mutation within the ligand-binding pocket converted the theophylline aptamer into a 3-methylxanthine (3MX)-responsive version. Notably, use of these agRNAs led to a drastic improvement in CRISPR-mediated recombineering throughput in E. coli, increasing transformation efficiency by approximately 10⁴-fold by preventing DSB (double-stranded break) lethality [86].

Recent studies reported that sgRNAs can be engineered into modular molecular “sensors” by incorporating ligand-responsive aptamers and self-cleaving ribozymes at the 5′ or 3′ ends of the scaffold. These riboswitch-like designs allow CRISPR–dCas9 complexes to be conditionally activated or repressed in response to both small molecules (e.g., tetracycline) and endogenous proteins (e.g., NPM, β-catenin, NF-κB), effectively functioning as endogenous protein sensors that couple intracellular signaling states to transcriptional control [150]. Nevertheless, the extent to which large protein ligands can reliably trigger sgRNA conformational switching remains mechanistically underexplored.

Scaffolds have also been engineered to contain G-quadruplex motifs that can be stabilized by the ligand pyridostatin (PDS). In HeLa cells, sgRNAs carrying these elements exhibited only slightly reduced editing efficiencies compared to controls, but their activity could be tuned in a dose-dependent manner, with higher PDS concentrations leading to progressively lower editing efficiencies. The same strategy was also tested in a CRISPRa context, where it was found to function as well. However, the dynamic range of this system was modest, with relatively limited ON/OFF ratios [151]. Another study from the same laboratory introduced NCD (N-acyl-2-amino-1,8-naphthyridine)-responsive sgRNAs, but this approach likewise suffered from a narrow dynamic range [152].

Another approach relies on toehold-mediated strand displacement, a process in which a DNA or RNA strand, guided by Watson–Crick base pairing, displaces another strand from a duplex. In the toehold-gated gRNA (thgRNA) system, CRISPR interference is activated upon sensing exogenous or endogenous RNAs in E. coli, leading to up to a 15-fold repression of target gene expression [153]. A similar system has been tested in mammalian cells, but here with an additional miRNA binding site so the sgRNA can only be activated when both the correct miRNAs and mRNA are present in the cells [154]. More sophisticated strategies have since been developed, including SI-sgRNAs (self-inhibitory sgRNAs), cgRNAs (conditional guide RNAs), and iSBH-sgRNAs (inducible spacer-blocking hairpin sgRNAs), as well as approaches based on modified crRNAs. Readers interested in a broader overview of conditional sgRNA and crRNA systems are referred to recent dedicated reviews on the topic [155,156].

Finally, sgRNAs can also be conditionally activated by introducing shRNAs and miRNA-complementary binding sites, but this will be discussed in Section 4.6.

4.6. Fusion with shRNAs, miRNA Elements, and Long Noncoding RNAs

Engineered sgRNAs have been designed to include RNA regulatory elements such as shRNAs or miRNA-responsive motifs, providing regulatory or conditional control.

One example is the incorporation of a short hairpin RNA (shRNA) into sgRNA arrays. These shRNAs form hairpin structures that silence target genes via RNA interference (RNAi). By including an shRNA against DNA ligase IV (LIG4), a key NHEJ factor, into an sgRNA–shRNA construct, a more than twofold increase in HDR-mediated genome editing efficiency was achieved [87]. This improvement results from the transient suppression of competing NHEJ repair. However, the exact structure of sgRNA–shRNA molecules after processing remains unclear, and additions at the 5′ or 3′ ends of the gRNA may affect activity. In multiplex assays, where transcripts encoding three sgRNA–shRNA pairs were expressed as a single unit, the middle sgRNA indeed showed reduced activity, limiting the overall triple-locus editing efficiency to just over 9% [87]. Similarly, another study placed a primary microRNA (pri-miRNA) between two sgRNAs, generating two guides and a pre-miRNA that was further processed into a functional miRNA capable of mediating target knockdown [157].

Additionally, miRNA-complementary binding sites have been exploited to confer conditional control over sgRNA maturation. In this strategy, sgRNA precursors are transcribed by RNA polymerase II and are thus capped and polyadenylated, preventing their direct use in Cas9 complexes. The inclusion of flanking binding sites complementary to endogenous or exogenous miRNAs allows Argonaute-2 (AGO2) to mediate precise cleavage, releasing functional sgRNAs only in the presence of the cognate miRNA or siRNA. This approach, termed the MICR platform, has been adapted into several variants: MICR-ON for miRNA-inducible transcriptional activation (via dCas9–VPR), MICR-i for repression (via dCas9–KRAB), and MICR-BE for base editing upon induction by specific miRNAs. These systems act as miRNA-responsive switches, enabling cell-type- or state-specific genome editing and regulation [88]. Another study also made use of miRNA-binding sites in mammalian cells to conditionally activate sgRNAs [154].

Finally, long noncoding RNAs (lncRNAs) have also been fused to sgRNA scaffolds to explore their biological functions. Constructs carrying inserts of up to 4.8 kb remained functional, producing reporter activation or repression patterns matching those of the native lncRNAs, but with only modest changes in expression [46].

4.7. Guide Barcoding

Pooled CRISPR screens are another area benefiting from scaffold modifications. Such experiments often rely on sequencing the spacer region to track sgRNAs, but this approach can be confounded by false positives and negatives, particularly at higher multiplicities of infection (MOI). To address this, iBAR sgRNA scaffolds were developed, in which random 6-nt barcodes are embedded within the tetraloop of the F + E scaffold. This design enables each gRNA to be represented by multiple independent barcoded variants. Importantly, incorporation of barcodes into the loop did not impair SpCas9 activity and resulted in high reproducibility across replicates [49].

In another approach, capture sequences were incorporated into sgRNA scaffolds to enable reverse transcription and compatibility with single-cell sequencing workflows. Two variants gave good results: sgRNA-CR1^cs1, with the capture sequence inserted into stem–loop 2 (SL2), and sgRNA-CR1^cs2, with the capture sequence appended to the 3′ end [89].

4.8. Mobile RNA Motifs

The phenomenon of mobile RNAs has been particularly well studied in plants [158,159,160,161,162], where these long-distance signaling molecules have been implicated in diverse biological processes, including gene silencing, development, and abiotic stress responses. Beyond their intrinsic biological significance, RNA mobility motifs have been repurposed to enable systemic genome editing by fusing them to sgRNAs.

One of the first sequences explored for this purpose was derived from the Flowering Locus T (FT) gene, whose mRNA is transcribed in leaf vascular tissue and transported to the shoot apical meristem to induce flowering [91,163]. When Ellison and colleagues (2020) [91] fused the FT sequence to the 3′ end of a virally delivered sgRNA, they observed heritable mutations at distal sites and transgenerational editing in Nicotiana benthamiana with higher efficiency than control constructs. The authors also tested mutant FT (mFT, carrying a start codon mutation) and truncated FT variants, both of which exhibited similar performance. The authors further demonstrated that tRNA fusions could produce comparable results. Earlier studies had already shown that both tRNAs and mRNAs with tRNA-like structures (TLSs) are present in the phloem and capable of long-distance transport [164]. Building on this, another study achieved systemic genome editing using Arabidopsis thaliana homografts and Brassica rapa heterografts. In these experiments, the A. thaliana rootstock expressed Cas9 and sgRNA transcripts, both fused to a tRNA^Met-derived motif or to a truncated version lacking two structural arms of the tRNA (the D and T loops). While the efficiency remained modest in both cases, the results nonetheless demonstrated the feasibility of graft-transmissible genome editing [92]. Moreover, tRNA–sgRNA fusions have also been successfully applied in A. thaliana using viral vectors for transcriptional activation and targeted DNA demethylation [165].

5. Conclusions

The sgRNA scaffold is a vital yet often underappreciated component of the SpCas9 CRISPR system. Swapping only the 20-nt spacer while keeping all other parts of the system unchanged enables retargeting to virtually any genomic locus, which has cemented the scaffold as an out-of-the-box default, sometimes at the expense of further optimization. However, it is now clear that the canonical sgRNA scaffold is merely one of many possible designs, and likely not the most efficient option for all applications. Despite the development of more potent scaffold variants, their adoption remains surprisingly limited. Most plasmids continue to use the original scaffold, presumably out of inertia. This reflects a common phenomenon in science, where established components persist simply because they are familiar, even when better alternatives exist [166,167,168,169].

It is already well accepted that there is no universal “best” SpCas9 protein. For example, no single high-fidelity variant offers optimal on/off-target performance across all loci, which led to the development of a panel containing more than a dozen SpCas9 variants, enabling researchers to match the best-suited enzyme to a given target [170]. Given that the sgRNA scaffold is a structured RNA whose stability and folding can be strongly affected by the spacer sequence, it is logical that no universal “best” scaffold exists either. As already mentioned for the optimized scaffold that was reported in 2015 [27], pooled screen data showed higher on-target activity, but also higher off-target effects [57]. Furthermore, among scaffolds obtained via directed evolution [74], some variants were broadly compatible, whereas others were highly spacer-specific. Likewise, in a screen of 995 scaffold variants across 36 spacers, the scaffold–spacer interaction emerged as a key determinant of efficiency [38]. In the absence of clear design rules or AI-assisted workflows, generating a bespoke scaffold for every target is impractical for most applications. A more realistic compromise is to provide a curated panel of pre-validated scaffold variants—optimized for parameters such as overall activity, multiplexing compatibility, reduced self-editing in base editors, or minimal inter-sgRNA interference—with the specific choice left to the user depending on application requirements. The collection assembled in this review represents an initial step toward such a resource, but additional systematic benchmarking will be required before a widely accepted reference set can be established. A key advantage of engineering the sgRNA scaffold rather than the protein component is that the scaffold is considerably smaller and already incorporates the targeting spacer, rendering it inherently more amenable to customization. This flexibility is particularly valuable in multiplexing contexts, where multiple guides can be deployed in parallel, whereas introducing multiple nuclease or effector variants would be impractical or impossible in many cases.

Beyond general scaffold variants, which have been adapted for temporary applications such as cutting, nicking, base editing, and prime editing, this review also covers aptamer-containing scaffolds for imaging and effector recruitment. The latter is particularly promising for combining multiple modalities using a single SpCas9 enzyme [44]. Numerous inducible scaffold systems have also been developed using both small-molecule and RNA-based triggers, although most still suffer from suboptimal on/off ratios. Additional specialized scaffold designs include those for pooled screening applications incorporating internal barcodes or capture sequences, as well as self-targeting scaffolds for molecular recording. Auxiliary elements, although not always part of the core scaffold structure, are frequently positioned either adjacent to it or, in some designs, embedded within it. For multiplex expression from a single transcript, three major processing strategies dominate: ribozymes, Csy4 cleavage sites, and tRNA-based auxiliary elements. Several stability-enhancing sequences and mobility motifs were also reviewed. Taken together, these developments illustrate how the sgRNA scaffold has shifted from a static support structure to a versatile regulatory module. In the near term, further progress is likely to come less from inventing new designs than from standardizing, benchmarking, and deploying existing ones in practice.

During the preparation of this review, many scaffold sequences were found to be inaccessible, buried in figures, embedded in unannotated plasmid maps, scattered across Supplementary PDFs, or in some cases impossible to reconstruct even when the materials and methods were available. If researchers want new tools to be adopted, key functional sequences must be reported in an accessible and machine-readable format. This review attempts to address this issue by compiling, in Supplementary Tables S8 and S9, the sequences of all discussed scaffolds, along with additional auxiliary elements. To further facilitate reuse, these sequences are also provided in a dedicated GitHub repository. With clearer reporting standards and openly accessible collections of validated RNA parts, scaffold engineering may finally transition from a niche pursuit to a central parameter in CRISPR system design.