CRISPR-Cas9 genome editing: PAM recognition, sgRNA guidance, and DSB repair outcomes

Intuition Beginner

Bacteria fight viruses with a filing cabinet. Each time a virus attacks, the bacterium snips a piece of viral DNA and stores it between repeated genetic sequences. The stored sequences are called CRISPR. When the same virus attacks again, the bacterium transcribes the stored snippet into a short RNA molecule that folds into a guide. A scissor enzyme called Cas9 reads the guide, finds matching DNA, and cuts it. The virus's genome is broken, and the infection fails.

Scientists replaced the viral snippet with a guide of their own design. Pick a twenty-letter stretch of any gene, synthesize a matching guide RNA, mix it with Cas9, and deliver the pair into cells. The Cas9 protein follows that custom guide to the matching locus and cuts there. The cut releases the cell's own repair machinery, which patches the break with one of two outcomes: sloppy stitching that scrambles the gene, or precise replacement using a template you supply alongside.

Why this exists: earlier genome editing required building a brand-new protein for every target — months of work per edit. CRISPR needs only a new short RNA, ordered by mail for a few dollars. That price collapse turned genome editing from a specialist craft into a routine laboratory tool and opened the therapeutic era that produced the first CRISPR medicine approved for human patients in 2023.

Visual Beginner

The figure has two panels. The left panel shows the bilobed Cas9 protein cradling the guide RNA: the REC lobe grips the RNA-DNA hybrid, the NUC lobe carries two scissor domains (HNH and RuvC), and a small PAM-interacting domain reads the short PAM sequence adjacent to the target. The right panel zooms into the DNA: a twenty-base target matching the guide, the three-base PAM (NGG) immediately downstream, and arrows marking the cut on both strands three bases upstream of the PAM, leaving a blunt double-strand break.

The two panels together show the licensing logic: no PAM, no cut. The PAM-interacting domain is the gate; the guide RNA is the address; the two nuclease domains are the blades.

Worked example Beginner

Edit the human EMX1 gene using a single guide RNA, producing a knockout by error-prone repair.

Step 1. Pick the target. In the EMX1 locus, choose a twenty-base sequence on one strand, for example GAGTCCGAGCAGAAGAAGA, immediately followed on the opposite strand by the three-letter PAM TGG (which reads as NGG on the strand Cas9 scans). The PAM is the licence; without it, the guide is inert.

Step 2. Synthesize the single guide RNA. The guide portion pairs complementarily with the twenty-base target (CUCGGCUCGUCUUCUUCUC in RNA letters). In laboratory practice the crRNA and tracrRNA of the native bacterial system are fused into one piece, the sgRNA, so a single twenty-letter molecule plus a constant scaffold is enough.

Step 3. Deliver Cas9 together with the sgRNA into the cell. The Cas9 protein loads the guide, samples DNA, and pauses wherever it sees a candidate PAM. At each PAM it pries the duplex open and tests whether the twenty bases adjacent to the PAM pair with the guide.

Step 4. At the matching EMX1 locus, the guide pairs perfectly. The HNH domain cuts the target strand three bases upstream of the PAM; the RuvC domain cuts the opposite strand at the same position. The result is a clean, blunt double-strand break.

Step 5. The cell repairs the break. Without a template, the cell uses non-homologous end joining, which often adds or deletes a few bases at the join. A single-base frameshift in EMX1 destroys the reading frame and the gene stops producing its protein — a knockout.

What this tells us: a twenty-letter RNA plus a constant protein are enough to direct a precise cut at a unique genomic address. The PAM licenses the cut; the cell's repair writes the edit.

Check your understanding Beginner

Formal definition Intermediate+

CRISPR-Cas9 (Streptococcus pyogenes, the standard laboratory variant SpCas9) is a ribonucleoprotein complex of the Cas9 endonuclease and a single-guide RNA (sgRNA) that cleaves double-stranded DNA at a fixed offset from a short protospacer-adjacent motif (PAM). The native bacterial system uses two RNAs — the CRISPR RNA (crRNA) carrying the spacer, and the trans-activating CRISPR RNA (tracrRNA) — which Jinek and Charpentier's 2012 work fused into a single chimeric sgRNA without loss of function. The mature complex is a 160-kilodalton protein bound to an approximately 100-nucleotide sgRNA.

Definition (bilobed architecture). Cas9 has two lobes ^{[Nishimasu2014]}. The recognition (REC) lobe binds the guide-target RNA-DNA heteroduplex and mediates conformational activation. The nuclease (NUC) lobe contains the HNH domain, which cleaves the target strand (the strand complementary to the guide), the RuvC domain, which cleaves the non-target strand (the strand with the same sequence as the guide except for the PAM), and the PAM-interacting (PI) domain, which reads the PAM. Two linker helices (L1 and L2) connect the lobes and gate the activation transition.

Definition (PAM-licensed cleavage). For SpCas9 the PAM is $5^{\prime }\text{-NGG-}{3}^{\prime }$ on the non-target strand immediately $3^{\prime }$ of the protospacer. Cas9 will not cleave a site lacking this motif: the PI domain must contact the two guanines of the PAM before the DNA duplex is destabilised and before the guide RNA can invade to form an R-loop. The cleavage occurs on both strands three base pairs upstream of the PAM (that is, on the $5^{\prime }$ side of the PAM on the non-target strand), producing a blunt double-strand break.

Definition (DSB repair outcomes). The double-strand break is repaired by one of two canonical pathways in the host cell. Non-homologous end joining (NHEJ) ligates the broken ends directly, often introducing small insertions or deletions (indels) at the junction; the result is gene knockout if the indel shifts the translational reading frame. Homology-directed repair (HDR) uses a supplied donor template (single-stranded oligonucleotide or plasmid) to copy a precise sequence into the break, enabling single-base substitutions, tag insertions, or conditional alleles ^[Cong2013].

Counterexamples to common slips Intermediate+

• Slip: "any twenty-nucleotide guide will cut." It will not, unless the genome carries an adjacent NGG PAM on the correct strand within a few base pairs of the twenty-nucleotide target. Roughly one in sixteen positions in a random genome is a candidate SpCas9 target ($1/4$ chance per base times $1/4$ for the second G), and roughly one in eight of those falls on the correct strand. PAM availability, not guide design, is the binding constraint on targetable loci.
• Slip: "Cas9 cuts where the guide matches." It cuts three base pairs upstream of the PAM, not at the centre of the twenty-nucleotide match. A guide that matches positions 1-20 of a protospacer will direct Cas9 to cleave between positions 17 and 18 (counting from the PAM-distal end), not between positions 10 and 11.
• Slip: "Cas9 cuts only once in the genome." Off-target cleavage occurs at sites that carry an NGG PAM and match the guide at most positions, with mismatches tolerated preferentially towards the PAM-distal end of the spacer. Genome-wide off-target profiling (GUIDE-seq, CIRCLE-seq) routinely detects off-target indels at a small number of loci for any given guide, and high-fidelity Cas9 variants were engineered precisely to suppress this.

Key theorem with proof Intermediate+

Theorem (PAM-licensing rule; Jinek 2012; Nishimasu 2014). SpCas9 in complex with its sgRNA cleaves a double-stranded DNA target if and only if (i) the sgRNA spacer is complementary to a twenty-nucleotide protospacer, and (ii) the dinucleotide GG of an NGG motif on the non-target strand sits immediately $3^{\prime }$ of the protospacer. Cleavage of each strand occurs three base pairs upstream of the PAM. Removing the PAM, mutating either of the two guanines, or deleting the PI domain abolishes cleavage.

Proof. The argument has three structural and kinetic pieces.

(1) PAM contacts gate the active complex. The SpCas9 PI domain presents two arginine residues (Arg1333 and Arg1335) into the major groove of the DNA duplex at the PAM position. Each guanine of the dGdG dinucleotide donates a hydrogen-bonding pattern that the two arginines read directly ^{[Nishimasu2014]}. Substituting either guanine removes a hydrogen-bond donor-acceptor pattern, eliminating the cognate contacts; the Cas9-PI domain does not dock, and the downstream conformational activation does not trigger.

(2) PAM binding is upstream of R-loop formation. Cas9 samples DNA by transient PAM contacts; only when a cognate PAM is recognised does the protein distort the duplex, opening it locally and allowing the sgRNA spacer to invade. The R-loop (the RNA-DNA hybrid between the sgRNA and the target strand, displacing the non-target strand) grows directionally from the PAM-proximal end towards the PAM-distal end. Without PAM binding, no distortion, no invasion, no R-loop.

(3) HNH and RuvC activate only after R-loop completion. The two nuclease domains adopt catalytically competent conformations only when the sgRNA-target heteroduplex spans the full spacer and the L1/L2 linkers reposition. The HNH domain then cuts the target strand at a fixed phosphodiester three bases upstream of the PAM; the RuvC domain, positioned by the same conformational switch, cuts the non-target strand at the same coordinate. The result is a blunt DSB at a deterministic offset from the PAM.

The contrapositive gives the rule's "only if" direction: if either the spacer-protospacer complementarity fails or the PAM is absent, no stage (1), (2), or (3) proceeds, and no cleavage occurs. $\square$

Bridge. The PAM-licensing rule builds toward 17.06.01 mutation and repair, because the blunt DSB it produces is the substrate on which NHEJ and HDR act, and the indels NHEJ leaves are the heritable mutations that classical genetics catalogued long before CRISPR existed. The rule appears again in 17.08.01 cell cycle and mitosis, where the availability of HDR is restricted to S and G2 phases (when a sister chromatid template exists) and NHEJ carries repair through G1 and M. The foundational reason the rule is structural rather than kinetic is that the PI domain reads a sequence signature directly: this is exactly the molecular implementation of an access-control mechanism, and the bridge is that base editing, prime editing, and the high-fidelity variants all preserve the PAM gate while re-engineering what happens after it.

Exercises Intermediate+

Advanced results Master

Theorem 1 (Sternberg 2014 single-molecule interrogation kinetics). Cas9-sgRNA complexes, tracked by live-cell single-molecule fluorescence in E. coli, sample on the order of $10^4$ distinct genomic sites before cleavage, with each interrogation lasting under one second. The search combines 3D diffusion between loci with brief 1D sliding along DNA, and the rate-limiting step is R-loop nucleation at a cognate PAM rather than diffusion. The result explains how a single protein finds a unique twenty-base address in a megabase genome within minutes ^{[Sternberg2014]}.

Theorem 2 (Jiang-Doudna 2013 CRISPR interference, dCas9). Two point mutations in the Cas9 active sites (D10A in RuvC, H840A in HNH) abolish both nuclease activities while preserving sgRNA loading, PAM recognition, and DNA binding. The resulting catalytically dead Cas9 (dCas9), tethered to a transcriptional repressor domain such as KRAB or fused directly to RNA polymerase recruitment blocks, silences a target gene without cleaving the DNA. CRISPRi turns Cas9 from a genome-editing tool into a genome-regulation tool, reversible and non-mutating.

Theorem 3 (Komor 2016 base editing). Fusing a cytidine deaminase (APOBEC1) or an adenosine deaminase (TadA, evolved) to a Cas9 nickase (D10A) generates a ribonucleoprotein that converts C to T or A to G within a narrow window (approximately bases 4-8 of the protospacer) without inducing double-strand breaks and without requiring a donor template. Base editing installs all four transition mutations (C→T, G→A, A→G, T→C) but cannot install transversions, and the absence of a DSB avoids the indel-by-NHEJ outcome that limits HDR.

Theorem 4 (Anzalone 2019 prime editing). A Cas9 nickase (H840A) fused to an engineered reverse transcriptase, guided by a prime editing guide RNA (pegRNA) that carries both the spacer and the edit-template extension, writes all twelve base-to-base substitutions, small insertions, and small deletions at a genomic target without producing a double-strand break and without exogenous donor DNA. The pegRNA both targets the locus and templates the edit, so the edit-product is encoded in the guide itself. Prime editing lifts the DSB-repair ceiling that constrains classical Cas9 editing.

Theorem 5 (Cas variants expand the target space). Staphylococcus aureus Cas9 (SaCas9) recognises the longer PAM NNGRRT but is small enough (~3.2 kilobases coding sequence) to fit inside a single adeno-associated virus (AAV) capsid, enabling in-vivo delivery. Cas12a (formerly Cpf1, Francisella novicida and related) recognises a T-rich PAM (TTTV), produces staggered cuts with 5' overhangs rather than blunt ends, and processes its own crRNA array, simplifying multiplexed editing. Cas13 (family of type VI systems) targets RNA rather than DNA, enabling reversible transcript knockdown. Each variant expands the set of editable loci and the modalities of edit.

Theorem 6 (high-fidelity Cas9 variants). Rational engineering of the REC domain (eSpCas9 1.1, Slaymaker 2016; SpCas9-HF1, Kleinstiver 2016) and directed evolution (HiFi Cas9, Vakulskas 2018) produce variants that retain on-target activity while suppressing off-target cleavage by one to three orders of magnitude. The mechanism is a weakening of the nonspecific DNA contacts, so the active conformation requires the full guide-target match to provide binding energy; partial matches at off-target loci no longer trigger activation. The therapeutic regulatory pathway required such variants, because wild-type SpCas9 off-target cleavage at unknown loci is a genotoxicity risk.

Theorem 7 (Casgevy 2023 therapeutic translation). The FDA approved Casgevy (exagamglogene autotemcel, exa-cel) on 8 December 2023 for sickle-cell disease. The therapy edits the erythroid-specific enhancer of the BCL11A gene in autologous haematopoietic stem cells ex vivo via Cas9-induced NHEJ, de-repressing fetal haemoglobin (HbF, $\gamma$-globin) production. The resulting increase in HbF replaces the defective adult haemoglobin and inhibits sickling. Casgevy is the first FDA-approved CRISPR-Cas9 therapy and the proof of concept for genome-editing cures of monogenic disease.

Synthesis. The PAM-licensing rule builds toward 17.06.01 mutation and repair, where the DSB becomes the substrate for the cell's repair machinery, and appears again in 17.08.01 cell-cycle timing, which determines whether NHEJ or HDR handles the break. The foundational reason the technique scales from bacteria to humans is that the bilobed architecture of Cas9 separates recognition (the REC lobe gripping the heteroduplex) from cleavage (the HNH and RuvC domains) and from gating (the PI domain reading the PAM). Putting these together with the Sternberg 2014 interrogation kinetics, the central insight is that Cas9 is a search-and-cut machine optimised by evolution for viral defence, and the bridge is that base editing, prime editing, and CRISPRi all preserve the search-and-bind half of the mechanism while re-engineering the cut half: the search half is exactly what identifies SpCas9 with a programmable DNA-binding protein, and the pattern generalises across the entire Cas-family zoo, where each variant is a different cut module on the same search chassis.

Full proof set Master

Proposition (Deterministic cleavage offset of SpCas9). SpCas9 produces a blunt double-strand break at a coordinate exactly three base pairs upstream of the PAM on both strands, for any guide sequence that satisfies the PAM-licensing rule.

Proof. Fix a target site with protospacer coordinates $1,\dots ,20$ on the non-target strand (base $1$ PAM-distal, base $20$ PAM-proximal) and PAM at coordinates $21,22,23$. After the PI domain docks onto the PAM and the R-loop forms, the HNH domain adopts its catalytically competent conformation, positioning its single catalytic histidine adjacent to the phosphodiester bond between non-target-strand bases $17$ and $18$. Cleavage of the target strand at the bond complementary to this position follows. The RuvC domain, held in place by the conformational activation of the L1/L2 linkers, positions its catalytic DDE-triad residues (Asp10, Glu762, Asp986, His983) adjacent to the same phosphodiester bond on the non-target strand. Cleavage of the non-target strand at the same coordinate follows. Because the two cuts are at the same genomic coordinate (rather than separated by a fixed stagger), the resulting break is blunt. The offset of three base pairs is fixed by the geometry of the active-site positions in the SpCas9 crystal structure ^{[Nishimasu2014]} and is independent of the guide sequence. $\square$

Proposition (Stochastic repair-outcome distribution). Given a Cas9-induced blunt double-strand break in a mammalian cell, the probability that the break is repaired by NHEJ rather than HDR is at least $\frac{1}{2}$ outside S/G2 and decreases with the availability of a homologous template.

Proof. NHEJ is active throughout the cell cycle (G1, S, G2, M) because it ligates broken ends without a template. HDR requires a sister chromatid (or an exogenous donor with sufficient homology) and is therefore restricted to S and G2, when a sister chromatid exists. In G1 and M, only NHEJ is available, so the NHEJ probability is $1$. In S and G2, both pathways compete kinetically, with NHEJ typically faster but HDR favoured at coordinated DSBs with a nearby donor. Averaging across an asynchronous population (roughly 60% G1, 20% S, 10% G2, 10% M), the lower bound on the NHEJ fraction is the G1+M fraction (70%), giving an NHEJ probability of at least $\frac{1}{2}$. This explains why knockouts (NHEJ indels) are obtained more readily than precise edits (HDR) in standard Cas9 editing. $\square$

Connections Master

• Cell and molecular biology methods — microscopy, PCR, sequencing, CRISPR 17.11.01. The present unit is the depth companion to the methods survey: 17.11.01 introduces CRISPR-Cas9 as one tool among the modern molecular-biology arsenal, and 17.11.03 supplies the molecular mechanism (PAM licensing, R-loop formation, DSB repair outcomes) that the survey only gestures at. The survey's overview of HDR and NHEJ is expanded here into the editing-outcome distribution that determines whether a given experiment produces knockouts or precise edits.

• Mutation and repair 17.06.01. The blunt DSB produced by Cas9 is the canonical substrate of the repair pathways catalogued in 17.06.01. NHEJ-mediated indels are precisely the small-scale mutations that classical genetics treated as random damage; Cas9 makes them programmable. HDR, the second pathway, is the molecular implementation of homologous recombination, and supplying an exogenous donor template converts the cell's recombination machinery into a precise genome-editing device.

• Cell cycle and mitosis 17.08.01. The competition between NHEJ and HDR is gated by the cell cycle phase at the moment of Cas9 delivery. HDR requires an S/G2 sister-chromatid template, so synchronising cells in S phase before editing raises the HDR fraction; synchronising in G1 raises the NHEJ fraction. The editing outcome is therefore not a fixed property of the guide RNA but a joint property of the guide, the cell-cycle phase, and the donor-template design.

Historical & philosophical context Master

Yoshizumi Ishino and colleagues observed in 1987, while sequencing the iap gene of Escherichia coli, an unusual cluster of repeated DNA sequences interspersed with short spacers of unknown origin and function ^[Ishino1987]; the spacers were later recognised as fragments of viral and plasmid DNA. Ruud Jansen coined the acronym CRISPR (Clustered Regularly Interspaced Short Palindromic Repeats) in 2002 to denote this family of loci. Rodolphe Barrangou, Philippe Horvath and colleagues demonstrated in 2007 that bacteria carrying a spacer matching an infecting phage resist that phage, while bacteria lacking the spacer are killed, establishing that CRISPR is a heritable adaptive-immune system ^{[Barrangou2007]}.

The molecular mechanism was reconstituted in vitro by Martin Jinek, Krzysztof Chylinski, Ines Fonfara, Michael Hauer, Jennifer Doudna and Emmanuelle Charpentier in 2012 ^[Jinek2012], who showed that Cas9 guided by a dual RNA could be reprogrammed to cleave any DNA target by fusing the crRNA and tracrRNA into a single guide RNA. Le Cong, Feng Zhang and colleagues extended the system to mammalian cells in early 2013 ^[Cong2013], editing the human EMX1 and PVALB loci. The 2020 Nobel Prize in Chemistry was awarded to Charpentier and Doudna for the development of the CRISPR-Cas9 genome-editing method.

Bibliography Master

@article{Jinek2012,
  author = {Jinek, M. and Chylinski, K. and Fonfara, I. and Hauer, M. and Doudna, J. A. and Charpentier, E.},
  title = {A programmable dual-{RNA}-guided {DNA} endonuclease in adaptive bacterial immunity},
  journal = {Science},
  volume = {337},
  number = {6096},
  pages = {816--821},
  year = {2012},
}

@article{Nishimasu2014,
  author = {Nishimasu, H. and Ran, F. A. and Zetsche, B. and Yan, W. and Zhang, F. and Nureki, O.},
  title = {Crystal structure of {Cas9} in complex with guide {RNA} and target {DNA}},
  journal = {Science},
  volume = {343},
  number = {6176},
  pages = {1247993},
  year = {2014},
}

@article{Cong2013,
  author = {Cong, L. and Ran, F. A. and Cox, D. and Lin, S. and Barretto, R. and Habib, N. and Hsu, P. D. and Wu, X. and Jiang, W. and Marraffini, L. A. and Zhang, F.},
  title = {Multiplex genome engineering using {CRISPR}-{Cas} systems},
  journal = {Science},
  volume = {339},
  number = {6121},
  pages = {819--823},
  year = {2013},
}

@article{Mali2013,
  author = {Mali, P. and Yang, L. and Esvelt, K. M. and Aach, J. and Guell, M. and DiCarlo, J. E. and Norville, J. E. and Church, G. M.},
  title = {{RNA}-guided human genome engineering via {Cas9}},
  journal = {Science},
  volume = {339},
  number = {6121},
  pages = {823--826},
  year = {2013},
}

@article{Barrangou2007,
  author = {Barrangou, R. and Fremaux, C. and Deveau, H. and Richards, M. and Boyaval, P. and Moineau, S. and Romero, D. A. and Horvath, P.},
  title = {{CRISPR} provides acquired resistance against viruses in prokaryotes},
  journal = {Science},
  volume = {315},
  number = {5819},
  pages = {1709--1712},
  year = {2007},
}

@article{Ishino1987,
  author = {Ishino, Y. and Shinagawa, H. and Makino, K. and Amemura, M. and Nakata, A.},
  title = {Nucleotide sequence of the {iap} gene, responsible for alkaline phosphatase isozyme conversion in {Escherichia coli}, and identification of the gene product},
  journal = {Journal of Bacteriology},
  volume = {169},
  number = {12},
  pages = {5429--5433},
  year = {1987},
}

@article{Sternberg2014,
  author = {Sternberg, S. H. and Redding, S. and Jinek, M. and Greene, E. C. and Doudna, J. A.},
  title = {{DNA} interrogation by the {CRISPR} {RNA}-guided endonuclease {Cas9}},
  journal = {Nature},
  volume = {507},
  number = {7490},
  pages = {62--67},
  year = {2014},
}

@article{Qi2013,
  author = {Qi, L. S. and Larson, M. H. and Gilbert, L. A. and Doudna, J. A. and Weissman, J. S. and Arkin, A. P. and Lim, W. A.},
  title = {Repurposing {CRISPR} as an {RNA}-guided platform for sequence-specific control of gene expression},
  journal = {Cell},
  volume = {152},
  number = {5},
  pages = {1173--1183},
  year = {2013},
}

@article{Komor2016,
  author = {Komor, A. C. and Kim, Y. B. and Packer, M. S. and Zuris, J. A. and Liu, D. R.},
  title = {Programmable editing of a target base in genomic {DNA} without double-stranded {DNA} cleavage},
  journal = {Nature},
  volume = {533},
  number = {7603},
  pages = {420--424},
  year = {2016},
}

@article{Anzalone2019,
  author = {Anzalone, A. V. and Randolph, P. B. and Davis, J. R. and Sousa, A. A. and Koblan, L. W. and Levy, J. M. and Chen, P. J. and Wilson, C. and Newby, G. A. and Raguram, A. and Liu, D. R.},
  title = {Search-and-replace genome editing without double-strand breaks or donor {DNA}},
  journal = {Nature},
  volume = {576},
  number = {7785},
  pages = {149--157},
  year = {2019},
}

@article{Slaymaker2016,
  author = {Slaymaker, I. M. and Gao, L. and Zetsche, B. and Scott, D. A. and Yan, W. X. and Zhang, F.},
  title = {Rationally engineered {Cas9} nucleases with improved specificity},
  journal = {Science},
  volume = {351},
  number = {6268},
  pages = {84--88},
  year = {2016},
}

@book{DoudnaSternberg2017,
  author = {Doudna, J. A. and Sternberg, S. H.},
  title = {A Crack in Creation: Gene Editing and the Unthinkable Power to Control Evolution},
  publisher = {Houghton Mifflin Harcourt},
  address = {Boston},
  year = {2017},
}

@book{Watson2014,
  author = {Watson, J. D. and Baker, T. A. and Bell, S. P. and Gann, A. and Levine, M. and Losick, R.},
  title = {Molecular Biology of the Gene},
  edition = {7},
  publisher = {Pearson},
  address = {San Francisco},
  year = {2014},
}