Gene Editing, Part Two

In Part One, we introduced CRISPR and its limitations. Conventional CRISPR-Cas9 editing produces a double-strand break in DNA whose repair can leave several different genetic changes. Following that break, one common response is non-homologous end joining, which reconnects the broken DNA ends without requiring a supplied repair template. The process can restore the original sequence, although processing of the ends can also introduce small insertions or deletions, collectively called indels. When an indel disrupts a gene’s function, the result can be a gene knockout. Such knockouts have allowed researchers to create animal models and investigate what happens when a particular gene loses its function. ^[1]

Restoring gene function, on the other hand, can require correcting an existing mutation or introducing functional DNA through a knock-in, a targeted insertion of a specified sequence. Researchers can attempt this by supplying a DNA template that carries the desired change between stretches matching the chromosome on either side. Through homology-directed repair, or HDR, the cell can copy that change into its own DNA as it repairs the double-strand break. ^[1]

However, supplying the template does not ensure that the cell will use it. HDR competes with other repair pathways, and its efficiency depends on the cell type and state, including whether the cell is dividing. Many cells relevant to treatment, such as mature neurons, are difficult to edit efficiently through this route. The double-strand break can also produce large deletions or rearrangements of chromosome material, even when Cas9 has recognized the intended target. ^[1]

To overcome some of these limitations, researchers have developed editing methods that build on CRISPR’s targeting system. Base editing combines that targeting system with enzymes that chemically convert individual DNA bases without requiring a double-strand break. ^{[2] [3]}

A base substitution replaces one of DNA’s four chemical bases, adenine (A), cytosine (C), guanine (G) or thymine (T), with another. Across the double helix, A pairs with T and C pairs with G. If a C is changed to a T on one strand, the complementary strand must eventually change from G to A for the two strands to agree. A permanent C-to-T substitution can therefore also be described as a G-to-A substitution when the same DNA is read from the opposite strand. ^{[2] [3]}

In a base editor, a deaminase is attached to a modified Cas9 protein. A deaminase is an enzyme that removes an amino group from a base, changing its chemical structure and how it pairs with other bases. Cas9 and its guide RNA position this enzyme at the desired sequence. As the guide pairs with one DNA strand, a short stretch of the opposite strand becomes exposed and available for the attached enzyme to act on. ^{[2] [3]}

The first cytosine base editors, reported by Alexis Komor and colleagues in Liu’s laboratory in 2016, use a deaminase to convert C into uracil, a base normally found in RNA that pairs like thymine. This creates a uracil-G mismatch within the DNA. The cell can remove uracil and reverse the intended change, so the researchers added a component that inhibits the enzyme responsible for initiating its removal. They also adapted Cas9 to nick the opposite, unedited DNA strand, breaking only one of the two backbones. That nick encourages repair to replace the unedited strand using the chemically altered strand as its template. Subsequent processing can turn the original C-G pair into a stable T-A pair. ^[3]

Adenine base editors apply a related strategy to A-to-G substitutions. Their deaminase produces inosine-containing DNA, which the cell’s copying machinery reads like guanine, and repair can convert the original A-T pair into a G-C pair. Developing this editor required an enzyme with a new activity. At the time, no natural enzyme was known to perform the required adenine deamination in DNA. Nicole Gaudelli and colleagues began with a bacterial enzyme that modifies RNA and used laboratory evolution and protein engineering to develop versions that could act on DNA. Their work, published in 2017, added A-to-G and T-to-C changes to the C-to-T and G-to-A changes available from cytosine editors. ^[4]

The substitutions available to these two classes follow directly from their chemistry. Each of the four DNA bases could, in principle, be replaced by any of the other three, giving twelve possible single-base substitutions. The original cytosine and adenine editors enable four of those substitutions. A mutation requiring an A-to-T correction, for example, cannot be directly corrected through their C-to-T or A-to-G chemistry. Converting bases alone also cannot insert a missing DNA segment or remove an unwanted insertion. A proposed treatment therefore has to begin by matching the required correction to an editor capable of making it. ^{[1] [2]}

Even when the required conversion is available, the surrounding sequence affects how cleanly it can be made. The deaminase acts within a short editing window, and several susceptible bases within that window can be modified together. Correcting one cytosine could therefore alter another nearby cytosine that should have been preserved. These additional changes are called bystander edits. Their consequences depend on where they occur, such as whether they change another amino acid in the resulting protein. Researchers can alter the enzyme’s editing window or choose another guide to improve the outcome, although guide placement must also satisfy Cas9’s requirement for a neighboring recognition sequence, the protospacer adjacent motif, or PAM. ^{[1] [2]}

Off-target editing presents another limitation because the Cas9-guide complex can recognize similar DNA sequences elsewhere in the genome. Some deaminases can also modify DNA or RNA independently of where the guide directs Cas9. These activities vary between editor versions, and engineering can reduce particular unwanted reactions. ^{[1] [2]}

Despite these limitations, base editing has produced substantial improvements in experimental disease models. Hutchinson-Gilford progeria syndrome causes severe premature aging and commonly results from a C-to-T mutation in LMNA, the gene that supplies the lamin A protein. Lamin A helps support the nucleus, and the mutation changes how the cell splices the gene’s RNA. During normal splicing, sections called introns are removed and the retained exons are joined to form the mature message. The progeria mutation creates an inappropriate splice site that removes part of the lamin A instructions, producing a shortened protein called progerin that disrupts nuclear structure and function. ^{[1] [5]}

An adenine base editor can reverse this mutation by changing the A opposite the mutated T into a base that is read as G, allowing repair to restore the original C-G pair. Luke Koblan and colleagues tested this approach in cells from children with progeria and in mice carrying the human disease mutation. Correction in the patient-derived cells reduced the abnormal RNA and progerin protein and improved nuclear shape. For the animal experiments, the researchers delivered genetic instructions for the editor using modified viruses. A single treatment at two weeks of age preserved smooth muscle cells in the aorta, the major artery leaving the heart. The ten treated mice had a median lifespan of 510 days, compared with 215 days for twelve saline-treated controls. ^[5]

Another development of CRISPR’s targeting system is prime editing, which uses an RNA template to let researchers specify virtually any replacement DNA sequence within the system’s edit-length limits. The sequence is supplied as information to be copied, allowing the correction to include substitutions, insertions and deletions beyond the chemical conversions available to the original base editors. Andrew Anzalone and colleagues first reported this approach in 2019. ^{[2] [6]}

To write the specified sequence, a prime editor combines a Cas9 nickase with an engineered reverse transcriptase. The nickase introduces a single-strand break, and the reverse transcriptase extends the resulting DNA end by copying an RNA template. Reverse transcription is also used in single-cell RNA sequencing to turn captured RNA into DNA suitable for sequencing. In prime editing, the newly copied DNA is attached directly to the chromosome being edited. ^{[1] [6]}

The prime-editing guide RNA, or pegRNA, provides both targeting and copying instructions. It retains the part of a CRISPR guide that recognizes the chosen DNA site and adds an extension with two further functions. One portion binds the DNA end exposed by the nick, providing a starting point for copying. The other carries the template containing the desired edit. Once the DNA end has paired with the pegRNA, the attached reverse transcriptase can extend it by copying that template. ^{[1] [6]}

This produces an extended DNA segment, often called a flap, that contains the edit and remains attached to the chromosome. The extension can pair with matching DNA on the opposite strand and displace the original segment. Cellular enzymes can then remove the displaced original DNA and join the newly written sequence into place. At this stage, the double helix contains one edited strand and one unedited strand, and further repair can make both strands retain the change. Some prime-editing systems introduce another nick in the unedited strand to encourage the cell to replace it using the edited strand as a template. ^{[1] [6]}

The template can encode any of the twelve types of single-base substitution, as well as small insertions, deletions and combinations of these changes. Conventional prime editing does not require a separate donor DNA molecule because the sequence to be copied is already contained in the pegRNA. Its practical reach depends on the length and structure of that template, the available target sites and the ability of the editor and cell to complete the reaction. ^{[1] [2] [6]}

One limitation is that the pegRNA must remain intact long enough to supply its template. Cellular enzymes can degrade its exposed extension, leaving shortened guides that still bind the editor or target DNA without being able to support productive copying. These incomplete guides can also occupy proteins and target sites that intact guides would otherwise use. To address this, researchers in Liu’s laboratory added compact RNA structures that protect the extension from degradation. The resulting engineered pegRNAs increased editing efficiency in the tested human cell lines and patient-derived fibroblasts. ^[1]

The cellular response also affects whether the new DNA is retained. Mismatch repair normally helps correct disagreements between the two DNA strands, and it can recognize a prime-editing intermediate as something that needs repair. If it removes the newly written strand before that strand has been sealed into place, the original sequence is restored. Temporarily inhibiting this response has improved editing in laboratory experiments. Unwanted indels can still arise during processing, particularly in systems that nick both strands, where concurrent nicks can sometimes result in a double-strand break. ^{[1] [6] [7]}

Larger sequence replacements introduce another challenge because prime editing initially writes the new sequence onto one DNA strand. The new segment must then be incorporated and its information retained across both strands. Twin prime editing extends this approach by using two pegRNAs to write complementary new sequences onto opposite strands. Those newly written segments can pair with each other, allowing cellular enzymes to remove the original intervening DNA and join the new segments into the chromosome. This arrangement can replace larger stretches without first copying the edit from one strand onto the other. ^{[1] [8]}

Twin prime editing can also prepare a target for another enzyme. Researchers can use it to install a short recognition sequence for a recombinase, an enzyme that joins DNA at particular sites. A separately supplied DNA molecule carries a corresponding recognition site and the larger sequence to be inserted. The recombinase then joins that cargo to the prepared genomic site. Combining these systems has enabled gene-sized insertions in cultured human cells. For diseases caused by different loss-of-function mutations in the same gene, inserting a functional sequence could offer a way to address several mutations with one general approach. It also adds a recombinase and donor DNA to the material that must enter the cell. ^{[1] [2] [8]}

Prime editing has also been applied to the much smaller insertion responsible for some cases of Tay-Sachs disease. Mutations in HEXA impair an enzyme needed to break down a fatty substance called GM2 ganglioside. The enzyme normally works within lysosomes, cellular compartments that digest and recycle material. When its activity is inadequate, GM2 accumulates to toxic levels, particularly in neurons, causing progressive neurological damage. One disease-causing mutation inserts four extra DNA bases into HEXA. In engineered human cells carrying this insertion, Anzalone and colleagues used prime editing to remove the additional bases and restore the original sequence. Under the best conditions tested, roughly a third of the target DNA copies were corrected. ^{[6] [9]}

Given these powerful tools for programming the genome, why do so many genetic diseases remain difficult to treat? Among the reasons is the difficulty of delivering an editor to enough of the cells responsible for a disease. Addressability is one of the three pillars of my framework for thinking about therapeutic development and progress, alongside observability and programmability.

Observability concerns our ability to measure cellular function and disease. Single-cell RNA sequencing has expanded that ability by distinguishing RNA profiles among individual cells, although those profiles remain partial measurements of a dynamic system. ^[10] Programmability concerns the biological changes that we can specify and induce, and base and prime editing expand this capability by allowing researchers to prescribe particular changes to DNA. Addressability concerns the cells in which those changes can actually be made. Here, reaching target cells with these large biomolecular therapies remains an overarching challenge.

For base and prime editing, the therapeutic cargo includes an editing protein and its guide RNA. The protein can be delivered directly or produced inside the cell from supplied genetic instructions, such as messenger RNA or DNA encoding the editor. An assembled protein-RNA complex can begin acting after it reaches the target DNA. Delivered messenger RNA has to be translated into protein. DNA instructions add an earlier step because they must first reach the nucleus and be transcribed into messenger RNA. The guide RNA must also be available when the editing protein is ready to act. ^{[11] [12]}

Once assembled, the guide-editor complex recognizes a DNA sequence rather than the identity of an organ or cell type. Many cells in a person contain that same sequence, including an inherited mutation, so the guide does not by itself provide access to the population that needs treatment. The administered cargo must reach the relevant tissue and enter its target cells. During uptake, a cell may enclose the material within an endosome, a membrane-bound compartment from which the cargo must escape before it can function. The editing protein and guide must ultimately gain access to nuclear DNA. Connie Cepko makes this availability of the target cells central to her presentation in Gene Therapy in Neuroscience 101, emphasizing that they must be reached in sufficient numbers to produce a therapeutic effect. ^{[11] [12]}

One way around much of the distribution problem is to remove the target cells for editing outside the body, or ex vivo. Blood-forming stem cells normally replenish the blood throughout life and can be collected for treatment under laboratory conditions. Techniques such as electroporation use electrical pulses to help introduce the editing material into those cells. Researchers can examine the resulting genetic changes and cellular behavior before returning the cells to the patient. If the edited stem cells successfully settle into the bone marrow and continue functioning there, they can supply new blood cells carrying the modification. ^{[11] [12]}

For cells that must remain in place, treatment requires delivery inside the body, or in vivo. Mature neurons illustrate the difficulty because they are embedded in existing tissue and connected to other cells through established circuits. Removing them for editing and returning them would require reestablishing those relationships. Delivery through the circulation also encounters the blood-brain barrier, where the properties of the brain’s blood vessels restrict the passage of many substances into the tissue. Direct injection can provide access to a local region, although the editor must still spread far enough to reach the required neuronal population. ^{[11] [16]}

Carrying the editing machinery into tissues therefore requires a delivery vehicle suited to the cargo and target. Two widely used approaches are lipid nanoparticles, which package material within lipid-based particles, and viral vectors, which adapt viruses’ ability to enter cells and transfer genetic material. Their behavior in the body determines which organs and cell populations receive the cargo, and how much of it becomes available inside those cells. ^{[11] [12]}

Lipid nanoparticles can carry messenger RNA encoding an editor together with its guide RNA. After uptake and release of the RNA, the cell produces the editing protein, creating a temporary supply that eventually disappears as the RNA and protein degrade. A successfully installed DNA change can persist after this exposure ends. Many nanoparticle formulations accumulate in the liver following administration through the bloodstream, making that organ relatively accessible to editing. This helps explain the prominence of liver targets among early in vivo applications. ^[12]

Viral vectors provide another route into cells. Adeno-associated virus, or AAV, has been adapted into a delivery vector whose protein shell carries DNA instructions for a therapeutic product. In gene augmentation, those instructions can supply a functional gene. For editing, they can instead encode the machinery that changes a sequence already present in the cell. Researchers can choose or engineer the AAV shell, called its capsid, and alter the administration route to influence which tissues receive the material. The viral components needed for replication are removed from the therapeutic vector, so the treated cells do not produce a spreading viral infection. ^{[11] [12]}

Once AAV is chosen, its cargo capacity constrains what can be delivered. A conventional prime editor’s encoding sequence is too large for the roughly 4,700-base capacity of one AAV vector. Researchers have therefore split the instructions between two vectors and arranged for the resulting protein parts to reassemble inside the cell. Both vectors must reach the same cell, and that cell must produce the parts in a form that allows the complete editor to be restored. ^[13]

AAV’s persistence introduces a further consideration. DNA delivered by the vector can remain active for a long time, which is useful when a therapy needs continued production of a missing protein. An editor usually needs enough time to install the intended change and can then disappear. Continued production extends the cell’s exposure to the editing machinery. The immune system can also recognize the capsid as foreign. Antibodies from previous exposure to related viruses may neutralize the vector before it enters the target cells, and an immune response to the first treatment can make subsequent dosing ineffective. ^{[11] [12] [15]}

Liu’s laboratory itself has developed engineered virus-like particles to deliver an assembled editing protein and its guide RNA. Viral components help the particles package their cargo and enter cells, without carrying a viral genome or DNA instructions for continued editor production. Developing this system required improving how much editor was packaged, how it was released and how it was directed to the nucleus. After delivery, the protein and RNA degrade, allowing the exposure to remain temporary even when the DNA change is lasting. ^{[1] [14]}

However, AAV vectors still have substantial drawbacks despite being among the most clinically established delivery platforms. There is currently no perfect delivery mechanism, and delivery remains a broad challenge for therapeutic development. ^{[11] [12] [15]}