Dr Giedrius Gasiūnas, CEO and Co-Founder of Caszyme, describes the company’s research into engineered CRISPR-Cas12l nuclease Asp2Cas12l M82, a potential new tool for genome editing and the development of gene therapies. 

CRISPR has become one of the most widely used tools for gene editing across research and therapeutic development, but the technology is not infallible. Based on a natural bacterial defense mechanism against viruses, CRISPR-associated (Cas) enzymes capture fragments of DNA and use them as molecular guides to recognise and cut matching DNA sequences. However, wielding these so-called ‘Cas’ enzymes as pharmaceuticals comes with trade-offs; they tend to be either sufficiently potent but too large to deliver to cells efficiently1, or compact but impaired in cutting efficiently1. Repair outcomes are also difficult to predict, as it is the cell rather than the enzyme that determines how a break is resolved.

Their efficacy at gene editing is closely linked to the protospacer adjacent motif (PAM). This is a short DNA sequence adjacent to the target sequence that the Cas enzyme recognises and requires for binding and cleavage to take place. The length and composition of the PAM therefore restricts the genomic sequence space that can be targeted, as the target sequence must be located next to a PAM with the specific features recognised by the Cas enzyme3. The PAM acts less like a key and more like a keyhole: the enzyme carries the right key within its guide RNA, but cleavage can occur only where an appropriate PAM keyhole sits directly adjacent to the target. 

Beyond PAM availability, vector delivery limits remain the predominant bottleneck in genetic medicine. Widely utilised enzymes such as SpCas9 (1,368 amino acids) or Cas12a (~1,200–1,300 amino acids; ~3.8 kb coding sequence) consume nearly the entire cargo capacity of adeno-associated viral (AAV) vectors, which are capped at approximately 4.7 kbp. This leaves virtually no room for tissue-specific promoters, polyadenylation signals, and guide RNA expression cassettes within a single vector – part of the reason why many companies have begun to investigate dual AAV-based therapies. While delivering Cas enzymes transiently as mRNA or ribonucleoprotein (RNP) complexes avoids viral packaging limits, enzyme size and stability still heavily dictate formulation efficiency, cellular uptake, and effective dosing windows in non-viral platforms like lipid nanoparticles (LNPs). 

A little-known bacterial family offers a new option 

In 2022, Lithuania-based Caszyme identified a novel family of Type V-L CRISPR candidates: Cas12l4. Discovered through metagenomic mining of wastewater bacteria within the phylum Armatimonadota, the Cas12l locus is characterised by an operon-like arrangement where a single RuvC endonuclease domain sits alongside cas1 and cas2 adaptation genes. 

Cas12l enzymes are some of the more compact enzymes, spanning fewer than 900 amino acids. This makes them much more suitable for viral delivery methods compared to Cas9 or Cas12a enzymes. They also present other important advantages including their ability to recognise and cleave next to cytosine-rich PAMs, a counterbalance to the thymine-rich PAMs preferred by most other Cas12 nucleases. Its N-terminal domain also extends across the full length of the bound DNA target, effectively locking it in place.  

However, wild-type of Cas12l exhibited modest baseline activity in mammalian cells (HEK293T), requiring systematic protein engineering to unlock their therapeutic potential. 

Improving Cas12l’s editing efficiency 

To optimise editing efficiency in human cells without increasing enzyme size, our team combined structure-guided rational design with protein language modelling, focusing initially on the Asp2Cas12l wild-type. 

First, we determined the ternary RNP-DNA target structure of Asp2Cas12l using cryo-electron microscopy to a resolution of 2.51 Å. The structure revealed key amino acid residues positioned near the negatively charged phosphate backbone of the DNA target strand, indicating an opportunity to strengthen the electrostatic protein-DNA interaction. 

We hypothesised that introducing positively charged arginine residues at these contact points would strengthen electrostatic protein-DNA interactions. Testing 34 site-directed arginine substitutions in human HEK293T cells yielded the engineered variant M67, which delivered up to a seven-fold increase in double-strand break indel rates over the wild-type enzyme. 

To advance beyond structure-guided engineering, we utilised ESM-2, a 15-billion-parameter protein language model trained on large-scale protein sequence data5.  

Then, using a zero-shot prediction approach, Caszyme used ESM-2 scored mutational tolerance to prioritise functionally viable substitutions. Following in silico screening, nine candidate mutations were introduced into the M67 backbone and tested in HEK293T cells. 

In particular, two mutations delivered significant enhancements: Q572R in the bridge-helix domain and F607S in the catalytic RuvC domain. By combining both of these mutations onto the same M67 backbone, Caszyme created Asp2Cas12l M82. M82 achieved indel rates of up to 99%, rescued efficiency at difficult genomic sites which previously edited poorly, and maintained robust activity when delivered transiently as either RNP or synthetic modified mRNA. 

M82 stands up to SpCas9 

Despite these structural improvements, Asp2Cas12l M82 retains a compact size of just 867 amino acids, roughly two-thirds the size of SpCas9 (1,368 aa), making it well suited to size-constrained delivery systems such as AAV. In both human HEK293T and retinal ARPE-19 cells, M82 achieved editing efficiencies comparable to SpCas9.  

Because Cas12l cleavage generates staggered 5′ overhangs, M82 is particularly adept at driving precise sequence insertion via homology-directed repair (HDR). At the human AAVS1 safe-harbor locus, M82 mediated up to 40% site-specific integration of a 1.6 kb double-stranded DNA donor template in HEK293T cells. For discrete single and dinucleotide substitutions, M82 mediated HDR rates of up to 56%. 

Comprehensive specificity profiling reinforced M82’s clinical translational potential. Genome-wide off-target cleavage was mapped via TEG-seq, an assay sensitive to cleavage events down to 0.0002% read abundance. Across four validated targets, 96.3% to 99.94% of tagged reads localised precisely to the intended on-target locus, with candidate secondary sites sitting near or below baseline detection thresholds. Orthogonal in silico prediction and targeted sequencing across 29 predicted loci confirmed no detectable off-target indels above background, demonstrating a clean safety profile. 

Not just one therapeutic route 

The combination of compact packaging, cytosine-rich PAM targeting and efficient HDR integration positions Asp2Cas12l M82 as a versatile chassis for both in vivo and ex vivo therapeutic programmes: 

  • Single-Vector Viral Delivery: By occupying only ~2.6 kb of packaging space, M82 frees up >2.0 kb within standard AAV limits, easily accommodating all necessary regulatory elements and gRNA cassettes within a single vector to bypass dual-vector efficiency losses. 
  • Transient Non-Viral Formats: M82’s stability as an RNP or mRNA transcript facilitates low-dose packaging into LNPs, restricting nuclease exposure to brief cellular windows and reducing systemic off-target liabilities. 
  • Targeted Gene Replacement: High HDR efficiency with staggered ends supports gene-correction strategies for monogenic disorders where simple gene disruption is therapeutically insufficient. 

Overcoming the physical vector cargo limit remains one of the final hurdles in bringing complex gene-editing architectures to clinic. Evolving naturally compact systems like Cas12l into high-efficiency, high-fidelity effectors can eliminate the compromise between enzyme potency and packaging feasibility. Nevertheless, expanding these tools into broader clinical pipelines will depend on collaborative partnerships across translational and clinical medicine, transforming theoretical genome engineering into scalable therapeutic reality. 

References

  1. Lino CA, Harper JC, Carney JP, et al. Delivering CRISPR: a review of the challenges and approaches. Drug Deliv. 2018;25(1):1234-1257. 
  2. Xin C, Yin J, Yuan S, et al. Comprehensive assessment of miniature CRISPR-Cas12f nucleases for gene disruption. Nat Commun. 2022;13:5623. 
  3. Gasiunas G, Barrangou R, Horvath P, Siksnys V. Cas9-crRNA ribonucleoprotein complex mediates specific DNA cleavage for adaptive immunity in bacteria. Proc Natl Acad Sci U S A. 2012;109(39):E2579-E2586. 
  4. Urbaitis T, Gasiunas G, Young JK, et al. A new family of CRISPR-type V nucleases with C-rich PAM recognition. EMBO Rep. 2022;23(12):e55481. 
  5. Lin Z, Akin H, Rao R, et al. Evolutionary-scale prediction of atomic-level protein structure with a language model. Science. 2023;379:1123-1130. 

About the author  

Dr Giedrius Gasiūnas is CEO and Co-Founder of Caszyme. With nearly 20 years in CRISPR-Cas research, he was part of the team that first demonstrated Cas9 is a programmable RNA-guided DNA endonuclease. He leads development of the company’s compact Cas12l nucleases as alternatives to Cas9 for therapeutic and drug discovery applications.