CRISPR Types Explained: Class, Type and Subtype Made Simple

 

I will be honest with you. The first time I tried to make sense of CRISPR classification, I closed the paper with a headache. Roman numerals, capital letters, “Cas” followed by a number, and what seemed like a new name for every protein. Does it really have to be this complicated? The short answer is no. It looks far worse than it is, and once you see the logic behind it, the whole system falls into place.

In this post I will walk through how CRISPR-Cas systems are classified, what the classification is actually based on, and where the famous Cas9 protein fits in. Along the way I will explain every term as we meet it, so you can follow even if you have never opened a microbiology textbook.

A quick bit of history

The story starts in 1987, long before anyone dreamed of gene editing. A Japanese team led by Yoshizumi Ishino was sequencing a gene in Escherichia coli and noticed something odd next to it: short, repeated DNA sequences separated by unique stretches of DNA. Nobody knew what they were for. For years they remained a curiosity, a puzzle sitting quietly in bacterial genomes.

Then the puzzle began to grow. Similar repeats turned up in other bacteria, and in archaea too. In 2002, Ruud Jansen and colleagues gave them the name we use today, CRISPR (clustered regularly interspaced short palindromic repeats), and noticed that a set of genes always travelled alongside them. They called those genes cas, for CRISPR-associated.

The real breakthrough came in 2007. Rodolphe Barrangou and colleagues, working with the yogurt bacterium Streptococcus thermophilus, showed that these repeats act as an adaptive immune system. When a virus that infects bacteria (a bacteriophage, or phage) attacks, the bacterium can save a small piece of the phage DNA between its repeats. That saved piece is called a spacer. Next time the same phage arrives, the bacterium recognizes it and destroys it. A memory of past infections, written into the genome. Beautiful, isn’t it?

After that, the field exploded, and so did the number of different CRISPR systems being discovered. That is exactly why a classification was needed.

How common is CRISPR in the first place?

Before we sort the systems into boxes, a fair question: how many microbes have one? Genome surveys suggest that roughly 40% of sequenced bacterial genomes carry a CRISPR-Cas system, while about 85% of archaea do. Treat these as rough figures. Databases over-represent bacteria that are easy to grow or medically important, while the many bacteria we cannot culture in the lab are under-represented. CRISPR systems are also gained and lost frequently, so even two strains of the same species can differ. And the exact number shifts a little depending on how a survey defines a system, for example whether a lone CRISPR array without cas genes counts.

What is the classification based on?

Every CRISPR-Cas system does three basic jobs.

• Adaptation: capturing new spacers from invaders. This is done by Cas1 and Cas2.
• Processing: turning the CRISPR array into short guide RNAs, called crRNAs.
• Interference: using a guide RNA to find the invader and destroy or disable it.

Cas1 and Cas2 are found in almost every system, so they cannot tell one system from another. Instead, the classification focuses on the interference step, and in particular on the effector module. The effector is the machinery that recognizes the target and deals with it. It is the part that makes each system different.

The classification then works as a hierarchy, like a family tree with three levels.

1. Class is decided by the architecture of the effector: is it a team of several proteins or a single large protein?
2. Type is decided by a signature gene, one diagnostic cas gene that is found only in that type.
3. Subtype is decided by the finer details, such as extra accessory genes, how the genes are arranged in the locus (the stretch of DNA where they sit), and the evolutionary relationships between the proteins, which is called phylogeny.

Wait, is the effector just the cutter?

Mostly, but not quite. The effector does the whole interference job, and that includes finding the target as well as cutting it. In some systems one protein does both. In others, the two jobs are split between different parts.

Think of it this way. In some systems the effector is a Swiss army knife that finds and cuts. In others it is a crew, where some members locate the target and others do the cutting. The cutter is part of the effector, but it is not the whole story.

Class 1: the team players

Class 1 systems use a multi-subunit effector complex, meaning several Cas proteins assemble together into one machine. They make up roughly 90% of CRISPR loci found in nature.

Type	Signature gene	Target	What to remember
I	Cas3	dsDNA	The most common type overall. A complex called Cascade finds the target, then Cas3 shreds it.
III	Cas10	RNA, and DNA while it is being transcribed	Works only when the target is being transcribed. Uses small signalling molecules called cyclic oligoadenylates.
IV	Csf1	Not well understood	Rare. Mostly found on plasmids and prophages, and often missing adaptation genes.

Type I is a lovely example of divided labour. Cascade is the finder: a guide RNA plus several Cas proteins that scan DNA for a match. Once Cascade locks on, it recruits Cas3, a protein that combines a helicase (an enzyme that unwinds DNA) with a nuclease (an enzyme that cuts nucleic acids). Cascade finds, Cas3 shreds. Type I also needs a short motif next to the target called a PAM (protospacer adjacent motif), which helps the system tell invader DNA from the bacterium’s own.

Type III works differently. Its complex recognizes RNA, and the Cas10 subunit makes cyclic oligoadenylate signals that switch on other nucleases such as Csm6. Because it needs the target to be transcribed, it checks that the invader is actively doing something before it strikes.

Class 2: the lone wolves

Class 2 systems use a single, large, multidomain protein that does everything the Class 1 team does. They are only about 10% of loci, but they are the stars of genome editing, precisely because one protein is so much easier to move into another cell than a whole complex.

Type	Signature gene	Target	What to remember
II	Cas9	dsDNA	Needs a tracrRNA as well as the crRNA. Blunt cut. Two nuclease domains, HNH and RuvC.
V	Cas12	dsDNA and ssDNA	RuvC nuclease only. Staggered cut. Cas12a needs no tracrRNA. Includes the miniature Cas12f.
VI	Cas13	RNA	Two HEPN domains. Collateral RNA cleavage, which is the basis of SHERLOCK diagnostics.

A few of those words deserve a plain explanation. A tracrRNA (trans-activating CRISPR RNA) is a second RNA that Cas9 needs, alongside the crRNA, to become active. HNH and RuvC are the two nuclease domains of Cas9: each one cuts one strand of the DNA double helix. HEPN domains are the RNA-cutting parts of Cas13. And collateral cleavage means that after Cas13 finds its target, it keeps chopping up any nearby RNA, which can be harnessed as a detection signal in a diagnostic test.

Where does Cas9 live?

Cas9 is the signature protein of Type II, and Type II is a minority, roughly 5 to 10% of CRISPR-bearing genomes in most surveys. It is essentially a bacterial protein and is very rare or absent in archaea. It turns up mainly in Firmicutes and Proteobacteria, with scattered occurrences in Bacteroidetes, Actinobacteria, Spirochaetes and Cyanobacteria. Even within one genus it is patchy, so having Cas9 in one species tells you little about its relatives.

The subtypes are defined by accessory genes.

• II-A carries Csn2 alongside Cas9. Examples include Streptococcus pyogenes (the source of SpCas9, the workhorse of the lab), S. thermophilus, Staphylococcus aureus (SaCas9), Listeriaand some Lactobacillus and Enterococcus species.
• II-B carries Cas4. Examples include Legionella pneumophila and Francisella novicida.
• II-C has only Cas1 and Cas2 alongside Cas9. It is the most widespread and diverse subtype, found in Neisseria meningitidis, Campylobacter jejuni, Pasteurella multocida, Treponema denticola, Corynebacterium diphtheriae and others.

Why does anyone care about the different versions? Because Cas9 proteins from different species, called orthologs, differ in size and in the PAM they recognize. SpCas9 is about 1,368 amino acids long and recognizes NGG. SaCas9 is smaller, at about 1,053, and recognizes NNGRRT. CjCas9 is smaller still, at about 984. Smaller orthologs are attractive when you need to fit the editor inside a delivery vehicle with limited space, such as an AAV (adeno-associated virus).

A simple way to remember it all

If the Roman numerals still feel slippery, try this.

• Class tells you one protein or many.
• Type is decided by a CRISPR system’s signature protein. A CRISPR system with Cas3 is Type I, one with Cas9 is Type II, one with Cas10 is Type III, one with Csf1 is Type IV, one with Cas12 is Type V, and one with Cas13 is Type VI.
• Subtype is the letter after the numeral, such as II-A or II-C, and it tells you which variant, mainly from the accessory genes.

And a handy shortcut for spotting Type II: if the system needs a tracrRNA, you are probably looking at Cas9 and not Cas12.

The classification keeps growing

This is not a finished map. Class 2 effectors appear to have evolved from nucleases carried by mobile genetic elements. Cas12, for example, is thought to descend from TnpB-like proteins found in transposons (jumping genes). Each time researchers discover a new family of these proteins, the classification has to stretch to fit it. The 2020 update recognized 2 classes, 6 types and 33 subtypes, and those numbers have likely moved since. More types have since been proposed, which tells you how much of the microbial world is still unexplored.

Why this matters for you

If you work in a lab, or are heading towards one, this classification is more than trivia. It explains why Cas9 needs a tracrRNA and Cas12a does not. It explains why Cas13 is used for RNA detection while Cas9 is used for DNA editing. And it points to where the next generation of tools might come from: the small, strange and little-studied systems sitting in the bacteria we have not yet learned to grow.

Watch the video

Prefer to see it explained? The Adwoa Biotech YouTube channel has a video on the basics of CRISPR-Cas9, covering the history, guide RNAs, PAM sequences and the sickle cell example. https://youtu.be/VlbGQiB83-Y?si=0gnpCu2gmcYZN0cR

Key takeaways

• CRISPR-Cas systems are classified mainly by their effector module, the machinery that recognizes and deals with the target.
• Class 1 uses a multi-subunit complex (Types I, III, IV). Class 2 uses a single protein (Types II, V, VI).
• The type is set by a signature gene, and the subtype by accessory genes, locus organization and phylogeny.
• Cas9 is the signature protein of Type II, found in a minority of CRISPR-bearing bacteria.
• The effector is the whole interference machinery, not just the cutter.

References

Barrangou, R., Fremaux, C., Deveau, H., Richards, M., Boyaval, P., Moineau, S., Romero, D. A., & Horvath, P. (2007). CRISPR provides acquired resistance against viruses in prokaryotes. Science, 315(5819), 1709-1712. https://doi.org/10.1126/science.1138140

Ishino, Y., Shinagawa, H., Makino, K., Amemura, M., & Nakata, A. (1987). Nucleotide sequence of the iap gene, responsible for alkaline phosphatase isozyme conversion in Escherichia coli, and identification of the gene product. Journal of Bacteriology, 169(12), 5429-5433. https://doi.org/10.1128/jb.169.12.5429-5433.1987

Jansen, R., van Embden, J. D. A., Gaastra, W., & Schouls, L. M. (2002). Identification of genes that are associated with DNA repeats in prokaryotes. Molecular Microbiology, 43(6), 1565-1575. https://doi.org/10.1046/j.1365-2958.2002.02839.x

Makarova, K. S., Haft, D. H., Barrangou, R., Brouns, S. J. J., Charpentier, E., Horvath, P., Moineau, S., Mojica, F. J. M., Wolf, Y. I., Yakunin, A. F., van der Oost, J., & Koonin, E. V. (2011). Evolution and classification of the CRISPR-Cas systems. Nature Reviews Microbiology, 9(6), 467-477. https://doi.org/10.1038/nrmicro2577

Makarova, K. S., Wolf, Y. I., Iranzo, J., Shmakov, S. A., Alkhnbashi, O. S., Brouns, S. J. J., Charpentier, E., Cheng, D., Haft, D. H., Horvath, P., Moineau, S., Mojica, F. J. M., Scott, D., Shah, S. A., Siksnys, V., Terns, M. P., Venclovas, C., White, M. F., Yakunin, A. F., ... Koonin, E. V. (2020). Evolutionary classification of CRISPR-Cas systems: A burst of class 2 and derived variants. Nature Reviews Microbiology, 18(2), 67-83. https://doi.org/10.1038/s41579-019-0299-x

Shmakov, S., Smargon, A., Scott, D., Cox, D., Pyzocha, N., Yan, W., Abudayyeh, O. O., Gootenberg, J. S., Makarova, K. S., Wolf, Y. I., Severinov, K., Zhang, F., & Koonin, E. V. (2017). Diversity and evolution of class 2 CRISPR-Cas systems. Nature Reviews Microbiology, 15(3), 169-182. https://doi.org/10.1038/nrmicro.2016.184

Related Content

• CRISPR-Cas9 basics: history, guide RNA, PAM and the sickle cell example https://adwoabiotech.blogspot.com/2026/11/crispr-cas9-explained-guide-rna-pam.html
• Specificity versus sensitivity in CRISPR diagnostics https://adwoabiotech.blogspot.com/2026/11/crispr-diagnostics-specificity.html
• Gene editing delivery and virus-like particles https://adwoabiotech.blogspot.com/2026/11/gene-editing-delivery-why-virus-like.html

How PCR Tells Plasmodium Species Apart: Reading the Bands

Image credit: @ATinyGreenCell 

I was looking at a photo of an agarose gel the other day. A ladder ran down each end, and between the two ladders sat a row of lanes, each with a bright band at a different height. My first thought was a lazy one: the bands must differ because we picked different primers for each lane, so we are really just labelling the result ourselves. Is the species written into the band, or are we writing it there?

The honest answer is a bit of both, and untangling the two is the whole point of this post. Once you see how the pieces fit, species-specific PCR stops looking like a magic trick and starts looking like good, careful design.

Figure: illustrative agarose gel with a DNA ladder (L) at each end and sample lanes between them. This image is for illustration only and does not show real Plasmodium data.

A band tells you size, and nothing else

Let's start with what a gel actually shows. When PCR copies a stretch of DNA, the product is called an amplicon, which simply means a copy of the target region made in large numbers. When you run that product through an agarose gel, shorter pieces travel farther and longer pieces lag behind. The ladder is a mix of DNA pieces of known lengths, so you can read off roughly how long your own band is.

That is all. A band has no species tag attached to it. It tells you how long the DNA is, and it tells you that the reaction found something to copy. Everything else, including the species name, comes from knowing which primers went into that tube.

Where the real differences live: the 18S rRNA gene

Now for the inherent part. Most species-specific Plasmodium PCR targets the 18S ribosomal RNA gene, often written 18S rRNA or SSU rRNA (small subunit ribosomal RNA). Ribosomes are the machines that build proteins, every cell needs them, and so every Plasmodium species carries this gene. That is why it makes such a handy target. It is also present in several copies per genome, roughly five to seven, which means there is more starting material to find and the test is more sensitive.

Here is the key feature. Along the length of this gene, some stretches are conserved, meaning they are almost identical in every species because changing them would damage the ribosome. Other stretches are variable, meaning evolution has been free to let them drift, so they differ from species to species. Picture a sentence in which most words are the same across four related languages but a few words are different in each. Those few words are what let you tell the languages apart.

These differences are real and built into the parasites. Nobody put them there. Primer design simply takes advantage of them.

Layer one: primers that can read the difference

A primer is a short piece of synthetic DNA, usually around 20 to 30 bases, that sticks to a matching sequence on the template and gives the polymerase a place to start copying. The sticking step is called annealing. The polymerase is the enzyme that builds the new DNA strand, and it can only add bases to the free end of a primer, known as the 3' end (said "three prime").

This is where specificity is won or lost. If the 3' end of the primer matches the template perfectly, the polymerase extends it and copying begins. If the very last base or two do not match, standard Taq polymerase, which has no proofreading ability, extends poorly or not at all. So designers place the 3' end of a species-specific primer right on a base that differs between species, and then use a stringent annealing temperature, meaning a temperature high enough that imperfect matches fall off. The result: the primer pair works on its own species and stays silent on the others.

That is the inherent layer. Without real sequence differences in the gene, there would be nothing for the primers to read.

Layer two: why the bands sit at different heights

Now back to my gel. If every species-specific pair worked on the same gene, why would the bands land at different heights?

Because in the classic assay each species gets its own forward and reverse primer pair, and each pair sits at a different position within the target region. Different positions mean different distances between the two primers, and that distance is the size of the product. This is where my original suspicion was partly right. The sizes are largely a consequence of where we chose to put the primers.

The widely used method is nested PCR, described by Snounou and colleagues (1993). Nested simply means two rounds. In the first round, genus-level primers sit in conserved regions and copy a long fragment from any Plasmodiumspecies present, around 1,100 bp in the commonly cited version of the protocol. A bp, or base pair, is the unit we use to count DNA length. In the second round, a small amount of that first product is moved into fresh tubes, one for each species, each with its own species-specific pair. The first round boosts sensitivity, and the second round supplies the species identity.

Species	Species-specific primer pair	Approximate product size
P. falciparum	rFAL1 and rFAL2	205 bp
P. vivax	rVIV1 and rVIV2	120 bp
P. malariae	rMAL1 and rMAL2	144 bp
P. ovale	rOVA1 and rOVA2	about 800 bp

Sizes are as commonly reported; a few papers give slightly different values for P. malariae, and P. ovale is now split into two species (curtisi and wallikeri) with a variant primer set for the second.

So a 120 bp band in the P. vivax tube and a 205 bp band in the P. falciparum tube are not the parasites announcing themselves in different sizes. They are the product of a careful design, in which sequence differences decide which pair works and primer placement decides how big the product is.

A small piece of history

Malaria parasites were first seen under the microscope by Alphonse Laveran in 1880, and for more than a century the blood smear stained with Giemsa remained the everyday way to name the species. It still is the front line in many clinics. Yet microscopy struggles when parasite numbers are low, when a mixed infection hides one species behind another, or when the smear is read by a tired eye at the end of a long shift.

PCR changed that. The Snounou paper appeared in 1993, the very year Kary Mullis received the Nobel Prize in Chemistry for inventing the technique. Within a few years, nested PCR had revealed something microscopy kept missing: mixed infections were far more common than anyone had assumed.

When size is the difference, and when it is not

In nested PCR, size is mostly a design choice. In other formats, size reflects a real difference in the DNA. If two species share the same primers but one has an insertion or deletion between them, the amplicons come out different lengths, and the gel shows it directly.

And some methods drop the gel altogether. In real-time PCR (qPCR), a fluorescent probe, a short DNA piece that glows only when it binds its exact target, reports which species is present. Different probes can carry different coloured dyes, so several species can be read in one tube. Another trick is melt curve analysis: after amplification you slowly heat the product, and the temperature at which the double strand falls apart depends on its sequence and GC content, which differs between species. Same idea, different readout.

What can go wrong

A few cautions are worth putting in any honest write-up of this method.

• Controls matter. A positive control for each species and a no-template control (a tube with water instead of DNA) tell you whether a band is trustworthy or just contamination.
• Isolates vary. A mutation under a primer can cause a false negative, so primers should be checked against parasites from the region where you work.
• Mixed infections can be missed. When one species is far more abundant, it can outcompete the others during amplification, which is a good reason to run separate species-specific reactions.
• Close relatives blur. Some closely related species are nearly identical at 18S, so a primer pair may not separate them. P. knowlesi also needs its own dedicated primer set, since the four-species panel above was not built for it.

The take-home

So, to answer my own question. The species is not written into the band height. It is written into the DNA sequence, and primers read it. The band height is something we then choose through primer placement, so that the answer is easy to read on a gel. Inherent differences decide who gets amplified, and design decides where the band lands.

Next time you look at a gel, ask two questions. What did the primers go looking for? And how far apart did we place them? Those two answers explain nearly every band on the picture.

Related content

• Check out What Is Giemsa Stain? The Chemistry and History Behind Malaria Diagnosis for the full chemistry breakdown.

• Explore LAMP Primer Design Explained to learn about primer design and assay optimisation.

• Read Sanger Sequencing Explained: How One of the Oldest DNA Sequencing Methods Still Works to understand the foundational method that long-read platforms eventually built on.

• Our guide to the P. falciparum life cycle breaks down the ring, trophozoite and schizont stages — the same ones you'll be identifying on your Giemsa smears.

References

Snounou, G., Viriyakosol, S., Zhu, X. P., Jarra, W., Pinheiro, L., do Rosario, V. E., Thaithong, S., & Brown, K. N. (1993). High sensitivity of detection of human malaria parasites by the use of nested polymerase chain reaction. Molecular and Biochemical Parasitology, 61(2), 315-320. 

Gene Editing Delivery: Why Virus-Like Particles Matter Now

Credit: Feng Zhang

 

The scissors are ready, the courier is not

For years the big question in gene editing was whether we could cut DNA precisely. We can. CRISPR-Cas9, base editors and prime editors are real tools that work in real cells, and the first CRISPR medicine is already approved. So the question has quietly changed into something less glamorous: how do we get these tools into the right tissue of a human body?

It matters because most diseases misbehave in one particular tissue. A liver problem needs the editor in the liver. A blood disorder needs it in blood stem cells. An editor that lands in the wrong place does nothing useful at best, and causes trouble at worst. Cutting is the solved part. Delivery is where a lot of the field is stuck right now.

Take the cells out, or go inside the body?

The approved route is the easy one. Casgevy, from CRISPR Therapeutics and Vertex, treats sickle cell disease and transfusion-dependent beta-thalassemia. Doctors collect blood stem cells (CD34-positive cells) from the patient, and in the lab the Cas9 protein and its guide RNA are pushed into those cells with a short electric pulse called electroporation. The edited cells then go back to the patient. This is ex vivo editing, which simply means editing outside the body, and it works well because the cells sit in a dish where you control everything.

The price of that control is cost and complexity. You need a specialised hospital, a conditioning regimen, and a long stay. Hence the dream of in vivo editing: one injection in an ordinary clinic, and the patient's own tissue is reprogrammed. In parts of West Africa where sickle cell disease is common, the gap between a specialised transplant centre and a local clinic is not a technicality. It decides who can be treated at all.

A quick tour of the delivery vehicles

AAV. Adeno-associated virus is the workhorse of gene therapy, and its problem is room. It carries only about 4.7 kilobases of DNA, and a Cas9 gene plus a guide RNA plus the control elements to run them usually overshoots that. AAV DNA also sits in the nucleus as an episome, a small ring of DNA separate from your chromosomes. It is stable, and it only fades as cells divide. For a nuclease that is the wrong rhythm, because the longer Cas9 hangs around, the more chances it has to cut places it was never meant to cut.

Lentivirus. Lentiviral vectors hold roughly 8 to 10 kilobases, enough for editing tools, and they have a proven track record in CAR T cell therapies for blood cancers. The catch is that they integrate into your chromosomes. A scissors gene stitched into the genome keeps making scissors. Lentiviruses are usually given a borrowed coat protein in a process called pseudotyping, and the most popular coat, VSV-G, is about to become a main character.

Lipid nanoparticles (LNPs). These are tiny fat bubbles that carry RNA. Injected into the bloodstream, they pick up a protein called ApoE and are swallowed by liver cells through the LDL receptor route. The cargo then has to escape the endosome, the little bubble the cell wraps around anything it swallows, before it can do anything. That is why LNPs love the liver, and why getting them anywhere else is hard.

Intellia's nexiguran ziclumeran (nex-z) is the flagship example. It uses an LNP to deliver CRISPR components that switch off the TTR gene in the liver, as a one-time treatment for transthyretin amyloidosis. People sometimes call this a liver disease, but it is really a body-wide one: the liver is simply the factory that makes the misfolded protein.

It is also a useful reminder that liver delivery is not risk free. In October 2025 the FDA paused the nex-z phase 3 trials after a patient developed Grade 4 liver enzyme elevations and raised bilirubin, and later died. Intellia pointed to complicating comorbidities, and the holds were lifted in early 2026 with enhanced liver monitoring added. The lesson I take from it: a vehicle that goes to the liver by default is both a gift and a liability.

Why a short stay is a feature

The ideal editor does its job and leaves. You want a tight window in which the nuclease is active, enough to make the intended cut and not so long that it wanders. This is the real appeal of delivering the editor as a protein rather than as a gene. A protein is already working the moment it arrives, and the cell breaks it down within a short time. No DNA means nothing to integrate and nothing to keep expressing.

How do you deliver a protein to a cell without a gene? That is where virus-like particles come in.

Enter the virus-like particle

A virus-like particle (VLP) is a virus shell with the dangerous parts removed. It keeps the outer membrane and the structural proteins that let a virus dock, enter a cell and release its cargo, but it carries no viral genome, so it cannot replicate or integrate. Picture a delivery van wearing a courier's uniform: it knows how to find the right door and hand over the parcel, but it has no way of building new vans.

The engineering in David Liu's lab turned this into something practical. In their engineered VLPs (eVLPs), the editor protein is fused to a retroviral structural protein called Gag. When the particle assembles and buds out of a producer cell, it traps the editor together with its guide RNA as a ready-made ribonucleoprotein, or RNP, which is just the protein and RNA already bound to each other. A built-in protease then cuts the editor free inside the particle. In the 2022 Cell paper, eVLPs delivering a base editor achieved about 63% editing of the Pcsk9 gene in mouse liver, and a single injection partially restored vision in a mouse model of Leber congenital amaurosis.

So a VLP borrows from both worlds. It has the viral talent for getting into cells and releasing cargo, and the LNP-like virtue of leaving no genetic footprint.

The coat decides the address

The protein on the surface of the particle decides which cells it can enter. VSV-G (vesicular stomatitis virus glycoprotein) is a favourite because it is superb at binding cells and triggering fusion with the endosome. But its tropism, meaning the range of cells it can enter, is very broad. That is brilliant in a lab dish and a headache in medicine, where you want a particle that goes to a chosen cell and not everywhere.

The fix is a two-step trick. First, mutate VSV-G so that it no longer binds its natural receptors while keeping its ability to fuse, giving a version called VSVGmut. Second, display an antibody fragment on the particle surface, a single-chain variable fragment (scFv), to supply the address. The glycoprotein opens the door, and the antibody fragment tells the particle which door to knock on.

Jennifer Doudna's group, with Jennifer Hamilton as first author, built on this idea with enveloped delivery vehicles (EDVs). In a culture test with a mix of cells that did and did not carry the B cell marker CD19, CD19-targeted EDVs carrying Cas9 knocked out the B2M gene in up to 74% of target cells, with little to no editing in bystander cells. In mice with a humanised immune system, EDVs went on to generate gene-edited CAR T cells in vivo, with no off-target delivery to liver hepatocytes. That last detail is what caught my eye, because the liver is exactly where most delivery vehicles end up. Hamilton and Doudna later co-founded Azalea Therapeutics, whose lead program is a CD19-targeted in vivo CAR T therapy heading toward the clinic.

Where do VLPs stand in the clinic?

As of November 2026, and to my knowledge, no VLP-delivered nuclease or base editor has been given to a patient. The closest example is Ensoma's EN-374, which uses VLPs to deliver a working copy of the CYBB gene to blood stem cells inside the body, for X-linked chronic granulomatous disease. That is gene addition rather than gene editing, but it is billed as the first in vivo blood stem cell gene insertion therapy in the clinic, and the first participant's safety data were presented in May 2026. If it stays safe, it tells regulators something useful about VLPs in general.

An old idea with a new job

VLPs were doing medicine long before CRISPR. The hepatitis B vaccine, licensed in 1986, is built from hepatitis B surface antigen that assembles into virus-like particles with no genome inside, and the HPV vaccines later followed the same logic. The immune system sees a convincing virus shape, builds memory, and nothing infectious was ever in the vial. What is new is not the particle. It is the idea of loading it with an editor and putting an address label on the outside.

What I am watching

I like this story because it is gloriously unfinished. Much of the current work is careful tinkering: the linkers that hold the antibody fragment on the surface, the glycoprotein mutations, the way the editor is packed inside. Three things will tell us whether VLPs grow up. First, clinical data from a VLP-delivered editor. Second, whether the targeting holds up in larger animals and not just mice. Third, whether production, which relies on transfected producer cells much like lentiviral manufacturing, can scale.

If they do, the next question will not be whether we can edit a gene, but which cell we choose to send the courier to.

Watch the video

This post is adapted from the third video in Kevin Curran's gene editing delivery series, with corrections and updates built in: https://www.youtube.com/watch?v=97EpLp1Crf8

Related content

• CRISPR-Cas9 basics: guide RNA, PAM and the sickle cell example
• Specificity vs sensitivity in CRISPR diagnostics: cry wolf, miss wolf
• Treatments that came from CRISPR

References

Banskota, S., Raguram, A., Suh, S., ... Liu, D. R. (2022). Engineered virus-like particles for efficient in vivo delivery of therapeutic proteins. Cell, 185(2), 250-265.e16. https://doi.org/10.1016/j.cell.2021.12.021

Ensoma. (2026, May). Ensoma to present clinical safety data from first participant dosed with in vivo HSC engineering therapy at ASGCT Annual Meeting [Press release]. https://www.biospace.com/press-releases/ensoma-to-present-clinical-safety-data-from-first-participant-dosed-with-in-vivo-hsc-engineering-therapy-at-asgct-annual-meeting

Hamilton, J. R., Chen, E., Perez, B. S., Sandoval Espinoza, C. R., Kang, M. H., Trinidad, M., ... Doudna, J. A. (2024). In vivo human T cell engineering with enveloped delivery vehicles. Nature Biotechnology, 42(11), 1684-1692. https://doi.org/10.1038/s41587-023-02085-z

Intellia Therapeutics. (2026, January 27). Intellia Therapeutics announces FDA lift of clinical hold on MAGNITUDE-2 Phase 3 clinical trial in ATTRv-PN [Press release]. https://ir.intelliatx.com/node/12431

Intellia Therapeutics. (2026, March 2). Intellia Therapeutics announces FDA lift of clinical hold on MAGNITUDE Phase 3 clinical trial in ATTR-CM [Press release]. https://www.sec.gov/Archives/edgar/data/1652130/000119312526084309/d101756dex992.htm

LAMP Primer Design Explained: Six Primers, One Temperature

Have you ever designed a PCR primer pair in five minutes and felt rather proud of yourself? Enjoy that feeling. LAMP is about to take it away.

LAMP stands for loop-mediated isothermal amplification, and it asks you to design six primers for one target, not two. Yet people love it, because it needs no thermocycler, gives positives in under 30 minutes and works in places where a PCR machine would never survive the trip. So how do you design those six primers without losing your mind? Let's go through it, term by term.

A short story about a very stubborn enzyme

In 2000, Tsugunori Notomi and colleagues at Eiken Chemical in Japan published a method that copied DNA at a single constant temperature (Notomi et al., 2000). That is what isothermal means: one temperature for the whole reaction, usually somewhere between 60 and 72 °C. A heat block or a water bath will do.

PCR needs a thermocycler because every cycle starts with denaturation, which means heating the DNA to about 95 °C so the two strands come apart. LAMP skips this step, and the reason is its enzyme.

LAMP uses a polymerase with strong strand displacement activity, such as Bst polymerase from Geobacillus stearothermophilus. Most polymerases stop, or chew up the DNA, when they meet double-stranded DNA ahead of them. A strand-displacing polymerase behaves like a snowplow. It keeps going and pushes the downstream strand off the template as it copies. In practice, that is why no heat cycling is needed, and why every new copy releases a single strand that can be primed again.

How does the reaction start without heating? At 60 to 65 °C the ends of double-stranded DNA keep opening and closing for a moment, which is called "breathing". An inner primer can slip into an open end, a step called strand invasion, and the polymerase takes over from there.

Meet the six primers

Here is the cast. Each name looks like alphabet soup until you see what the primer actually does.

• F3 and B3 (outer primers): the forward and backward primers sitting furthest out on the target. Their job is to displace the strand made by the inner primers, which releases it for the next step. They are used at the lowest concentration.
• FIP and BIP (forward and backward inner primers): the stars of the show. They are long, 40 to 50 nucleotides, because each one is two pieces joined together. One piece (F2 or B2) matches the target. The other (F1c or B1c) is the reverse complement of a region further inside the target. The "c" means complementary. That second piece is what lets the product fold back on itself.
• LF and LB (loop primers): optional, but I would not skip them. They bind the single-stranded loops between F2 and F1c, and between B2 and B1c. They give the polymerase more places to start, so the reaction speeds up. The webinar transcript calls them "blue primers", which is a speech-to-text mishearing of "loop primers". Nagamine and colleagues showed that loop primers accelerate the reaction (Nagamine et al., 2002).

Together, these primers cover a stretch of target called the amplicon footprint. For six primers you need about 280 base pairs or more, and you must leave a gap between F2 and F1c so that the loop primer fits there.

Why LAMP products look like a bowl of spaghetti

Once the inner primer has extended and the outer primer has displaced the new strand, that strand has complementary sequences at both ends. So it folds into a stem-loop, also called a dumbbell: each end pairs with a region within the same strand. The folded 3' end then acts as its own primer, and the strand extends itself with no new primer needed. This is the engine of LAMP.

The result is concatemers: long molecules made of the target unit repeated end to end, built from inverted repeats (a sequence followed by its reverse complement, which is what allows the folding). On an agarose gel they appear as a ladder or smear of bands. That pattern is the signature of a true LAMP product, and it is a useful way to tell it apart from background.

The webinar describes this growth as "factorial". Most sources call it exponential. Whichever word you prefer, the practical point is the same: you get an enormous amount of DNA, in micrograms rather than the milligrams the transcript suggests, which is why you can sometimes see the result by eye.

The design rules, with the reasons

None of these rules is arbitrary, so I will give the reason for each.

Primer length and orientation. Most primers (F3, B3, F2, B2, F1c, B1c, loops) are 15 to 25 nucleotides, much like a PCR primer. Primers are always written, ordered and synthesized 5' to 3', so you can paste the sequence straight into an order form.

GC content: 40 to 60%. G-C pairs have three hydrogen bonds and A-T pairs have two, so GC-rich DNA binds more tightly. Too little GC and the primer binds weakly. Too much encourages secondary structures.

Tm (melting temperature). The temperature at which half of a DNA duplex has separated, so it tells you how tightly a primer binds. Primers in a pair (F3 and B3, F2 and B2, and so on) should be within 5 °C of each other, so both work equally well at the reaction temperature.

No runs of three or more identical bases. Stretches like AAA, or dinucleotide repeats like ATAT, cause mispriming (binding slightly off target) and make the primer harder to synthesize and to clean up by HPLC (a purification method that removes truncated, failed copies of the oligo).

Avoid secondary structure. This is the big one for LAMP, because six primers means a lot of possible partners. The three troublemakers are:

• Hairpin: one primer folds back and pairs with itself.
• Self-dimer: two copies of the same primer stick together.
• Primer-dimer (cross-dimer): two different primers stick together. This is the main cause of non-specific background.

ΔG (Gibbs free energy) is how we score these. A more negative ΔG means a more stable structure. Here the webinar's own explanation tangled itself, and the speaker corrected the slide in the Q&A, so here is the version to remember:

• For dimers and hairpins, you want weak, unstable structures, so a ΔG that is not very negative. The webinar's guide is about −3.5 kcal/mol as the limit.
• For the ends of primers, you want the opposite: stable ends, with a ΔG of about −4 kcal/mol or lower, so the primer reliably starts and holds an extension. The ends that matter are the 3' end of F2, B2, F3 and B3, and the 5' end of F1c and B1c.

Poly-T linkers. A short run of T bases between the two halves of FIP or BIP gives the loop more flexibility. The speaker says it has been reported to add speed, but he has not been able to prove it himself. Treat it as something to try, not a rule.

Choosing the target (this is where false results are born)

A false negative is a real target you miss, and a false positive is a signal from something that is not your target. Both start at target selection.

Pick a conserved region, a stretch of sequence identical across all the strains you want to detect, found by multiple sequence alignment (lining up many strain sequences to see where they agree). The webinar showed seven Ebola strains with a short, perfectly conserved region, and eleven Drosophila strains too variable to design against.

If variation is unavoidable you have two options. You can use degenerate nucleotides, bases that stand for a mix using IUPAC codes (R means A or G, Y means C or T), so one primer covers both variants. Or you can design two primer sets for two populations.

Then check cross-reactivity, meaning your primers amplifying a related organism. BLAST, the database search tool, is a good start, but it only knows what is in its database. The speaker's advice is to also pick real near-neighbour organisms that could turn up in your sample, and test against them in the lab.

Software: PrimerExplorer and its quirks

The free workhorse is PrimerExplorer v5, from Eiken. It has real limits: sequences up to 2,000 base pairs, only A, C, G and T (so no degenerate codes, which you must add manually later), and results in HTML only. The most surprising quirk is that it designs the four main primers first, and the loop primers in a second, separate run.

A few practical points from the walkthrough:

1. Upload your conserved sequence and highlight any known mutation sites.
2. Default mode avoids mutations completely. Common mode buries them where a mismatch is tolerated (the 5' ends of F3, B3, F2, B2 and the 3' ends of F1c, B1c), so wild type and mutant both amplify. Specific mode places the mutation at the 3' end of F2, B2, F3, B3 (and the 5' end of F1c, B1c), so only the mutant amplifies, because a 3' mismatch blocks extension. Do not expect a perfect light switch. Test it.
3. From the five sets generated, choose one with the highest ΔG for dimerization, check the end stability, and make sure there is enough space between F2 and F1c for loops.
4. Save the set, reopen PrimerExplorer, load it, and run the loop primer design in the gaps.

If you get thousands of designs, tighten the settings. If you get none, widen the distances, or go back and find another conserved region. The software sorts your sequence into a GC-rich, AT-rich or normal bucket and sets defaults accordingly, which you can change under detailed settings.

A paid alternative from Premier Biosoft does alignments, interprets BLAST results automatically and calculates energies for multiplexing, which means detecting several targets in one tube.

The tip I found most useful: match the Tm to your enzyme

PrimerExplorer's defaults were built around Bst, which prefers 60 to 65 °C. Lucigen's LavaLAMP enzyme prefers 68 to 74 °C. The webinar showed what happens when you design with Bst defaults and run at the higher temperature: some published primer sets worked beautifully, and others gave zero separation between positives and negatives.

So they raised the design Tm set-point. The result was faster, more consistent time to result (TTR), which is when a reaction first crosses the detection threshold, and a larger gap between positives and negatives. Increasing the Tm by up to 6 °C helped. Some designs only allowed 4 or 5 °C, because the primers became too long, but even that improved things. Note that this data comes from one enzyme and the vendor's own experiments, so it is a lesson in thinking, not a universal law.

Optimization: screen first, then be clever

LAMP design is less predictable than PCR. The speaker's advice is blunt: design three to five sets, and be ready to go to six, because many will fail. His example showed three sets: one total failure, one with fast but noisy background, and one that gave positives in under ten minutes with almost no background. And do this with software, not by hand.

Then tune the reaction:

• Primer ratio: a common starting point is 1:4:8 (outer : inner : loop). If one primer causes trouble, lower its concentration.
• Reaction temperature: the biggest lever.
• Mg²⁺ (magnesium): the cofactor the polymerase needs. More speeds the reaction but also raises background. Lowering it is a fix for non-specific amplification.
• Additives: betaine reduces secondary structure and Triton is a detergent, but check whether they survive lyophilization, the freeze-drying that lets a reaction be shipped as a stable pellet for field use.

For serious optimization, use design of experiments (DoE), which changes several factors together in a planned way instead of one at a time. Start with a screening design to see which factors matter, then run a factorial design to see interactions, since the best Mg²⁺ level depends on temperature and enzyme amount.

Detection and its catch

You can read LAMP in real time with a fluorescent dye that lights up when bound to double-stranded DNA, which shows exactly when positives and background rise. You can also read it at the end by colour, using a pH-sensitive dye that changes colour as amplification acidifies the reaction (Tanner et al., 2015), or by turbidity, since the by-product pyrophosphate binds magnesium and forms a white precipitate (Mori et al., 2001).

Here is the catch. You cannot use hydrolysis (TaqMan) probes, because Bst lacks the exonuclease activity that cleaves them. So most LAMP detection is not sequence-specific, and primer-dimer background looks exactly like real signal. That is why no-template controls and gel checks matter so much.

The takeaway

LAMP gives you speed and simplicity at the bench, and charges you in the design stage. Use a conserved target, use software, make several designs, match your Tm to your enzyme, and let real-time data decide. A practical protocol is covered in detail in the Tomita et al. paper (Tomita et al., 2008).

Watch the video

This post is adapted from the Lucigen webinar on LAMP primer design and assay optimization: https://www.youtube.com/watch?v=GJkvQqDufh0

References

Mori, Y., Nagamine, K., Tomita, N., & Notomi, T. (2001). Detection of loop-mediated isothermal amplification reaction by turbidity derived from magnesium pyrophosphate formation. Biochemical and Biophysical Research Communications, 289(1), 150-154. https://doi.org/10.1006/bbrc.2001.5921

Nagamine, K., Hase, T., & Notomi, T. (2002). Accelerated reaction by loop-mediated isothermal amplification using loop primers. Molecular and Cellular Probes, 16(3), 223-229. https://doi.org/10.1006/mcpr.2002.0415

Notomi, T., Okayama, H., Masubuchi, H., Yonekawa, T., Watanabe, K., Amino, N., & Hase, T. (2000). Loop-mediated isothermal amplification of DNA. Nucleic Acids Research, 28(12), e63. https://doi.org/10.1093/nar/28.12.e63

Tanner, N. A., Zhang, Y., & Evans, T. C. (2015). Visual detection of isothermal nucleic acid amplification using pH-sensitive dyes. BioTechniques, 58(2), 59-68. https://doi.org/10.2144/000114253

Tomita, N., Mori, Y., Kanda, H., & Notomi, T. (2008). Loop-mediated isothermal amplification (LAMP) of gene sequences and simple visual detection of products. Nature Protocols, 3(5), 877-882. https://doi.org/10.1038/nprot.2008.57

Related Content

• Specificity vs sensitivity in CRISPR diagnostics: cry wolf, miss the wolf
• Sequencing platforms series: Sanger, Illumina, Nanopore, PacBio, Ion Torrent
• RNA extraction from Tempus blood tubes

 

Cry Wolf or Miss the Wolf? Specificity and Sensitivity in CRISPR Diagnostics

I have run enough diagnostic assays to know one thing for certain: a result on the screen is never the whole story. Somewhere behind every positive and every negative sits the question of whether the test got it right.

Think of a village watching for a wolf. Every time the village relies on a lookout, that lookout gets two chances to be wrong. The lookout can cry wolf, or the lookout can miss the wolf. A diagnostic test is no different.

Crying wolf: the false positive

There was no wolf, but the test said it saw one. That is a false positive, and the property that protects you from it is called specificity.

A highly specific test only raises the alarm when the target is truly there. A poorly specific one panics over shadows. In a clinic, that can mean unnecessary treatment, anxiety, and a lot of wasted resources.

Missing the wolf: the false negative

Now the opposite. The wolf was right there, but the test did not detect it. That is a false negative, and the property that protects you from it is called sensitivity.

This is the quieter failure. Nobody panics, which is exactly the problem. A patient who is infected goes home reassured, and the infection moves on.

How CRISPR helps with both

In a recent interview, Feng Zhang, one of the pioneers of CRISPR technology, explained that CRISPR-based diagnostics can improve both metrics at once. His reasoning is worth unpacking, because the two improvements come from two different steps.

Better sensitivity (fewer missed wolves). The sample is first concentrated and then amplified, usually by PCR or an isothermal method. CRISPR then adds a second layer of amplification on top, called collateral activity. Once the CRISPR enzyme recognises its target, it starts cutting nearby nucleic acids indiscriminately, including a reporter molecule that gives off a signal. The result is a stronger signal from the same amount of starting material. It is a bit like turning up the volume on a whisper.

In the SHERLOCK platform that Zhang’s group developed, this enzyme is Cas13, which targets RNA (Gootenberg et al., 2017). Zhang does not name the enzyme in this interview, so that detail comes from the original paper.

Better specificity (fewer false alarms). Amplification already checks for the viral sequence once. CRISPR then checks again, because the guide RNA has to recognise the sequence before the signal is produced. A false result has to survive both checks to count. It is the friend who asks “are you sure?” before you hit send.

Why this matters outside the lab

Zhang points out that CRISPR diagnostics are being developed for point-of-care and at-home use. For anyone working in places where a full molecular lab is not around the corner, that is the part that excites me most. A test that is both sensitive and specific, and that does not need a lab full of equipment, changes who gets an answer and how quickly.

Quick summary

The wolf	What it means	Technical term	How CRISPR helps
Cry wolf (Specificity)	It was not there, but the test said it saw it	False positive (specificity problem)	Target checked twice: amplification, then CRISPR detection
Miss the wolf (Sensitivity)	It was there, but the test did not detect it	False negative (sensitivity problem)	Amplify first, then Cas13 collateral cutting boosts the signal

Related Content

CRISPR-Cas9 Explainer: How It Works, Guide RNA and PAM: https://adwoabiotech.blogspot.com/2026/11/crispr-cas9-explained-guide-rna-pam.html

Reference

Gootenberg, J. S., Abudayyeh, O. O., Lee, J. W., Essletzbichler, P., Dy, A. J., Joung, J., Verdine, V., Donghia, N., Daringer, N. M., Freije, C. A., Myhrvold, C., Bhattacharyya, R. P., Livny, J., Regev, A., Koonin, E. V., Hung, D. T., Sabeti, P. C., Collins, J. J., & Zhang, F. (2017). Nucleic acid detection with CRISPR-Cas13a/C2c2. Science, 356(6336), 438–442. https://doi.org/10.1126/science.aam9321