After I finished 8th grade, I started a “job” for a professor researching PCR techniques. I say “job” because I wasn’t really expected to do anything productive; it was more, charity in the form of work history.
Recently, I was thinking back on how PCR and my thinking have changed since then.
The polymerase chain reaction (PCR) is a method widely used to make millions to billions of copies of a specific DNA sample rapidly, allowing scientists to amplify a very small sample of DNA (or a part of it) sufficiently to enable detailed study.
Specifically, it copies a region of DNA with segments at the start + end that match some added DNA pieces made chemically. Mostly, this is used to detect if certain DNA is present in a sample.
how PCR works
First, you need to get DNA out of some cells. This can be done with chemicals or ultrasound.
Then, you need to separate DNA from other stuff. This can be done by adding beads that DNA binds to, washing the beads, and adding some chemical that releases the DNA.
Now, you can start the PCR. You mix together:
the DNA
primers: short synthesized DNA sequences that bind to the start and end of your target sequence
nucleoside triphosphates to make DNA from
a polymerase: an enzyme that binds to a double-stranded region and extends it into a single-strand region
Then:
Heat the DNA until it “melts” (the strands separate).
Cool the solution so primers can bind to the released single strands.
Wait for the polymerase to extend the primers.
Repeat the process.
Obviously, a polymerase that can survive high enough temperatures to melt DNA is needed. So, discovery of Taq polymerase was key for making PCR possible.
better enzymes
These days, there are better enzymes than Taq, which go faster and have lower error rates. Notably, KOD and Q5 polymerase. A lot of labs still seem to be using outdated polymerase choices.
real-time PCR
There are some fluorescent dyes that bind to double-stranded DNA and change their fluorescence when they do. If we add such dye to a PCR solution, we can graph DNA strand separation vs temperature. Different DNA sequences melt at slightly different temperatures, so with good calibration, this can detect mutations in a known DNA sequence.
multiplex PCR
Instead of adding a dye that binds to DNA, we can add fluorescent dye to primers that gets cleaved off by the polymerase, increasing its fluorescence. Now, we can add several primer pairs for different sequences, each labeled with different dyes, and what color is seen indicates what sequence is present.
However, due to overlap between different dye colors, this is only practical for up to about 6 targets.
Obviously, you could do 2 PCR reactions, each with 36 primers, and determine which sequence is present from a single color from each reaction. And so on, with targets increasing exponentially with more reactions. But massively multiplex PCR is limited by non-specific primer binding and primer dimers.
There are other ways to indicate specific reactions, such as probes separate from the primers, but the differences aren’t important here.
PCR automation
Cepheid
Cepheid makes automated PCR test machines. There are disposable plastic cartridges; you put a sample in 1 chamber, the machine turns a rotary valve to control flow, and drives a central syringe to move fluids around. Here’s a video.
So, you spit in a cup, put the sample in a hole, run the machine, and an hour later you have a diagnosis from several possibilities, based on DNA or RNA. It’s hard to overstate how different that is from historical diagnosis of diseases.
SiTime
The Cepheid system seemed moderately clever, so I looked up the people involved, and noticed this guy. Kurt Petersen, also a founder of SiTime, which is a company I’d heard of.
Historically, oscillators use quartz because it doesn’t change much with temperature. The idea of SiTime was:
use lithography to make lots of tiny silicon resonators
measure the actual frequency of each resonator, and shift them digitally to the desired frequency
use thermistors to determine temperature and digitally compensate for temperature effects
As usual, accuracy improves when you average more oscillators, as sqrt(n). Anyway, I’ve heard SiTime is currently the best at designing such systems.
alternatives are possible
“Moderately clever” isn’t Hendrik Lorentz or the guy I learned chemistry from. I could probably find a design that avoids their patents without increasing costs. In fact, I think I’ll do that now.
...
Yep, it’s possible. Of course, you do need different machines for the different consumables.
BioFire
Another automated PCR system is BioFire’s FilmArray system. Because it’s more-multiplex than Cepheid’s system, they need 2 PCR stages, and primer-primer interactions are still a problem. But still, you can do 4x the targets as Cepheid for only 10x the cost. For some reason it hasn’t been as popular, but I guess that’s a mystery to be solved by future generations.
droplets
Suppose you want a very accurate value for how much of a target DNA sequence is in a sample.
If we split a PCR solution into lots of droplets in oil, and examine the droplets individually, we can see what fraction of droplets had a PCR reaction happen. That’s usually called digital droplet PCR, or ddPCR.
Another way to accomplish the same thing is to have a tray of tiny wells, such that liquid flows into the wells and is kept compartmentalized. Here’s a paper doing that.
mixed droplets
It’s obviously possible to:
make many different primer mixtures
make emulsions of water droplets in oil from each of them
mix the emulsions
use microfluidics to combine each primer droplet with a little bit of the sample DNA
do PCR on the emulsion
Is anybody doing that? I’m guessing it’s what “RainDance Technologies” is doing...yep, seems so.
Of course, if we re-use the microfluidic system and have even a tiny bit of contamination between runs, that ruins results. So, I reckon you either need very cheap microfluidic chips, or ones that can be sterilized real good before reuse. But that’s certainly possible; it’s just a manufacturing problem.
my thoughts at the time
Back then, while my “job” was about regular PCR, I was more interested in working on something else. My view was:
Testing for a single disease at a time is useful, but the future is either sequencing or massively parallel testing. Since I’m young, I should be thinking about the future, not just current methods.
My acquaintance Nava has a similar view now. Anyway, I wasn’t exactly wrong, but in retrospect, I was looking a bit too far forward. Which I suppose is a type of being wrong.
non-PCR interests
I’d recently learned about nanopore sequencing and SPR, and thought those were interesting.
nanopore sequencing
Since then, Oxford Nanopore sequencers have improved even faster than other methods, and are now a reasonable choice for sequencing DNA. (But even for single-molecule sequencing, I’ve heard the fluorescence-based approach of PacBio is generally better.)
Current nanopore sequencers are based on differences in ion flow around DNA depending on its bases. At the time, I thought plasmonic nanopore approaches would be better, but obviously that hasn’t been the case so far. That wasn’t exactly a dead end; people are still working on it, especially for protein sequencing, but it’s not something used in commercial products today. I guess it seemed like the error rate of the ion flow approach would be high, but as of a few years ago it was...yeah, pretty high actually, but if you repeat the process several times you can get good results. Of course current plasmonic approaches aren’t better, but they do still seem to have more room for improvement.
Why did I find nanopore approaches more appealing than something like Illumina?
Fragmenting DNA to reassamble the sequence from random segments seemed inelegant somehow.
Enzymes work with 1 strand of DNA, so why can’t we?
Illumina complex, make Grug brain hurty
surface plasmon resonance
SPR (Wikipedia) involves attaching receptors to a thin metal film, and then detecting binding to those receptors by effects on reflection of laser light off the other side of the metal film. Various companies sell SPR testing equipment today. The chips are consumables; here’s an example of a company selling them.
But those existing products are unrelated to why I thought SPR was interesting. My thought was, it should be possible to make an array of many different receptors on the metal film, and then detect many different target molecules with a single test. So, is anybody working on that? Yes; here’s a recent video from a startup called Carterra. I don’t see any problems without simple solutions, but they’ve been working on this for 10 years already so presumably there were some difficulties.
electrical DNA detection
While working at that lab, I had the following thought:
The conformation of DNA should depend on the sequence. That should affect its response to high-frequency electric fields. If you do electric testing during PCR, then maybe you could get some sequence-specific information by the change in properties during a later cycle. If necessary, you could use a slow polymerase.
So, when I later talked to the professor running the lab, I said:
me: Hey, here’s this idea I’ve been thinking about.
prof: Interesting. Are you going to try it then?
me: Is that...a project you want to pursue here, then?
prof: It might be a good project for you.
me: If you don’t see any problems, I’d be happy to discuss it in more detail with more people when you’re available.
prof: Just make it work, and then you won’t have to convince me it’s good.
me: I...don’t have the resources to do that on my own; you’re the decision-maker here.
prof: We, uh, already have enough research projects, but you should definitely try to work on ideas like that on your own.
me: …I see.
In retrospect, was my idea something that lab should’ve been working on? Working on droplet PCR techniques probably would’ve been better, but on the other hand, the main thrust of their research was basically a dead end and its goal wasn’t necessary.
papers on EIS of DNA
Electric impedance spectroscopy (EIS) involves measuring current with AC voltage, for multiple frequencies, and detecting phase of current relative to voltage.
Here’s a 2020 paper doing EIS on PCR solutions after different numbers of cycles. It finds there’s a clearly detectable signal! There’s a bigger effect for the imaginary (time delay) than the real (resistance) component of signals. They used circuit boards with intermeshing comb-like electrodes to get a bigger signal.
It’d be easy to say “the idea worked, that’s gratifying” and conclude things here. But taking a look at that graph of delay vs PCR cycle, apparently there’s a bigger change from the earlier PCR cycles, despite the increase in DNA being less. And the lower the frequency, the more of the change happens from earlier cycles. So, that must be some kind of surface effect: DNA sticking to a positively charged surface and affecting capacitance but with a slight delay because DNA is big. And that means the effect will depend on length, but not significantly on sequence.
Looking at some other papers validates that conclusion; actually, most papers looking at EIS of DNA used modified surfaces. If you bind some DNA sequence to a metal surface, and then its complement binds to that, you can observe that binding from its electrical effects. There’s a change in capacitance, and if you add some conductive anions, having more (negative) DNA repels those and reduces conductivity. Using that approach, people have been able to detect specific DNA sequences and single mutations in them. The main problem seems to be that you have to bind specific DNA to these metal surfaces, which is the same problem SPR has. Still, it’s a topic with ongoing research; here’s a 2020 survey paper.
electrochemical biosensors
Electrochemical biosensors are widely used today, less than PCR but more than SPR. Some of them are very small, the size of a USB drive. The sensor chips in those, like SPR chips, are disposable.
The approach I described above is sometimes called “unlabeled electrochemical biosensors”, because they don’t use “labels” in solution that bind to the target molecules to increase signal. Here’s a survey describing various labels. I think most electrochemical sensors use labels. Needing to add an additional substance might seem like a disadvantage, but changing the detection target by adding something to a liquid is often easier than getting a different target-specific chip. On the other hand, that means you can only detect 1 target at a time, while unlabeled sensors could use multiple regions detecting different targets.
isothermal DNA amplification
PCR uses temperature cycling, but if you use a polymerase that displaces bound DNA in front of it, you can do DNA amplification at a constant temperature. The main approach is LAMP; here’s a short video and here’s wikipedia.
LAMP is faster and can sometimes be done with simpler devices. PCR is better for detecting small amounts of DNA, is easier to do multiplex detection with, and gives more consistent indications of initial quantity. Detection of DNA with LAMP is mostly done with non-specific dyes...which is why I’m mentioning LAMP here.
If you use a parallel droplet approach, with a single dye to indicate amplified DNA plus a fluorescent “barcode” to indicate droplet type, then the difficulty of multiplex LAMP doesn’t matter. The same is true if you use a SPR chip with a pattern of many DNA oligomers on its surface. So, if those approaches are used, LAMP could be attractive.
PCR retrospective
Link post
my history
After I finished 8th grade, I started a “job” for a professor researching PCR techniques. I say “job” because I wasn’t really expected to do anything productive; it was more, charity in the form of work history.
Recently, I was thinking back on how PCR and my thinking have changed since then.
what PCR does
Wikipedia says:
Specifically, it copies a region of DNA with segments at the start + end that match some added DNA pieces made chemically. Mostly, this is used to detect if certain DNA is present in a sample.
how PCR works
First, you need to get DNA out of some cells. This can be done with chemicals or ultrasound.
Then, you need to separate DNA from other stuff. This can be done by adding beads that DNA binds to, washing the beads, and adding some chemical that releases the DNA.
Now, you can start the PCR. You mix together:
the DNA
primers: short synthesized DNA sequences that bind to the start and end of your target sequence
nucleoside triphosphates to make DNA from
a polymerase: an enzyme that binds to a double-stranded region and extends it into a single-strand region
Then:
Heat the DNA until it “melts” (the strands separate).
Cool the solution so primers can bind to the released single strands.
Wait for the polymerase to extend the primers.
Repeat the process.
Obviously, a polymerase that can survive high enough temperatures to melt DNA is needed. So, discovery of Taq polymerase was key for making PCR possible.
better enzymes
These days, there are better enzymes than Taq, which go faster and have lower error rates. Notably, KOD and Q5 polymerase. A lot of labs still seem to be using outdated polymerase choices.
real-time PCR
There are some fluorescent dyes that bind to double-stranded DNA and change their fluorescence when they do. If we add such dye to a PCR solution, we can graph DNA strand separation vs temperature. Different DNA sequences melt at slightly different temperatures, so with good calibration, this can detect mutations in a known DNA sequence.
multiplex PCR
Instead of adding a dye that binds to DNA, we can add fluorescent dye to primers that gets cleaved off by the polymerase, increasing its fluorescence. Now, we can add several primer pairs for different sequences, each labeled with different dyes, and what color is seen indicates what sequence is present.
However, due to overlap between different dye colors, this is only practical for up to about 6 targets.
Obviously, you could do 2 PCR reactions, each with 36 primers, and determine which sequence is present from a single color from each reaction. And so on, with targets increasing exponentially with more reactions. But massively multiplex PCR is limited by non-specific primer binding and primer dimers.
There are other ways to indicate specific reactions, such as probes separate from the primers, but the differences aren’t important here.
PCR automation
Cepheid
Cepheid makes automated PCR test machines. There are disposable plastic cartridges; you put a sample in 1 chamber, the machine turns a rotary valve to control flow, and drives a central syringe to move fluids around. Here’s a video.
So, you spit in a cup, put the sample in a hole, run the machine, and an hour later you have a diagnosis from several possibilities, based on DNA or RNA. It’s hard to overstate how different that is from historical diagnosis of diseases.
SiTime
The Cepheid system seemed moderately clever, so I looked up the people involved, and noticed this guy. Kurt Petersen, also a founder of SiTime, which is a company I’d heard of.
Historically, oscillators use quartz because it doesn’t change much with temperature. The idea of SiTime was:
use lithography to make lots of tiny silicon resonators
measure the actual frequency of each resonator, and shift them digitally to the desired frequency
use thermistors to determine temperature and digitally compensate for temperature effects
As usual, accuracy improves when you average more oscillators, as sqrt(n). Anyway, I’ve heard SiTime is currently the best at designing such systems.
alternatives are possible
“Moderately clever” isn’t Hendrik Lorentz or the guy I learned chemistry from. I could probably find a design that avoids their patents without increasing costs. In fact, I think I’ll do that now.
...
Yep, it’s possible. Of course, you do need different machines for the different consumables.
BioFire
Another automated PCR system is BioFire’s FilmArray system. Because it’s more-multiplex than Cepheid’s system, they need 2 PCR stages, and primer-primer interactions are still a problem. But still, you can do 4x the targets as Cepheid for only 10x the cost. For some reason it hasn’t been as popular, but I guess that’s a mystery to be solved by future generations.
droplets
Suppose you want a very accurate value for how much of a target DNA sequence is in a sample.
If we split a PCR solution into lots of droplets in oil, and examine the droplets individually, we can see what fraction of droplets had a PCR reaction happen. That’s usually called digital droplet PCR, or ddPCR.
Another way to accomplish the same thing is to have a tray of tiny wells, such that liquid flows into the wells and is kept compartmentalized. Here’s a paper doing that.
mixed droplets
It’s obviously possible to:
make many different primer mixtures
make emulsions of water droplets in oil from each of them
mix the emulsions
use microfluidics to combine each primer droplet with a little bit of the sample DNA
do PCR on the emulsion
Is anybody doing that? I’m guessing it’s what “RainDance Technologies” is doing...yep, seems so.
Of course, if we re-use the microfluidic system and have even a tiny bit of contamination between runs, that ruins results. So, I reckon you either need very cheap microfluidic chips, or ones that can be sterilized real good before reuse. But that’s certainly possible; it’s just a manufacturing problem.
my thoughts at the time
Back then, while my “job” was about regular PCR, I was more interested in working on something else. My view was:
My acquaintance Nava has a similar view now. Anyway, I wasn’t exactly wrong, but in retrospect, I was looking a bit too far forward. Which I suppose is a type of being wrong.
non-PCR interests
I’d recently learned about nanopore sequencing and SPR, and thought those were interesting.
nanopore sequencing
Since then, Oxford Nanopore sequencers have improved even faster than other methods, and are now a reasonable choice for sequencing DNA. (But even for single-molecule sequencing, I’ve heard the fluorescence-based approach of PacBio is generally better.)
Current nanopore sequencers are based on differences in ion flow around DNA depending on its bases. At the time, I thought plasmonic nanopore approaches would be better, but obviously that hasn’t been the case so far. That wasn’t exactly a dead end; people are still working on it, especially for protein sequencing, but it’s not something used in commercial products today. I guess it seemed like the error rate of the ion flow approach would be high, but as of a few years ago it was...yeah, pretty high actually, but if you repeat the process several times you can get good results. Of course current plasmonic approaches aren’t better, but they do still seem to have more room for improvement.
Why did I find nanopore approaches more appealing than something like Illumina?
Fragmenting DNA to reassamble the sequence from random segments seemed inelegant somehow.
Enzymes work with 1 strand of DNA, so why can’t we?
Illumina complex, make Grug brain hurty
surface plasmon resonance
SPR (Wikipedia) involves attaching receptors to a thin metal film, and then detecting binding to those receptors by effects on reflection of laser light off the other side of the metal film. Various companies sell SPR testing equipment today. The chips are consumables; here’s an example of a company selling them.
But those existing products are unrelated to why I thought SPR was interesting. My thought was, it should be possible to make an array of many different receptors on the metal film, and then detect many different target molecules with a single test. So, is anybody working on that? Yes; here’s a recent video from a startup called Carterra. I don’t see any problems without simple solutions, but they’ve been working on this for 10 years already so presumably there were some difficulties.
electrical DNA detection
While working at that lab, I had the following thought:
So, when I later talked to the professor running the lab, I said:
In retrospect, was my idea something that lab should’ve been working on? Working on droplet PCR techniques probably would’ve been better, but on the other hand, the main thrust of their research was basically a dead end and its goal wasn’t necessary.
papers on EIS of DNA
Electric impedance spectroscopy (EIS) involves measuring current with AC voltage, for multiple frequencies, and detecting phase of current relative to voltage.
Here’s a 2020 paper doing EIS on PCR solutions after different numbers of cycles. It finds there’s a clearly detectable signal! There’s a bigger effect for the imaginary (time delay) than the real (resistance) component of signals. They used circuit boards with intermeshing comb-like electrodes to get a bigger signal.
It’d be easy to say “the idea worked, that’s gratifying” and conclude things here. But taking a look at that graph of delay vs PCR cycle, apparently there’s a bigger change from the earlier PCR cycles, despite the increase in DNA being less. And the lower the frequency, the more of the change happens from earlier cycles. So, that must be some kind of surface effect: DNA sticking to a positively charged surface and affecting capacitance but with a slight delay because DNA is big. And that means the effect will depend on length, but not significantly on sequence.
Looking at some other papers validates that conclusion; actually, most papers looking at EIS of DNA used modified surfaces. If you bind some DNA sequence to a metal surface, and then its complement binds to that, you can observe that binding from its electrical effects. There’s a change in capacitance, and if you add some conductive anions, having more (negative) DNA repels those and reduces conductivity. Using that approach, people have been able to detect specific DNA sequences and single mutations in them. The main problem seems to be that you have to bind specific DNA to these metal surfaces, which is the same problem SPR has. Still, it’s a topic with ongoing research; here’s a 2020 survey paper.
electrochemical biosensors
Electrochemical biosensors are widely used today, less than PCR but more than SPR. Some of them are very small, the size of a USB drive. The sensor chips in those, like SPR chips, are disposable.
The approach I described above is sometimes called “unlabeled electrochemical biosensors”, because they don’t use “labels” in solution that bind to the target molecules to increase signal. Here’s a survey describing various labels. I think most electrochemical sensors use labels. Needing to add an additional substance might seem like a disadvantage, but changing the detection target by adding something to a liquid is often easier than getting a different target-specific chip. On the other hand, that means you can only detect 1 target at a time, while unlabeled sensors could use multiple regions detecting different targets.
isothermal DNA amplification
PCR uses temperature cycling, but if you use a polymerase that displaces bound DNA in front of it, you can do DNA amplification at a constant temperature. The main approach is LAMP; here’s a short video and here’s wikipedia.
LAMP is faster and can sometimes be done with simpler devices. PCR is better for detecting small amounts of DNA, is easier to do multiplex detection with, and gives more consistent indications of initial quantity. Detection of DNA with LAMP is mostly done with non-specific dyes...which is why I’m mentioning LAMP here.
If you use a parallel droplet approach, with a single dye to indicate amplified DNA plus a fluorescent “barcode” to indicate droplet type, then the difficulty of multiplex LAMP doesn’t matter. The same is true if you use a SPR chip with a pattern of many DNA oligomers on its surface. So, if those approaches are used, LAMP could be attractive.