We’re planning on submitting some dual indexed, paired read, amplicon sequencing samples, and depending on how many we have, we may submit for sequencing on a Miseq or Nextseq. Since this is all custom (and we generated the indices), I had to figure out how exactly how the process plays out for the two instruments to see what we need to the sample sheet for demultiplexing. I figure I’d illustrate it out for people in the lab so they understand the process as well.

The common steps

Everything starts with bridge amplification, where both the forward and reverse strands are physically bound to the flow cell by their 5′ ends. I’m denote where the DNA strands are physically bound to the flow cell with the grey circles on the ends.

Next is denaturing the strands so they are no longer bound, and also cleaving the strand that is bound to the flow cell via the p5 sequence. The result is something that looks like this.

Now the single stranded molecule is ready for sequencing from the read 1 primer. The light blue box is showing the nucleotides we’ll be reading in this particular library.

Once that read is done, next is reading the first index.

We didn’t do a ton of dual indexing of the libraries in my postdoc (and since Jason used to run all of the kits anyway), I didn’t really need to know these steps, but I’ve had to figure them out now setting everything up in my own lab. I’ll go over what happens with the Nextseq first, since that’s conceptually a bit easier. Here’s what the Illumina docs say.

The Nextseq way

So this setup allows for dual indexing read off of two custom primers. So for this specific library, this means that the complementary sequence is going to be synthesized, making a double-stranded bridged molecule again.

And then after everything is denatured and the original template strand removed (presumably by cleaving at the p7 sequence this time), the second index can now be sequenced.

Followed by sequencing of the second read.

The Miseq way

Looking at what Illumina says in their documents for dual indexing on the Miseq, it looks a little different:

Note: I didn’t realize this until we did the exercise, but the p5 cluster generator we used to use in the Fowler lab is longer than the actual p5 sequence Illumina gives in their manuals. Not quite sure for this discrepancy, though I’m assuming that may mean that the sequences immediately after the p5 oligo during this step won’t be our rather variable index sequences, and instead may be some constant bases preceding the indexes. I’m guessing this is not a deal-killer, but something we’ll still have to be cognizant of when determining the run programming.

After that, is the complete second strand synthesis, resulting in a double stranded bridged molecule again.

Followed by denaturing and cleavage (again, presumably of p7 sequence) as described above, followed by annealing of the read2 primer.

So why does this matter?

Well, I think the practical implications are a few-fold. Firstly, Miseq dual indexing won’t need that second custom index read primer, since it will be reading off of the p5 sequence. And this is further complicated by the fact that the p5 adapter sequence we added onto the amplicons may perhaps be a bit longer than what’s actually on the chip, so we may have to factor this into the run parameters. Secondly, the strand that is reading that second index is different. With the Miseq, it’s being read off that first strand and thus in the same orientation as index 1. On the Nextseq, it’s being read on the second strand, so it will be read in the opposite order as index 1. Thus, I think this matters in terms of whether we’re putting the forward or reverse complement in the sample sheet, which will differ depending on whether we’re going Miseq or Nextseq with these samples.

We were recently doing some small scale Gibson-based NNK site saturation mutagenesis PCR reactions. In this scheme, we are independently transforming each position separately, so the number of transformants (ie. colonies) on a given plate should be directly related to the likelihood that all of the desired variants that we want to see are there at least once.

In fact, there are three parameters that factor into how good the variant coverage is at a given position. This is going to be 1) nucleotide biases in the creation of the NNK degenerate region of the primer, 2) the number of transformants, and 3) the fraction of the number of total transformants are actually variants, rather than undesired molecules such as carryover of the WT plasmid used for the template.

For any given experiment, you’re not going to know what the nucleotide bias is like until you actually Illumina sequence your library…. but at that point, you’ll already know the variant coverage of your library, so no need to estimate it anymore. On the other hand, if you know the nucleotide biases you observed for similar libraries, then you can do this estimation far before you get around to Illumina sequencing. Based on previous libraries, I have a pretty good idea of what the biases from machine-mixed NNK primers from IDT are like. For simplicity sake, I’m using 40% G, 20% C, 20% A, and 20%T as a rough estimate for the nucleotide bias I saw in the most biased NNK libraries.

The other two parameters are going to be very much experiment specific, and can be determined shortly after generating the library. The number of transformants can be determined by counting colonies from the transformation. And the amount of template contamination can be roughly determined by performing Sanger sequencing on a handful of colonies from those plates. Thus, I chose a few reasonable values for each: colony counts ranging from the very small (10 and 20) to quite large (400 and 1000), and template contamination percentages from almost impossibly low (0%) and much more likely (10 or 20%) all the way to possibly prohibitively high (50% and 75%). I then simulated the entire process, bootstrapped 20 times to get a sense of the average output, and made a plot showing what types of variant coverages you get depending on the combinations of those observed parameters. This is what the plot looks like:

So there you go. In a reasonable condition where you have, let’s say 10 or 20% template contamination, then you’d really be hoping to see at least 200 colonies, and hopefully around 400, where you can then really pat yourself on the back. If things went awry with the DPNI step, for example, and you were getting between a quarter to a half of colonies being template, then you’d minimally want 400 or so colonies and don’t feel too safe until you got a fair bit more than that. Though that’s only to make sure you at least have one copy of every variant at that position. If your library is half template, then chances are you’ll be running into a bunch of other problems down the line.

Due to popular request, I’m going to put my *most current* version of the HEK 293T Landing Pad recombination protocol here for others’ benefit. Much of the credit goes to Sarah Roelle, who wrote up this current version of the protocol.

Recombination:

[This protocol is for a 24-well plate:]
Day 1:

1) Make 2 transfection mixtures per sample:
Tube 1: 23 μL Opti-MEM + 1 μL Fugene6

Tube 2 (If using cells that don’t already express the Bxb1 recombinase enzyme): μL of DNA corresponding to 16 ng Bxb1 plasmid + μL corresponding to 224 ng attB plasmid and OPTI-MEM to a final tube volume of 24 μL.
-OR-
Tube 2 (If using cells that already express the Bxb1 recombinase enzyme): μL of DNA corresponding to 240 ng attB plasmid and OPTI-MEM to a final tube volume of 24 μL.

2) Mix Tube 2 into Tube 1 for each sample. Mix up and down a couple times with pipet. Then let the mixtures sit for 15-30 min while you get the cells ready (Unless you trypsinized and counted the cells first, to know how many you had in case they were limiting).

4) [Meanwhile] Trypsinize and count the cells. Add 120,000 cells per well in a final volume of 300 μL media to each well.

5) Once at least 15 minutes have passed since mixing, add the mixtures dropwise throughout the well of cells being transfected.

Day 2:
Add at least 500 μL media to each well.

Notes about scaling up / down:
We typically pilot new plasmids in a 24-well scale, and transfect larger populations of cells by transfecting 6-wells and increasing the number of plates / wells as needed.

Negative selection (if applicable):

Negative selection with iCasp9 is wonderful, and hopefully you are using one of those landing pad versions. I typically use 10nM final concentration of AP1903, although I’m fairly certain 1nM is just as effective. Death occurs in a couple of hours, so you can come back to your plate / flask later on in the day and change media to get rid of the dying cells. I typically wait until at least 72 hours after recombination to perform this step.

Important note: If you’re doing a library-based experiment, then make sure you leave some cells aside (even 100k to 500k cells will be plenty) that DO NOT go through any selection steps, since this will allow us to estimate the number of recombined cells (and thus estimate the coverage of your library; see below for the calculation). If you’re just doing individual samples, this isn’t nearly as big of a deal, since you’ll be able to visually inspect the number of recombinants (ie. do you see only a handful of surviving cells, or is it 100+ individual cells surviving?).

Positive selection (if applicable):

I tend to do the positive selection step AFTER negative selection, since the negative selection step has usually thinned the number of cells down enough such that the positive selection is going to be most effective (I find that over-confluent wells tend not to do super-grant with positive selection). Thus, while it can be done as soon as 72 hours post recombination, I tend to do this a week or later after recombination. You can find effective concentrations for positive selection of landing pad HEK 293T cells here.

Some additional notes:

1. Due to a phenomenon of (annotated) promoter-less expression from the thousands of un-recombined plasmids that remain in the cell following transfection, I typically wait ~ 5 to 7 days before running any cells in the flow cytometer, to wait for that background signal to die down.
2. Other transfection methods may work, but will need to be optimized. For example, I have observed transfection with lipofectamine 3000 to result in prolonged promoter-less expression, probably due to increased transfection and greater toxicity preventing cell division and thus plasmid dilution.
3. Calculating the number of recombinants. OK, so as I alluded to above, it’s definitely worth estimating the number of recombinants of any library-based experiments. To do this, take your observed recombination rate from running flow on your unselected cells (say, you see 5% of cells as being mCherry+ / recombined when you ran flow on unselected cells 7 days after transfection), and multiply that fraction with the number of cells you transfected. So, if we pretend that you transfected 20 million HEK 293T cells and you had observed 5% of cells recombined in your unselected well, then the rough calculation is 2e7 cells * 0.05, amounting to roughly 1 million cells estimated to have been recombined. Of course, directly sequencing what’s there in the recombined library is a more direct measurement of library coverage at the recombined cell step, but doing this calculation is still usually worthwhile as another line of evidence.