Must I HPLC-purify my PCR Primers for amplicon sequencing with PacBio SMRT?

Question

I would like to order some primers for amplicon sequencing. I am using universal tag primers (unusually long 30-mers recommended by PacBio for SMRT, both F&R) + barcodes (16-mers).

The resulting order of 192x46bp oligos is prohibitively expensive, because of the recommended HPLC purification. The increased expense is related to purification and to the increase in minimum scale (amount) required for ordering.

My question is this: what are the worst-case scenarios (and likelihood thereof) if I do not HPLC-purify primers for PacBio SMRT sequencing?

I am generally looking to make the process of sequencing heterozygous RT-PCR products less expensive, so additional advice in that vein would be terrific, especially from experienced users.

Additional, minor and clarifying side-questions. The PacBio recommended primer, the Universal Tag 30-mer primer seems a bit long to me, and possibly unnecessarily so. Does anyone know why they would insist on a 30-mer? It's a bit irritating that there is no specific reason given, and combined with the unnecessarily long symmetric 16-mer barcodes, and 5’-block (5’ NH₄-C6), they are pushing up the cost for little transparently good reason. Perhaps I am wrong. Any offered reasoning on these points would be appreciated.

Answer 1

I am at the same stage as you right now. I am assuming that you are planning a 2-step amplification: first one with the universal primer and second one with the barcoded primers. So, this has been my rationale and experience on this.

HPLC purification

If you are generating your own primers and not buying the plate from PacBio it is better to get HPLC purified oligos. The idea behind using HPLC purified primers is that the oligo synthesis may produce truncated products (primers with deletions usually towards the 5' region) in little concentrations. According to IDT, each base addition has an approximate error rate of 1%. Therefore, longer products have a higher probability of deletions; the number of expected errors would be 0.01 × length of the primer. For a 46nt primer (16nt barcode + 30nt universal sequence), the percentage of deletion-free products would be ~63% (Poisson probability of no deletions). So, you can expect 27% truncated products with approximately 1.2 deletions per product^†. Therefore there is some likelihood that the truncated products may mess up your barcode. However, the probability of the deletions specifically in the barcode regions would be less than this estimate.

5' NH₄-C6 modification

This modification makes prevents the ligation of the PCR products. You generate the final library by re-amplifying the PCR products of universal primer, with the barcoded primers. The product will also contain a little amount of the template, which you do not want to be ligated to the sequencing adapters. Again, this would be very less: for example if you take 2ng of the template and your yield is ~1.5µg (I typically get this much yield even with a colony PCR of 25 cycles) then the template contamination is less than 0.2% (ideally you should calculate this with number of molecules instead of mass but since the template and product would have almost equal molecular weight you can use the latter also).

Why 16-mer barcodes

You can use shorter barcodes too but you should make sure that they are different enough such that any sequencing error would not make it difficult for you to distinguish between your barcodes. PacBio has an error rate of 13% for a single read; it reduces if you do the Circular Consensus Sequences (0.13^n where n is the number of rounds). When you have a large number of samples then it is better to use longer barcodes as you can reliably distinguish between different samples. They also recommend that the barcoded primers have a 5' phosphorylation. This will further increase your cost. The 5' phosphorylation will increase the ligation efficiency. It is technically possible to use polynucleotide kinase to phosphorylate the products but the efficiency of T4-PNK varies with the 5' base (Houten et al, 1998). In this case I would recommend pre-phosphorylated primers.

Why 30-mer universal primer

I don't have a good knowledge about this but I guess the reasons have to probably do with the molecular biology rather than the data analysis part of it. Demultiplexing can be done manually and it is not very difficult. Technically, a 20nt sequence should also suffice (usual length of PCR primers). Under usual circumstances, I don't think you are going to save a lot by using a 10nt smaller primer. The total length of primer (universal sequence + the template specific sequence) may go down from ~50+ to ~40+. Many oligo-synthesis services have an upper bound for HPLC purification; they would PAGE purify the longer ones. For example, the company that I order from has a cutoff of 50. So, longer primers have to be PAGE purified which would reduce your yield and increase the price. But usually you can negotiate with the company to allow your 1-5nt longer-than-cutoff product to be HPLC purified.

---

Summary

Some of these modifications would just improve the efficiency marginally and may be skipped if your data is not likely to be affected by a slight drop in efficiency.

So this is my final recommendation:

• HPLC purification: not necessary but recommended for long primers (at least the barcodes)
• NH₄-C6 modification: not necessary
• 5' phosphorylation of barcoded primers: recommended
• Length of universal primer sequence: can possibly be shortened by 10nt but probably that is not likely to reduce the cost significantly.

However, you can significantly reduce your cost by asymmetric barcoding. For example, if you have 100 samples then the minimum number of barcode primers needed would be 20 (10 pairs). You can shuffle the forward and reverse primers to create 100 combinations.

---

^† Explanation of the error rate probability

Let $\lambda$ be the error rate per base (0.01) and $N$ be total number of bases (46). The error will follow Poisson distribution and the probability of errors would be:

$$P(X=k)=\frac{(\lambda N)^k e^{-\lambda N}}{k!}$$

For zero error i.e. $k=0$ the probability would be simply $e^{-\lambda N}$ which would be ~0.63

The mean number of deletions for the 27% that contain errors would be (see zero-truncated Poisson distribution):

$$E(X|X>0)=\frac{\lambda N}{1-e^{-\lambda N}} \approx 1.25$$