Insights & Solutions

Practical notes from real projects: PCR troubleshooting, platform-specific quirks, and classifiers worth knowing about.

This page collects practical experience that doesn't fit neatly into the more structured pages elsewhere on this site. It's the kind of thing that usually only comes up once you've already started generating data. For primer placement, degeneracy, and 3' end protection, see Primer Design. For long-read vs. short-read platform tradeoffs, see Long-Read vs. Short-Read Sequencing. For SINTAX, reference database choice, and custom database building, see Data Prep Short-Reads (PE), Step F.

PCR Troubleshooting

A poorly prepared library cannot be rescued by sequencing, and the first targeted PCR often determines whether a project succeeds: perfect primers alone won't help if the PCR conditions themselves aren't optimised. Annealing temperature, cycle number, reagent concentrations, and polymerase choice (see Primer Design for the polymerase and 3' protection side of this) all matter. In silico planning helps, but at some point optimisation has to move into the lab, tested and fine-tuned empirically.

Avoiding Excessive PCR Cycling

Running extra PCR cycles to increase product yield may seem like a simple fix, but it carries real risks. Once amplification reaches the plateau phase, additional cycles mostly re-amplify molecules that are already amplified, rather than the original template, which is exactly why bias increases with cycle number: some fragments amplify more efficiently than others, and the effect compounds each extra cycle, overrepresenting some sequences while others are lost.

Two further risks come from the same place. PCR errors accumulate: even high-fidelity polymerases make occasional mistakes, and more cycles means more chances for them to occur. Excessive cycling also raises the risk of chimera formation, where incomplete DNA strands anneal to unrelated fragments and create artificial sequences that were never present in the original sample.

A better approach is fewer cycles, with several independent reactions pooled together. This minimises amplification bias, reduces errors, and preserves low-abundance sequences, while still producing enough product for downstream use.

Diagnosing False Priming

False priming can be a real challenge, and at the GDC we strongly recommend a pilot study before committing to a large-scale sequencing project. One of the most effective ways to investigate false priming afterward is to examine mismatches at the primer binding sites in your raw reads and compare them to other regions in the same reads. Our pipeline reports these as a mismatch profile. Each line shows one observed primer-site variant, how frequently it occurred, and where it differs from the expected primer sequence. Dots indicate matches; letters indicate mismatches. For example:

R1 Primer-Site Mismatches (top 10):

 ...
 0.3% 5'-.............S...-3'
 0.3% 5'-.......G.........-3'
 0.3% 5'-............W.CA.-3'
 0.4% 5'-.....G...........-3'
 0.4% 5'-...............A.-3'
 0.5% 5'-................G-3'
 0.5% 5'-......G..........-3'
 0.5% 5'-..............CAG-3'
 0.6% 5'-...............AG-3'
65.1% 5'-.................-3' (no mismatches)

R2 Primer-Site Mismatches (top 10):

 ...
 0.4% 5'-..........G.........-3'
 0.4% 5'-.................AA.-3'
 0.4% 5'-................TAA.-3'
 0.5% 5'-...........G........-3'
 0.5% 5'-................TA.T-3'
 0.5% 5'-.....A..............-3'
 0.7% 5'-...................T-3'
 0.9% 5'-................T.AT-3'
 1.3% 5'-.................AAT-3'
63.5% 5'-....................-3' (no mismatches)

Two things stand out here. First, the frequency of perfect primer sites is relatively low, around 65%. Second, diversity in the primer binding region is unexpectedly high, a further warning sign. On top of that, mismatches cluster at the 3' end, a pattern we often observe when endonuclease activity contributes to false priming (see Primer Design for the mechanism behind why 3' mismatches specifically matter). Patterns like these give you real insight into PCR-related problems and help you refine your approach before scaling up.

Platform-Specific Caveats

Heteroduplex Formation in PacBio CCS

PacBio long-read sequencing delivers highly accurate, full-length reads through Circular Consensus Sequencing (CCS), but a persistent challenge, particularly for amplicon sequencing, is heteroduplex formation: mismatches between the forward and reverse strands of a DNA molecule, often from genetic variation, PCR errors, or incomplete strand separation. When mismatched strands are sequenced together, they can produce ambiguous base calls and reduce the accuracy of variant detection.

On earlier PacBio platforms such as the Sequel, a significant difference between the forward and reverse strands would produce a separate CCS read for each. With Revio, the consensus process instead collapses this into a single consensus sequence representing the most likely base at each position. That's beneficial for many genomic applications, but for amplicon sequencing, where precise allele resolution is the whole point, it can mask genuine genetic variation instead of just noise.

Good Library Prep Is the Real Fix, Not a PacBio-Specific Workaround

Heteroduplexes form during PCR itself, when single strands from slightly different templates, different alleles, different closely related community members, denature and reanneal with each other instead of with their true complementary strand. This becomes far more likely in exactly the situation described in Avoiding Excessive PCR Cycling above: once primer becomes depleted relative to accumulated product in late PCR cycles, a denatured strand is less likely to encounter a primer to extend from and more likely to reanneal with another single strand instead, including an imperfectly matched one. Thompson, Marcelino, and Polz (2002) Nucleic Acids Research 30(9):2083-2088 documented this directly: heteroduplexes became increasingly prevalent specifically as primers became limiting late in the reaction, and demonstrated that a low-cycle "reconditioning PCR" (re-amplifying a diluted aliquot of the product for a few more cycles with fresh primer and reagents, restoring primer excess) substantially reduces heteroduplex content. This isn't guaranteed to eliminate the problem entirely, later work found reconditioning insufficient on its own in some cases, but it's a genuinely effective, well-documented mitigation, not just a theoretical fix.

In other words, this is fundamentally a wet-lab problem with a wet-lab-stage fix, keeping cycle number reasonable and primer in excess through the reaction, rather than something to work around after the fact on Revio's consensus output. It happens to matter more for PacBio amplicon sequencing because Revio's consensus calling can silently mask a heteroduplex as a single "most likely" sequence rather than flagging it, but the heteroduplex itself was already formed in the PCR tube, long before it reached the sequencer.

What To Do About It

Keep PCR cycle numbers low and primer concentration in excess through the end of the reaction (see Avoiding Excessive PCR Cycling above), consider a reconditioning PCR step if heteroduplex formation is a known concern for your samples, pool replicate PCR reactions rather than over-cycling a single one, and inspect CCS quality metrics for signs of unusually high heterogeneity at specific positions. For samples expected to be highly polymorphic, where allele-level resolution really matters, it's worth discussing library preparation strategy with us before committing to a full run, since the fix belongs upstream of sequencing, not after it.

Background Notes: Classifiers and Databases

See Data Prep Short-Reads (PE), Step F for how we use SINTAX with SILVA or UNITE, and for building custom reference databases. This section is background, not a recommendation to switch: a couple of alternatives worth knowing about, not currently part of our standard pipeline but sometimes relevant depending on the study.

Alternative classifiers:

RDP Classifier: a naive Bayesian classifier, widely used and generally reliable, though performance can degrade for sequences distantly related to anything in the reference or when the database itself is incomplete.
QIIME 2's taxonomy plugins: QIIME 2 doesn't use one fixed classifier, it supports several depending on the plugin, including a scikit-learn naive Bayes classifier, VSEARCH-based consensus classification, and BLAST-based classification. Which one a given QIIME 2 analysis actually used is worth checking rather than assuming.

Additional databases:

PR2: tailored to eukaryotic microbes, particularly protists, useful for environmental studies with a protist focus but not a good fit outside that group.
MIDORI: a curated database of metazoan mitochondrial gene sequences, COI most prominently, also 16S, 12S, and cytochrome b, built from GenBank. It's widely used for animal eDNA and metabarcoding studies generally (plankton, meiofauna, gut content, environmental water samples), not specific to any one habitat type.

As with any reference database, coverage and primer compatibility matter more than raw size (see the reference database note on the Overview Workflows page). A database that's a poor match for your primers or your organismal group will underperform a smaller, better-targeted one.

Blackman, Walser, Rüber, Brantschen, Villalba, Brodersen, Seehausen, and Altermatt (2023) General principles for assignments of communities from eDNA: Open versus closed taxonomic databases. Environmental DNA 5(2):326-342.