Rates of recombination obtained for Phusion are highly variable across conditions. Previous research has proposed that enzymes with higher processivity yield more chimeric sequences (26). This is confirmed at high cycle/high template concentration conditions, with an average yield of 71% chimeras (Figure 2). However, the surprising result is that the same enzyme yields absolutely no chimeras when the initial template concentration is sufficiently low (1.4 × 10³ starting molecules) and when the cycle number is reduced to 30 cycles (Figure 2). The processivity-enhanced enzyme makes up for its high rate of chimera formation by being able to amplify initial concentrations that are four to five orders of magnitude lower than what the strict proofreading polymerase can amplify (Table 1), effectively reducing chimera formation to zero (Table 2). Non-proofreading polymerases like Taq might also benefit from less concentrated initial templates but they are less desirable for molecular evolution studies. Furthermore, we attribute Phusion's ability to amplify low template concentrations to the enhanced processivity. Ordinary Taq polymerases might not be able to amplify concentrations that are low enough to reduce artifact formation.

The percentages of chimeras formed per reaction are higher for certain conditions in the present survey than have been reported in much of the literature, which we believe is due to the greater complexity of our starting templates (i.e., eight actin haplotypes). The rates of recombination for a proofreading enzyme (Vent_R) under standard conditions (30 cycles) averaged ~40% recombinants in the present study, while the highest available literature reports are 32% for Taq polymerase after 30 cycles on 7 distinct initial haplotypes (30), 16% for the proofreading Expand H-F system (Boehringer, Mannheim, Germany) after 25 cycles using 8 initial haplotypes (27), and 31% recombinants using Taq polymerase across multiple loci in polyploid cotton (41). Since the reported chimera formation rates in other available literature ranges 1–5% across a variety of enzymes (15,16,17,18,19)(24,26), we attribute the higher rates in our experiment [as well as in two others (27,30)] to the higher number of initial haplotypes: most studies used 2–4 initial haplotypes to test chimera formation. In contrast, ≤35% recombinants were reported for only two MHC loci (31), which may indicate a possible influence of the template itself. Furthermore, we demonstrated that there is an increase in chimera formation as diversity in the original sample increases. This reinforces the idea that molecular environmental studies might be plagued with a slew of artificial sequences (20). Moreover, initial template concentrations (i.e., abundances) in environmental samples will most likely be unequally distributed, which might influence the formation of artifacts. However, differential abundance of templates probably has a larger impact on the detection of true diversity and our analyses indicate that multiple (>2) PCR reactions will be required to capture the true diversity of a sample even when abundances of templates are equivalent.

We find that the majority of chimeras contain more than one breakpoint, indicating that more than two parental sequences can be involved in PCR-mediated recombination. This high rate of crossover is independent of cycle number or initial template concentration (Figure 3). This observation will create problems for chimera-detecting software that base search criteria in finding one breakpoint per sequence. For example, while using the online software Bellerophon (40) to detect the chimeras in the present data set, only an average of 65 ± 18% of chimeras were detected. Even more worrisome, there is a false-positive rate of 40 ± 31% (see Supplementary Material 3 for details).

Capturing the full diversity within a sample requires a combination of multiple PCR reactions that have been performed under chimera-reducing conditions. On average, PCR reactions at high cycle numbers are unable to recover all diversity (average 4 ± 1 out of 8 starting haplotypes, Figure 2), even if all three replicates are combined (Supplementary Material 1), and have the added bias of generating false haplotypes. While low cycle number improves recovery (7 ± 1 out of 8 starting haplotypes; Figure 2), it is likely that a single PCR experiment will not capture all the diversity. For example, in chimera-reducing conditions with low cycle number and lowest initial template concentration possible, individual PCR reactions detected an average of four out of the eight haplotypes, but performing three replicates will certainly describe all diversity in this eight-haplotype system (Supplementary Material 1). Hence we reiterate the necessity of replicating PCR reactions to assess biodiversity (19,24,26), and we add the recommendation that these replicate PCR runs be undertaken with minimal DNA concentrations and cycle numbers, which need to be established on a sample-by-sample basis.