PCR amplification of 16S rDNA was found to be highly biased, so that the rDNA from one species out of four was preferentially amplified. We present evidence that the observed PCR bias most likely occurs because the genomic DNA of some species contains segments outside the amplified sequence that inhibit the initial PCR steps. Attempts to overcome this bias by use of a ‘touch down’ PCR procedure or by performing PCR in the presence of denaturants or cosolvents such as acetamide, DMSO, or glycerol were unsuccessful. Since the PCR inhibiting interference from template flanking DNA segments evidently is dependent on the position of the primer sites, we suggest that community diversity analysis based on PCR amplification of 16S rDNA can be improved by extending the procedure from comparative analysis of 16S rDNA amplified by use of only one primer set to a procedure involving at least two different 16S rDNA PCR amplifications performed with different primer sets.
The use of molecular approaches in microbial ecology has revealed a tremendous prokaryotic diversity of which only minor fractions have been discovered previously by traditional culture enrichment techniques (e.g. [1–5]). Often these molecular methods include comparative analyses of 16S rDNA segments PCR-amplified by the use of primers targeting conserved sequences (e.g. [3–6]). Such use of PCR in community diversity analysis is based on the assumption that PCR amplification of the homologous but non-identical rDNA sequences occur with equal efficiency. Recent reports suggest, however, that this assumption is not always fulfilled. Reysenbach et al.  reported that rDNAs of extreme thermophilic archaea could not be amplified by the usual PCR methods, but that preferential amplification of yeast DNA in a mixture of archaea and yeast DNA could be prevented if the PCR was performed with 5% acetamide in the PCR mixture. Wilson and Blitchington  studied the human colonic biota by rDNA analysis, and reported that the representation of the ecosystem was dependent on the number of PCR cycles used for amplification of the 16S rDNA. The representation of the ecosystem after 35 PCR cycles was, in comparison to the representation after nine PCR cycles, distorted and lacked diversity. Chandler et al.  reported that the PCR template concentration effects the composition and distribution of total community 16S rDNA clone libraries. It appears, therefore, that a more fundamental understanding of the causes of PCR bias would be useful in order to improve PCR-based methods for community diversity analysis.
In the present communication we show that PCR amplification of 16S rDNA from four bacterial species isolated from a waste gas biofilter is highly biased, and present evidence that the observed biases are most likely occurring because the genomic DNA of some species contain segments outside the amplified region that inhibit the initial phase of the PCR.
Materials and methods
Strains and growth conditions
The strains of Pseudomonas putida, Microbacterium arborescens, Nocardioides simplex, and the β-subgroup proteobacterium used in the present investigation were all isolated from a biofilter treating toluene-containing waste gas . Cultivations were performed in Luria-Bertani broth (LB) .
DNA from pure or mixed cultures was prepared essentially as described by Grimberg et al. . One ml of culture was harvested by centrifugation, and washed by resuspension in 1 ml of TNE buffer (10 mM Tris-HCl, pH 8.0, 10 mM NaCl, 10 mM EDTA, pH 8.0) followed by centrifugation. The pellet was resuspended in 270 μl of TNEX buffer (10 mM Tris-HCl, pH 8.0, 10 mM NaCl, 10 mM EDTA, pH 8.0, 1% (v/v) Triton X-100), and 10 μl lysozyme solution (5 mg/ml) as well as 7.5 μl proteinase K solution (20 mg/ml) were added. The suspension was incubated at 37°C for 2 h followed by incubation at 65°C for 2 h. After this, the DNA was precipitated by addition of 15 μl NaCl solution (5 M), 1.0 ml of 96% ethanol, and centrifugated for 20 min at 15 000×g and 4°C. Then the pellet was washed with 1.0 ml of 96% ethanol, dried, and resuspended in 500 μl TE buffer (10 mM Tris-HCl, pH 8.0, 1.0 mM EDTA, pH 8.0).
The forward and reverse primers used were: 515F (5′GTGCCAGCAGCCGCGGTAA3′) and 1408R (5′TGACGGGCGGTGTGTACAAGGC3′) targeting conserved regions 515–533 and 1408–1387 (IUB nomenclature for Escherichia coli); GM5F (5′CCTACGGGAGGCAGCAG3′) and DS907R (5′CCCCGTCAATTCCTTTGAGTTT3′)  targeting conserved regions 341–357 and 928–907; 9F (5′GAGTTTGATCCTGGCTCAG3′) and 1512R (5′ACGGCTACCTTGTTACGACTT3′) targeting conserved regions 9–27 and 1512–1492. The primers used were purchased from Hobolth DNA Syntese.
PCR was performed on a GeneAmp PCR System 9600 (Perkin Elmer) in 50 μl 1×PCR buffer, containing 200 μM each of dATP, dCTP, dGTP, and dTTP, 1 μM forward primer, 1 μM of reverse primer, 2 mM MgCl2, 1.25 U Taq DNA polymerase (Pharmacia Biotech, Allerød, Denmark), and approximately 60 ng genomic DNA of each species (or mixtures and dilutions as indicated in the text). Except for the experiment where ‘touch down’ PCR was used, we used the following PCR procedure: Initial denaturation at 94°C, then 35 (or as indicated) two-step cycles with denaturation at 94°C and annealing/elongation at 68°C, then termination of the amplification process with 10 min at 68°C followed by cooling to 4°C. The same results (with respect to PCR bias) were obtained regardless of whether (i) the initial denaturation time was 2 min or 10 min, (ii) the denaturation time used in the temperature cycles was 30 s, 2 min or 5 min, or (iii) the time for primer annealing and elongation was 1 min, 2 min or 5 min. In addition, substitution of the two-step cycles described above with a three-step cycle with denaturation at 94°C for 30 s, primer annealing at 68°C for 1 min, and elongation at 75°C for 1 min gave the same results. In the ‘touch down’ PCR experiment  the initial annealing temperature was set at 75°C and decreased 1°C for ten cycles until a touch down of 65°C, at which annealing temperature an additional ten cycles was carried out. For ‘touch down’ PCR the initial denaturation was carried out at 94°C for 10 min, the cycle denaturation was carried out at 94°C for 1 min, the annealing time was 1 min, and elongation was carried out at 70°C for 3 min.
PCR product (10 μl) was digested with five units of restriction enzyme in a total of 14 μl 1×ReAct buffer no. 1 or 2 (depending on which restriction enzyme was used) by incubation for 2 h at 37°C or 60°C (depending on which restriction enzyme was used). The digests were analyzed by electrophoresis in 3.5% (w/v) agarose gels in 1×TBE buffer (89 mM Tris, 89 mM boric acid, 1 mM EDTA) both containing ethidium bromide.
Sequencing of 16S rDNA
Pre-sequencing PCR amplification of the 16S rDNA was performed on a GeneAmp PCR System 9600 (Perkin Elmer) in 50 μl 1×PCR buffer containing 200 μM each of dATP, dCTP, dGTP, and dTTP, 1 μM of each primer, 1.25 U of Taq polymerase (Pharmacia Biotech, Allerød, Denmark), and approximately 10 ng template DNA. The PCR was done with 30 temperature cycles consisting of 1 min at 94°C (denaturation), 1 min at 52°C (annealing), and 1 min at 72°C (extension). A final extension step at 72°C was performed for 3 min. The 5′ and 3′ primers used were 5′GAGTTTGATCCTGGCTCAG3′ and 5′ACGGCTACCTTGTTACGACTT3′, targeting conserved regions 9 to 27 and 1512 to 1492 (IUB nomenclature for E. coli). The 16S rDNA PCR products were purified by electrophoresis in 1.5% (w/v) agarose gels, followed by a GeneClean (BIO 101, Vista, CA, USA) procedure according to the manufacturer's instructions. Sequencing was carried out by using a Taq DyeDeoxy terminator cycle sequencing kit (Applied Biosystems, Foster City, CA, USA) according to the manufacturer's instructions. The sequence primers used were those described by Lane . Sequence products were analyzed with a model 373A DNA sequencer (Applied Biosystems).
16S rDNA sequence analysis
16S rDNA sequences were aligned with published sequences from EMBL, DDBJ, and Genbank by the use of basic local alignment search tool (BLAST) .
Results and discussion
The four bacterial species used in the present investigation were all isolated from a waste gas biofilter . They constitute part of a model microbial consortium that we study in our laboratory, and they were chosen for the present study because we wished to develop a fast molecular method to monitor changes in species compositions in our microbial consortia. Based on 16S rRNA sequence analysis these four species were identified as strains of P. putida, M. arborescens, N. simplex, and a β-subgroup proteobacterium (closest to Burkholderia cepacia).
Initially we established a method to distinguish between the four bacterial species on the basis of differences in their 16S rDNA. DNA from pure cultures was used as template for PCR amplification of 16S rDNA fragments using the eubacterial targeting primers 515F and 1408R, and subsequently a number of tetrameric restriction enzymes were screened for their ability to generate unique restriction fragment length polymorphism (RFLP) patterns. As shown in Fig. 1 (lanes 3–6), restriction of the 16S rDNA PCR fragments with HinfI gave a unique RFLP type for each of the four species. Next, we investigated if use of the PCR-RFLP method allowed simultaneous detection of more than one species. The PCR-RFLP procedure was performed on (1:1) mixtures of DNA from the pure cultures in all combinations, and on a (1:1:1:1) mixture of DNA from all four strains. In all mixtures where P. putida DNA was present the PCR was highly biased so that only the P. putida RFLP type could be seen (see Fig. 1, lanes 7–10). If the pure cultures were mixed prior to the DNA extraction, and the PCR-RFLP procedure was performed on DNA extracted from the mixed cultures, we observed the same PCR bias (data not shown).
Judged from the intensities of the bands in an ethidium stained agarose gel, and from spectrophotometric measurements, the DNA preparation procedure used gave similar amounts of chromosomal DNA from the four species (data not shown). However, as suggested by Farrelly et al. , the number of chromosome equivalents, the size of chromosomes, and the number of rrn operons per chromosome may differ among species, and therefore the observed differences in the amount of PCR product could be caused by differences in the amount of PCR template. In order to investigate this, the RFLP-PCR procedure was performed on mixtures of DNA from P. putida and each of the other three species in the ratios 1:1, 1:10, and 1:100. If the observed PCR bias was caused by the relatively small differences in template concentration that would be expected from differences in the number of rrn operons among the four species, then this should be revealed when the P. putida DNA was present in an amount ten times lower than the other template DNA. As shown in Fig. 2 (lanes 3, 6, and 9), the PCR bias was still pronounced in the 1:10 mixtures, suggesting that the PCR bias is not due to template concentration differences. When the P. putida DNA amount was 100-fold less than the DNA of the other species, the PCR amplification bias occurred with varying degrees dependent on which template the P. putida DNA competed with. In combination with M. arborescens and β-subgroup proteobacterium DNA, the PCR amplification still resulted in high amounts of P. putida DNA (Fig. 2, lanes 4 and 7), whereas in combination with N. simplex DNA the PCR amplification resulted in small amounts of P. putida DNA (Fig. 2, lane 10).
Preferential PCR amplification of low-G+C rDNA templates from a mixture of low-G+C and high-G+C rDNA has previously been reported . Although the G+C contents of the 16S rDNA of the P. putida, M. arborescens, N. simplex, and β-subgroup proteobacterium used here (53.8%, 56.5%, 57.8%, and 54.5%) differ only moderately, we nevertheless made an attempt to investigate if the observed PCR bias could be explained by differences in G+C contents. Reysenbach et al.  reported that the PCR bias observed after PCR amplification from a mixture of low-G+C and high-G+C rDNA was prevented when the PCR was performed with 5% acetamide in the PCR mixture. In our case, however, the PCR bias was unchanged when 5% acetamide was included in the PCR mixtures (data not shown). This result makes it unlikely that differences in G+C contents should cause the observed bias.
Since Wilson and Blitchington  showed that the PCR bias observed when 35 PCR cycles were used could be reduced if the PCR was run with nine cycles, we investigated if the PCR bias observed in our case could be reduced by reducing the number of PCR cycles. When PCR amplifications with mixed templates were performed using 10, 20, and 30 PCR cycles (35 PCR cycles were used in all other experiments except for the ‘touch down’ PCR), the resulting RFLP patterns again showed only the P. putida RFLP type (data not shown), suggesting that the observed PCR bias was not dependent on the number of PCR cycles.
As the experiments indicated that the observed PCR bias was not caused by differences in template concentration, or differences in G+C contents, and was independent on the number of PCR cycles, the bias most likely originates from the initial PCR steps. Such PCR bias would be observed if the PCR primers do not match the primer sites of the four different 16S rDNAs equally well, or if the single stranded template and flanking region DNA of the four species form different secondary structures (affecting primer site accessibility or elongation efficiency) when the temperature is lowered after the first and second denaturation steps. The possibility that the bias was caused by primer mismatches was excluded, however, since sequencing of 16S rDNA showed that the primers completely matched the P. putida, M. arborescens, and N. simplex species, while for the β-subgroup proteobacterium there was a single mismatch at position 20 of the 1408R primer (data not shown). If the observed PCR bias was caused by secondary structures involving single stranded template and flanking region DNA, it would be expected that restriction of the genomic DNA with endonucleases cutting outside the template DNA can affect the PCR bias. As shown in Fig. 3 (lanes 8–10), digestion of the DNA mixtures with restriction enzymes HindIII and PvuII (which do not cut in the template DNA) prior to the PCR did indeed reduce the PCR bias in the P. putida/M. arborescens mixture, but not in the P. putida/β-subgroup proteobacterium and P. putida/N. simplex mixtures. The fact that restriction with HindIII+PvuII reduces the PCR bias in only one case, strongly suggests that the PCR bias depends on specific DNA sequences outside the template region. To assure that the observed effect on the PCR bias was caused specifically by the enzymatic restriction and not by the dilution of the DNA in restriction buffer, or following ethanol precipitation and resolution, we included controls without the restriction enzymes. As shown in Fig. 3 (lanes 5–7) these controls showed the usual PCR bias.
If the observed PCR bias is caused by interference from template flanking DNA regions during the initial PCR steps there should not be bias if the 16S rDNA PCR fragments are used initially as templates. In order to investigate this, the PCR-RFLP procedure was performed on 104 dilutions of (1:1) mixtures of 16S rDNA PCR fragments from each species, and on a 104 dilution of a (1:1:1:1) mixture of 16S rDNA PCR fragments from all four strains. As shown in Fig. 4, the PCR-RFLP procedure now showed the true composition of the original template DNA without any bias. In an attempt to map the hypothesized interfering DNA regions we amplified 9–1512 (IUB numbering  is used throughout) rDNA fragments from each species using the primers 9F and 1512R, and used these PCR fragments in the usual PCR-RFLP from mixed templates. These PCR amplifications were unbiased (data not shown) suggesting that the hypothesized interfering regions are located outside the 9–1512 rDNA region.
The suggestion that a PCR interfering region is located in the M. arborescens genomic DNA outside the 9–1512 rDNA region agrees with the finding that the putative interfering DNA to some extent is removed by a HindIII+PvuII digestion, and the fact that no HindIII or PvuII sites are found in the 9–1512 rDNA region of any of the four species. The observation that the PCR bias is independent on the number of PCR cycles in all cases strongly suggests that the bias originates from the initial PCR phase, and is caused by interference from a DNA segment outside the 515–1408 rDNA template region present also in the β-subgroup proteobacterium and N. simplex genomic DNA. However, as PCR amplification from chromosomes and PCR fragments differ most in the initial PCR steps, the evidence for location of interfering DNA outside the 9–1512 rDNA region in the case of the β-subgroup proteobacterium and N. simplex genomic DNA should be regarded with some reserve. Suzuki and Giovannoni  reported that template annealing may inhibit the formation of template-primer hybrids and lead to a bias towards a final 1:1 mixture. Such a bias mechanism may prevail in the initial PCR steps when PCR products are used as initial templates, whereas when genomic DNA is used in the PCR, template annealing in the initial PCR phase may be hindered by non-template genomic DNA. When genomic DNA is used in the PCR, a strong bias occurring in the initial PCR steps because of folding of single stranded DNA may overrule bias towards a 1:1 mixture that could occur in the later phase of the PCR where the concentration of template DNA is high. Notably, the PCR experiments underlying the hypothesis of Suzuki and Giovannoni  were done on rDNA PCR fragments; not genomic DNA.
Bias caused by folding of single stranded DNA that affects either primer accessibility or elongation efficiency should be absent or changed if a set of primers complementary to other positions on the template is used. We therefore used the eubacterial targeting primers GM5F and DS907R  for amplification of 16S rDNA. The primers GM5F and DS907R anneal to positions 341–357 and 928–907 of the rDNA, while the primers 515F and 1408R anneal to positions 515–533 and 1408–1387 of the rDNA. PCR-RFLP performed on mixtures of template DNA using the GM5F and DS907R primers, and the restriction enzyme DdeI, showed that preferential amplification of P. putida rDNA was now not occurring (Fig. 5). In this case, however, there was a preference for amplification of rDNA from the β-subgroup proteobacterium (Fig. 5, lanes 6 and 8).
Since the observed PCR bias most likely is caused by the formation of secondary structures between single stranded template and flanking DNA, we made attempts to overcome this bias by the use of methods known to reduce the formation of secondary structures in DNA. Presumably because of the general property of organic solvents to destabilize DNA in solution, cosolvents such as DMSO and glycerol can be used to enhance PCR amplification . It is therefore possible that cosolvents in PCR mixtures can remove the PCR bias observed in the present study. As described previously, the PCR bias was unchanged when the PCR was performed with 5% acetamide in the PCR mixture. When the PCR was performed in the presence of 2% DMSO the bias was slightly reduced in the P. putida/M. arborescens mixture, and when the PCR was performed in the presence of 10% glycerol the bias was slightly reduced in all three mixtures (Fig. 6). As seen in lane 1 of Fig. 6A and B, the PCR-RFLP pattern of M. arborescens contains an additional small band (in comparison to the RFLP pattern seen in Figs. 1 and Figs. 3) when the PCR is performed in the presence of either DMSO or glycerol. This band was present when the PCR-RFLP analysis was performed on M. arborescens genomic DNA or 515–1408 rDNA fragments with 4 mM MgCl2 in the PCR mixture (data not shown), but was not present when the PCR-RFLP analysis was performed with 2 mM MgCl2 in the PCR mixture (Figs. 1 and Figs. 3). These observations, and the fact that according to the sequence of the M. arborescens 16S rDNA the PCR-RFLP type should be as shown in Figs. 1 and Figs. 3, suggest that the additional band is due to unspecific amplification from the 515–1408 rDNA region. It appears that the synthesis of this PCR side-product in the P. putida/M. arborescens mixture is less suppressed than synthesis of the correct product when DMSO is present in the PCR mixture (Fig. 6A, lane 5), but more suppressed than synthesis of the correct product when glycerol is present in the PCR mixture (Fig. 6B, lane 5). Since the synthesis of this side-product most likely is guided by one unspecific priming inside the 515–1408 rDNA region and one correct priming, it will most likely be subject to inhibition similar to the synthesis of the correct product.
As an alternative to adding cosolvents to the PCR mixture, it has been reported that special PCR procedures in some cases can remove PCR inhibiting secondary structures in DNA. In the so-called ‘touch down’ PCR procedure  the annealing temperature is set to 10°C above the expected annealing temperature, and decreased 1°C every cycle for 10 cycles, after which the ‘touch down’ temperature is used for primer annealing for an additional 10 cycles. ‘Touch down’ PCR, however, could not prevent preferential amplification of the P. putida DNA (data not shown).
The present work strongly suggests that biased PCR amplification of 16S rDNA can occur because the DNA of different species contains segments outside the template region that inhibit the initial phase of the PCR to different degrees. This PCR bias could not be eliminated by performing PCR in the presence of cosolvents, or by using special PCR procedures such as ‘touch down’ PCR. The PCR bias was, however, dependent on the position of the primer sites. That is, the 16S rDNA of one species was preferentially amplified from DNA mixtures when we used one set of primers, while the 16S rDNA of another species was preferentially amplified when we used another set of primers. We suggest, therefore, that community diversity analysis based on PCR amplification of 16S rDNA can be improved by extending the procedure from comparative analysis of 16S rDNA amplified by use of only one primer set to a procedure involving at least two different 16S rDNA PCR amplifications performed with different primer sets.
We thank Ashraf Ibrahim for helping M.C.H. with sequencing of the four 16S rDNA genes. This work was supported by a grant from the Danish Research and Development Program for Food Technology (FØTEK2).