Microarray-Based Evaluation of Whole-Community Genome DNA Amplification Methods

Pure cultures, environmental samples, and DNA extraction.DNA was extracted from Shewanella oneidensis MR-1, Desulfovibrio vulgaris Hildenborough, Rhodopseudomonas palustris CGA009, and Thermoanaerobacter ethanolicus X514 as previously described (30). To evaluate the performance of WCGA methods on community DNA, a soil sample from the USDA-ARS High Plains Grasslands Research Station (Cheyenne, WY) was used. DNA was extracted from 5 g of soil as previously described (29).

DNA amplification, labeling, and microarray hybridization.The outlines of all experiments are shown in Fig. 1. Three MDA methods, Bst (14) and two commercial phi29-based kits, REPLI-g (REPLI-g ultrafast minikit; Qiagen, Valencia, CA) and Templiphi (Illustra Templiphi amplification kit; GE Healthcare, Piscataway, NJ), were evaluated for whole-genome or whole-community genome amplification. An aliquot of pure-culture DNA (10 ng) or community DNA (100 ng) was amplified with both methods as described by Lage et al. (15) for Bst, by Wu et al. (26) for Templiphi, and by the manufacturer's manual for REPLI-g. DNA concentrations were determined by PicoGreen (Invitrogen, Carlsbad, CA). For pure cultures, the same amounts (1 μg) of unamplified genomic DNA (gDNA) and amplified DNA (aDNA) were fluorescently labeled with Cy3 and Cy5, respectively, as previously described (24). Labeled aDNA was mixed with the gDNA and cohybridized with whole-genome microarrays at 45°C for 10 h. For the community sample, aDNA was fragmented by sonication to an average size of ∼600 bp. gDNA or aDNA (3.0 μg) with or without sonication was labeled with Cy5 as described above, and labeled community DNA was hybridized to Geochip 3.0 (10), which contains over 24,000 probes for microbial functional genes involved in C, N, S, and P cycling, metal reduction, and organic remediation at 42°C for 10 h. All experiments were performed in triplicate using an HS4800Pro hybridization station (TECAN, San Jose, CA). Microarray scanning and data processing were performed as described previously (26) except that the signal-to-noise ratio (SNR) was set at 2.0 (12) for the selection and removal of bad spots. The cluster analysis was performed using the unweighted pair-wise average-linkage hierarchical clustering algorithm (7) with R project (www.r-project.org).

• Open in new tab
• Download powerpoint

Fig. 1.

Outline of evaluation of three MDA methods using pure cultures (a) and a community sample (b). gDNA, unamplified genomic DNA; aDNA, amplified DNA.

DNA yield.MDA is performed under isothermal conditions. The incubation temperatures for Bst, REPLI-g, and Templiphi were 50°C, 30°C, and 30°C, respectively. Amplification with Bst produced larger amounts of DNA than both phi29 kits. For pure cultures, Bst yielded 6.7 to 11.0 μg of DNA, while REPLI-g and Templiphi produced 5.7 to 8.3 μg and 1.6 to 2.9 μg, respectively. For communities, Bst produced 13.0 μg, while REPLI-g and Templiphi produced 9.0 μg and 3.8 μg of DNA, respectively (Fig. 2 ). The DNA yield was affected by the amount of DNA template, the incubation time, and the composition of the reaction buffer. The amounts of deoxynucleoside triphosphates (dNTPs) in Bst, REPLI-g, and Templiphi were about 24 μg, 20 μg, and 8 μg, respectively, a possible reason for the yield differences. Considering the incubation times (10 h for Bst, 1.5 h for REPLI-g, and 3 h for Templiphi), REPLI-g demonstrated the highest amplification efficiency, with fold increases of approximately 3.8 × 10² to 5.5 × 10² per hour in DNA for pure cultures and 60 per hour for community DNA.

• Open in new tab
• Download powerpoint

Fig. 2.

Yields of the three amplification methods for pure culture and community DNA. Error bars represent the standard deviations of the results for 3 replicates.

Amplification representativeness for pure cultures.The representativeness of pure-culture aDNA was determined using whole-genome open reading frame (ORF) microarrays. All three methods showed a high genome coverage (>99%) but a higher representational bias than was found for gDNA (Table 1; also see Fig. S1 in the supplemental material). The representational biases for gDNA and aDNA amplified by Bst, REPLI-g, and Templiphi were 0.02 to 0.04, 0.20 to 0.27, 0.08 to 0.11, and 0.09 to 0.20. Wu et al. (26) evaluated Templiphi for whole-genome amplification (WGA) using three pure cultures and showed lower but comparable biases. The proportion of genes whose hybridization signal ratios (aDNA/gDNA) showed 1.5-fold or greater differences was 0 for REPLI-g. For Templiphi, 0 to 1.4% of the genes showed a 1.5-fold difference, and none of the genes showed greater differences. However, Bst resulted in a higher percentage of genes showing differences, with 0.9 to 8.6%, 0 to 1.2%, and none of the genes showing >1.5-, 2.0-, and 3.0-fold differences, respectively (Table 1). The MDA methods showed different levels of bias with each pure culture. This could be due to uneven amplification in the regions of enzyme displacement, or errors in DNA quantitation may have resulted in different amounts of aDNA and gDNA being used for hybridization. REPLI-g showed the lowest coefficient of variation of representational biases, indicating that it performed similarly with different genomes. These results suggest that for amplification of pure cultures, all three MDA methods showed high genome coverage and a <3-fold bias. However, REPLI-g and Templiphi worked similarly and showed better results than Bst in terms of representational bias. Furthermore, no significant correlation was observed between GC contents and the hybridization signal ratios. We also found that the biased genes in aDNA produced by the different MDA methods were different and the biased genes within replicates were consistent (see Fig. S3 and S4 in the supplemental material). Thus, we inferred that the bias was a combination of random processes, such as the initial random priming and/or DNA breakage (28) and the preferences of the reaction systems of the MDA methods.

Amplification representativeness for a community sample.To evaluate amplification representativeness for environmental samples, in which the gene diversity is more complex than in pure or mixed cultures, DNA from a soil sample was used (Table 2; also see Fig. S2 in the supplemental material). In total, 781 genes were detected using gDNA. For REPLI-g aDNA, 91% of the genes detected using gDNA were positive, with an additional 733 genes detected; for Templiphi, 67% of genes detected by gDNA were positive, with an extra 701 genes detected. These results indicate that MDA-based methods increase detection sensitivity for functional genes within bacterial communities. However, Bst aDNA showed fewer (51%) genes detected by gDNA, with an additional 85 genes detected. REPLI-g showed the highest number of different genes and the highest genome coverage (the percentage of genes detected in aDNA that were detected in gDNA). However, in considering the redundant genes (genes detected in aDNA but not in gDNA), REPLI-g and Templiphi showed higher numbers of redundant genes than Bst, and Bst showed the lowest percentage of redundant genes (number of redundant genes/number of genes detected in aDNA). REPLI-g and Bst showed similar representational biases, and both methods showed no genes whose hybridization signal ratios had more than a 1.5-fold difference, while Templiphi showed a higher representational bias than REPLI-g or Bst.

Sonication was tested to determine if this increased labeling efficiency; however, no significant difference in the amount of fluorescent dye incorporated (measured by Nanodrop; data not shown) was observed with or without sonication. While sonication after Bst and REPLI-g amplification appeared to reduce amplification bias, it also lowered gene detection. Sonication may break the complex structure of aDNA formed by MDA, resulting in more even labeling. At the same time, some information may be lost due to DNA breakage within ORFs. Sonication did not help reduce amplification bias for Templiphi, although the number of genes detected decreased dramatically. Cluster analysis of microarray data showed that aDNA amplified by the same method, with or without sonication, clustered together and that aDNA amplified with REPLI-g had a gene structure more similar to that of the gDNA than was seen for either Bst or Templiphi (Fig. 3). aDNA produced by all the MDA methods showed relative abundances in gene categories and phylogenetic profiles that were similar to the results for gDNA (Fig. 4), suggesting that it would be reliable to use an MDA-based microarray method for biological interpretation.

• Open in new tab
• Download powerpoint

Fig. 3.

Hierarchical cluster analysis of amplified and unamplified community DNA based on GeoChip results. gDNA, unamplified DNA; Bst, amplified with Bst; Bst_S, amplified with Bst and sonicated before labeling; REPLI-g, amplified with REPLI-g; REPLI-g_S, amplified with REPLI-g and sonicated before labeling; Templiphi, amplified with Templiphi; Templiphi_S, amplified with Templiphi and sonicated before labeling.

• Open in new tab
• Download powerpoint

Fig. 4.

Relative abundance of all functional gene categories (a) and microbial classes (b) detected by GeoChip in gDNA and aDNA. Relative abundance was calculated based on the gene number. Bars: 1, gDNA; 2, aDNA amplified with Bst; 3, aDNA amplified with Bst and sonicated before labeling; 4, aDNA amplified with REPLI-g; 5, aDNA amplified with REPLI-g and sonicated before labeling; 6, aDNA amplified with Templiphi; 7, aDNA amplified with Templiphi and sonicated before labeling.

The quantitation feasibility of microarray hybridization has been demonstrated, and good correlations of quantitation based on microarray hybridization and real-time PCR were obtained. For instance, a significant linear relationship (r² = 0.96; P < 0.01) was observed between signal intensity and target DNA concentration within a concentration range of 60 to 1,000 ng with a 50-mer functional gene array (23). The expression of Shewanella oneidensis MR-1 genes under different environmental stresses was evaluated with the whole-genome microarray, and the results were validated with real-time quantitative reverse transcription-PCR. High correlations (r = 0.95 [n = 8] in reference 9; r² = 0.91 [n = 7] in reference 19; r = 0.95 [n = 6] in reference 4; and r = 0.93 [n = 10] in reference 17) were obtained. For gene detection in community samples, Rhee et al. (21) used real-time PCR analysis with several representative genes and showed that the results of microarray-based quantification were very consistent with those of real-time PCR (r² = 0.74 [n = 6]). He et al. (11) also showed that the gene copy number measured in a soil sample by quantitative real-time PCR was well correlated with the signal intensity detected by GeoChip 3.0 (r = 0.724 [n = 91], P = 0.0001). These results suggest that microarray hybridization data are consistent with real-time PCR data and, hence, that the microarray hybridization signals appear to be reliable.

Bst seems to have a lower requirement for DNA quality than phi29, as Bst DNA polymerase was shown to be effective in WGA of DNA from formalin-fixed, paraffin-embedded tissues (1), while a restriction enzyme fragmentation step was needed for phi29 to amplify DNA derived from formalin-fixed, paraffin-embedded tissue (25). We observed a higher success rate with Bst amplification than with either REPLI-g or Templiphi, especially when the DNA quality (i.e., absorbance ratios of 260/280 and 260/230) was low (data not shown), which may be an advantage for environmental samples. Bst has been used to amplify yeast, human, and formalin-fixed, paraffin-embedded tissue DNA (1, 15). phi29 has been proven suitable for WGA of a single bacterium (20) and for microbial communities (26). phi29 provided more representational results than Bst for WGA of human DNA at the single-cell level, as judged by locus-specific PCR (22), while Bst was shown to be relatively unbiased compared to phi29 for yeast genome analysis by array comparative genome hybridization (15). However, all of these studies used DNA from single organisms. For individual genomes, all three MDA methods showed a high genome coverage and a relatively low amplification bias with pure culture DNA, as shown in this study. For complex microbial community samples, however, 10,000 genomes may be present in a single sample. Our previous study showed that the amount of DNA template would influence the representativeness of MDA products (26). Though we used 10× DNA template for MDA of the community sample, the MDA products for the community sample still showed less genome coverage than was the case for pure cultures. This increased complexity may explain why we observed such differences in representational bias using different MDA methods. The higher error rate of Bst may explain the lower gene detection and higher amplification bias, especially when the amount of template is limited (2). Though both kits are phi29-based, REPLI-g and Templiphi showed quite different results in WCGA. Templiphi was originally designed to amplify circular DNA, such as plasmids, while REPLI-g is designed for genomic DNA. The reaction buffer composition (e.g., random primer, dNTPs, enzyme concentration, and other assisting enzymes) may affect amplification results, as different phi29 commercial kits perform differently.

In conclusion, three MDA methods, Bst, REPLI-g, and Templiphi, were evaluated for use in WCGA of microbial DNA with microarrays in this study. The amplification biases of all methods were less than 3-fold. REPLI-g, a phi29-based kit, showed less bias and higher gene detection than the other two kits, while Bst had a higher success rate and might be more suitable for low-quality DNA. Sonication after WCGA by Bst and REPLI-g reduced the representational bias while also reducing gene detection.