Amplification of ORFs by PCR from a normalized genomic template

S. cerevisiae served as a test case for genome-scale ORF cloning in our group, which has educated our production of other genome-scale ORF collections. The overall strategy was to use gene-specific PCR to capture the ORFs from a genomic DNA template, which provided a single, quality controlled, and normalized template. This was feasible because only 4% of the predicted ORFs contain introns. These ORFs were maintained in the target list as the presence of introns was not considered problematic for expression in yeast by scientists in this field and because we plan to return to these using cDNA methods. The cloning effort was carried out in four distinct phases, each characterized by production of a group of successfully cloned and verified ORFs, and a set of cloning and/or sequence failures that were passed forward to the next phase. For all four phases, ORFs were amplified by PCR and cloned using nonrestriction enzyme-mediated strategies into a Gateway recombination-based cloning vector (Fig. 1).

Open in a separate window

Figure 1.

Workflow diagram of clone production. The entire process from the design of primers to production of clone stocks is shown for the four production phases. The process started by designing primers for every ORF in the genome. The primers were used to amplify the ORFs from the genome. Subsequent amplifications with universal primers generated ORF sequences flagged by recombinational cloning sites at either end and were monitored by a diagnostic gel. The product was cloned into a recombinational cloning vector via a BP clonase reaction or In-Fusion reaction. Competent bacterial strains were transformed with the reaction mix to yield colonies that were isolated robotically, cultured in liquid medium, and stored as 15% glycerol stocks.

The overall failure rate during the first two phases was ~70% and failures were primarily due to quality and design issues pertaining to the RG primers and polymerase choice. For Phases Three and Four, we used higher fidelity polymerases than those used for previous phases. Together with the inclusion of newly designed ORF-specific primers, both polymerases improved overall cloning success (Table 1).

Table 1.

Summary of the major differences and results of the four cloning phases

Open in a separate window

^aNumber of reads includes only the successful sequencing reads.

Capture of PCR products in a vector compatible with cloning via enzyme-mediated, site-specific recombination

PCR products were initially captured in the Gateway entry vector pDONR201 and later in pDONR221. Using this system, ORFs are captured in the correct orientation via subtle but noncompatible differences between the 5′ and 3′-flanking att site sequences. Capture was initially done using the BP reaction, a method well suited to high-throughput cloning. However, we found that approach did not efficiently capture fragments 2.5 kb or longer. For this reason, in our last cloning phase, a linearized derivate of pDONR221 was used in conjunction with Clontech’s In-Fusion method for ORFs longer than 2.5 kb. Capture of the PCR amplified fragment in the vector was defined as positive when colonies were detected after transformation, thus allowing single-colony isolation on solid agar (Fig. 1).

Initially, we selected four colonies per ORF and maintained them separately, as we expected that this would increase the likelihood of obtaining at least one mutation-free clone. With experience, our methods have improved such that the benefits of choosing multiple isolates no longer outweigh the costs, as 80% of ORFs can be accepted based on a single isolate. Thus, by Phase Four, we revised our strategy to isolate one colony per ORF. After capture into the vector, half of the capture reaction was plated on solid agar and the remaining transformation mix was stored at −80°C, allowing us to return to the frozen transformation mix without the need to repeat the entire cloning procedure.

DNA sequencing reveals high-fidelity capture of 87% of known and predicted yeast ORFs

To provide a well-annotated collection, we sequenced the complete insert (including the 5′ and 3′ attB sequences), assembled sequence contigs, and compared the assemblies to corresponding reference sequences. Our strategy was to first perform end-read sequencing of all clones using vector-specific (universal) primers followed by insert-specific (internal) primers designed to cover the gaps. Moreover, sequence data were managed, quality-checked, assembled, and analyzed using the Automated Clone Evaluation (ACE) suite of software tools. An assembled sequence contig was considered to be full length if it covered the entire ORF and the flanking sequences relevant for site-specific recombination. After contig assembly, the sequences were compared with their corresponding reference sequences in ACE at both the nucleotide and amino acid levels. For this cloning project, insertion, deletion, and nonsense mutations were defined as nonacceptable “discrepancies” where sequence confidence was high (Phred quality score >25). Low confidence regions were resequenced to obtain better quality. Clones were accepted if the assembled experimental sequences matched the corresponding reference sequences perfectly or contained discrepancies deemed acceptable (any number of silent changes and up to two amino acid changes). The majority of clones had no differences with the reference peptide sequence (82%), and all differences at both the nucleotide and amino acid levels are carefully documented in our distribution database (http://plasmid.hms.harvard.edu) and in Supplemental Table 1. Full-length sequences for all clones are also available at GenBank.

ORF size, PCR primer attributes, and GC content contribute to cloning failure

Despite repeated attempts, there were a number of clones for which we were never successful at creating acceptable clones. We analyzed these recalcitrant ORFs and identified several factors that may have contributed to cloning failure. Clearly, large genes were more difficult to clone: the yeast ORF collection we describe here covers more than 93% of ORFs <1 kb, 85% of ORFs 1–4 kb, but 36% of ORFs >4 kb (Fig. 2). Among the 128 large ORFs (>4 kb) for which we failed to obtain a qualified clone, 24 failed at the capture step of clone production, and 104 failed at sequence validation. In general, GC content did not seem to be a contributing factor, except in the extreme cases where ORFs with very low GC content were more likely to fail. Because there are only 39 ORFs in the yeast genome with GC content <30%, this factor did not have a big impact on our overall cloning efforts.

Open in a separate window

Figure 2.

Size distribution of the yeast gene clone collection as compared with the sizes of the predicted protein-coding sequences as defined by the current annotation at SGD. The cloning success rate was more than 90% for genes smaller than 1 kb and about 85% for medium sized ORFs (1–4 kb). For large ORFs (>4kb), only 36% were cloned and accepted after sequence analysis.

In our amplification strategy, primers target gene-specific regions of ~20–30 nucleotides in length that correspond to the extreme 5′ and 3′ ends of the coding sequence. We were surprised to find that many different genes share identical 5′ and 3′ ends, making it difficult to amplify all desired ORFs. To determine the extent to which primer specificity contributed to cloning failure, we compared primer sequences for all ORFs with one another. Matching primer pairs typically causes favored amplification of the shortest gene sharing the primer sequences. We also examined whether primer sequences could bind elsewhere in the genome (not necessarily at the ends of other genes). This situation leads to failed amplification or amplified junk sequence. We found that high primer sequence similarity with other ORFs, as well as high primer stickiness to genomic DNA, reduced the cloning success rate from 87% to 70%. Details are listed in Supplemental Table 2 and Supplemental Figure 1.

Failure to clone some sequences could reflect errors in the target ORF sequences

For 31 ORFs, we observed that multiple independent clones had the same frameshifting nucleotide discrepancy compared with the genome reference. In these cases, failure to obtain an isolate matching the genome sequence may be due to an error in the genome sequence, rather than an error in the amplification and cloning process. ORF YBR078W provides an instructive example. For this ORF, we analyzed a total of seven isolates from three cloning efforts with different PCR primers and enzymes. All clones carry the same single base-pair insertion at position 1609 of the coding sequence, which is unlikely to be an artifact of cloning. This and similar cases are summarized in Table 2 and may be helpful in annotation of the yeast genome.

Table 2.

Persistent differences between experimental and reference sequences that may point to errors in genome sequence or annotation

Open in a separate window

Primer design, synthesis, and PCR amplification

For phases one and two, first-step PCR primers were designed and synthesized by Research Genetics as Yeast Gene Pair Primers (SGD website). Gene-specific primer sequences were designed based on the initial S. cerevisiae ORF annotation. For Phases One and Two, second-step PCR was performed with universal primers as follows: forward primer, 5′-GGGGACAAGTTTGTACAAAAAAG CAGGCTTCCAGCTGACCACCATG; reverse primer (under-line, attB sequence; bold, RG tag sequence), 5′-GGGGAC CACTTTGTACAAGAAAGCTGGGTATC CCCGGGAATTGCCA.

For Phases Three and Four, first-step PCR, optimal gene-specific primers were designed using a modified nearest-neighbor algorithm (Sugimoto et al. 1995). The Tm was set to 60°C. attB sequences were appended to the gene-specific region as follows: forward primer, 5′-TACAAAAAAGCAG GCTCCACC-atg then gene specific sequences; reverse primer, 5′-GTACAAGAAAGCTGGGTC-normalized STOP codon then gene-specific sequences (underline, partial attB sequences). For second-step PCR, universal primers were as follows: forward primer, 5′-GGGGACAAGTTTGTACAAAAAAGCAGGCTCC; reverse primer, 5′-GGGGACCACTTTGTACAAGAAAGCTGGGTC (underline, partial attB sequences). For all phases, a Kozak consensus sequence was included. PCR amplification conditions were set up according to product specification from manufacture (Table 1).

Capture of amplified ORFs

For capture of amplified ORFs in Phase One, PCR fragments were gel purified prior to BP, followed by transformation into DH5α, plated into 6-well plates. Four single colonies were picked by hand for each ORF. After growing in liquid culture and making glycerol stocks, DNA was prepared for LR transfers into pBY011 (Butcher et al. 2006) for sequencing. For capture of amplified ORFs in Phase Two, the size of PCR fragments after step one PCR was checked via agarose gel electrophoresis and PCR fragments from step two PCR were gel purified prior to BP capture into pDONR201, followed by transformation into DH5α (T1-resistant) and plating into 48-well plates for robotic colony isolation. Four individual colonies per ORF were picked. End-read sequencing was done on entry clones (pDONR201) at Agencourt. For capture of amplified ORFs in Phase Three, capture was done as described for Phase Two, except that the vector was pDONR221. End-read sequencing was done on entry clones (pDONR221) at the Dana Farber/Harvard Cancer Center DNA Resource Core (Harvard Medical School). Internal-read sequencing was done at Ludwig Institute for Cancer Research (University of California San Diego). For Phase Four, there were two possible workflows for PCR amplification and ORF capture. For ORFs <2500 bp, second-step PCR was used to generate att sites. Next, ORFs were captured by direct BP reaction. For ORFs >2500 bp, after the first-step PCR, products were separated by agarose gel electrophoresis, isolated, and purified. Next, we performed In-Fusion cloning into linearized pDONR221 and 50% of the transformation mix was plated to LB/agar (the remainder was frozen at −80°C with 15% glycerol). One clone per ORF was isolated for glycerol stock production and sequencing. End-read and internal-read sequencing were both done at Harvard Medical School.

To obtain a linear pDONR221 version that would allow for directional cloning, we introduced unique restriction sites in the 5′ (NcoI) and 3′ (XhoI) att sequences by site-directed mutagenesis, and sequence verified the correct insertion after In-Fusion reactions.

Identification of known or predicted transcription factor genes

A total of 421 S. cerevisiae ORFs were identified as candidate transcription factors (TFs) according to their annotations in several curated databases. A total of 329 ORFs were identified as candidate TFs based on at least one of the following three criteria: (1) annotated in Gene Ontology as “DNA-binding” and one of “Transcription Factor,” “Transcriptional Activator,” or “Transcriptional Repressor” in the Yeast Proteome Database (Costanzo et al. 2000); (2) annotated as “Transcription Factor” in the Munich Information for Protein Sequences (MIPS) Database (Mewes et al. 2002); (3) selected for genome-wide location analysis (ChIP-chip) in a prior global study (Harbison et al. 2004). An additional 92 ORFs were added to this list, as lower confidence TFs candidates, based on their curation in the Pfam database as containing at least one of 44 known DNA-binding domains (Bateman et al. 2004).

Subcloning of ORFs into expression vector, bacterial expression, and purification of proteins

All of the 96-well purifications were performed robotically with optimized protocols on the Biomek Fx (Beckman Coulter) from cell lysis to elution of the bound protein. For GST affinity purification, the robotic deck was setup using 25 mL of lysis buffer (100 mM NaH₂PO_4, 10 mM Tris at pH 8.0), 25 mL of wash buffer (100 mM Tris, 500 mM NaCl, 2 mM EDTA, 0.1% Triton X-100, 10% glycerol), and 15 mL of elution buffer (wash buffer plus 1% glutathione). All buffers are maintained in chilled conditions on ice and contain complete protease inhibitors (Roche). For purification, cell pellets were thawed for 15 min at room temperature and placed on the deck of the robot at the appropriate location. The cells were lysed in 200 μL of lysis buffer and resuspended thoroughly by repeated pipetting with the help of the 96-well pod of the robot. Ten microliters of DNase mix (10 mg/mL DNase [Sigma] in 900 mM MnCl₂, 100 mM MgCl₂) was added to the lysate and incubated with shaking (900 rpm) for 10 min. Next, the lysate was allowed to bind to 30 μL of MagneGST (Promega) with shaking at 900 rpm for 10 min in the clockwise direction and 10 min in the anticlockwise direction. The beads were separated with the help of a magnabot, and the remaining lysate was removed and discarded with the help of the 96-well head of the robot. The MagneGST beads with bound protein were washed three times with wash buffer by shaking at 900 rpm for a total of 5 min each time, 2.5 min in the clockwise direction and 2.5 min in the anticlockwise direction. The bound protein was eluted in 50 μL of elution buffer.

Immunoblotting

Proteins were analyzed on precast 4%–12% XT Criterion gradient gels (BioRad) according to the manufacturer’s protocols. Immunoblots were probed with HRP-conjugated rabbit anti-GST antibody (Sigma) at 20 ng/mL final concentration and developed using SuperSignal West Femto Maximum Sensitivity Substrate (Pierce) according to the manufacturer’s protocols.

Protein-binding microarray experiments and data analysis

Whole-genome yeast intergenic microarrays were synthesized essentially as described previously (Ren et al. 2000). PBM experiments and data analysis were performed essentially as described previously, using Rap1 at a final concentration of 20 nM in the protein-binding reaction mixture (Mukherjee et al. 2004; Berger and Bulyk 2006). After quality control filters were applied to the data from the single PBM experiment, 5987 spots (92%) remained for subsequent analysis. A total of 77 features were significantly bound at a Bonferroni-corrected P value significance threshold of 0.001. To search these sequences for over-represented DNA sequence motifs, we used BioProspector (Liu et al. 2001) at widths ranging from 6 to 18 bp (Huber and Bulyk 2006). Motifs were then evaluated according to their group specificity scores (Hughes et al. 2000).

We thank all past and present members of the Harvard Institute of Proteomics who have contributed to the development of the techniques that made this work possible. Special thanks to Dr. Leonardo Brizuela, Dr. Joeseph Pearlberg for helpful discussions and advice, and Stephanie Ness (LICR, San Diego) for her efforts in performing DNA sequencing. We thank Katrina Saulrieta and Zachary Smith for technical assistance. This work has been supported by grants R01 HG002923 (J.L.) and R01 HG003420 (M.L.B.) from the National Institutes of Health, and funds from the Ludwig Institute for Cancer Research.