Intrinsic and selected resistance to antibiotics binding the ribosome: analyses of Brucella 23S rrn, L4, L22, EF-Tu1, EF-Tu2, efflux and phylogenetic implications

Background Brucella spp. are highly similar, having identical 16S RNA. However, they have important phenotypic differences such as differential susceptibility to antibiotics binding the ribosome. Neither the differential susceptibility nor its basis has been rigorously studied. Differences found among other conserved ribosomal loci could further define the relationships among the classical Brucella spp. Results Minimum inhibitory concentration (MIC) values of Brucella reference strains and three marine isolates to antibiotics binding the ribosome ranged from 0.032 to >256 μg/ml for the macrolides erythromycin, clarithromycin, and azithromycin and 2 to >256 μg/ml for the lincosamide, clindamycin. Though sequence polymorphisms were identified among ribosome associated loci 23S rrn, rplV, tuf-1 and tuf-2 but not rplD, they did not correlate with antibiotic resistance phenotypes. When spontaneous erythromycin resistant (eryR) mutants were examined, mutation of the peptidyl transferase center (A2058G Ec) correlated with increased resistance to both erythromycin and clindamycin. Brucella efflux was examined as an alternative antibiotic resistance mechanism by use of the inhibitor L-phenylalanine-L-arginine β-naphthylamide (PAβN). Erythromycin MIC values of reference and all eryR strains, except the B. suis eryR mutants, were lowered variably by PAβN. A phylogenetic tree based on concatenated ribosomal associated loci supported separate evolutionary paths for B. abortus, B. melitensis, and B. suis/B. canis, clustering marine Brucella and B. neotomae with B. melitensis. Though Brucella ovis was clustered with B. abortus, the bootstrap value was low. Conclusion Polymorphisms among ribosomal loci from the reference Brucella do not correlate with their highly differential susceptibility to erythromycin. Efflux plays an important role in Brucella sensitivity to erythromycin. Polymorphisms identified among ribosome associated loci construct a robust phylogenetic tree supporting classical Brucella spp. designations.


Background
Brucellosis is a zoonotic disease caused by the Gram-negative bacterium Brucella. It is taxonomically related to plant pathogens and other animal symbionts and is transmitted to humans from infected domestic animals and wildlife through contact during animal husbandry practices, meat production, or by ingestion of unpasteurized milk products. The genus Brucella contains six classical species reflecting host preferences [1,2], and additional species have been proposed to include marine isolates from seal, dolphin, and porpoise [3]. The classical species and their hosts are: B. abortus, bovine; B. melitensis, caprine; B. suis, porcine; B. ovis, ovine; B. canis, canine; and B. neotomae, desert wood rat. However, B. suis and B. canis have similar metabolic profiles [4] and genomic maps [5], supporting their close relationship. Similarly, the metabolic characteristics and phage susceptibility of B. suis biovar 5 are more like that of B. melitensis rather than B. suis [6,7].
The classical Brucella spp. designations are still widely used to emphasize important pathogenicity, virulence, and host preference differences among the Brucella even though similarity among the ribosomal RNA loci led to the designation of Brucella as a monospecific species [8,9], B. melitensis. Brucella speciation may have arisen as a result of their isolation due to different preferred hosts and to divergence of the host species [10] even though their 16S rrn loci are identical [11][12][13][14]. In any case, discordant genotype/phenotype may require the use of other widely conserved loci to define bacterial species [15,16].
Meyer [17] found differences in sensitivity to erythromycin among the classical species of Brucella and their biovars by measuring inhibition of growth using high and low concentration antibiotic discs. Brucella abortus biovars except biovar 2 were resistant to erythromycin, and B. ovis, B. melitensis, and B. canis were intermediate in resistance between B. abortus and B. suis. Only B. suis strains were sensitive to the high concentration antibiotic discs. Meyer argued that investigating the ribosomal structure could explain these differences in sensitivity and generate critical knowledge "to account for and recapitulate the lineage of species and biotypes of Brucella".
Bacterial susceptibility to macrolide and lincosamide antibiotics results from their binding to 23S rRNA, inhibiting protein synthesis by blocking the peptide exit tunnel [18][19][20]. Bacteria can become resistant to macrolides and lincosamides by spontaneous mutations of ribosomal associated loci or by increased efflux. Resistance to macrolides and lincosamides is commonly due to (i) mutation of 23S rrn causing a reduction in the binding of the antibiotics to the peptidyl transferase center [21,22], typically nts A2058, A2059, A2062, and C2611, Escherichia coli (Ec) 23S rRNA numbering, (ii) mutation of ribosomal proteins L4 or L22 leading to widening the entrance to the peptide exit tunnel allowing access to the tunnel even in the presence of the antibiotics [18,20,[23][24][25], (iii) methylation of ribosomal 23S rRNA [26], or (iv) increased efflux [27,28]. Bacterial resistance to synthetic macrolides or ketolides can be conferred by mutation of ribosome associated factor EF-Tu [29].
There are several families of efflux pumps, though few non-RND (resistance nodulation division) family efflux pumps cause intrinsic or spontaneous resistance of Gramnegative bacteria to common clinical antibiotics [27,28,30,31]. Inability to demonstrate efflux activity however does not necessarily mean a lack of efflux. Efficiency of efflux of antibiotics is variable, being dependent on the structure of the antibiotic. Antibiotic resistance can be complex as observed for Haemophilus influenzae L22 mutant HMC-C [32]. For this mutant, an increase in macrolide MIC values was only shown in the presence of efflux [33].
Here, we show that the large differences in relative intrinsic susceptibilities of reference strains of Brucella and three marine isolates to macrolide antibiotics and a lincosamide do not correlate with ribosomal associated polymorphisms. We establish that antibiotic efflux plays an important role in differential antibiotic susceptibility in Brucella. A robust phylogenetic tree constructed from concatenation of ribosome associated polymorphisms illuminates relationships among the Brucella.

MIC determination by Etest
The relative MIC values of the classical Brucella spp. and three marine isolates (Table 1) to macrolides and a lincosamide were determined by use of the Etest. Log-fold differences in MIC values were found (Fig. 1). The susceptibility of Brucella was similar for the three macrolides erythromycin, azithromycin, and clarithromycin. Only B. abortus, except biovar 2, and B. melitensis had MIC values of ≥ 16 μg/ml. The pattern of sensitivity of Brucella to the lincosamide clindamycin differed from that of the macrolides. Generally, MIC values were higher for clindamycin than for the macrolides. Brucella abortus, except biovar 2, was the most resistant to clindamycin, having MIC values of ≥ 128 ug/ml. Only B. melitensis biovars 2 and 3 had lower MIC values for clindamycin than for erythromycin. For B. suis, clindamycin MIC values ranged from a low of 3 μg/ml to a high of 24 μg/ml. The other Brucella with the exception of the seal isolate, ranged from 2 to 64 μg/ml. The seal isolate was resistant to clindamycin.

23S rrn sequence comparisons
Sequences of two regions of the Brucella 23S rrn encoding 2498 nts (69 to 1678 and 1920 to 2807), including sites of 23S rrn mutations known to increase bacterial resistance to macrolides and clindamycin were determined and compared ( Table 2). Mixtures of cells or DNA (1:3) with disparate 23S rrn sequences were amplified to demonstrate that heterogeneity among the three 23S rrn copies would be detectable (data not shown). Though the distal portion of 23S rrnC from the genomic B. suis 1330 sequence could not be amplified with either of two primer pairs that were complementary to the published B. suis 23S rrnC genomic sequence, amplification was successful using primers homologous to internal, conserved genomic rrnC sequences from all three Brucella genomes and sequences flanking rrnC from B. abortus and B. melitensis. The amplified distal portion of rrnC from B. suis 1330 was identical in sequence to that of rrnA and rrnB from B. suis. Among the 23S rrn sequences from Brucella, three polymorphic and three monomorphic sites were identified. In addition, three monomorphisms were identified in the 23S rrn intervening sequences.  Table 2). Note that all the 23S rrn Brucella positions are numbered based on B. abortus 23S rrnA, including the intervening sequence. The only polymorphism that occurred in the peptidyl transferase center was nt 2632 (2610, Ec). No correlation could be made between the polymorphisms and relative antibiotic susceptibility. Other sites known to affect susceptibility to macrolides and clindamycin were not polymorphic.
Monomorphisms were found in both 23S rrn and in the 23S rrn intervening sequences. Monomorphisms were identified in 23S rRNAs from dolphin (A955G); B. neotomae (insertion of a C between nt 1002-1006); and B. suis biovar 5 (T2090C). Several intervening sequences of the Brucella 23S rrn loci varied from the consensus sequence reported by Bricker [34]. The C indel in the intervening sequences of B. melitensis 16 M, forming a string of six Cs instead of five beginning at nt 222, reported by Bricker was confirmed. The other two monomorphisms occurred in B. suis biovar 5 (C219T) and B. melitensis biovar 3 (C206T).

L4 analyses
Though the GenBank Brucella genomic rplD sequences encoding ribosomal protein L4, differed due to an indel in rplD found only in the B. melitensis 16 M genomic sequence, we did not observe this indel in our sequence of rplD from B. melitensis 16 M. We found the three rplD genes from the genomes were identical. Though no polymorphisms were identified among the rplD sequences, three monomorphisms were discovered among their amino termini. Two sequence transitions, B. neotomae (G108A) and B. suis biovar 2 (C213T), were found. Both of which were silent. A transversion identified in the porpoise isolate (G314T) would replace an Arg, a charged amino acid (aa), with Leu, a noncharged one.

L22 polymorphisms
Putative L22 sequences from the Brucella reference strains and three marine isolates ( Fig. 2) were determined and their tertiary structures predicted and compared by using Swiss-Pdb Viewer (Fig. 3). Among the Brucella putative L22 sequences, all variations except one occurred in the βhairpin loops or near the carboxy termini. Brucella suis biovar 5 alone had an alternate Ala codon at aa 44. The Brucella β-hairpin loops were polymorphic and variable in length due to variable copy numbers of a two aa motif, Gly-Arg. The lengths of all the β-hairpin loops of putative L22 peptides except those from B. neotomae and B. suis biovars 2 and 3 were equal, 11 aa. The β-hairpin loops of L22 from B. suis biovars 2 and 3 were shorter due to a net two aa (Gly-Arg) deletion, while the β-hairpin loop from B. neotomae was longer due to a net two aa (Gly-Arg) insertion. Within the loop of the β-hair pin at aa 101, there was either a Gly, Val, or an Asp. While the variation of sequence at aa 101 of L22 did not greatly affect the predicted tertiary structures, the indels did (Fig. 3). A polymorphism was also identified very near the 3'-end of rplV. L22 polymorphic sites grouped the Brucella differently than other loci in this study. Putative L22 sequences from B. abortus and B. melitensis were identical. Brucella suis biovars 1, 4 and 5, and the marine isolates were identical. Putative L22 sequences from B. suis biovars 2 and 3 differed from those of biovars 1, 4 and 5 in having one rather than two Gly-Arg motifs. Putative L22 sequences from both B. ovis and B. neotomae were unique. No correlation could be made between relative antibiotic susceptibilities of the Brucella strains and their L22 sequences.
Though MIC values of the ery R mutants of the classical strains B. suis, B. canis, and B. neotomae increased, they did not increase as much as those of the marine isolates ( Table   Ribosomal protein L22 polymorphisms among Brucella reference strains and three marine isolates Figure 2 Ribosomal protein L22 polymorphisms among Brucella reference strains and three marine isolates. The putative peptide sequence of L22 is underlined, parentheses bracket polymorphic sites and list amino acids found among L22 peptides at that site. In regions where sequence was variable, the sequence for each putative Brucella L22 is given below. Amino acids occurring in the stalk of the β-hairpin appear in bold-italics and amino acids occurring in the loop of the β-hairpin are in bold.
Beneath amino acids that are double underlined is a list of Brucella strains containing those aa. The single letter code is used to denote the putative aa sequence of the peptides; b = biovar and numbers following "b" designate biovar numbers. Accession numbers of rplV sequences for each strain are deposited in GenBank and are listed in Materials.

4).
While MIC values increased for the ery R mutants of the classical Brucella strains, the increases were only 2 to 6fold compared to 15 to > 256-fold for the marine Brucella.
Over half of the marine ery R mutants had erythromycin MIC values of 128 μg/ml or higher, and all the dolphin mutants had MIC values greater than 256 μg/ml. The clindamycin MIC values of the ery R mutants were similar to those of the parental strains except for porpoise b which had a mutation in 23S rrn and the seal isolates a and c.
Mutations among the marine ery R mutants were found in two ribosomal associated loci, 23S rrn and rplD. Only a single mutation was identified in 23S rrn. Porpoise isolate b had a mutation within the peptidyl transferase center of 23S rrn, nt 2058 (Ec), but all three 23S rrn copies were not mutated, as the signal was mixed. This mutant was resistant to both erythromycin and clindamycin, MIC values >256 μg/ml. Most ribosomal associated mutations occurred in rplD (Table 4), and these were only found among the marine isolates. The mutations were not random. Several ery R isolates had mutations at nt 209 or nt 217. At nt 209, porpoise isolates a and d and dolphin isolates a and c had an A instead of a G, substituting an Asp for a Gly. Dolphin ery R mutants isolates b and d had a T instead of a C at nt 217 of rplD, resulting in the incorporation a Cys of rather than an Arg. Seal ery R isolate c had an A instead of a C at nt 217 which resulted in the incorporation of a Ser rather than an Arg. Two of the porpoise ery R isolates c and e had deletions in rplD, resulting in the loss of 18 or 30 codons. The deletion of 18 aa in L4 of por-poise c is consistent with recombination between two copies of 5'GGG-CCG-CGC-3' occurring between nt 153-161 and 207-215.
Only one ribosomal associated loci mutation was identified among the B. suis, B. canis, and B. neotomae ery R mutants by analyzes of 23S rrn, rplD, and rplV. A duplication of a six-bp repeat in the β-hairpin loop of L22 of B. neotomae isolate d expanded the number of Gly-Arg aa repeats from three to four (Fig. 2, Table 4). This mutant had a slightly higher MIC value for erythromycin.

Efflux
Erythromycin and clindamycin MIC values of the reference strains and the ery R mutants were analyzed in the presence of the efflux inhibitor PAβN (Tables 4 and 5  Structures were predicted based on coordinates of L22 from Thermus thermophilus [52] and prepared using Swiss-Pdb viewer [50,51]. Region containing β-hairpin loops (→).

Phylogenetic tree
A phylogenetic tree was constructed using concatenated 23S rrn, rplV, tuf-1, and tuf-2 (Fig. 4). Brucella formed a node with the closest clades being other α-Proteobacteria, Agrobacterium, Mesorhizobium, and Caulobacter followed by Leptospira and γ-Proteobacteria, Xylella, Acinetobacter, and the facultative intracellular animal pathogen The tree constructed from a concatenated sequence, i.e. a supergene or supermatrix, was consistent with a concatenated tree calculated from individual loci (data not shown). Both trees supported classical classification, clustered the marine isolates with B. melitensis, and indicated intrinsic differences among marine Brucella. Bootstrap numbers (Fig. 4) were robust for all nodes (99 or 100) except B. ovis, which clustered with B. abortus; in the additive tree, B. ovis formed a unique branch. Though B. suis   were low and more similar to those of B. suis than to those of either B. melitensis or B. abortus. The patterns of relative sensitivity to macrolides of the reference Brucella were similar for erythromycin, clarithromycin, and azithromycin but differed from that for clindamycin. The susceptibility of B. suis to relatively low concentrations of the macrolide azithromycin suggests that this antibiotic may be a beneficial treatment for B. suis infections as it has a long in vivo half-life (50 hours), concentrates in macrophages, and lacks uptake saturation [35].
Ribosomal associated loci 23S rrn, rplD, rplV, tuf-1, and tuf-2 were analyzed for polymorphisms. Three monomorphisms were identified among rplD loci, but only one of them resulted in a difference among the putative L4 sequences. Although polymorphism was high among the tuf-1 and tuf-2 loci, all were silent. Sequences among 23S rrn and rplV loci were polymorphic.
The three polymorphic sites identified among the Brucella 23S rrn loci separated them into three groups ( Table 2). The only sequence difference among the 23S rrn peptidyl transferase centers of the reference Brucella strains was at nt 2610 (Ec), where there was either a T or a C. Many nucleotides in the peptidyl transferase center are conserved among bacteria and other organisms, but nt 2610 (Ec) is not. Either a T or C is common in bacteria. In any case, a T2610C (Ec) mutation in 23S rrn from S. pneumo-nia only slight affected its MIC values for macrolides and clindamycin [36]. Mutation of the peptidyl transferase center of 23S RNA (A2058G, Ec) of porpoise ery R mutant isolate b increased the erythromycin and clindamycin MIC values from 1.5 and 16 μg/ml, respectively, to >256 μg/ml. These MIC values were unaffected by the presence of efflux inhibitor PAβN. Concurrent appearance of resistance to erythromycin and clindamycin by mutation of nt 2058 (Ec) is observed in other bacteria [23]. Methylation of either nt 2059 or 2058 (Ec) of the peptidyl transferase center reduces the sensitivities of bacteria to macrolides and lincosamides [26]. We were unable to identify homologs of any 23S erm methylation genes by BLAST [37], but, then, methylation of ribosomal rRNA is much more widely described in Gram-positive clinical isolates [26].
The rplV sequences of the reference Brucella strains and marine Brucella were polymorphic, resulting in the differences among their putative L22 peptide sequences and lengths of the L22 β-hairpin loops. This was unexpected because L22 peptide sequence is conserved within a bacterial species [18,38] and the length of the L22 β-hairpin loop is highly conserved across biological kingdoms [38]. Differences in β-hairpin loop lengths among the Brucella L22 peptides were due to variable numbers of Gly-Arg repeats (Fig. 2). Though B. neotomae ery R isolate d had four Gly-Arg repeats, due to a six base insertion, the mutant's Brucella strains are as listed in Table 1; MIC, minimum inhibitory concentration; Ery, erythromycin; CL, clindamycin; PAβN, efflux inhibitor Lphenylalanine-L-arginine β-naphthylamide; b, biovar; ab, abortus; mel, melitensis; neo, neotomae.
Phylogeny of Brucella calculated using highly conserved ribosomal associated loci Figure 4 Phylogeny of Brucella calculated using highly conserved ribosomal associated loci. Shown is the single optimization alignment tree based on rplV, tuf-1, tuf-2, and 23S rrn sequences from 28 taxa consisting of the 21 Brucella strains (see Table 1 The single amino acid difference found among the putative L4 sequences of the reference and marine strains could not be correlated with a difference in MIC values. Among the ery R mutants, all but two of the mutations were identified in rplD, and, interestingly, they only occurred among the marine ery R isolates. All erythromycin MIC values that increased among the ery R marine isolates were lowered by the efflux inhibitor PAβN. Nevertheless, some of the MIC values remained relatively high in the presence of PAβN. The L4 peptides of these mutants may work in conjunction with or be dependent on specific efflux RND pumps as shown for Haemophilus influenza HMC-C [32,33].
The tuf-1 and tuf-2 loci were the most polymorphic of the ribosomal associated loci examined, yet their putative peptide sequences were identical. Strain sequence differences between tuf-1 and tuf-2 were confined to the borders. This is consistent with gene conversion occurring more efficiently within conserved sequences rather than near the borders. Given that B. melitensis and B. abortus genomes have fewer single nucleotide polymorphisms (SNP) between them than either has with B. suis, tuf-1 and tuf-2 from B. abortus and B. melitensis were expected to be highly similar. This was not the case. Brucella abortus and B. suis tuf-1 and tuf-2 had few sequence differences (  [40]. Many clinical isolates are resistant to antibiotics due to increased efflux as a result of mutations of efflux promoters and global and physically linked regulator genes or mobilization of insertion sequences (for a review see [28]). Most of the ery R strains had increased antibiotic efflux, though the marine ery R strains had larger increases in efflux than those of the classical reference strains of B. suis biovar 1, B. canis, and B. neotomae (Table 4). This suggests a fundamental biological difference between these groups. It is known that the marine Brucella have a high copy number [41] of the insertion sequence IS711 [42]. IS711 has been shown to mobilize in Brucella under stress or selective pressure [43,44] and could be a source of instability [42] in marine Brucella.
Brucella phylogenetic trees and dendrograms have been constructed based on genomics maps [3,5,6], amplified fragment length polymorphisms (AFLP) [45], multilocus enzyme electrophoresis (MLEE) [6], and outer membrane proteins omp2a/omp2b [3]. Now, other universally conserved loci, especially 23S rrn, EF-Tu, rpoB, and gyrase, are increasingly being used to establish relationships among highly similar bacteria with important phenotypic differences to determine their relationships [16]. We constructed a phylogenetic tree based on concatenated sequences of ribosomal associated loci. Most phylogenetic trees and dendrograms, including ours, place B. abortus, B. suis/B. canis, and B. melitensis on separate branches, supporting alternative evolutionary paths. Recently, it was shown that Brucella isolates could be identified at the species level using 21 variable number tandem repeats (VNTR) [46]. The neighbor joining tree based on VNTR data produced major clusters that encompassed the classical Brucella spp. On this tree, the reference B. suis biovar 5 strain, which appears as a unique branch on our tree, was shown to be only distantly related to all other reference strains and isolates by VNTR analyses [46]. Though B. ovis formed a single cluster by VNTR analyses, it clustered, albeit with a low bootstrap value, with B. abortus on our tree. Significant sequence differences have been reported between B. ovis and other classical Brucella spp. reference strains [47,48]. Brucella neotomae grouped with B. melitensis here but was on a separate node. Based on VNTR data, B. neotomae occurs on a unique branch but groups with B. abortus on a AFLP generated dendrogram [45]. Marine isolates are not found on many Brucella phylogenetic trees. Ours grouped the marine Brucella and B. neotomae with B. melitensis but on separate branches. This is in agreement with the genetic diversity observed among the marine isolates and proposals that marine isolates may comprise more than one species [3,41].

Conclusion
Ribosomal associated polymorphisms among the reference Brucella spp. did not correlate with differential intrinsic antibiotic resistance to erythromycin or clindamycin. Efflux is an important mechanism of resistance to macrolides and the lincosamide clindamycin in Brucella and can be inhibited by the RND efflux inhibitor PAβN. A phylogenetic tree constructed based on concatenated ribosomal associated loci supports alternative evolutionary paths for B. melitensis, B. abortus, and B. suis, and clustered the marine Brucella with B. melitensis, and B. canis with B. suis. It also supports the doubtful close relationship of B. suis biovar 5 with B. suis.

Bacterial strains and growth conditions
Bacterial strains (Table 1) were obtained from our laboratory collection for this study. Bacteria were grown at 37°C on tryptose agar (DIFCO Laboratories, Detroit, MI) containing 5% bovine serum in the presence of 7.5% CO 2 . Cells were suspended in saline (10 10 CFU/ml), mixed with two volumes of methanol, and stored at 4°C until needed.

Etest
In vitro activities of azithromycin, clarithromycin, erythromycin, and clindamycin were determined by the Etest (AB Biodisk, Piscataway, NJ). The highest MIC determination for these antibiotics using the Etest is 256 μg/ml. The preformed gradient of the Etest strips covers a continuous MIC range corresponding to 15 two-fold dilutions with a precision of 0.5 dilution. Bacterial inocula were prepared by adjusting the turbidity of a 48 h culture to a 0.5 McFarland standard (5 × 10 8 CFU/ml). The suspension was streaked onto Difco™ Mueller Hinton agar (Becton, Dickinson and Company, Sparks, MD) in the presence or absence of 25 μg/ml of efflux inhibitor PAβN [49] (Sigma Chemical Co., St. Louis, MO) using a cotton swab, and the Etest strips applied. Plates were incubated (37°C) in 7.5% CO 2 . Results were read after 48 h.

Selection of erythromycin mutants
Brucella strains having erythromycin MIC values <5 μg/ml were suspended in saline and plated (10 8 cfu) in triplicate onto Difco™ Mueller Hinton agar containing 5, 10, or 20 μg/ml of erythromycin (Sigma Chemical Co.) and incubated at 37°C in the presence of CO 2 . Five colonies were selected from plates with the highest concentration of erythromycin supporting growth, and subsequently streaked onto tryptose serum agar and Difco™ Mueller Hinton agar containing erythromycin.

PCR amplification
Master mixes for PCR reactions were prepared by use of the Fast Start Taq DNA polymerase kit (Roche Molecular Biochemicals, Indianapolis, IN) according to manufacturer's instructions. Methanol treated cells were diluted 1/ 10 in water and used immediately or stored at 4°C up to 2 months. One μL was added per 25 μL of reaction mixture. Reactions were 50 or 100 μL. Cells were disrupted and amplification initiated by heating the reactions to 95°C for 5 min. Melting, annealing, and elongation temperatures and times were 95°C for 15 sec, 60°C for 30 sec, and 72°C for 90 sec, respectively. After amplification for 35 or 40 cycles, elongation was extended by 4 min.