Abstract
Rolling circle amplification has been useful for detecting point mutations in isolated nucleic acids, but its application in cytological preparations has been problematic. By pretreating cells with a combination of restriction enzymes and exonucleases, we demonstrate that rolling circle amplification in situ can detect gene copy number and single base mutations in fixed cells with efficiencies up to 90%. It can also detect and quantify transcribed RNA in individual cells, making it a versatile tool for cell-based assays.
Rolling circle amplification (RCA) is a molecular cytogenetic technique used with a padlock oligonucleotide probe to detect single base changes in isolated nucleic acids (1–5). Padlock probes are composed of ≈100 nucleotides that hybridize to targets of ≈30 bases. The 30-base target-binding region of the probe is split into two 15-base segments placed in opposite orientation at each end of the linear probe so that a circle must be formed for hybridization to occur (6, 7). At 10 bases per helical turn, the hybridized probe wraps around its target three times, and the remaining 70 bases form an unhybridized single-stranded loop. Posthybridization DNA ligation connects the two ends of the probe in the middle of the 30-base binding region. The unbound 70-base loop facilitates probe circularization and permits ≈20 bases to serve as a primer recognition site for DNA polymerase to replicate the circle. RCA is an isothermal process in which the polymerase progresses continuously around the loop until the 100 bases have been replicated hundreds or thousands of times. Incorporating a labeled nucleotide during the RCA reaction produces sufficient signal for easy visualization of the target.
Application of RCA to in situ targets in fixed or permeabilized cells has not been uniformly successful to date. Whereas recent work has demonstrated that the concept is viable (8), DNA detection efficiencies of 20–30% lessen the utility of RCA as an assay. Lack of success has been attributed to possible blocking of the polymerase by the target strand, and it was suggested that this problem might be overcome by cutting the target DNA strand near the RCA probe's hybridization site (5). Under these conditions, DNA polymerase could free the probe from the target, in effect spinning the probe away from the target, keeping the polymerase from being blocked during the amplification process. Here, we report that in addition to restriction enzyme digestion of DNA, additional steps were required to achieve consistent and satisfactory results for RCA in situ. Whereas heat denaturation is typically used to render the target DNA single stranded, we found that complete removal of the nontargeted DNA strand by digestion with exonuclease III significantly increased the efficiency of the process.
We also demonstrate the use of RCA to detect mRNA in cytological preparations. Using appropriate image analysis techniques, the RCA assay is sufficiently quantitative to enable transcriptionally mediated dose-response curves to be generated. Increasing DNA detection efficiency to ≈90% and developing a means to use RNA as a template significantly increases the effectiveness of RCA as an assay.
Materials and Methods
DNA Detection.
Target preparation.
Two cell lines were used in these experiments. One was a human lymphoblastoid (HLB) line (Coriell Cell Repositories, Camden, NJ), putatively normal with regard to karyotype and gene expression. The other was a Molt-4 lymphoid cell line (American Type Culture Collection, ATCC) derived from a patient with acute lymphoblastoid leukemia. HLB cells were expected to have two normal copies of the Tp53 gene and to be normal with regard to Tp53 expression. Molt-4 cells are reported by ATCC to express no normal Tp53 and to have one normal and one or more abnormal copies of the Tp53 gene, in which there is a G→A transition in codon 248 of exon 7. Cells were prepared for DNA detection by first incubating in a hypotonic solution (0.075 M KCl) for 30 min at 37°C followed by three fixations in methanol/acetic acid (3:1 vol/vol) and dropped on clean glass microscope slides. Fixed cells on slides were covered with 50 μl of ribonuclease A (500 μg/ml, Roche Biochemicals) under a glass coverslip. Slides were incubated 1 h at 37°C and then rinsed with sterile water. Restriction enzymes were used to cut ≈20 base pairs either 3′ or 5′ of the sequence of interest, Tp53 in this case. Either AflIII or BbsI (0.1 unit/μl, New England Biolabs) was applied for 12 h at 37°C. Cells were treated with exonuclease III (1.3 units/μl, Life Technologies, Grand Island, NY) in 1× exonuclease III buffer (50 mM Tris, pH 8.0/5 mM MgCl2/1 mM DTT), then incubated 1 h at 37°C and rinsed with sterile water.
Single color reaction.
Simultaneous hybridization and ligation were performed with 0.8 μM of probe, 20 units of Ampligase thermostable DNA ligase (0.43 unit/μl, Epicentre Technologies, Madison, WI), and 1× Ampligase buffer. The 50-μl reaction was placed on the slide, covered with a glass coverslip, and sealed with rubber cement. The slide was heated to 94°C for 10 min to ensure that both probe and target DNA were single stranded and then lowered to 42°C for 1 h to allow hybridization and ligation of the probe. Slides were washed in 2 × SSC at 42°C for 15 min, rinsed in sterile water, and blown dry. The RCA reaction mixture consisted of 4 μM of T7 primer, 200 μM of each dNTP (Roche), either 63 nM digoxigenin-11-dUTP (Roche) or 63 nM biotin-dUTP (Roche), 2 units of ThermoSequenase DNA polymerase with pyrophosphatase (United States Biochemical), and 1× ThermoSequenase buffer, which was added to the slide, which was then covered with a coverslip, sealed with rubber cement, and heated 12 h at 54°C. Slides were washed in 2 × SSC at 45°C for 5 min, 1 × PBS at 45°C for 5 min, and rinsed in sterile water at room temperature. Anti-digoxigenin–fluorescein antibody or Texas Red Avidin (Roche, 200 ng/μl) was incubated on the slide at 37°C for 10 min and washed 2 × 5 min in 1 × PBS at room temperature. Slides were mounted in 4′,6-diamidino-2-phenylindole in anti-fade and viewed with an Axiophot fluorescence microscope (Zeiss).
Dual color reaction.
Two probes were used, one complementary to the normal Tp53 gene and the other complementary to the mutated form found in the Molt-4 cells. Each probe was primed with the T7 oligonucleotide and also contained a separate promoter sequence used to hybridize a fluorochrome-tailed oligonucleotide included in the reaction to its complementary amplified sequence. The SP6 promoter was used to bind to the reaction products from the mutant probe, GGTTCATGCCGCCCttttttttTATTTAGGTGACACTATAGttttttttCCCTATAGTGAGTCGTATTAttttttttGGTGAGGATGGGCCTCT, and the T3 promoter was used to bind to the reaction products of the normal probe (GGTTCATGCCGCCCtttttttATTAACCCTCACTAAAGGGAttttttttCCCTATAGTGAGTCGTATTAttttttttGGTGAGGATGGGCCTCC). The underlined portions of each probe sequence are the SP6 and T3 sequences, respectively; thymidines used as spacers are indicated by lowercase letters. The procedure for ligation and rolling circle amplification is similar to the procedure for a single color reaction. Differences include ligating 0.4 μM of each probe (mutated and normal) as well as incorporating 10 μM of fluorochrome tailed oligonucleotide (T3 and SP6) in place of digoxigenin-dUTP in the RCA reaction. The tailed oligonucleotides were used to obtain different colors for the mutant and normal Tp53 gene sequences. Fluorochrome tailing was achieved with a 10 μM solution of the promoter oligonucleotide with digoxigenin-dUTP (T3 oligo) or biotin-dUTP (SP6 oligo) (100 nM, Roche) using terminal deoxynucleotidyl transferase (1.5 units/μl, Life Technologies) and 1× terminal deoxynucleotidyl transferase buffer. The reaction was incubated 1 h at 37°C.
RNA Detection.
Qualitative detection.
RCA using the human Tp53 mRNA complementary probe (CGGTTCATGCCGCCCtttttttttCCCTATAGTGAGTCGTATTAtttttttAGGGAAATCACTCCCAATTAtttttttGGTGAGGATGGGCCTC) was performed, and digoxigenin-dUTP was incorporated during the reaction. Slides were incubated with fluorescein-conjugated anti-digoxigenin antibody as described. All slides were prepared by centrifuging live cells suspended in PBS onto glass slides followed by fixation in 100% ethanol for 5 min. The cells prepared for this experiment were not treated with restriction enzymes or exonuclease III, nor were they heat denatured. Consequently, no nuclear DNA signal was evident. T4 RNA ligase (Epicenter, 20 units) was used for the ligation under identical conditions as described above. The slides were stained for 3 min with 10 μg/ml acridine orange (AO) in Sorenson's buffer at room temperature, rinsed in PN buffer and sterile water at room temperature, then mounted in Sorenson's buffer and stained with 4′,6-diamidino-2-phenylindole.
Quantitative detection.
Probe sequences for the RNA detection were as follows: keratin 10, TGTGAGAGCTGCACAttttttttCCCTATAGTGAGTCCTATTAtttttttttTATTTAGGTGACACTATAGttttttttATCTGGGCCTGAATC; GSTT2, CATTCTTCTTGGCGAttttttttCCCTATAGTGAGTCCTATTAtttttttttATTAACCCTCACTAAAGGGAttttttttctctaaggggatgc; chromosome 18 alpha satellite, GAATTGAACCACCGTAttttttCCCTATAGTGAGTGAGTCGTATTAttttttAAATATCATCTTTGGTGTTTCCTAtttttttGTACTCACACTAAGA; Tp53, see above; p72, GCTACTAGCTCCATtttttttttCCCTATAGTGAGTCCTATTAttttttttATTAACCCTCACTAAAGGGAtttttttttCCAGTTGAGGTGGT; and vimentin, GGAAGCGCACCTTGtttttttttCCCTATAGTGAGTCCTATTAttttttttTATTTAGGTGACACTATAGtttttttttTATTCTGCTGCTCCA.
All reactions were carried out as described above. Irradiations were carried out in a Mark 1 137Cs source (J.L. Shepherd and Associates). Pixel intensities for each image were established using scientific imaging software from IPLabs (Scanalytics, Billerica, MA).
Results and Discussion
To detect the Tp53 gene in HLB cells, cells were fixed on slides and treated with two nucleases before RCA (Fig. 1). DNA was cut with a restriction endonuclease, and the double helix was then rendered single stranded by digestion with exonuclease III. Padlock probe hybridization and RCA followed. Endonuclease digestion was carefully selected to place a cut within ≈20 bases of the probe's binding site but without actually being within that site. The substrate for exonuclease III is double-stranded DNA, which it digests from a 3′ end leaving a single strand of DNA in its wake, proceeding until it reaches a region where the DNA is already single stranded (9, 10). Thus, the entire genome will still be represented following exonuclease digestion, but for any given locus only one of the two strands will be present.
Figure 1.
Flow diagram of RCA in situ. Pretreatment prior to in situ RCA detection of the Tp53 gene. Restriction enzymes were used to cut ≈20 base pairs either 3′ or 5′ of the probe binding site. AflIII was used for digestion 5′ of the binding site, and BbsI was used for digestion 3′ of the target. Cells were then treated with exonuclease III, which digests DNA 3′ → 5′ starting with 3′ hydroxyl left by the endonuclease, resulting in staggered single-stranded DNA. The DNA strand remaining following AflIII digestion in this case is the complement to the sense probe sequence. DNA remaining following BbsI digestion is the complement to the antisense probe sequence. BbsI digestion constitutes a negative control for the RCA process using the sense probe, and AflIII constitutes a negative control for the antisense probe. The two ends of the sense probe create an incomplete circle as they anneal to the complementary site on the DNA digested with BbsI. The DNA strand digested with AflIII is complementary to the sense probe and allows it to anneal and ligate, completing the circle and locking the probe onto the target. Targets other than Tp53 may require different endonucleases.
Success of the RCA reaction depended on the probe's target sequence remaining intact following digestion. To ensure that the target strand was not removed by the exonuclease treatment, the endonuclease had to cut 5′ of the site on the strand to which the probe bound, and a second cut must not occur too near the 3′ end on the target strand. Identical RCA reactions were performed with two probes that were complementary to the Tp53 gene present in two copies in the target cells. One of the probes was complementary to the coding strand, the other to the noncoding strand. In one set of reactions, the target strand was cut 3′ to the binding site; in the other set, the cut was 5′ to the site. Each reaction was carried out using either the probe complementary to the transcribed strand or the probe complementary to the nontranscribed strand; only one probe was present in each reaction. Reaction products were labeled by incorporating a hapten-conjugated nucleotide (either biotin or digoxigenin) in the reaction mixture. The results were detected by a subsequent treatment with a fluorochrome-conjugated (Texas red or fluorescein) antibody to the hapten, producing fluorescent-labeled signals at the site of the reaction. In every test, reactions with 5′ nicks successfully produced labeled signals, whereas reactions with nicks 3′ to the target never produced signals.
To determine gene copy number, RCA using the Tp53 probe was performed on methanol/acetic acid (3:1)-fixed cells on glass slides subsequently treated with enzymes as described above. We scored 200 interphase cells for the number of fluorescent signals present, with the criteria that two labeled spots indicated a normal cell and three or more spots indicated aneuploidy (Fig. 2). Approximately 10% of the cells that were scored showed one or no spots; we counted these as failed in reaction. Of the cells that showed two or more signals, 6% showed three or more (scored as aneuploid), whereas 94% had two signals. Scoring metaphase chromosomes showed a 5% aneuploid rate; these HLB cells are known to develop a small degree of aneuploidy in culture, but are generally considered to be normal nonprimary cell lines. We repeated these studies using probes for other genes and for an alpha satellite repeat, and we obtained similar results (not shown). We also tested the notion that it was the breaks in the target strand rather than the single-stranded target that was responsible for increasing the efficiency of RCA. Treating only with a restriction enzyme, and not with exonuclease III, produced no significant increase in efficiency over only heat denaturation and no enzymatic treatment at all. We also studied other ways of rendering target DNA single stranded. Simply cutting the target DNA with restriction enzymes produced no signal at all, and following the restriction enzyme treatment with heat denaturation produced efficiencies of ≈25%. Thus, rendering the target DNA unifilar at the binding site seems to be responsible for increasing the detection efficiency of RCA.
Figure 2.
Copy number detection. RCA in HLB cells in the presence of digoxigenin-dUTP (A) or biotin-dUTP (B) unambiguously identifies very short DNA sequences in the Tp53 gene. In B, both normal and aneuploid cells are evident, consistent with their known karyotypic variability.
To demonstrate the application of RCA in detecting single base changes in nucleic acid targets in situ, Molt-4 cells were examined. Two separate probes complementary to a 30-base region of this exon were constructed. One probe contained the complement to the normal base, and the other contained the complement to the mutant base, with the 3′ terminal base of each probe corresponding to the site of the mutation. If the terminal base in the probe is not complementary to the target, that base will not hybridize, preventing ligation and blocking the polymerase from progressing continuously around the loop with the result that no fluorescent signal will be generated. Consequently, the mutant and normal target sequences will only be detected by their respective probes. To differentiate between the reaction products the sequences of two bacteriophage promoters were incorporated into the RCA probes. Because the T7 promoter primer is incorporated into the probes to initiate the RCA reaction, we incorporated either the T3 (for the normal sequence probe) or SP6 (for the mutant sequence probe) bacteriophage promoter sequences into the RCA probes. Oligomers corresponding to the two promoter sequences were then included in the RCA reaction and were differentially labeled, T3 with digoxigenin and SP6 with biotin. During the RCA reaction, each replication of the probe produces a single-stranded sequence complementary to the promoter sequence contained within the probe. The T3 or SP6 oligomers should hybridize to these sites as they are produced, labeling each product with either digoxigenin or biotin. Amplification products of Tp53 in Molt-4 cells were detected using a fluorescein-conjugated anti-digoxigenin antibody and Texas red-conjugated avidin to produce green and red signals at the sites of the normal and mutant alleles, respectively (Fig. 3). It is important to note that the binding sequences of the mutant-complement and normal-complement probes differ by only a single base. Whereas this difference is placed so as to prohibit the two ends of the probe from being ligated and amplified should they anneal to the incorrect site (mutant to normal site or the reverse), it is insufficient to stop such misbinding from occurring at all loci. Some fraction of the time, this mishybridization will occur, and although improperly bound probe will be washed away before amplification, no signal will be produced at that site. Thus, for simultaneous two-probe binding, to prevent such false negatives each probe should be hybridized, ligated, and washed off sequentially. The wash step will not remove properly bound and ligated probe, but it will remove unligated material. This process will ensure the maximum possible efficiency of detection. Even with this precaution, efficiencies for two-color detection were considerably lower than for single-color RCA (≈30% as compared with greater than 90%). We attribute this to the method of fluorescence labeling, as it is similarly inefficient when used for only single color detection.
Figure 3.
Allele discrimination. A Molt-4 cell in which a single nucleotide (G to A) difference in two alleles of the Tp53 gene was detected by RCA in situ. Two probes were used, which differed in that the 3′ termini of the normal and the mutant versions were complementary to the normal (green signal) and the mutant (red signal) sequences, respectively.
By designing an RCA probe's binding site to be complementary to a transcribed mRNA sequence, gene expression could also be detected. The method of cell fixation for RNA detection was considerably more important than for DNA detection. Although various methods of cell preparation, including conventional acid/alcohol fixation and alcohol fixation yielded similar results for DNA-based RCA in situ, routine detection of RNA was only made possible by centrifuging the cells onto slides in culture media followed by an alcohol wash. In these experiments, a probe with a 30-base binding site complementary to a transcribed region of the Tp53 gene was used to determine the presence of Tp53 mRNA in Molt-4 and HLB cell lines. Unlike HLB cells, Molt-4 cells produce no normal Tp53 transcripts (11). Single-color RCA was performed both before and after treatment of fixed cells with RNase. HLB and Molt-4 cells were also stained with AO following all treatments before the RCA step to ensure that the results of the RCA reaction corresponded with the actual status of RNA in the cells. In each case, AO staining detected RNA before, but never following, RNase treatment. RCA performed on HLB cells that had not been treated with RNase showed considerable labeling in the cytoplasmic region surrounding the nucleus (Fig. 4). In the Molt-4 cells, however, no RCA products were detected whether the reaction was performed preRNase or postRNase. In the HLB cells, only the postRNase-treated RCA reaction was null. As an additional control, an RCA probe with a binding site that was a copy of the Tp53 mRNA, rather than a complement to it, was tested and produced no signal.
Figure 4.
mRNA detection. Normal HLB cells (A) and Molt-4 cells (B) stained with AO, which labels single-stranded nucleic acid (RNA) red and double-stranded nucleic acid (nuclear DNA) yellow. (A and B) The presence of RNA in each of the cell types. RCA was performed on replicate cell preparations using a probe with 3′ and 5′ DNA binding sites complementary to the probe described in Fig. 3. (C) The green fluorescence signal surrounding the nuclei of the HLB cells demonstrates the presence of Tp53 transcript detected by RCA. (D) No such signal is seen in the Molt-4 cells, demonstrating the lack of normal Tp53 transcript.
RCA probes were constructed to bind to the mRNA of several genes known to be radiation dose responsive (http://rex.nci.nih.gov/RESEARCH/basic/lbc/patent/web6kinduced.htm), including Tp53, human DEAD-box protein p72, vimentin, keratin 10, and glutathione S-transferase theta 2 (GSTT2). HLB cells were exposed to 137Cs gamma rays at doses up to 2 Gy, then fixed and evaluated by RCA. For each RCA probe, three different ligases were used: Ampligase, T4 DNA ligase, and T4 RNA ligase, the last of which has been reported as an effective ligating agent for short DNA fragments (12). All produced similar results, but the T4 RNA ligase products had the highest signal to background ratio. It is important to note that RNA serves only as a hybridization template for the DNA ligase and that only DNA is ligated, not RNA. Use of DNA ligases to join single-stranded fragments of DNA hybridized to an RNA target has been described in the literature (13). As a negative control, each RCA reaction was also run without ligase, the results of which were used to normalize the results of the experiments. A probe complementary to untranscribed alpha satellite DNA was used as an additional negative control. Neither negative control produced a response signal; results are shown in Fig. 5. The fact that cell lines are frequently unique in their gene expression patterns makes comparison with the literature difficult, but curve shapes for Tp53 and p72 expression in HLB cell lines were confirmed by microarray analysis using the Affymetrix array system (M. A. Coleman and A. J. Wyrobek, unpublished data).
Figure 5.
Radiation dose-response curves. Normal HLB cells were irradiated, left in culture medium for 2 h, and then fixed and analyzed. T4 RNA ligase (20 units in 4 μl of 10× buffer, Epicentre Technologies) was used to ligate the probes. Each experiment was replicated from a new stock of frozen cells and produced curves of the same shapes. Fifty to 150 cells were analyzed per data point; error bars represent ± SEM. Cells were analyzed by measuring mean pixel intensity of green fluorescence using IPLabs image analysis software (Scanalytics).
We have shown that RCA in situ is useful for discriminating alleles, determining gene copy number, and quantifying gene expression in single cells. The sensitivity, specificity, and speed of RCA may also allow it to be used for focused investigations of cell and tissue responses to drugs of pharmaceutical importance, for evaluation of adverse environmental exposure to humans by ionizing radiation and chemicals, and for clinical purposes such as prenatal diagnosis and pathological characterization of tumors. The exquisite sensitivity of in situ RCA may add an entirely new dimension to the fields of genomics, pathology, mutagenesis, and cytogenetics.
Acknowledgments
HLB cells were kindly provided by Matthew A. Coleman and Andrew J. Wyrobek (Biology and Biotechnology Research Program, Lawrence Livermore National Laboratory). This work was performed under the auspices of the U.S. Department of Energy by the University of California, Lawrence Livermore National Laboratory, under Contract No. W-7405-Eng-48 with support from National Institutes of Health Grant CA55861 and Department of the Energy Grant DOE KP110202.
Abbreviations
- RCA
rolling circle amplification
- HLB
human lymphoblastoid
- AO
acridine orange
Footnotes
This paper was submitted directly (Track II) to the PNAS office.
References
- 1.Baner J, Nilsson M, Mendel-Hartvig M, Landegren U. Nucleic Acids Res. 1998;26:5073–5078. doi: 10.1093/nar/26.22.5073. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Schweitzer B, Wiltshire S, Lambert J, O'Malley S, Kukanskis K, Zhu Z, Kingsmore S F, Lizardi P M, Ward D C. Proc Natl Acad Sci USA. 2000;97:10113–10119. doi: 10.1073/pnas.170237197. . (First Published August 22, 2000; 10.1073/pnas.170237197) [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Lizardi P M, Huang X, Zhu Z, Bray-Ward P, Thomas D C, Ward D C. Nat Genet. 1998;19:225–232. doi: 10.1038/898. [DOI] [PubMed] [Google Scholar]
- 4.Lizardi P M, Ward D C. Nat Genet. 1997;16:217–218. doi: 10.1038/ng0797-217. [DOI] [PubMed] [Google Scholar]
- 5.Thomas D C, Nardone G A, Randall S K. Arch Pathol Lab Med. 1999;123:1170–1176. doi: 10.1043/1543-2165-123.20.1170. [DOI] [PubMed] [Google Scholar]
- 6.Nilsson M, Antson D O, Barbany G, Landegren U. Nucleic Acids Res. 2001;29:578–581. doi: 10.1093/nar/29.2.578. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Nilsson M, Krejci K, Koch J, Kwiatkowski M, Gustavsson P, Landegren U. Nat Genet. 1997;16:252–255. doi: 10.1038/ng0797-252. [DOI] [PubMed] [Google Scholar]
- 8.Nilsson M, Malmgren H, Samiotaki M, Kwiatkowski M, Chowdhary B P, Landegren U. Science. 1994;265:2085–2088. doi: 10.1126/science.7522346. [DOI] [PubMed] [Google Scholar]
- 9.Goodwin E, Meyne J. Cytogenet Cell Genet. 1993;63:126–127. doi: 10.1159/000133516. [DOI] [PubMed] [Google Scholar]
- 10.Meyne J, Goodwin E H. Methods Mol Biol. 1994;33:141–145. doi: 10.1385/0-89603-280-9:141. [DOI] [PubMed] [Google Scholar]
- 11.Rodrigues N R, Rowan A, Smith M E, Kerr I B, Bodmer W F, Gannon J V, Lane D P. Proc Natl Acad Sci USA. 1990;87:7555–7559. doi: 10.1073/pnas.87.19.7555. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Troutt A B, McHeyzer-Williams M G, Pulendran B, Nossal G J. Proc Natl Acad Sci USA. 1992;89:9823–9825. doi: 10.1073/pnas.89.20.9823. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Zhong X B, Lizardi P M, Huang X H, Bray-Ward P L, Ward D C. Proc Natl Acad Sci USA. 2001;98:3940–3945. doi: 10.1073/pnas.061026198. [DOI] [PMC free article] [PubMed] [Google Scholar]