A method for detecting single mRNA molecules in Arabidopsis thaliana
© Duncan et al. 2016
Received: 30 November 2015
Accepted: 25 January 2016
Published: 5 August 2016
Despite advances in other model organisms, there are currently no techniques to explore cell-to-cell variation and sub-cellular localization of RNA molecules at the single-cell level in plants.
Here we describe a method for imaging individual mRNA molecules in Arabidopsis thaliana root cells using multiple singly labeled oligonucleotide probes. We demonstrate detection of both mRNA and nascent transcripts of the housekeeping gene Protein Phosphatase 2A. Our image analysis pipeline also enables quantification of mRNAs that reveals the frequency distribution of transcripts per cell underlying the population mean.
This method allows single molecule RNA in situ to be exploited as a powerful tool for studying gene regulation in plants.
KeywordsRNA FISH Gene expression Fluorescence microscopy Arabidopsis Transcription
Quantitative real-time PCR is commonly used to analyze plant gene expression, but this method lacks potentially important information relating to sub-cellular localization of RNA and masks cell-to-cell variation [1, 2]. To effectively study these aspects of gene regulation, it is necessary to study RNA at the cellular level.
A method that has achieved this aim is in situ hybridization followed by microscopic analysis. Initially, researchers performed in situ hybridizations using radioactive probes . Early improvements involved linking the probes to enzymes that catalyze chromogenic or fluorogenic reactions [4–6]. In Arabidopsis mRNA in situ hybridization has been routinely used for detailed visualization of gene expression patterns [7–9]. While this method gives good semi-quantitative spatial information, it produces images with limited cellular resolution. More recently plant researchers have used fluorescently labeled probes to directly label transcripts. This has improved cellular resolution, but relatively poor sensitivity has resigned it mainly for detection of highly repetitive RNAs [10, 11]. Single molecule fluorescence in situ hybridization (smFISH) was developed to maximize both sensitivity and specificity by using multiple singly labelled probes to visualize RNA molecules as discrete spots of fluorescence . A recent version of this method uses 48 fluorescently labeled DNA oligonucleotides (20mers) to hybridize to different portions of each transcript. This provides a balance between probe length and number that effectively reduces false positive signals (due to off-target binding) whilst maintaining single molecule sensitivity .
Establishment of smFISH in other model systems has led to greater understanding of transcriptional regulation for many genes [14–17]. In addition to quantifying mRNA at the single cell level, this detection method can be used to visualize sites of transcription  and long non-coding RNAs .
Optical properties of plant cells and tissues provide significant challenges for fluorescence microscopy . Inherent light scattering adversely affects both the excitation and the detection efficiency; moreover plants contain many native molecules that emit high levels of background auto fluorescence compared to other organisms . We chose to develop a smFISH method in fixed Arabidopsis root cells as they typically allow clearer imaging than leaves or other above ground tissue.
We established our method by probing a widely expressed housekeeping gene At1G13320—the A2 scaffolding subunit of Protein Phosphatase2A (PP2A) . Unlike several environmentally regulated phosphatase subunits, it exhibits mRNA levels that are relatively unperturbed by a range of abiotic and biotic stresses and is transcribed evenly across many tissue types throughout development. These robust properties led to PP2A being identified as a superior gene for qPCR normalization . To validate our method we used smFISH to detect PP2A mRNAs and used an image analysis pipeline to automate transcript counting within cells. Together our smFISH protocol and image analysis algorithm provides a straightforward framework for other plant researchers to study gene expression at the single-cell and single-molecule resolution.
Results and discussion
We had equal success in imaging RNA labelled with Quasar®570 and 670 dyes, but we were unable to observe RNA labelled with FITC (data not shown). We found super-resolution structured illumination microscopy (SIM) produced high quality images of our samples; therefore it may be possible to overcome this multiplex limitation through the detection of spectrally barcoded smFISH probes  (Additional file 5).
We chose to perform the analysis on a projection of the z-stack because it simplifies the processing considerably. For data with a higher density of mRNAs or situations where the position of the mRNAs in the z direction is of interest, the spot detection and segmentation algorithms could be implemented in three dimensions. However, for these data, spot density was not high enough to make this necessary.
In this report we present a FISH method that allows for gene-expression profiling of transcripts in Arabidopsis roots. By characterizing cell-to-cell transcriptional variability of the housekeeping gene PP2A we demonstrate that smFISH can be combined with automated image analysis to quantify single RNA transcripts for the first time in plants. As smFISH has been used extensively for RNA analysis in many model organisms [25–28] we believe that this root squash protocol can be easily adapted to suit other plant species with amendments made to the fixation and permeabilization steps as necessary. However adapting smFISH for use in green tissues is likely to represent a greater challenge due to high levels of autofluorescence. Similar issues have been overcome in other organisms through the application of tissue clearing  and cryosectioning . We believe that similar approaches may also be employed to enable transcript imaging in other plant tissues.
In addition to quantifying mRNA and visualizing active sites of transcription at the single cell level, corroboratory qRT-PCR data has shown that smFISH can be used to calculate mRNA fold changes at the cellular level . This method can also identify RNA derived from maternal and paternal gene copies  and, in conjunction with masking oligonucleotides, it can even distinguish RNA transcripts that differ by only a single nucleotide polymorphism . Our adaptation of smFISH for use in Arabidopsis thaliana now opens up these exciting opportunities to the plant research community.
Plant material and growth conditions
Col-0 seeds were surface sterilized in 5 % v/v sodium hypochlorite for 5 min and rinsed three times in sterile distilled water before being sown on MS media minus glucose. They were stratified for 3 days at 5 °C before being transferred to a growth cabinet (Sanyo MLR-351H) 16 h light, 100 μmol m−2 s−1, 22 °C ± 1 °C.
Reagents and solutions
smFISH probe sequences used to detect PP2A mRNA
PP2A exon probes
smFISH probe sequences used to detect PP2A nascent transcripts
PP2A intron probes
Nuclease-free water—not DEPC treated (Qiagen, Cat. No. 129117).
Paraformaldehyde (Sigma, Cat. No. P6148) freshly depolymerized, 4 % w/v in water.
Nuclease-free 10× Phosphate Buffered Saline (Thermo Scientific, Cat. No. AM9624).
70 % Ethanol (freshly made using nuclease free water).
Nuclease-free 20× saline-sodium citrate (20× SSC, Thermo Scientific, Cat. No. AM9763).
RNase A (Sigma, Cat. No. R4642) diluted to 100 μg/ml.
T10E1 buffer (10mM Tris-HCl, 1mM EDTA, pH 8)—Sigma, Cat. No. 93283-100mL
Deionized Formamide (Sigma, Cat. No. F9037).
Dextran Sulphate (Sigma, Cat. No. Res2029D).
Nuclease free Tris HCl buffer 1 M pH8 (Thermo Scientific, Cat. No. AM9855G).
Glucose oxidase (Sigma,Cat. No. G0543).
Bovine Live Catalase (Sigma, Cat. No. C3155).
Wash buffer (50 ml)
5 ml nuclease free 20× SSC mixed with 5 ml nuclease free deionized formamide and nuclease free water up to 50 ml final volume. (Final composition: 10 % formamide, 2× SSC).
DAPI (4′, 6-Diamidino-2-phenylindole; Sigma cat. no. D9564) Diluted to 100 ng/μl in wash buffer (Final composition: 100 ng/μl, 10 % formamide, 2× SSC).
Hybridization solution (10 ml)
Dissolve 1 g dextran sulfate in 1 ml nuclease free 20× SCC, 1 ml deionized formamide and nuclease free water up to 10 ml final volume. (Final composition: 100 mg/ml dextran sulfate and 10 % formamide in 2× SSC).
Anti-fade GLOX buffer minus enzymes (1 ml)
40 μl 10 % glucose in nuclease-free water, 10 μl 1 M Tris–HCl, pH 8.0 and 100 μl 20× SCC was mixed with 850 μl nuclease-free water. (Final composition: 0.4 % glucose in 10 nM Tris–HCl, 2× SSC).
Anti-fade GLOX buffer containing enzymes (100 μl)
1 μl glucose oxidase and 1 μl mildy vortexed catalase suspension added to 100 μl GLOX minus enzyme solution.
Poly-l-Lysine slides (Sigma, Cat. No. PO425 or similar NOTE: these are not essential but the samples adhere better to these than untreated slides).
Low stender-form preparation dishes (VWR, Cat. No. 470144-866 or similar).
22 mm × 22 mm No.1 glass coverslips (Fisher Scientific, Cat. No. 12333128 or similar).
Coplin jar (Sigma, Cat. No. S6016 or similar).
Parafilm® M sealing film (Bemin, Cat. No. PM992).
Hybridization chamber (or a suitable dark box with a layer of tissue moistened with water will suffice).
37 °C incubator
Zeiss Elyra PS1 inverted microscope with cooled EM-CCD Andor iXon 897 camera.
smFISH probe design
Since designing smFISH probes is similar to designing PCR primers most primer design software packages can be used  but we used the online program Stellaris® Probe Designer version 2.0 from Biosearch Technologies (http://singlemoleculefish.com). Input of PP2A coding sequence into the program automatically generates a set of probes complementary to the PP2A mRNA, optimized for binding to the target sequence. Before ordering our pre-labelled probes from Biosearch Technologies we completed a TAIR BLAST query for each sequence to ensure target specificity (https://www.arabidopsis.org/Blast/). Tables 1 and 2 list the oligonucleotide sequences used to detect PP2A mRNA and nascent transcripts respectively.
Sample preparation (timing: 2 h)
Seedlings were removed from the media 10 days after germination. Root tips were dissected using a razor blade and forceps and placed into a glass dish containing 4 % paraformaldehyde to fix for 30 min at room temperature. The roots were removed from the fixative and washed twice with 1× PBS. 3–4 roots were then arranged on a slide and covered by a glass coverslip and the meristems were squashed manually by applying pressure through the coverslip. The slide, together with the sample and coverslip, were then submerged briefly in liquid nitrogen (~5 s) to adhere the roots to the slide. The coverslip was then flipped off with a razor blade and the samples were left to dry at room temperature for a minimum of 30 min. Tissue permeabilization was then carried out by immersing the samples in coplin jars containing 70 % ethanol and left to shake gently for a minimum of 1 h.
Note: We ensured coplin jar lids were sealed with parafilm to prevent evaporation during the ethanol incubation period.
Hybridization (timing: 4 h—overnight)
Residual ethanol was left to evaporate at room temperature for 5 min before 2, 2-min washes were carried out with wash buffer. 100 μl of hybridization solution with probes at a final concentration of 250 nM was then added to each slide. Coverslips were laid over the samples to prevent buffer evaporation and the probes were left to hybridize in a humid chamber at 37 °C overnight in the dark.
Sample mounting (timing: 2 h)
Hybridization solution containing unbound probes was removed using a pipette in the morning. Each sample was then washed twice with 200 μl wash buffer and finally immersed in coplin jars containing wash buffer for 30 min at 37 °C. 100 μl of the nuclear stain DAPI was then added to each slide and left to incubate at 37 °C for 30 min. Following DAPI removal, 100 μl 2× SSC was added samples and removed. 100 μl GLOX buffer minus enzymes was added to the samples and left to equilibrate for 2 min and then replaced with 100 μl of anti-fade GLOX buffer containing enzymes. The samples were then covered by coverslips sealed. Excess GLOX buffer was wicked away using tissue before the coverslips were sealed with nail varnish. We immediately imaged our samples as we observed a noticeable reduction in image quality around 4 h after mounting.
Note: Oxygen-scavenging GLOX buffer maximised the stability of our smFISH fluorophores and we observed rapid bleaching when it was substituted with the commercial anti-fade mounting media Vectorshield (data not shown).
A Zeiss Elyra PS1 inverted microscope was used for imaging. A 100X oil-immersion objective (1.46 NA) and cooled EM-CCD Andor iXon 897 camera (512 × 512 QE > 90 %) was used to obtain all images in the standard, rather than super-resolution mode. The following wavelengths were used for fluorescence detection: for probes labeled with Quasar®570 an excitation line of 561 nm was used and signal was detected at 570–640 nm; for probes labeled with Quasar®670 an excitation line of 642 nm and signal was detected at 655–710 nm; for DAPI an excitation line of 405 nm and signal was detected at wavelengths of 420–480 nm. For all experiments exposure times between 200–250 ms were used and a series of optical sections with z-steps of 0.2 μm were collected.
Note: When establishing this technique for the first time we recommend that the following controls be carried out: no probe (where probes are omitted from the hybridization solution) Additional file 1, and RNase A treatment (Additional file 2). To confirm RNA specificity we incubated samples with RNase for 1 h at 37 °C in a humid chamber after the ethanol permeabilization step, rinsed in 10 mM HCl for 5 min, washed twice with 2× SSC for 5 min before the protocol was continued.
Z-stacks were deconvolved using AutoQuant X2 (Media Cybernetics). Projections and analysis of 3D pictures were performed using Fiji (an implementation of ImageJ, a public domain program by W. Rasband available from http://rsb.info.nih.gov/ij/). Typically from 4 to 6 roots more than 300 cells can be obtained by this method, which were then suitable for further analysis using our automated mRNA counting programme.
Structured illumination microscopy
A Zeiss Elyra PS1 inverted microscope was used for imaging using a 63X water objective (1.2 NA) to match samples mounted in GLOX buffer. The SIM camera used was an EM-CCD Andor iXon 885. We collected ×5 phases at ×3 angles total 15 images per plane. Series of optical sections with z-steps of 0.2 μm were collected.
The following wavelengths were used for fluorescence detection: for probes labeled with Quasar®570 an excitation line of 561 nm was used and signal was detected at 570–640 nm; for probes labeled with Quasar®670 an excitation line of 642 nm and signal was detected at 655–710 nm; for DAPI an excitation line of 405 nm and signal was detected at wavelengths of 420–480 nm. For all experiments series of optical sections with z-steps of 0.2 μm were collected.
Images were processed using Zen Black default parameters. The images were also colour aligned using Zeiss “channel aligned” tool. Reference images of multiple coloured beads were collected in SIM mode then processed. Then an alignment matrix was generated using the SIM bead data and this was applied to the experimental SIM data.
We have made our mRNA counting programme publically available at: https://github.com/JIC-CSB/FISHcount. Our smFISH image analysis consists of two components—cell segmentation and mRNA counting. These combine into an overall workflow that results in an image where each cell is annotated with the number of mRNA located within it (Figs. 4, 5). Bioformats  is used to convert the microscope image into individual z-stacks for each channel. The analysis pipeline then processes these z-stacks to produce the annotated image, and is implemented in the Python programming language .
The Watershed algorithm is used to segment the image into regions representing cells, using the implementation provided by the scikit-image library . Segmentation using the Watershed algorithm requires an input image denoting gradient magnitude, and a set of seeds for initialising the flood filling of the input image.
To identify the seeds for the Watershed algorithm each plane in the DAPI stack is normalised for intensity, then a maximum intensity projection taken. Contrast Limited Adaptive Histogram Equalization (CLAHE, ) is used to locally equalize the intensity of the projection. A Sobel filter is applied to the projection to find nuclear edges. Otsu’s thresholding is then applied to select the nuclei. Each detected nucleus is reduced to its centroid for use as a seed for the segmentation.
The gradient magnitude input for the Watershed algorithm is generated by taking a minimum intensity projection of the probe channel, which represents the background auto-fluorescence of each cell. This projection is equalized with CLAHE and smoothed with a Gaussian filter. Taking this image as the basis for the Watershed algorithm and applying the seeds derived from the DAPI channel yields a segmented image.
To locate the spots representing RNA molecules, each z-slice in the probe channel stack is normalised, and a maximum intensity projection of the stack taken. A Sobel filter is applied to the projection to detect edges. We use scikit-image’s implementation of fast normalised cross-correlation template matching to find the probe locations. This algorithm tests the correlation between a given template and the equivalently sized section of a larger image for each point in that image. It produces another image, the intensity values of which correspond to the degree to which the template correlates with the image (so that the maximum intensity value corresponds with perfect correlation, and the minimum with perfect anti-correlation). We initially apply this algorithm using a template constructed as an annular element sized to the diffraction radius of the microscope. The single closest match to this template is then taken as a second template to re-apply the correlation. We then apply a correlation threshold, correlation values above which corresponded to identified mRNA spots, yielding their locations. This threshold was chosen based on comparison to manual spot counting in test data sets, such that it gave an optimum balance between false negatives and false positives.
For validation of the results, identified spot locations and the segmentation derived from the DAPI nuclear stain and probe autofluorescence is used to produce an annotated image. This image overlays probe counts and segmentation boundaries on the projection of the probe autofluorescence channel. Each image is manually inspected to ensure that the image analysis workflow has not generated spurious results.
Graphs presented in Fig. 6 were created using GraphPad Prism 6 for Mac OS X software version 6.0 g (La Jolla, California).
SD, SR and CD designed the experiments. SD and SR developed the smFISH method and acquired the data. TO and MH developed the method for automated image analysis. SD, SR and CD analyzed the data. SD, TO, MH CD and SR wrote the manuscript an all authors approved the final version. All authors read and approved the final manuscript.
We acknowledge Peter Shaw and Silvia Costa for comments on the manuscript, Ali Pendle for advice, as well as all members of the Dean and Howard groups for discussions. This work was supported by the UK Biotechnology and Biological Sciences Research Council (BBSRC) Grant BB/K00008X/1 and the Earth and Life Systems Alliance (a collaborative venture between John Innes Centre and University of East Anglia). C.D. acknowledges support from European Research Council Advanced Grant MEXTIM and BBSRC Institute Strategic Programme Grant BB/J004588/1.
The authors declare that they have no competing interests.
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
- Lecuyer E, et al. Global analysis of mRNA localization reveals a prominent role in organizing cellular architecture and function. Cell. 2007;131(1):174–87.View ArticlePubMedGoogle Scholar
- Raj A, et al. Stochastic mRNA synthesis in mammalian cells. PLoS Biol. 2006;4(10):e309.View ArticlePubMedPubMed CentralGoogle Scholar
- Gall JG. Differential synthesis of the genes for ribosomal RNA during amphibian oogenesis. Proc Natl Acad Sci U S A. 1968;60(2):553–60.View ArticlePubMedPubMed CentralGoogle Scholar
- Akam ME. The location of Ultrabithorax transcripts in Drosophila tissue sections. EMBO J. 1983;2(11):2075–84.PubMedPubMed CentralGoogle Scholar
- Raap AK, et al. Ultra-sensitive FISH using peroxidase-mediated deposition of biotin- or fluorochrome tyramides. Hum Mol Genet. 1995;4(4):529–34.View ArticlePubMedGoogle Scholar
- Tautz D, Pfeifle C. A non-radioactive in situ hybridization method for the localization of specific RNAs in Drosophila embryos reveals translational control of the segmentation gene hunchback. Chromosoma. 1989;98(2):81–5.View ArticlePubMedGoogle Scholar
- Hejatko J, et al. In situ hybridization technique for mRNA detection in whole mount Arabidopsis samples. Nat Protoc. 2006;1(4):1939–46.View ArticlePubMedGoogle Scholar
- Brewer PB, et al. In situ hybridization for mRNA detection in Arabidopsis tissue sections. Nat Protoc. 2006;1(3):1462–7.View ArticlePubMedGoogle Scholar
- Javelle M, Marco CF, Timmermans M. In situ hybridization for the precise localization of transcripts in plants. J Vis Exp. 2011;57:e3328.PubMedGoogle Scholar
- Pontes O, et al. RNA polymerase V functions in Arabidopsis interphase heterochromatin organization independently of the 24-nt siRNA-directed DNA methylation pathway. Mol Plant. 2009;2(4):700–10.View ArticlePubMedPubMed CentralGoogle Scholar
- Pontvianne F, et al. Subnuclear partitioning of rRNA genes between the nucleolus and nucleoplasm reflects alternative epiallelic states. Genes Dev. 2013;27(14):1545–50.View ArticlePubMedPubMed CentralGoogle Scholar
- Femino AM, et al. Visualization of single RNA transcripts in situ. Science. 1998;280(5363):585–90.View ArticlePubMedGoogle Scholar
- Raj A, et al. Imaging individual mRNA molecules using multiple singly labeled probes. Nat Methods. 2008;5(10):877–9.View ArticlePubMedPubMed CentralGoogle Scholar
- Castelnuovo M, et al. Bimodal expression of PHO84 is modulated by early termination of antisense transcription. Nat Struct Mol Biol. 2013;20(7):851–8.View ArticlePubMedGoogle Scholar
- Neuert G, et al. Systematic identification of signal-activated stochastic gene regulation. Science. 2013;339(6119):584–7.View ArticlePubMedPubMed CentralGoogle Scholar
- Yang B, et al. Single-cell phenotyping within transparent intact tissue through whole-body clearing. Cell. 2014;158(4):945–58.View ArticlePubMedPubMed CentralGoogle Scholar
- Ji N, et al. Feedback control of gene expression variability in the Caenorhabditis elegans Wnt pathway. Cell. 2013;155(4):869–80.View ArticlePubMedGoogle Scholar
- Levesque MJ, Raj A. Single-chromosome transcriptional profiling reveals chromosomal gene expression regulation. Nat Methods. 2013;10(3):246–8.View ArticlePubMedPubMed CentralGoogle Scholar
- Cabili MN, et al. Localization and abundance analysis of human lncRNAs at single-cell and single-molecule resolution. Genome Biol. 2015;16:20.View ArticlePubMedPubMed CentralGoogle Scholar
- Frost F. Fluorescent microscopy autofluorescence: plants, fungi, bacteria. Cambridge: Cambridge University Press; 1995.Google Scholar
- Lillo C, et al. Protein phosphatases PP2A, PP4 and PP6: mediators and regulators in development and responses to environmental cues. Plant Cell Environ. 2014;37(12):2631–48.View ArticlePubMedGoogle Scholar
- Czechowski T, et al. Genome-wide identification and testing of superior reference genes for transcript normalization in Arabidopsis. Plant Physiol. 2005;139(1):5–17.View ArticlePubMedPubMed CentralGoogle Scholar
- Raj A, Tyagi S. Detection of individual endogenous RNA transcripts in situ using multiple singly labeled probes. Methods Enzymol. 2010;472:365–86.View ArticlePubMedGoogle Scholar
- Lubeck E, Cai L. Single-cell systems biology by super-resolution imaging and combinatorial labeling. Nat Methods. 2012;9(7):743–8.View ArticlePubMedPubMed CentralGoogle Scholar
- Little SC, Tikhonov M, Gregor T. Precise developmental gene expression arises from globally stochastic transcriptional activity. Cell. 2013;154(4):789–800.View ArticlePubMedPubMed CentralGoogle Scholar
- Raj A, van Oudenaarden A. Nature, nurture, or chance: stochastic gene expression and its consequences. Cell. 2008;135(2):216–26.View ArticlePubMedPubMed CentralGoogle Scholar
- Raj A, van Oudenaarden A. Single-molecule approaches to stochastic gene expression. Annu Rev Biophys. 2009;38:255–70.View ArticlePubMedPubMed CentralGoogle Scholar
- To TL, Maheshri N. Noise can induce bimodality in positive transcriptional feedback loops without bistability. Science. 2010;327(5969):1142–5.View ArticlePubMedGoogle Scholar
- Lyubimova A, et al. Single-molecule mRNA detection and counting in mammalian tissue. Nat Protoc. 2013;8(9):1743–58.View ArticlePubMedGoogle Scholar
- Hansen CH, van Oudenaarden A. Allele-specific detection of single mRNA molecules in situ. Nat Methods. 2013;10(9):869–71.View ArticlePubMedPubMed CentralGoogle Scholar
- Levesque MJ, et al. Visualizing SNVs to quantify allele-specific expression in single cells. Nat Methods. 2013;10(9):865–7.View ArticlePubMedPubMed CentralGoogle Scholar
- Linkert M, et al. Metadata matters: access to image data in the real world. J Cell Biol. 2010;189(5):777–82.View ArticlePubMedPubMed CentralGoogle Scholar
- van der Walt S, Colbert SC, Varoquaux G. The NumPy array: a structure for efficient numerical computation. Comput Sci Eng. 2011;13(2):22–30.View ArticleGoogle Scholar
- van der Walt S, et al. scikit-image: image processing in Python. PeerJ. 2014;2:e453.View ArticlePubMedPubMed CentralGoogle Scholar
- Pizer SM, et al. Adaptive histogram equalization and its variations. Comput Vis Graph Image Process. 1987;39(3):355–68.View ArticleGoogle Scholar