OUP user menu

Fluorescent hybridisation combined with flow cytometry and hybridisation of total RNA to analyse the composition of microbial communities in human faeces using 16S rRNA probes

Lionel Rigottier-Gois , Anne-Gaëlle Le Bourhis , Gramet Geneviève , Violaine Rochet , Joël Doré
DOI: http://dx.doi.org/10.1111/j.1574-6941.2003.tb01063.x 237-245 First published online: 1 March 2003


To determine the structure of human faecal microbiota, faecal samples from 23 healthy individuals were analysed with a similar set of probes targeting six phylogenetic groups using rRNA dot-blot hybridisation and whole cell fluorescent in situ hybridisation (FISH) combined with flow cytometry. When microbiota compositions derived by each method were compared, the results were not statistically different for Clostridium coccoides, Fusobacterium prausnitzii, Bifidobacterium spp. and Enterobacteria. Conversely, the proportions were significantly different for Bacteroides and Atopobium (P<0.05). The metabolic state of these bacteria within the colon could explain the discrepancy observed between the rRNA level and the actual cell proportion. However, both approaches supplied consistent and complementary information on the structure of the faecal microbiota. FISH combined with flow cytometry appears best suited to future high throughput analysis.

  • Human fecal microbiota
  • Dot-blot hybridisation
  • Fluorescent in situ hybridisation
  • Flow cytometry


The human colon is colonised by a wide range of bacterial communities amounting to 1014 bacteria and playing an important role in the health status of the host due to their involvement in nutrition, immunology and pathology [1,2]. To acquire more knowledge concerning the role of these bacterial communities in human health, epidemiological investigations are required. The attention paid to dietary modulation of the colonic microbiota using functional foods such as probiotics has further increased the need to better understand the structure and activities of the colonic microbiota. To be able to conduct epidemiological studies, high throughput methods are required to collect qualitative and quantitative information from large numbers of samples. These methods should also take into account the dynamics of the colonic microbiota. Traditionally, the faecal microbiota has been studied using bacteriological culture methods based on anaerobic selective media [3,4]. Culture-based studies have shown that faecal bacteria comprise 400 distinct species, but 70% of these bacteria belonged to the following six genera: Bacteroides, Eubacterium, Clostridium, Ruminococcus, Fusobacterium and Bifidobacterium. However, most bacteria from the faecal microbiota are strict anaerobes and thus difficult to culture. Analyses based on microscopic examination have shown that 60 to 70% of the faecal bacteria cannot be cultured [5,6]. Culture-based studies have thus only made it possible to partially identify the composition of the faecal microbiota. The current development of molecular methods in microbial ecology is now enabling the culture-independent analysis of the faecal microbiota. Molecular analyses have mainly been targeted at ribosomal RNA and, more specifically, at 16S rRNA [7]. Applied to faecal microbiota, molecular approaches based on the direct study of 16S rRNA genes [5] or using 16S rRNA probe hybridisation [8,9] have revealed the predominance of four phylogenetic groups gathering the six dominant cultivable genera. The Bacteroides group comprises the genus Bacteroides and also encloses the genera Prevotella and Porphyromonas. The Clostridium coccoides group includes species of Clostridium, Eubacterium, Ruminococcus and Butyrivibrio and corresponds to the Clostridium rRNA sub-cluster XIVa defined by Collins et al. [10]. The Clostridium leptum group contains members of Clostridium, Eubacterium, Ruminococcus and Anaerofilum genera as well as the Fusobacterium prausnitzii species and corresponds to the Clostridium rRNA cluster IV of Collins et al. [10]. The Bifidobacterium genus represents the fourth predominant group of the faecal microbiota. In addition to these four phylogenetic groups, sub-dominant groups, like the Enterobacteria, including Escherichia coli [9], and the Atopobium cluster, including the Coriobacterium group [11], were frequently detected in human faeces. Although molecular methods were consistent in terms of dominant groups [12], a more thorough analysis of the data acquired independently revealed differences in the proportions of each phylogenetic group, depending on the approach used. For instance, the Bacteroides group is by far the main group in terms of rRNA index when using dot-blot hybridisation [9,13], while the C. coccoides group is the main one in terms of the number of cells when using fluorescent in situ hybridisation (FISH) [8]. These inconsistencies between studies could be explained by different probe specificities or sampling strategies. To conclusively determine which is the main group in human faecal microbiota, we used the same faecal samples and two direct approaches based on 16S rRNA probe hybridisation with a similar set of probes targeting the main components of the faecal microbiota. Dot-blot hybridisation represented in our group the reference method to characterize the composition of faecal microbiota in healthy individuals. It was compared to FISH analysis adapted to flow cytometry detection, and applied in this study to analyse the composition of the faecal microbiota in humans.

Materials and methods

faecal samples

Faeces from 23 healthy human subjects (12 men and 11 women) between 3 and 68 years of age were collected. Donors were on a West European diet. None had any history of digestive pathology or received antibiotic treatment within six months prior to the study. Faecal samples were collected as described in Rochet et al. [14] in sterile plastic boxes and kept under anaerobic conditions using an anaerocult®A (Merck, Nogent sur Marne, France) and stored at 4°C for a maximum of 4 h before processing.


The probes used in this study targeted the small sub-unit rRNA. The sequences, reference strains and references of the control and group-specific probes are presented in Table 1.

View this table:

16S rRNA-targeted oligonucleotide probes, experimental wash temperatures and strains used as reference in rRNA dot-blot hybridisation

ProbeSequence from 5′ to 3′ endOPD codeaReference RNA extracted fromWash temperature (°C)Reference
EUB 338GCTGCCTCCCGTAGGAGTS-D-Bact-0338-a-A-18Escherichia coli MRS600 (Roche)54[16]
Bacto 1080GCACTTAAGCCGACACCTS-*-Bacto-1080-a-A-18Bacteroides vulgatus ATCC848250[13]
Bac 303CCAATGTGGGGGACCTTS-*-Bacto-0303-a-A-17NAbNAb[27]
Erec 482GCTTCTTAGTCARGTACCGS-*-Erec-0482-a-A-19Ruminococcus productus ATCC 2734047[8]
Fprau 645CCTCTGCACTACTCAAGAAAAACS-*-Fprau-0645-a-A-23F. prausnitzii L2–6c50[28]
Bif 228GATAGGACGCGACCCCATS-G-Bif-0228-a-A-18Bifidobacterium longum ATCC 1570753.5[30]
Bif 164CATCCGGCATTACCACCCS-G-Bif-0164-a-A-18NAbNAb[29]
Enter 1432CTTTTGCAACCCACTS-*-Ent-1432-a-A-15Escherichia coli MRS600 (Roche)43[9]
Ato 291GGTCGGTCTCTCAACCCS-*-Ato-0291-a-A-17Collinsella aerofaciens ATCC 2598656.5[11]
  • a OPD code: oligonucleotide probe database code

  • b NA: not applicable

  • c Strain F. prausnitzii L2–6 kindly provided by H. Flint, Rowett Research Institute, UK.

Total RNA extraction and dot-blot hybridisation

Total RNA was extracted from 0.2 g of frozen faecal material as described by Stahl et al. [15] as modified by Doré et al. [13]. rRNA standards were prepared by extracting RNA from the reference strains in Table 1. Determination of total RNA extracted from faeces was performed using the universal probe Univ 1390 against the rRNA standard of E. coli MRS600 (Roche Molecular Diagnostics). An equivalent of 200 ng of total RNA from each faecal sample were blotted in triplicate or duplicate on Nylon membranes (one membrane per probe). The hybridisation was performed overnight at 42°C in 10 ml of hybridisation buffer (900 mM NaCl, 19.5 mM NaH2PO4, 30.5 mM Na2HPO4, 5 mM EDTA, 0.5% sodium dodecyl sulfate (SDS), 10×Denhart's solution, 0.5 mg ml−1 PolyA, pH 7.2) containing 40 nmol of radioactively labelled probe. Two washing steps were performed for 30 min each in 200 ml 1×SSC, 1% SDS at the appropriate washing temperatures of the group-specific probe (Table 1). The degree of hybridisation on dot-blots was quantitated by radio-imaging using the Instant Imager (Packard Instrument). Results are expressed as rRNA indexes representing the group-specific rRNA as a percentage of the total bacterial rRNA assessed with the EUB 338 probe targeting a region conserved within the domain bacteria [16].

Cell fixation, permeabilisation and in situ hybridisation

Faeces were homogenised by mechanical kneading for 3 min and 0.5 g aliquots (wet weight) added to 4.5 ml of sterile brain heart infusion (BHI) broth. The suspension was mixed 3–5 min in a 50-ml stoppered sterile glass jar fitted with a magnetic stirrer. One volume of the suspension was added to 3 volumes of 4% paraformaldehyde (PFA) in phosphate-buffered saline (PBS; 130 mM NaCl, 3 mM NaH2PO4 2H2O, 7 mM Na2HPO4 12H2O, pH 7.2). After overnight fixation at 4°C, the fixed suspension in PFA was stored at −70°C until hybridisation. The EUB 338 probe was used as the positive control probe. The NON 338 probe designed by Wallner et al. [17] was used as the negative control probe. Both control probes were covalently linked at their 5′ end either to fluorescein isothiocyanate (FITC) or to the sulfoindocyanine dye indodicarbocyanine (Cy5; Interactiva). The group-specific probes were only labelled at their 5′ end with Cy5. 100 μl of fixed suspension was mixed into 1.0 ml of PBS. Before hybridisation, cells were always pelleted at 8000×g for 3 min in a microcentrifuge tube and resuspended in a volume of 1 ml. After one wash in Tris–EDTA buffer (100 mM Tris–HCl, pH 8.0, 50 mM EDTA), pellets were resuspended in Tris–EDTA buffer containing 1 mg ml−1 lysozyme (Serva, Heidelberg, Germany) and incubated for 10 min at room temperature. Cells were then washed in PBS to remove lysozyme and equilibrated in the hybridisation solution (900 mM NaCl, 20 mM Tris–HCl, pH 8.0, 0.1% SDS, 30% formamide). A 50-μl aliquot of this suspension was used for FISH with control and group-specific probes. Hybridisation was performed in a 96-well microtiter plate overnight at 35°C in the hybridisation solution containing 4 ng μl−1 of the appropriate labelled probe(s). Following hybridisation, 150 μl of hybridisation solution was added in each well and the cells were pelleted at 4000×g for 15 min. Non-specific binding of the probe was removed by incubating the bacterial cell suspension at 37°C for 20 min in a washing solution (64 mM NaCl, 20 mM Tris–HCl, pH 8.0, 0.1% SDS). Cells were finally pelleted and resuspended in PBS. Aliquots of 100 μl were added to 1 ml of Facs Flow (Becton Dickinson) for data acquisition by flow cytometry.

Flow cytometry

Data acquisition was performed using a Facs Calibur flow cytometer (Becton Dickinson) equipped with an air-cooled argon ion laser providing 15 mW at 488 nm combined with a 635-nm red-diode laser. All the parameters were collected as logarithmic signals. The 488-nm laser was used to measure the forward angle scatter (FSC, in the 488-nm band-pass filter), the side angle scatter (SSC, in the 488-nm band-pass filter), and the green fluorescence intensity (FL1, in the 530-nm band-pass filter) conferred by the FITC-labelled probes. The red diode laser was used to detect the red fluorescence conferred by the Cy5-labelled probes (FL4, in a 660-nm band-pass filter). The acquisition threshold was set in the side scatter channel. The rate of events in the flow was generally lower than 3000 events s−1. With faecal suspensions, a total of 100 000 events were stored in list mode files. Subsequent analyses were conducted using CellQuest software (Becton Dickinson). Cell enumeration was performed by combining, in one hybridisation tube, one group Cy5-probe with the EUB 338-FITC probe. An FL1 histogram (green fluorescence) was used to evaluate the total number of bacteria hybridising with the EUB 338-FITC probe. A gate was designed in this histogram representing the total number of bacterial cells in the sample and was used to build an FL4 histogram (red fluorescence) to directly estimate the proportion of cells targeted by the group Cy5-probe in the sample. The proportion of cells was corrected by eliminating background fluorescence, which was measured using the negative control NON 338-Cy5 probe. Results were expressed as cells hybridising with the group-Cy5 probe as a proportion of the total bacteria hybridising with the EUB 338-FITC bacteria domain probe.

Statistical analysis

The mean rRNA indexes determined by dot-blot hybridisation and the mean cell proportions estimated by FISH were calculated using the results of duplicates with FISH and of triplicates or duplicates with rRNA dot-blot hybridisation. Only means with a standard error of less than 6% with RNA and 2% with FISH were accepted. The Mann–Whitney (Wilcoxon) W-test to compare the medians was performed using StatGraphics® (Manugistics, Rockville, MD, USA) to determine whether there was a significant difference between the proportions of the bacterial groups determined by the two methods, at a confidence level of 95% (P<0.05).


Composition of faecal microbiota assessed by rRNA hybridisation

RNAs were successfully extracted from the 23 samples tested. Yields ranged from 1.5 to 258 μg g wet weight−1. The rRNA indexes obtained from the groups targeted with the set of probes are detailed in Table 2. A large variation was observed between individual faecal samples. Bacteroides represented the main rRNA index with a relative proportion of 41.7%±13.5% and a range from 12.1% to 64.2%. The probe for C. coccoides detected 21.9%±10.2% of the total rRNA and was the second highest rRNA index (ranging from 11.8% to 54.6%). The third highest was observed with the probe for F. prausnitzii, which represented on average 9.2%±7.4% of total rRNA hybridised (range from 2.0% to 33.9%). When probes Bif 228, Enter 1432 and Ato 291 were used, the rRNA indexes were 2.9%±4.2%, 1.0%±2.7% and 0.3%±0.5%, respectively. When the rRNA indexes of the six groups were added together, a total of 76.9%±19.8% was found (from 46.4% to 125.8%). Surprisingly, the total of the rRNA indexes for four individuals was higher than 100% (samples 3, 5, 17 and 25).

View this table:

Proportions of Bacteroides, C. coccoides, F. prausnitzii, Bifidobacterium, Enterobacteria and Atopobium groups in healthy humans measured by rRNA dot-blot hybridisation using EUB 338 as reference compared to Bacto 1080, Erec 482, Fprau 645, Bif 228, Enter 1432 and Ato 291

IndividualBacterial group
Bacto 1080Erec 482Fprau 645Bif 228Enter 1432Ato 291Total

Composition of faecal microbiota assessed by FISH combined with flow cytometry

Typical flow cytometry histograms and dot-plots are presented in Fig. 1. All group-specific probes gave a shift in signal of 1 log unit or more, allowing the specific detection and enumeration of the corresponding cells. The composition of the faecal microbiota of the 23 samples was successfully analysed by FISH adapted for detection by flow cytometry. The proportions of cells hybridised with the group-specific probes among the bacteria detected with the EUB 338 probe are presented in Table 3. The most abundant group was detected with the C. coccoides group probe and represented 22.0%±7.6% of cells with a range from 10.4% to 38.5%. The F. prausnitzii group was the second most represented with 11.3%±5.9% of cells detected (ranging from 1.3%–25.5%). The Bacteroides group came third and accounted for 9.1%±6.7% of bacterial cells (ranging from 0.4%–26.1%). The probes for Bifidobacterium, Enterobacteria and Atopobium groups gathered 4.1%±3.9%, 1.0%±2.8% and 3.7%±2.8% of bacterial cells, respectively. When the proportions of bacterial cells were added together, a mean of 51.0%±14.4% was obtained with the six group-specific probes (ranging from 26.9% to 75.5%).


Flow cytometry dot-plots and histograms obtained by FISH analysis of sample 1. Fixed cells were hybridised in (A) with either NON 338-FITC in FL1 or NON 338-Cy5 in FL4, (B) with a combination of EUB 338-FITC and Erec 482-Cy5, (C) with a combination of EUB 338-FITC and Fprau 645-Cy5 and in (D) with a combination of EUB 338-FITC and Bac 303-Cy5. FL1 histograms show green fluorescence intensities conferred by EUB 338-FITC. The events under the region R1 corresponded to bacterial cells hybridised with the probe EUB 338-FITC. This region was designed according to the level of background when NON 338-FITC was used. FL4 histograms were gated on the region R1 corresponding to M1 of the FL1 histogram. FL4 histograms show red fluorescence intensities conferred by the Cy5 probes. The events under M2 represented the proportion of bacterial cells hybridised with the group-specific probe within the total bacterial cells hybridised with the bacteria domain probe EUB 338-FITC. The proportion of cells was corrected by eliminating background fluorescence, which was measured using the negative control NON 338-Cy5 probe.

View this table:

Proportions of Bacteroides, C. coccoides, F. prausnitzii, Bifidobacterium, Enterobacteria and Atopobium groups in healthy humans assessed by FISH combined with flow cytometry detection using Bac 303, Erec 482, Fprau 645, Bif 164, Enter 1432 and Ato 291

IndividualBacterial group
Bac 303Erec 482Fprau 645Bif 164Enter 1432Ato 291Total

Correspondence between compositions of faecal microbiota assessed with the two 16S rRNA probing methods

The distribution of the proportions of each group is represented in Fig. 2. Statistical analyses were performed to determine whether there were statistically significant differences between the relative proportions of the groups according to the molecular approach used. Statistically significant differences (P<0.05) were found between the proportions of Bacteroides and Atopobium groups. When the proportions of each phylogenetic group were added together and compared, a statistically significant difference was observed between the totals obtained by each method. Conversely, no statistically significant differences were found between the proportions derived by the two methods for the C. coccoides, F. prausnitzii, Bifidobacterium and Enterobacteria groups.


Distribution of the proportion of bacterial groups in human faeces assessed by relative rRNA dot-blot hybridisation and by relative cell enumeration by FISH combined with flow cytometry.


Two 16S rRNA probing methods were used in this study to characterise the composition of faecal microbiota in 23 healthy humans. The rRNA dot-blot hybridisation determined the relative proportion of the rRNA of a phylogenetic group among the total bacterial rRNA. FISH combined with flow cytometry estimated the relative proportion of the bacterial cells of a group within the total number of bacterial cells. When comparing faecal microbiota composition, results were not statistically different for the following four groups: C. coccoides, F. prausnitzii, Bifidobacterium spp. and Enterobacteria. The results were, however, significantly different for the Bacteroides and Atopobium groups.

When the results of the two methods were compared independently, the composition of the faecal microbiota of our samples were consistent with several studies based on the same molecular methods. The rRNA indexes of the six predominant groups of the faecal microbiota were consistent with the rRNA indexes estimated by Doré et al. [13] and Sghir et al. [9]. In particular, with an rRNA index of more than 37%, the Bacteroides group was the major group in the faecal microbiota of healthy humans. This result was consistent with the observation that Bacteroides was the most common cultivable group and represented 30% of the total cultivable bacteria [4]. The composition of the faecal microbiota obtained by FISH adapted to flow cytometry detection correlated well with the composition observed when FISH was combined with microscopic detection and automated image analysis [8,18,19]. Using FISH, the C. coccoides group was the main group representing more than 22% of the total bacteria in the faecal microbiota. The 9% of bacterial cells hybridising with the probe Bac 303 was consistent with the proportion of Bacteroides cells estimated when the same group probe was used previously (e.g. 4.6% in Tannock et al. [19]). However, this proportion was lower than in Franks et al. [8], when detected with the probes Bfra 602 and Bdis 656, where it amounted to 20%.

In the present study, analysing one set of faecal samples using two molecular methods with a similar set of probes showed similar proportions of C. coccoides, F. prausnitzii, Bifidobacterium spp. and Enterobacteria groups. The C. coccoides and F. prausnitzii groups are predominant phylogenetic groups in the faecal microbiota, and both methods supplied the same information, as previously reported [8,9]. Concerning the Bifidobacterium group, although no statistically significant difference was observed between the two methods, a bimodal distribution of the frequency of proportions was evident in the total RNA hybridisation, but not with FISH. Among the individuals analysed, RNA from Bifidobacterium was either expressed poorly, with 18 out of 23 individuals exhibiting rRNA indexes below 2.6%, or substantially, with five individuals exhibiting rRNA indexes higher than 9%. This suggests that the Bifidobacterium species have different levels of adaptation to the distal colon. This could be linked to the metabolic heterogeneity of the species within the genus Bifidobacterium[20]. Alternatively, external parameters such as the diet of each individual can positively influence the Bifidobacterium spp., as has been shown with dietary fibres [21,22]. Regarding Enterobacteria, a good correlation between the two methods was observed. The best example concerns faecal sample 14, which presented an unusually high level of Enterobacteria with 13.9%±4.8% detected by rRNA dot-blot hybridisation and 13.6%±0.5% by FISH.

The rRNA and cell proportions of the Bacteroides group are consistent with molecular studies using either one or other of the two probing strategies [12]. Discrepancy between the proportions of the Bacteroides group had already been observed when the results of the two probing techniques were compared but it could be explained by the samplings which were different. In the present study we observed that, for the same set of faecal samples, this difference still exists. The specificity of the Bacteroides group probes used could be one reason for the difference. Two probes were used for the following reasons: (i) the probe Bac 303 targeting the genera Bacteroides and Prevotella was difficult to label radioactively and thus not reliable for evaluating the rRNA index (G. Gramet, personal communication); (ii) the probe Bacto 1080, which in addition to Bac 303 detects the Porphyromonas genus, targeted a region not accessible in FISH technology (L. Rigottier-Gois, unpublished observation). The importance of the accessibility of the probe to its target has been shown by Fuchs et al., for the 16S and 23S rRNA of E. coli [23,24]. However, the Porphyromonas genus could hardly explain the 31% difference in the proportion of Bacteroides estimated by the two methods. Culture or even molecular-based studies [5] showed that Porphyromonas members represented less than 2% of the total bacteria in human faeces. Our main hypothesis is that the difference relates to the actual parameters measured by the two methods. rRNA dot-blot hybridisation gave an rRNA index that reflects the number of ribosomal operons, the ribosomal content of cells, and the general metabolic activity, while the FISH analysis measured a proportion of cells. The amount of rRNA per cell is different according to the type of species and the metabolic state of the bacterial cell [25]. We observed that the rRNA indexes and cell proportions were consistent for most of the phylogenetic groups in human faeces, with the exception of Bacteroides, where the rRNA content was significantly higher than the relative cell proportion. This could indicate that the Bacteroides are metabolically more active than the other groups in the human colon. This observation could be related to the great nutritional ability and versatility of the bacteria in the Bacteroides group [26]. Complex exogenous and endogenous substrates are their main source of energy in the distal colon. Their high metabolic activity may result from the ability of Bacteroides to degrade endogenous muco-polysaccharides and glyco-polysaccharides which are produced in the whole colon. Conversely, the high relative proportion of Atopobium cells compared to their small rRNA index could be due to a less appropriate nutritional environment for the bacteria of this group. In spite of the variations observed for these two groups, it is important to note that the differences observed represent less than 0.5 log unit of population equivalent, which is the level of precision usually offered by culture-based methods.

On average, more than 50% of the bacterial cells and 70% of the total RNA were detected with our set of group-specific probes. This means that new probes have to be developed and validated to detect other bacteria present in the faecal microbiota and not yet enumerated with the current set of probes. Efforts now have to be made to further analyse the 16S rDNA from clone libraries of total DNA or from cultivated members of the faecal microbiota to enlarge the set of phylogenetic groups to detect and better describe the bacterial composition of human faeces.

This study offers a solid basis of information concerning the reliability of FISH combined with flow cytometry as used to characterise the composition of faecal microbiota as it gave consistent and complementary data with respect to the reference method of rRNA dot-blot hybridisation. In the near future, if large scale analyses are to be conducted, we will use FISH combined with flow cytometry because it is undoubtedly a high throughput method. For instance, it could be applied to samples from healthy humans from all over the world to provide a geographic description of the composition of the faecal microbiota. With this method it will be possible to analyse on a larger scale the colonisation of the gut in infants or the impact of ageing on intestinal microbiota. It will also be possible to perform nutritional studies to assess the potential benefits of functional foods on faecal microbiota. This high throughput method might also contribute to the identification of changes in the composition of the intestinal microbiota in patients suffering from inflammatory bowel diseases.


We thank P. Pochart and V. Chmiliewski for their helpful advice and comments during the set-up of the flow cytometry detection of microorganisms. We gratefully acknowledge the human volunteers who took part to the study. This study was carried out with financial support from the Commission of the European Communities, specifically the RTD programme ‘Quality of Life and Management of Living Resources’, QLK1-2000-108, ‘Microbe Diagnostics’, coordinated by Professor Michaël Blaut (Dife, Germany).


  1. [1]
  2. [2]
  3. [3]
  4. [4]
  5. [5]
  6. [6]
  7. [7]
  8. [8]
  9. [9]
  10. [10]
  11. [11]
  12. [12]
  13. [13]
  14. [14]
  15. [15]
  16. [16]
  17. [17]
  18. [18]
  19. [19]
  20. [20]
  21. [21]
  22. [22]
  23. [23]
  24. [24]
  25. [25]
  26. [26]
  27. [27]
  28. [28]
  29. [29]
  30. [30]
View Abstract