OUP user menu

Phase variation in meningococcal lipooligosaccharide biosynthesis genes

A.W. Berrington, Y.-C. Tan, Y. Srikhanta, B. Kuipers, P. van der Ley, I.R.A. Peak, M.P. Jennings
DOI: http://dx.doi.org/10.1111/j.1574-695X.2002.tb00633.x 267-275 First published online: 1 December 2002


Neisseria meningitidis expresses a range of lipooligosaccharide (LOS) structures, comprising of at least 13 immunotypes (ITs). Meningococcal LOS is subject to phase variation of its terminal structures allowing switching between ITs, which is proposed to have functional significance in disease. The objectives of this study were to investigate the repertoire of structures that can be expressed in clinical isolates, and to examine the role of phase-variable expression of LOS genes during invasive disease. Southern blotting was used to detect the presence of LOS biosynthetic genes in two collections of meningococci, a global set of strains previously assigned to lineages of greater or lesser virulence, and a collection of local clinical isolates which included paired throat and blood isolates from individual patients. Where the phase-variable genes lgtA, lgtC or lgtG were identified, they were amplified by PCR and the homopolymeric tracts, responsible for their phase-variable expression, were sequenced. The results revealed great potential for variation between alternate LOS structures in the isolates studied, with most strains capable of expressing several alternative terminal structures. The structures predicted to be currently expressed by the genotype of the strains agreed well with conventional immunotyping. No correlation was observed between the structural repertoire and virulence of the isolate. Based on the potential for LOS phase variation in the clinical collection and observations with the paired patient isolates, our data suggest that phase variation of LOS structures is not required for translocation between distinct compartments in the host.

  • Meningococcus
  • Phase variation
  • Lipooligosaccharide
  • Lipopolysaccharide
  • Pathogenesis
  • Neisseria meningitidis

1 Introduction

Neisseria meningitidis is a common colonist of the human nasopharyngeal epithelium, with carriage rates exceeding 20% in certain populations [1]. In contrast, meningococcal disease is rare and plays no part in disseminating the organism, but when tissue invasion occurs the consequences can be grave. Meningococcal meningo-encephalitis and meningococcal bacteraemia carry a mortality approaching 10% [2].

Many Neisseria genes are subject to phase variation, which can be defined as the high frequency, reversible on–off switching of gene expression. Phase variation is a feature of genes associated with a variety of meningococcal antigens including pili [3], capsule [4], outer membrane proteins [5,6], haemoglobin receptors [7], and lipooligosaccharide [8,9]. In Neisseria sp., phase variation is commonly mediated by homopolymeric tracts, which act as sites of hypermutation: tracts situated within promoters can influence the degree of gene expression by regulating transcription (e.g. Opc [5]), while tracts within coding regions result in on–off switching according to whether the downstream sequences are moved in or out of frame for expression (e.g. lgtA[8,9]).

Phase variation has been proposed to underlie the capacity of the organism to cause invasive disease. In cases of invasive disease the meningococcus translocates between and survives within a variety of anatomical sites including the nasopharyngeal epithelial surface, the blood, the cerebrospinal fluid (CSF), and the tissues between them. Phase variation acts as a generator of diversity to maximise the chance of survival during host colonisation. There are three lines of evidence in favour of this hypothesis. First, phase variation is a feature of antigens, often surface associated, which are known to have key roles in colonisation and invasion (for example the Opa proteins), and for which the ability to vary gene expression facilitates differences in tissue tropism [10]. Second, there are well-documented examples in which phase variation, in some cases of multiple antigens [11], has been shown to influence the interaction between organism and host cells in vitro. Third, there are a few in vivo data demonstrating phase variation between different sites of infection in human cohorts or individual patients. For example, group B capsular polysaccharide is subject to on–off switching of expression via a poly-cytidine tract within siaD, the gene encoding the α-2,8-sialyltransferase responsible for capsule assembly in group B strains [4]. In vitro studies demonstrate that although the capsule can hinder initial colonisation, it is an important mediator of serum resistance during bacteraemia [12], and indeed the isolation of non-capsulate organisms from the carrier sites of patients suffering from invasive disease caused by capsulate strains is well recognised. Two such throat/blood pairs have been investigated at the molecular level, and in both instances phase variation of capsule expression by siaD frame-shift mutation was confirmed [4].

The lipid A component of meningococcal LOS (endotoxin) plays a crucial role in the pathogenesis of sepsis following invasion [13], but the importance of the oligosaccharide moiety in adherence and serum resistance is increasingly recognised. Meningococci are capable of expressing a range of LOS structures many of which may be distinguished immunologically. This facilitates division into at least 13 immunotypes (ITs), L1–L13 [14]. These oligosaccharide moieties are synthesised by glycosyltransferases encoded by a family of genes designated lgt, of which six, lgtABCDE[9] and the unlinked gene lgtG[9,15], have been described. lgtA, encoding a β1,3-N-acetylglucosamine transferase, and lgtB, encoding a β1,4-galactosyl transferase, are responsible for the addition of the terminal sugars of the lacto-N-neotetraose (LNT: Galβ1–4GlcNAcβ1–3Galβ1–4Glc) structure that is a feature of several ITs including L2, 3, 4, 7, and 9. lgtC, encoding an a1–4-galactosyl transferase, is required for synthesis of the L1 structure. lgtD is highly homologous to lgtA and encodes another β1,3-N-acetylgalactosamine transferase. This gene has been well described in Neisseria gonorrhoeae[16] but has been found in only a single strain of N. meningitidis[9]. The lgtE gene encodes a further β1,4-galactosyl transferase, while lgtG encodes a putative a1,3-galactosyl transferase.

lgtA, lgtC, lgtD and lgtG are subject to phase variation of expression mediated by homopolymeric tracts within their coding regions. This allows an individual strain to vary the LOS it expresses such that a population bearing these genes should be regarded as capable of producing a repertoire of ITs rather than just one [9]. For example, phase variation of lgtA mediates switching between ITs L3 and L8 in the meningococcal strain MC58 by mediating loss or gain of the LNT terminal structure of the L3 IT (Fig. 1) [8].

Figure 1

Primary structures of meningococcal oligosaccharides. ITs are labelled to the left of each structure, and grey arrows show phase variation between them. Only the terminal structures of L1 and L3 are shown as they are extensions of the L8 structure. L6 has the same basal structure as L4, but the terminal two sugars are missing. The activities of the lgtA, lgtB, lgtC, lgtE and lgtG gene products are indicated. lgt genes present in each of the type strains are indicated with arrowed lines on the left of the figure. Based on Scholten et al. [26].

Phase variation of lgtC expression facilitates switching between L1 and L8 (Fig. 1), and this has been observed to occur in strains 126E and M978 [9]. In our hands, the influence of lgtC switching is restricted to isolates in which lgtA is switched off: where lgtA is active its activity overrides that of lgtC and the L3 IT results. Phase variation of lgtG expression, mediating switching between L2 and L3 (Fig. 1), has also been demonstrated [9].

Several lines of evidence suggest that L3–L8 IT switching might be important in pathogenesis. First, strains expressing the L3 IT are serum resistant in comparison to those expressing the L8 IT [17]. LNT is thought to contribute to serum resistance indirectly through its capacity to become sialylated at the terminal β-d-galactose residue, which decreases the deposition of C3b and IgM on the cell surface leading to increased resistance to classical and alternative complement-mediated killing [12,18]. In contrast, the non-sialylated structures may be better adapted to initial adherence and invasion. In an in vitro model, Opc-mediated invasion of acapsulate MC58 is reduced for organisms expressing L3 compared to L8 bacteria [19]. In addition, phase variation of LOS in N. gonorrhoeae affects bacterial invasion of cultured cells and serum sensitivity [20]. Studies using mice have shown that L8 strains (presumably serum sensitive) predominate in the nasopharynx while L3 organisms are most commonly found in blood [21]. The same may apply in human infection [22].

Although intriguing, these observations are inconclusive regarding the relevance of phase variation of LOS to the pathogenesis of invasive disease. In part this reflects the paucity of data relating to clinical strains, particularly strains from multiple sites within individual patients. Accordingly, with a view to increasing our understanding of the role of phase variation, we have investigated two collections of meningococcal isolates. The first was a global collection of serogroup B strains. These had been previously characterised by other workers using multi-locus enzyme electrophoresis (MLEE) and multi-locus sequence typing [23], and are known to include representatives of the hypervirulent lineages associated with epidemic and hyperendemic disease, as well as representatives of less virulent lineages [23]. This collection was surveyed for the distribution of lgt genes in order to determine the IT repertoire of the individual strains. The second was a local collection derived from patients in South East Queensland, Australia with invasive meningococcal disease. The latter group of isolates was investigated with particular emphasis on the predicted and actual ITs of paired isolates from different body compartments within individual patients.

2 Materials and methods

2.1 Media and growth conditions

Meningococcal strains were grown at 37°C in 5% CO2 on brain-heart infusion agar (Oxoid) made with 1% agar and supplemented with 10% Levinthal base [24]. Blood clinical isolates were sub-cultured once from the broth (detected by automated Bactec) onto a solid agar then frozen. CSF, throat and sputum samples were plated directly onto solid agar, sub-cultured once then frozen.

2.2 LOS characterisation

LOS was harvested by lysis of organisms in SDS/proteinase K solution. Tricine–SDS–polyacrylamide gel electrophoresis and silver staining of LOS have been described previously [25]. Immunotyping was performed by whole-cell ELISA using monoclonal antibodies as previously reported [26].

2.3 Southern blotting and hybridisation

Cla I-digested genomic DNA from control and experimental isolates was separated on 0.7% agarose gels and transferred to PolyScreen PVDF membrane (NEN Research Products, Boston, MA, USA) essentially as described in Sambrook et al. [27]. Probes were generated by PCR amplification from appropriate control strains using the primer pairs Lic2 and Lic23 for lgtA (strain MC58, 550 bp), Lic13 and Lic1 for lgtB (strain MC58, 804 bp), LgtCF and LgtCR for lgtC (strain 126E, 945 bp), F104 and Xmore37 for lgtD (strain 126E, 338 bp), LG2 and LG3 for lgtG (strain 35E, 603 bp) (Table 4). Probes were labelled with digoxigenin using the DIG High Prime DNA Labelling and Detection Starter Kit 1 (Boehringer Mannheim) and tested in dot-blot experiments. Membranes were pre-hybridised in a buffer comprising 5× SSC, 0.02% SDS, and 1× blocking solution (Boehringer Mannheim) for 30 min, then hybridised to probe for at least 16 h in the same solution. For lgtA, lgtB and lgtE, hybridisation was carried out at 68°C followed by two 15-min washes at 68°C with pre-warmed 0.1× SSC/0.1% SDS. For lgtC and lgtG, hybridisation was carried out at 65°C followed by two 5-min washes at room temperature with 2× SSC/0.1% SDS and two 15-min washes at 65°C with pre-warmed 0.1× SSC/0.1% SDS. Specific detection of lgtD required higher stringency conditions: hybridisation was carried out at 68°C followed by two 15-min washes at 68°C with pre-warmed 0.05× SSC/0.05% SDS. After blocking, bound probe was detected by incubation with alkaline phosphatase (AP) labelled anti-digoxigenin antibody (anti-digoxigenin AP Fab fragments, Boehringer Mannheim), followed by a colourimetric AP substrate.

View this table:
Table 4

Primers used


2.4 PCR amplification

The homopolymeric tract regions of lgtA, lgtC and lgtG were amplified from boiled whole-cell lysates essentially as described in Sambrook et al. [27]. lgtA was amplified using primers Lic31ext and Lic16ext (355-bp amplicon), lgtC was amplified using primers LgtCF and LgtCR (946 bp), lgtG was amplified using primers LG1 and LG2 (1056 bp) (Table 4). To address the potential problem of PCR errors introducing misleading frame-shift mutations into the homopolymeric tracts, all PCRs were performed in triplicate. PCR products were purified from agarose using the QIAex II gel extraction kit (Qiagen) and pooled for use as sequencing template.

2.5 Sequencing

Sequencing was carried out using the BigDye Ready Reaction premix (PE Applied Biosystems), and performed on the pooled triplicate PCR products in both forward and reverse directions. The primers used were Lic31ext (lgtA forward), Lic16ext (lgtA reverse), LgtCF (lgtC forward), R99 (lgtC reverse), LG6 (lgtG forward) and LG8 (lgtG reverse) (Table 4). Samples were amplified using 25 cycles of 96°C ×10 s, 50°C ×5 s and 60°C for 4 min. Amplified DNA was recovered by ethanol precipitation and submitted to the Australian Genome Research Facility for automated sequencing.

3 Results

3.1 LOS structural variation repertoire in the global collection

We determined the presence of the LOS biosynthesis genes lgtA–E and lgtG by Southern blot analysis, to indicate the potential LOS expression of each strain. We also sequenced the homopolymeric tract of lgtACG (when present) to predict the expression status of these genes and the predicted IT.

The distribution of the lgt genes in the 37 members of the global collection are shown in Table 1. lgtA and lgtB are present in all of the strains tested, as is lgtE in all except NGH38. lgtC and lgtG are present in fewer strains and lgtD is not present in any of the strains tested. The poly-G tracts of lgtA range in size from 5 to 17 bp (Table 1). In all but three cases lgtA is in-frame for expression (tract length 5, 8, 11, 14 or 17 bp). The poly-G tracts of lgtC are between 9 and 14 bp in length. lgtC is in-frame for expression (tract length 10 or 13 bp) in six of the 14 strains carrying this gene. The poly-C tracts of lgtG are between 9 and 13 bp. lgtG is in-frame for expression (11-bp tract) in six of the 19 strains carrying the gene. There was no obvious association between lineage and either gene distribution or pattern of predicted gene expression.

View this table:
Table 1

Summary of Southern hybridisation, tract sequencing and immunotyping results for the global collection

MLEE clonesStrainSouthern hybridisation and sequencing resultsPredicted ITActual IT
A4 ClusterBZ10√5(√14)×(√10)L1,L2,L3,L4L3/L4L3
Lineage 3400√11√10××L1,L3,L4,L8L1/L3/L4L1,3
M101/93√8√13××L1,L3,L4,L8L1/L3/L4L1,3, L6
Sub-group 1BZ133√11×××L3, L4, L8L3/L4L3, L6
BZ149√5×××L3,L4L3/L4L3; L2,5
  • √ and × indicate the presence or absence of a gene. The numbers indicate the length of any associated homopolymeric tract — those in parentheses are predicted to cause the gene to be switched off. For simplification, ITs L3, L7 and L9, all of which terminate in the same LNT structure, are collectively referred to as L3. NI=not interpretable.

From the gene complement data, we could predict the potential range of ITs that each strain could express (‘predicted potential IT’. Tables 1–3), and from sequence analysis of the homopolymeric tracts we could predict the IT currently expressed by that strain (‘predicted current IT’. (Tables 1–3). The IT L2, L3 and L4 molecules all contain LNT, with the differences being in substitutions of the second heptose of the core region. Switching between these ITs is in part mediated by phase variation of lgtG, but the position of a phosphoethanolamine (PEA) also differs between L2/4 and L3. It is unclear whether separate enzymes with differing specificity catalyse addition of PEA, or whether a single enzyme with relaxed substrate specificity occurs whose activity is dependent on the presence or absence of the LgtG-added glucose residue. Until this is resolved, and the identity of the PEA transferase determined, it is not possible to predict whether L3 or L4 will be produced in the absence of LgtG activity. Consequently all strains containing lgtABE are shown as potentially expressing L3 and L4. Interestingly, of 33 strains capable of producing L4 (those with lgtABE, and no predicted LgtG activity), only one strain (MPJ361) actually does. Similarly the molecular basis for L5 and L6 production is not known.

To determine whether the genotype (gene complement and predicted expression status) correlated with phenotype, the IT of each strain was analysed using a panel of monoclonal antibodies, as previously described [26]. The two methods were in accord, apart from five strains in which IT could not be determined (see Table 1).

3.2 LOS structural variation repertoire in the local collection

The distribution of the lgt genes in the local collection are shown in Table 2. With the exceptions of the paired isolates (see below), this study was limited to the phase-variable meningococcal genes lgtA, lgtC and lgtG. As with the global collection, lgtA is universally present and in most cases in-frame for expression, while lgtC and lgtG are less uniformly distributed. With the proviso that the distribution of lgtB and lgtE was not determined in the unpaired isolates, once again the predicted ITs were generally in accordance with the observed ITs.

View this table:
Table 2

Summary of Southern hybridisation, tract sequencing and immunotyping results for the local collection

SiteStrainSouthern hybridisation and sequencing resultsPredicted ITActual IT
MPJ36IB√5(√11)×(√10)L3, L4L4
MPJ42IB√5√10×(√12)L1, L3L2,9
  • √ and × indicate the presence or absence of a gene, a blank indicates not tested. The numbers indicate the length of any associated homopolymeric tract — those in parentheses are predicted to cause the gene to be switched off. For simplification, ITs L3, L7 and L9, all of which terminate in the same LNT structure, are collectively referred to as L3.

3.3 Paired isolate studies

The local collection contained six nasopharyngeal/blood isolate pairs. Interestingly, none of these pairs showed any discordance with respect to their capsular expression — all were capsulate (data not shown). These isolates were investigated both phenotypically and genotypically with particular emphasis on the demonstration of lgt phase variation within pairs.

3.3.1 Genotypic analysis

For the 12 strains, lgtA, lgtC and/or lgtG tract sequencing were performed (Table 3). lgtA, present in all six pairs of isolates, showed no evidence of phase variation between the members of an individual pair. Of four pairs containing lgtC, one pair, 42B and 42T, showed apparent phase variation in that the poly-G tract present in the blood isolate was 10 bp long (predicted lgtC expression on), while that in the throat isolate was 11 bp long (predicted lgtC expression off). lgtG was present in four pairs of isolates without any evidence of phase variation.

View this table:
Table 3

Analysis of throat/blood pairs from the local collection

StrainSouthern Blot And Sequencing ResultsPredicted ITActual IT
MPJ15I blood√5×√11L2, L3, L4, L8L2L2
MPJ15I throat√5×√11L2, L3, L4, L8L2L3
MPJ33I blood√5×(√12)L2, L3, L4, L8L3/L4L3
MPJ33I throat√5×(√12)L2, L3, L4, L8L3/L4L3
MPJ35I blood(√9)√13×L1, L3, L8L1L1, L8
MPJ35I throat(√9)√13×L1, L3, L8L1L1, L8
MPJ36I blood√5(√11)(√10)L1, L2, L3, L4, L8L3/L4L4
MPJ36I throat√5(√11)(√10)L1, L2, L3, L4, L8L3/L4L4
MPJ42I blood√5√10(√12)L1, L2, L3, L4, L8L1/L3/L4L2, L9
MPJ42I throat√5(√11)(√12)L1, L2, L3, L4, L8L3/L4L2, L9
MPJ501 blood√8(√9)×L1, L3, L8L3/L4L3
MPJ501 throat√8(√9)×L1, L3, L8L3/L4L3
  • √ and × indicate the presence or absence of a gene. The numbers indicate the length of any associated homopolymeric tract — those in parentheses are predicted to cause the gene to be switched off.

3.3.2 Phenotypic analysis

LOS preparations from these isolates were analysed by Tris–tricine gel electrophoresis and silver staining. For each pair, the LOS band migrated equally for both the throat and the blood isolate (for example see Fig. 2). These strains were further investigated using conventional immunotyping. In four of the six pairs examined the immunotyping results agreed with the predicted phenotype. In the case of pair 42 a genotypic change was phenotypically silent (see above) and in pair 15 there was a phenotypic change between the pairs that had identical genotypes in the genes examined (Table 3).

Figure 2

Tris–tricine SDS–PAGE gel showing equal migration of LOS band from each member of a throat and blood pair. Lanes 1–4: LOS size ladder derived from lgt mutants of MC58¢3; lane 1: lgtE; lane 2: lgtA; lane 3: lgtB; lane 4: MC58¢3; lanes 5 and 6: LOS bands from throat and blood isolate pair from patient sample 33, demonstrating equal migration.

4 Discussion

Phase variation of LOS gene expression and the phase-variation repertoire of individual isolates has been investigated in laboratory strains of N. meningitidis, but its precise role in virulence and the disease process has not been defined. One reason for this is that isolates from multiple sites within individual patients, that could be used to examine LOS variation in translocation from one host compartment to another, are difficult to acquire. In part this reflects an increasing emphasis on pre-admission administration of antibiotics, which although reducing the rate of bacterial isolation has been shown to save lives. Moreover, in the acute clinical situation accurate diagnosis and treatment take priority over epidemiological investigations, so in many cases throat swabs are not taken, or are taken following antibiotic exposure. Finally, in the face of positive cultures from sites such as blood or CSF, many routine diagnostic laboratories would not pursue or retain associated carriage isolates. The local collection described here therefore provided a rare opportunity to investigate such sets of isolates.

Our survey of a global collection of clinical strains demonstrated the wide LOS repertoire that is available to the meningococcus. A good correlation between predicted and experimentally determined IT was observed, with >94% of samples having a phenotypic IT consistent with the genotype. The phase-variable nature of the biosynthesis of the LOS molecule will obviously produce a mixed population of organisms with respect to genotype and phenotype. In many cases, the antibody reactivity indicates that more than one IT is being produced. The strains of the global collection more obviously express heterogeneous LOS (multiple ITs) compared to the recently isolated strains of the local collection. This may be a function of the time and number of sub-cultures since original isolation from the selective environment of the human host. These results also suggest that sequencing of homopolymeric tracts from a population of organisms is less sensitive than antibody reaction in showing actual IT expression. This once again emphasises the importance of both genotypic analysis to determine potential LPS structures, and conventional phenotypic immunotyping to determine the structures actually expressed by the organism.

Our analysis of the strains of the global collection indicated that all but one strain possessed each of lgtA, lgtB and lgtE, with lgtA usually switched ‘on’, consistent with expression of the terminal LNT structure. In keeping with the fact that these strains were isolated from patients with invasive disease, the serum-resistant L3 IT predominated. The lgtC and lgtG genes were less frequently found. There was no obvious association between genotype, IT, phase-variation repertoire and hypervirulent lineages.

Our investigations in locally derived strains extended these observations and confirmed their applicability to this population. Moreover, the availability of nasopharyngeal and blood isolate pairs from six individuals with invasive disease allowed us to investigate directly the role of phase variation between these two compartments (nasopharynx and blood). lgtA was present in all six pairs and all cases had a low number of G residues not expected to phase vary at a high frequency (four strains had five G resides, one strain had eight G resides and one strain had nine G resides). Conventional immunotyping confirmed that the LNT structure was expressed by both blood and nasopharyngeal isolates except in pair 35 in which lgtA was off in both sites. In pair 42 the homopolymeric tract contained 11 bases in the throat isolate (lgtC off) and 10 in the blood isolate (lgtC on). However, the two isolates were immunologically identical, and we hypothesise that the activity of lgtA in the blood isolate overrode that of lgtC such that the phase-variation event was rendered phenotypically silent. A similar phenomenon has been observed in N. gonorrhoeae strain F62, in which the activity of lgtC is apparent only if lgtA is switched off [28]. Finally, lgtG was present in four pairs with no evidence of genotypic phase variation between them, but pair 15 differed in phenotype, with the blood isolate expressing L2, and the throat isolate expressing L3. This phenotype may be explained by variation in the presence and position of the PEA residue.

Previous work is consistent with a switch from non-sialylated LOS in the nasopharynx as a colonising phenotype (e.g. L8 IT) to a sialylated, serum-resistant LOS in the blood (e.g. L3 IT) [1,21,22]. In vitro studies have provided a rationale to explain this observation in which strains expressing non-sialylated LOS are more adherent and invasive than strains expressing sialylated, serum-resistant LOS in meningococcus [11,19] and also in gonococcus [20]. In this study we have revealed that half of the strains examined have a lgtA gene with only five G resides, therefore phase variation from LNT+(sialylated) to LNT- (non-sialylated) LOS is unlikely. From these observations we concluded that phase variation, altering LOS sialylation, is not a prerequisite for meningococcal translocation in the host. Alternatively, in strains with lgtA genes that are unlikely to phase vary, the alteration in LOS sialylation, facilitating invasion, may be operating at the level of gene regulation rather than selection of a random event (phase variation).


This work was funded by grants from the Meningitis Research Foundation (UK) and a Project Grant from the NHMRC (9938007). The authors thank the following people for provision of strains in the Australian sample clinical collection: Narelle George and Joan Faoagali, Royal Brisbane Hospital; Robin Kelly and David Alfredson, Gold Coast Hospital; and Martin Tlise and Theo Mollee Mater Hospital, Brisbane.


  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].
View Abstract