Hypertension resistance polymorphisms in ROMK (Kir1.1) alter channel function by different mechanisms

Liang Fang, Dimin Li, Paul A. Welling


The renal outer medullary K+ (ROMK) channel plays a critical role in renal sodium handling. Recent genome sequencing efforts in the Framingham Heart Study offspring cohort (Ji W, Foo JN, O'Roak BJ, Zhao H, Larson MG, Simon DB, Newton-Cheh C, State MW, Levy D, and Lifton RP. Nat Genet 40: 592–599, 2008) recently revealed an association between suspected loss-of-function polymorphisms in the ROMK channel and resistance to hypertension, suggesting that ROMK activity may also be a determinant of blood pressure control in the general population. Here we examine whether these sequence variants do, in fact, alter ROMK channel function and explore the mechanisms. As assessed by two-microelectrode voltage clamp in Xenopus oocytes, 3/5 of the variants (R193P, H251Y, and T313FS) displayed an almost complete attenuation of whole cell ROMK channel activity. Surface antibody binding measurements of external epitope-tagged channels and analysis of glycosylation-state maturation revealed that these variants prevent channel expression at the plasmalemma, likely as a consequence of retention in the endoplasmic reticulum. The other variants (P166S, R169H) had no obvious effects on the basal channel activity or surface expression but, instead, conferred a gain in regulated-inhibitory gating. As assessed in giant excised patch-clamp studies, apparent phosphotidylinositol 4,5-bisphosphate (PIP2) binding affinity of the variants was reduced, causing channels to be more susceptible to inhibition upon PIP2 depletion. Unlike the protein product of the major ROMK allele, these two variants are sensitive to the inhibitory affects of a G protein-coupled receptor, which stimulates PIP2 hydrolysis. In summary, we have found that hypertension resistance sequence variants inhibit ROMK channel function by different mechanisms, providing new insights into the role of the channel in the maintenance of blood pressure.

  • inward rectifier
  • potassium channel
  • blood pressure
  • inward rectified potassium channel 1.1
  • Bartter
  • phosphotidylinositol 4,5-bisphosphate
  • misfolded protein

in addition to its critical function in potassium homeostasis, the renal outer medullary K+ (ROMK) channel (also known as Kir1.1, product of the KCNJ1 gene) is well appreciated to play a critical role in renal sodium handling (23). As the major pathway for potassium exit into the thick ascending limb (TAL) lumen (13), ROMK channels supply a sufficient amount of K+ to the apical Na+-K+-2Cl cotransporter (NKCC2) to maintain the reabsorptive transport of NaCl in this segment. In fact, because the reabsorptive salt transport through NKCC2 is rate-limited by the availability of luminal potassium in the TAL (6), changes in the activity of ROMK can have dramatic effects on renal salt reabsorption. A human disease underscores the importance of this process and its role in blood pressure. Indeed, loss-of-function mutations in ROMK cause Bartter's syndrome (20), a relatively rare recessive nephropathy characterized by a loss of sodium reabsorption in the TAL, severe salt wasting, and hypotension.

Recent intriguing genetic and epidemiological studies raise the possibility that more prevalent loss-of-function mutants in ROMK may also play a role in health. Indeed, large-scale sequencing efforts in the Framingham Heart Study linked heterozygous sequence variants in ROMK with a significant reduction in blood pressure and protection from hypertension (9). While strongly suggestive, the consequences of these sequence variants on the ROMK function have, thus far, only been inferred. It has been reasonably suggested that the variants might impair ROMK function because they affect phylogenetically conserved residues, but, with the exception of one variant, which was previously linked to Bartter's syndrome (18), direct experimental evidence has been wanting. Furthermore, the mechanisms underlying altered channel function remain completely unknown. In the present study, we explore these unresolved issues.


Molecular biology.

All studies were performed with ROMK cDNA, containing an external hemagglutinin (HA) epitope tag as described before (27). Site-directed mutagenesis was performed using a PCR-based strategy with PfuTubo DNA polymerase (QuikChange; Stratagene). The sequence of all modified cDNAs was confirmed by dye termination DNA sequencing (University of Maryland School of Medicine Biopolymer Core). All constructs used for studies in Xenopus oocytes were subcloned between the 5′- and 3′-untranslated region (UTR) of the Xenopus globin gene in the modified pSD64 vector for optimal translation. This vector also contains a polyadenylate sequence in the 3′-UTR (dA23dC30). M1 receptors were expressed in Xenopus oocytes as previously described (14). Oocytes were coinjected with M1 and channel cRNA at a 5:1 ratio.

cRNA synthesis.

Complementary RNA was transcribed in vitro in the presence of capping analog from linearized plasmids containing the cDNA of interest using SP6 RNA polymerase (mMessage Machine; Ambion). cRNA was purified by spin column chromatography (MEGAclear; Ambion). Yield was quantified spectrophotometrically and confirmed by agarose gel electrophoresis.

Xenopus oocyte isolation and injection.

Xenopus laevis (Xenopus Express, Homosassa, FL) oocytes were isolated using a protocol approved by the Institutional Animal Care and Use Committee at the University of Maryland Medical School as described previously (4). Oocyte aggregates were dissected from the ovarian lobes and then incubated in OR-2 medium (in mM: 82.5 NaCl, 2 KCl, 1 MgCl2, and 5 HEPES, pH 7.5) containing collagenase (type 3; Worthington) for 2 h at room temperature. Oocytes were stored at 19°C in OR-3 medium (50% Leibovitz's medium, 10 mM HEPES, pH 7.4). Later (12–24 h), healthy-looking Dumont stage V-VI oocytes were injected with 50 nl of DEPC-treated water containing 250 pg of ROMK cRNA and were then incubated in OR-3 medium at 19°C. Experiments were performed 3 days after injection.


Whole cell currents in Xenopus oocytes were monitored using a two-microelectrode voltage clamp (OC-725; Warner). Voltage-sensing and current-injecting microelectrodes had resistances of 0.5–1.5 MΩ when backfilled with 3 M KCl. Data were collected using an ITC16 analog-to-digital, digital-to-analog converter (Instrutech), filtered at 1 kHz, and digitized on line at 2 kHz using Pulse software (HEKA Electronik) for later analysis. Once a stable membrane potential was attained, oocytes were clamped to a holding potential at the predicted potassium equilibrium potential (i.e., near zero current value), and currents were recorded during 500-ms voltage steps, ranging from −100 to +40 mV in 20-mV increments. ROMK potassium currents are taken as the barium-sensitive current (1 mM barium acetate). For initial functional screens, oocytes were bathed in a 90 mM KCl solution (in mM: 90 KCl, 1 MgCl2, 1 CaCl2, and 5 HEPES, pH 7.4), and inward currents at −100 mV are reported. To study M1 receptor-dependent regulation of ROMK, outward potassium currents (at 0 mV) were measured under more physiological potassium (1 mM) concentrations (in mM: 1 KCl, 89 N-methyl-d-glucamine gluconate, 1 MgCl2, 1 CaCl2, and 5 HEPES, pH 7.4). pH-dependent gating was accessed from whole cell inward potassium currents in oocytes perfused with a series of different pH solutions containing butyrate, following the same methods, analysis, and pHi measurements as before (5).

For the giant excised patch-clamp experiments, the vitelline membrane was removed as described before (4) to gain access to the oocyte plasma membrane with large, fire-polished electrodes (0.4–0.9 MΩ, backfilled with 96 mM KCl, 1 mM MgCl2, and 5 mM HEPES buffer, adjusted to pH 7.4 with KOH/HCl). Electrodes were made from borosilicate glass capillaries (World Precision Instruments) using a Narishige PP-83 puller. Oocytes were continuously perfused with “FVPP” buffer (in mM: 96 KCl, 5 EDTA, 10 HEPES, 5 NaF, 3 Na3VO4, and 10 Na4O7P2 and were adjusted to pH 7.4 with NaOH/HCl), which prevents spontaneous inward-rectifier potassium (Kir) channel rundown upon excision (8), in a small (100 μl)-volume chamber equipped with a rapid solution exchange apparatus (Bioscience Tools; PC-15). Inward macroscopic potassium currents were recorded in the inside-out configuration at membrane potential of −100 mV before and after abruptly adding polylysine (300 μg/ml; Sigma-Aldrich). Currents were acquired and digitized at 1 KHz from the Axopatch 200B patch-clamp amplifier using the ITC-16 computer interface (Instrutech, Long Island, NY) and HEKA Pulse software (version 8.31) for latter analysis (IGOR software; Wave Metrics).

Surface expression.

Plasmalemma expression of the external HA-tagged ROMK channel was measured in single oocytes following procedures described before (27). Briefly, oocytes were washed in cold OR-2 medium, fixed with 4% formaldehyde (15 min at 4°C), washed in OR-2, and then incubated for 1 h at 4°C in OR-2 containing 1% BSA. Exposed HA epitopes on the surface of intact oocytes were labeled with a mouse monoclonal anti-HA antibody (Covance) for 1 h, and then oocytes were washed and incubated with horseradish peroxidase (HRP)-coupled goat anti-mouse (30 min). Individual oocytes were placed in 50 μl of Enhanced Chemiluminescence Substrate (Amersham) and incubated for 5 min at room temperature. Luminescence was measured from individual oocytes for 10 s in a Sirius luminometer and reported as relative light units per second.

Western blot analysis.

Oocytes (∼100/reaction) were washed in homogenization buffer (in mM: 80 sucrose, 5 MgCl2, 5 NaH2PO4, 1 EDTA, and 20 Tris, pH 7.4) containing a protease inhibitor cocktail (5 μg/ml leupeptin, 1 mM phenylmethylsulfonyl fluoride, and 5 μg/ml pepstatin A) and then broken by trituration with a 25-gauge needle. After low-speed centrifugation (100 g) for 10 min, supernates were spun at 14,000 g (20 min, 4°C) to collect the total membrane fraction. Pellets were washed in the homogenization buffer, spun again (10 min), placed in solubilization buffer (4% sodium deoxycholate, 20 mM Tris, pH 8.0, 5 mM EDTA, and 10% glycerol, containing the protease inhibitors), and rocked for 2 h (4°C). Particulate material was pelleted (14,000 g, 20 min, 4°C), and the solublized proteins (∼10 μg/lane) were resolved by SDS-PAGE and transferred to nitrocellulose membranes.

Blots were blocked in TBS with Tween 20 (0.1%) (TBS-T) containing 5% nonfat dry milk (NFDM) for 45 min at room temperature. A mouse monoclonal anti-HA antibody (Covance) was diluted in 5% NFDM to 1:1,000 and incubated at 4°C overnight, washed for 10 min in TBS-T three times, incubated with goat anti-mouse antibody coupled to HRP (1:5,000, 5% NFDM), and washed extensively for 10 min in TBS-T three times. Bound antibodies were then revealed using enhanced chemiluminescence reagent (Pierce) and fluorography (Kodak).

The glycosylation state of ROMK was assessed by glycosidase digestion. Solubilized membrane protein (∼40 μg protein/15 μl reaction volume) was incubated in glycoprotein denaturing buffer (New England Biolabs) at 65°C for 10 min and then for 1 h at 37°C with either N-glycosidase F (PNGaseF, 2,000 units; NEB) or endoglycosidase H (Endo H, 2,000 units; NEB) in a 20-μl reaction containing 1× of the corresponding enzyme buffer as recommended by the manufacturer. Lysates treated without enzyme served as the control. Western blot analysis was performed as described above.

Statistical analysis.

Data are presented as means ± SE. Statistical analysis was performed using GraphPad PRISM version 5. Statistical significance was determined by t-test when comparing two groups and by one-way randomized ANOVA followed by Bonferroni's post hoc test when comparing multiple groups or Dunnett's post hoc test when multiple test groups were compared with the control. P < 0.05 was considered significant.


To begin to clarify the mechanisms by which all reported hypertension resistance (HR) sequence variants (P166S, R169H, R193P, H251Y, and T313FrameShift, ROMK2 amino acid numbering) in ROMK might affect blood pressure, we characterized the functional properties and surface expression of channels bearing each of the polymorphisms. An HA epitope tag was incorporated into an external site that does not perturb channel activity for quantitative measurements of plasmalemma expression (27). Biophysical properties of the channels were assessed in Xenopus oocytes by two-electrode voltage clamp, under conditions in which wild-type ROMK K+ current is easily detected and robust (Fig. 1A). In contrast to the wild-type channel (wild type is used throughout the manuscript to refer to the protein product of the major ROMK allele), three of the mutant channels (R193P, H251Y, and T313FrameShift) exhibited no whole cell potassium current above background. As measured by HA antibody binding and analytical luminometry, the R193P, H251Y, and T313FS mutations also completely inhibited surface expression, indicative of a trafficking and/or misfolding defect (Fig. 1B). By contrast, the other two mutant channels (P166S, R169H) displayed no difference in current and a slight increase in surface expression compared with the wild-type channel. The increase in number of channels at the cell surface without a concomitant increase in whole cell K+ current may be consistent with subtle alterations in channel gating or single channel conductance.

Fig. 1.

Effects of hypertension resistance (HR) polymorphisms on renal outer medullary K+ (ROMK) channel activity and surface expression. A: whole cell potassium currents were measured by two-microelectrode voltage clamp in Xenopus oocytes following the injection of cRNA encoding wild-type (WT) or HR variant ROMK (P166S, R169H, R193P, H251Y, and T313FS), or water (control). Summary of average inward potassium currents ± SE (at −100 mV, 90 mM external potassium, n = 15 oocytes from 4 frogs, P < 0.05) B: cell surface expression of external HA epitope-tagged channels, assessed by hemagglutinin (HA) antibody binding and luminometery, is indicated in relative light units (RLU) (n = 15, #P < 0.01, *P < 0.05).

R193P, H251Y, and T313FS affect channel trafficking in the biosynthetic pathway.

To determine how trafficking of the R193P, H251Y, and T313FS channel might be blocked, an analysis of glycosylation state maturation was conducted, taking advantage of the progressive modification of N-linked oligosaccharide chains on proteoglycans as they pass from the endoplasmic reticulum (ER) through the Golgi. ROMK channels acquire N-linked glycosylation at a single extracellular asparagine residue and then are further processed to a larger Endo H resistant form, indicative of maturation beyond the medial Golgi (2, 19, 25). Consistent with this, and as shown in Fig. 2, the wild-type ROMK migrates as the following three protein species: a 65-kDa PNGase F-sensitive, Endo H-insensitive band; a 44-kDa PNGase F- and Endo H-sensitive band; and a lower band (42 kDa) that is not altered by either PNGase F or Endo H. Channels bearing the P166S and R169H mutations displayed the same pattern as the wide type, corroborating the surface expression studies. On the other hand, channels containing the R193P, H251Y, or T313FS mutations did not exhibit mature glycosylation. In fact, with channels bearing either R193P or H251Y variants, the absence of mature glycosylation was largely offset by an increase in the core glycosylated species (e.g., Endo H sensitive), indicative of a trafficking arrest somewhere between ER and medial Golgi. The T313FS mutant was not glycosylated at all. Taken together, these observations strongly suggest that the R193P, H251Y, and T313FS mutations disrupt trafficking at an early step in the secretory pathway, and thereby prevent plasma membrane delivery.

Fig. 2.

Effects of HR polymorphisms on ROMK glycosylation state maturation. Western blot analysis (anti-HA) of indicated HA-tagged ROMKs expressed in oocytes and resolved by SDS-PAGE following the treatment with no glycosidase (A) or endoglycosidase H (Endo H, B) or N-glycosidase F (PNGase F, C). C: analysis of glycosylation state maturation. Relative contributions of each band, as determined by densitometry, are shown. IB, immunoblot. Data are means ± SE (n = 5). #P < 0.01 and *P < 0.05 compared with control species.

P166S and R169H affect phosphotidylinositol 4,5-bisphosphate-dependent gating.

ROMK requires the binding of phosphotidylinositol 4,5-bisphosphate (PIP2) to stay open (8), and, significantly, P166 and R169 are located within the cytoplasmic domain near the putative PIP2-binding surface (23). To explore whether P166S and R169H alter the PIP2-dependent gating process, we employed a well-established method (8, 12, 17), measuring the decline in channel activity in excised patches upon exposure to polylysine, a positively charged reagent that competes with the channel for PIP2. As shown in Fig. 3, the wild-type channel exhibited a slow decrease in channel activity, consistent with previous reports that PIP2 binds tightly to ROMK and disassociates slowly (17). By contrast, the mutants displayed a very rapid decrease in channel activity. In fact, the half current decay time of these HR mutant channels was at least an order of magnitude faster than the wild type (Fig. 3A). By contrast, as measured by the apparent pKa of gating (Fig. 3B), neither mutation changed how the channel is regulated by intracellular pH, a process that occurs by a separate but interrelated mechanism as PIP2-dependent gating (11). Taken together, the observations are indicative of a dramatic reduction in the apparent PIP2 binding affinity.

Fig. 3.

The HR polymorphisms P166S and R169H affect phosphotidylinositol 4,5-bisphosphate (PIP2)-dependent gating. A: potassium currents in giant excised patches from oocytes injected with ROMK (WT, P166S, and R169H) cRNA upon exposure to polylysine, a positively charged reagent that competes with the channel for PIP2. Left, representative tracings of the WT channel compared with the HR variants P166S and R169H. Right, T50 is the time required for polylysine to inhibit ROMK currents to half of the initial amplitudes (n = 10 P < 0.05). B: mutants exhibit no alterations in pH sensitivity; shown are pHi titration curves of ROMK inward potassium currents, I, relative to basal current, Imax, in Xenopus oocytes (WT, pKa = 6.68; P166S pKa = 6.71; R169H pKa = 6.64; each data point is the average ± SE of n = 9 oocytes from 4 frogs). C: cell potassium currents in oocytes injected with the M1 receptor and the indicated ROMK channel before (open bars) and after (filled bars) addition of ACh (25 μM) (*P < 0.05, n = 12 oocytes from 4 frogs). PLC, phospholipase C; DG, diacylglycerol; IP3, inositol trisphosphate.

Normally, ROMK gating is relatively resistant to physiological changes in PIP2 levels because of the high affinity of PIP2 binding (17). To test whether P166S and R169H mutations might decrease PIP2 affinity enough to confer a new physiological modality of negative regulation, we tested whether these channels acquire sensitivity to G protein-coupled receptor (GPCR)-mediated changes in PIP2 levels (Fig. 3C). For these studies, we coexpressed ROMK with the M1 muscarinic receptor. This GPCR activates phospholipase C (PLC), induces PIP2 hydrolysis, and has been used widely to study PIP2 regulation of Kir channels (1, 12, 14). Outward potassium currents were measured under physiological conditions in intact oocytes before and after addition of ligand (25 μM ACh). In contrast to wild-type ROMK, which is not affected by ACh, potassium currents carried by ROMK P166S or R169H rapidly declined following application of ligand. Thus these HR variant channels acquire sensitivity to M1 receptor activation, suggesting they may be inhibited by physiological attenuation of PIP2.


In this study, we found the reported HR polymorphisms in ROMK (9) do, in fact, cause a loss-of-channel function. Two different underlying molecular defects were observed. One group of variants, R193P, H251Y, and T313FS, blocks cell surface localization and alters glycosylation state maturation, consistent with a trafficking defect at an early step in the biosynthetic pathway. By contrast, the P166S and R169H HR variants produce a more subtle defect, altering PIP2-dependent gating such that channels become susceptible to inhibition upon PIP2 depletion.

Mistrafficking of R193P, H251Y, and T313FS.

The failure of the R193P and H251Y to efficiently reach the cell surface is likely to be the result of misfolding. A number of trafficking signals have been identified in ROMK that control channel delivery to (15, 25, 26) and from (3) the cell surface, but R193P and H251Y are not located near any of them. Instead, structural homology modeling of ROMK (23) predicts that R193 and H251 are deeply buried at separate sites within the cytoplasmic domain and not readily accessible to the solvent-exposed surface. Thus it seems probable that these highly conserved residues are important for determining the general structural integrity of the channel rather than being essential residues in trafficking signals. The finding that R193P and H251Y channels are only core glycosylated indicates that the block in trafficking occurs at an early step in the biosynthetic pathway. Although further studies are required to rigorously examine the biochemical mechanism, the biosynthetic trafficking arrest is most likely a consequence of the quality control process in the ER that prevents forward traffic of misfolded proteins through the Golgi, where Endo H-sensitive core oligosaccharides are normally processed into complex oligosaccharides.

The T313 frame shift variant, originally indentified as a Bartter's syndrome mutation (20), shares similarities and differences with R193P and H251Y. Early characterization of ROMK T313FS channels by immunocytochemistry in insect cells (Sf9) (18) strongly suggested that the mutation induces a trafficking defect. Our studies corroborate and extend this observation. By directly measuring channels at the plasmalemma, we confirmed that T313FS blocks cell surface expression entirely. Unlike R193P and H251Y, we found that the frame-shift mutation also completely blocks glycosylation. Because T313FS affects residues in the cytoplasmic COOH-terminus, far from the sole NH2-terminal glycoslyation site, the complete lack of glycosylation provides evidence that the mutation has global effects. Perhaps, this shouldn't be too surprising since the frame shift replaces the last 60-amino acid COOH-terminus with a completely novel stretch of 35 amino acids. Because truncation of the COOH-terminal region does not alter surface expression (4), it seems likely that the addition of the new sequence, rather than removal of the wild-type sequence, is responsible. Structural homology modeling of ROMK indicates that part of the affected wild-type COOH-terminal region forms a key contact with the cytoplasmic NH2-terminus. Deleterious effects of packing the frame-shift structure against the cytoplasmic NH2-terminus may, thus, be transmitted far into the channel architecture.

Conditional loss-of-function in P166S and R169H.

In contrast to the dramatic effects of the mistrafficking mutations, P166S and R169H affect ROMK function in a subtle manner. Both mutations impinge on the putative PIP2-binding site and alter PIP2-dependent gating. Although PIP2-dependent regulation of ROMK shares overlapping mechanisms with the pKa- and pH-dependent gating processes, the mutations do not alter pKa of gating and do not affect residues in the phosphorylation motifs, making it highly probable that they disrupt the PIP2 gating process directly, rather than indirectly. In fact, they make the channel become susceptible to inhibition upon depletion of PIP2. Unlike the wild-type channel, the P166S and R169H variants are sensitive to the inhibitory affects of a GPCR, which stimulates PIP2 hydrolysis through activation of PLC. While further studies are required to make definitive conclusions about the exact signaling mechanism, the response likely reflects reduced PIP2 affinity of the mutant channels and inhibition by physiological attenuation of PIP2 levels.

Such a gain-of-inhibitory regulation may be particularly important for ROMK channels in the TAL, where ROMK activity and salt transport may be controlled by GPCR-mediated signaling pathways. Different GPCR, including the M1 receptor, have been implicated. While it remains to be determined which receptors, if any, are involved, a role of the calcium-sensing receptor (CaSR) should be considered (7). Activation of CaSR suppresses salt reabsorption in the TAL (16), and this has been suggested to occur, at least in part, through inhibition of ROMK (16). Cytochrome P-450 (21) and phospholipase A2-dependent signaling (22) pathways have been implicated in communicating supraphysiological activation of CaSR to inhibition of wild-type ROMK. Because CaSR can also activate PLC, and suppress PIP2 levels (10), the P166S and R169H ROMK variants are predicted to be more sensitive to physiological stimulation of the receptor than the wild-type channel. Obviously, further studies are required to test these ideas. In an age of personalized medicine, it will be interesting to learn if subjects with these particular HR variants exhibit especially strong blood pressure-lowering responses to dietary calcium supplementation or CaSR activators.

It should be pointed out that the P166S and R169H mutants are surprisingly more robustly expressed at the cell surface than the wild-type channel. While further studies are required to elucidate the mechanistic basis, the glycosylation status provides a potential clue. Because the mutations do not reduce the amount of unglycosylated ROMK, as might be expected if they increased forward trafficking through the biosynthetic pathway, one would suspect that these two variants may, instead, slow endocytosis. In this regard, it is interesting that ROMK endocytosis is governed by binding to a clathrin adaptor, autosomal recessive hypercholestrolemia (3), which also interacts with PIP2. Perhaps, endocytosis is favored when cargo and endocytotic adaptor share a common affinity for the same PIP2-rich microdomains.

In summary, hypertension-resistant polymorphisms in ROMK inhibit channel function, reinforcing the important involvement of these channels in renal salt absorption and maintenance of blood pressure.


These studies were funded by grant support from the National Institute of Diabetes and Digestive and Kidney Diseases to P. A. Welling (DK-54281 and DK-63049).


No conflicts of interest are declared by the authors.


  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. 25.
  25. 26.
  26. 27.
View Abstract