In vitro and in vivo models for the study of oral delivery of nanoparticles ^☆

5.1. Simulated gastric fluids

The GI tract presents a unique microenvironment of enzymes and ionic strength, which impact both the chemical and colloidal stability of nanoparticles. The typical PBS used to conduct in vitro studies for oral delivery may therefore be inadequate. Simulated gastric and intestinal fluids provide a medium that is representative of the fluid found in the stomach and intestine and thus are better suited to study and analyze the release profiles of controlled release nanoparticles. In the in vitro fluid model, simulated gastric fluid medium at a pH of 1.2 and simulated intestine fluid medium at a pH of 6.8 without enzymes are prepared according to some standard, often times the USP XXVI recommendations or the British Pharmacopoeia [41] with controlled release experiments being performed in a test tube [42,43]. These experiments are typically performed at 37 °C and some are done with agitation (100 rpm) [44] to model the in vivo movement of the gastric fluids. Samples of the medium are withdrawn from the test tube at desired intervals, the nanoparticles are removed by centrifugation, and the amount of drug or protein remaining in the supernatant is analyzed [42].

Simulated intestinal fluids can also be altered to create intestinal fluid models to be used in cell culture, specifically with CaCo-2 cells (cell line details in 5.3). Several studies show the viability of the CaCo-2 monolayer when incubated with modified simulated intestinal fluid [45,46]. These solutions are divided into two categories including the fasted state (FaSSIF) and the fed state (FeSSIF). FeSSIF has approximately 2–5 times greater amounts of bile salts than in the FaSSIF [45]. Notably, the SIF for cell culture does not contain acetic acid, potassium phosphate dibasic, or any of the enzymes given in the British Pharmacopoeia including pancreatin. In one study, taurocholate is identified as the least toxic bile salt for the FaSSIF and the FeSSIF [45], which allows for higher CaCo-2 cell viability. The specially modified SIF has an osmolality similar to that of human cells (275–299 mOsm) at 336–343 mOsm, thus preventing cell swelling or shrinking and making this fluid a reasonable model for transport across CaCo-2 monolayers in intestinal fluid.

5.2. Drug release studies using a membrane

The dialysis membrane model is used for in vitro analysis in experiments that aim to determine peptide and drug transport properties. In this technique, nanoparticle dispersions are added to dialysis bags or a dialysis tube [6] and suspended in the desired buffer. For example, in the case of estradiol release from PLGA nanoparticles, ionic strength phosphate buffer at pH 7.4 is used as the buffer. To simulate GI conditions, the membrane system is maintained at 37 °C and movement is facilitated by shaking at 100 rpm [47,48]. The model can be made even more accurate in representing gastric and intestinal fluids with the use of low pH buffers that model the acidic conditions of the gut. To analyze the drug release, aliquots of the exterior membrane buffer are removed at pre-determined time intervals for measure of the drug content by methods such as high-performance liquid chromatography (HPLC) [47,49].

5.3. CaCo-2 monolayer

The human epithelial colorectal adenocarcinoma cell line, CaCo-2 is typically used to represent the predominant cell type in the gut, the gut epithelial cells [50]. CaCo-2 monolayers can either be grown in a monolayer on a single cell culture well plate or in a Transwell ® setup to model the transport of drug or passage of nanoparticles across the epithelial layer. With the CaCo-2 monolayer grown on a single well cell culture plate, nanoparticle uptake by the cells or transfection of the CaCo-2 cell line can be determined through various methods [34,51]. In the Transwell ® model, the CaCo-2 monolayer is grown on an inserted porous membrane and a different cell line can be seeded on the bottom chamber of the plate. For gene delivery studies, the model works by placing the nanoparticle dispersion on the monolayer on the membrane while nourishing the cells beneath the membrane with nanoparticle-free medium. Transfection of the cells seeded in the bottom well indicates that the nanoparticles have effectively transported through the CaCo-2 monolayer. The CaCo-2 monolayer is typically used between day 15 and day 25 after initial seeding [44,52]. The expected TEER, transepithelial electrical resistance, is a measurement of the confluence and viability of the monolayer. An appropriate TEER value for the experimental period is between 260 and 450 Ω cm²[53]; a TEER of 300 Ω cm² is representative of that found in the colon while the stripped large intestine has a TEER ranging from 100 to 275 Ω cm²[55]. Typically, a higher TEER value is representative of a more compact cell layer while a decrease in TEER causes greater cell-layer permeability [56]. TEER is an important parameter characterizing the integrity of the cell layer [54] or the tissue [55].

Although this system is often used to model intestinal transport, one disadvantage of the CaCo-2 monolayer is its impermeability to hydrophilic or paracellular transport; therefore making it a better model for colonic tissue and restricting the reliability of the technique as a model for the small intestinal tissue [50]. However, it is not a good model for transport and absorption across the jejunum. Another disadvantage of the CaCo-2 monolayer is that it only models epithelial cells in the intestinal epithelium, however, there are many other cell types including mucosal cells and M cells. Cocultures are therefore an addition to the CaCo-2 monolayer culture and they are used to represent the multicellular intestinal epithelium.

5.4. CaCo-2 and HT29 coculture

Mucus is significant in limiting transport from the gut fluids to the epithelial cells. The potential trapping of the nanoparticles by mucus lining of the epithelial layer also has an effect on the retention time of the nanoparticles in the system. The coculture of CaCo-2 and HT29, a human adenocarcinoma cell line, provides a model for the second major cell type in the gut epithelium, the mucus-secreting goblet cells (Fig. 1). HT29 cells differentiate into mature goblet cells under the influence of methotrexate (MTX) making this cell line important in the mucus layer formation. The coculture of CaCo-2 and HT29-MTX is typically optimized by mixing the two cell lines in appropriate ratio and plating them on the porous membrane of the Transwell ®. The viability of the CaCo-2 monolayer with the mucus secretion is determined the same way as CaCo-2 monolayer alone, using TEER [50]. The expected TEER values of the coculture decrease as the percentage of HT29-MTX increases; for example TEER ranges from 0 to ∼790 Ω cm² from time 0 to 23 days for 25% HT29-MTX, while it ranges from 0 to ∼310 Ω cm² for 75% HT29-MTX [57]. An alternative way to determine the viability of the CaCo-2 and HT29-MTX coculture is demonstrated by Mahler et al. and involves mucus layer characterization. The cocultures are seeded on the Transwell ® membrane inserts and after 16 days a slice of the insert is cut out and stained with 0.25% toluidine blue for microscopic examination [58].

In one experiment, HT29 sub-clones are tested for preparation of the most uniform mucus monolayer. With some HT29 sub-clones, the problem of multiple layer cell growth results in only the top layer of HT29 maturing into goblet-secreting cells, leaving the CaCo-2 and HT29 as multicellular layers below and limiting the accuracy of the model. However, the HT29-D1 and HT29-E12 clones, both stimulated by MTX, were found to completely mature into goblet cells and form a uniform mucus layer with a CaCo-2 monolayer beneath it [50]. Different sub-clones exhibit varying depths of the mucus layer and therefore they can be used to model different segments of the GI tract. HT29-D1 develops into mucus layers of 53±52 μm and the HT29-MTX-E12 results in mucus layers of 142±51 μm [50]. One limitation to the CaCo-2 HT29-MTX model is that the active transport of drugs is underestimated by this model compared to the actual results in vivo[57].

5.5. Coculture for M cell formation

In the intestine, the lymphoid structures, such as the Peyer's Patch, are separated from the lumen by the follicle-associated epithelium (FAE), which contains enterocytes and M cells with few goblet cells. M cells have the capability of transporting particulate matter, including nanoparticles, through transcytosis from the gut contents to the underlying mucosal immune system [59,60].

The initial method for the formation of M cells in the Transwell ® system involved the use of lymphocytes taken from the Peyer's Patch of mice mixed with the CaCo-2 cell line [60,61]. In this system, the CaCo-2 monolayer was allowed to form and then the isolated lymphocytes were cultured on the basolateral side of the monolayer in the Transwell® plate. The Peyer's Patch lymphocytes converted the endothelial cells into M cells (Fig. 1); in the presence of the lymphocytes, the CaCo-2 acquired transcytotic activity and were able to transfer fluorescin isothiocyanate (FITC)-conjugated latex beads through the filter to the bottom well. Kernéis et. al. also proved the presence of M cells in the coculture by showing that the Transwell ® system could transport the Vibrio cholerae bacteria, which are uniquely taken-up and transported by M cells [61].

Since the aforementioned model lacked uniformity and did not use solely human cell lines, another method is described using human Burkitt's lymphoma Raji B cells. In this method, CaCo-2 cells are cultured for 14 days, then Raji B are added to the bottom chamber, and the cocultures are maintained for 4–6 days. It is desirable that the epithelial layer lose its microvilli thus demonstrating the CaCo-2 cell line conversion into M cells; however, the loss of microvilli is shown to be nonuniform in this coculture method, suggesting that some of the epithelial cells remained unconverted [60]. While this nonuniform conversion may actually be beneficial in representing the distribution of M cells and enterocytes in the GI tract, the M cell formation occurs on both the apical and basolateral sides of the membrane insert, which is not representative of their presence in the in vivo model [60].

A more physiological, functional, and reproducible model to study the mechanisms of M cell transport is the inverted co-culture model. The inverted model encourages closer contact between the CaCo-2 cell monolayer and the Raji B cells or B lymphocytes. This method involves seeding the CaCo-2 cells on the membrane inserts, inverting the inserts 3–5 days later, placing a piece of silicon tube on the basolateral side of the insert, and adding Raji B cells to the basolateral side of the insert 9–11 days later. The cocultures are maintained for 5 days, the silicon tubing is removed, and the inserts are placed back into the original orientation [59]. The inverted FAE model appears to be the state-of-the-art, with further improvement to be expected with the incorporation of flow using a microfluidic approach.

5.6. Gut-on-a-chip

A new technique termed gut-on-a-chip provides an in vitro model that mimics the intestine's structure and transport properties while sustaining microbial flora. The microdevice is fabricated with upper and lower microchannels using polydimethylsiloxane (PDMS) with a 30 μm thick porous membrane separating them. Fig. 2 illustrates the microdevice design; vacuum sources are used to control the fluid flow through the microchannels and vacuum chambers on either side of the microdevice provide cyclic strain under computer control to mimic intestinal peristalsis. The microchannel is treated with an extracellular matrix solution containing rat type I collagen, and CaCo-2 cells are cultured on the ECM-coated porous microchannel. The CaCo-2 cells experiencing fluid flow in the microdevice form a monolayer in only 3 days, as compared to the 21 days that it takes in the static medium Transwell ® system. The fluid flow during CaCo-2 culture in monolayers allows for the cells to grow in columnar forms with microvilli folds, tight junctions, and crypt formation that mimic in vivo human intestinal epithelium. Introducing peristalsis to the system provides increased paracellular permeability while maintaining TEER, which is more comparable to the transport in in vivo human intestine. Because microbial flora is an important part of intestinal physiology, Lactobacillus rhamnosus GG, a bacterium originally isolated from the human gut, is cultured on top of the CaCo-2 monolayer and a better mimic of the human intestine. The gut-on-a-chip device is an interesting model that matches the physiology, morphology, transport, and pathophysiology characteristics of the human intestine by combining CaCo-2 cells and microbial flora, mechanical stimulants, and fluid flow onto a controllable microfluidic device [62].

6.1. Everted intestinal sac

Everted rat intestine sacs have been used to measure the mucoadhesive properties, transport and cellular uptake of some nanoparticle assemblies [6,28,63], including drugs and peptides conjugated with nanoparticles [64]. In one protocol, the frog intestine is removed from the dead animal and portions of the intestine are cut to test permeability in various segments of the intestine. The segments of intestine are silk sutured at one end, filled with nanoparticle solution, sutured at the open end, and immersed in Frog Ringer (FR) solution. Solution is then withdrawn from the exterior of the intestine segment and analyzed for the nanoparticle contents [42]. In an alternative method, a rat's intestine is removed, washed, segmented, and fixed onto glass cannulas with rubber rings [64]. A study of drug delivery using a lipid-based particle showed that ex vivo permeation studies failed to predict what occurred in vivo[65], presumably due to the differences in the ex vivo and in vivo extracellular environment as well as the difference in intestinal segment cells ex vivo, thus showing the potential inaccuracy of this model.

7.1. Intestinal loop model

In the intestinal loop model, the animal, typically a rat, rabbit [53], or mouse, is fasted for approximately 3 days and then put under anesthesia for the entirety of the surgical procedure. An incision is made in the midline of the abdomen to expose the small intestine. The desired segments are sliced, held with a clamp, and then washed to remove excess food particles; all of this is done while keeping the intestine attached and in close proximity to the body cavity. The segment is then ligated using suture to form a loop. Careful consideration is made to maintain the normal blood flow and muscle movement to the looped segment [66]. An alternative way to perform the initial preparation is to make an incision at the desired location and then ligating that end. To determine particle transport across the intestine, the particle suspension is injected directly into the loop, and the intestinal loop is returned to the body cavity of the animal for up to 2 h while the animals are maintained under anesthesia [47,53]. After a chosen time period, the animal is sacrificed and the intestinal loop is removed from the body cavity of the animal [67].

A majority of the time, the closed loop intestine is returned to the body cavity for 2 h and then the entire intestinal segment is removed. Another way to use the closed loop method is to inject the nanoparticle solution into the closed loop, remove the suturing after 2 h, and return the intact intestine to the body [66]. This allows for the animal to resume normal digestive processes and movement. After no more than 24 h, the animal is killed and the intestines removed for analysis.

Typically, after performing a closed loop experiment, the tissue is flash frozen and fluorescence or morphology of the intestine is determined by microscopy [53]. For a more quantitative result, the frozen tissue can be homogenized for protein analysis [66]. The tissue can also be removed and dissolved in acetonitrile after homogenization and then evaluated for drug content [48], or blood can be drawn from the animal and analyzed for the presence of the administered drug [68]. To measure the uptake of non-fluorescent peptides or drugs, FITC is often used to label the peptide or drug and a fluorescent microscope measures the fluorescence of the cryo-frozen tissue samples taken from the intestine [69]. A disadvantage of this model is the variation between nanoparticle uptake in the small intestine of different animals: for instance, one study shows an order of magnitude difference of uptake between the rat intestinal loop model and that of the in vivo murine model [70].

8.1. Organ analysis

In organ analysis, the animals are sacrificed after oral feeding and the organs are removed and fixed using 4% (or less) paraformaldehyde solution [41]. This technique of histology analysis is useful for semiquantification but is not useful for counting particles that are smaller than 0.5 μm due to confusion between particles and small aggregates. The semi-quantification is done with gel permeation chromatography (GPC), which measures the distribution of the nanoparticle size after in vivo oral delivery; the harvested tissue is dissolved and the solubilized polymer is extracted for GPC analysis [71].

In studies of gene delivery, a reporter gene is commonly used. For instance, in one study, the LacZ gene is delivered using chitosan nanoparticles and after three days the mice are sacrificed, and the stomach and small intestine are surgically removed and stained with X-gal overnight. This process stains for the presence of β-galactosidase to indicate the effective transcription of the orally delivered chitosan/DNA nanoparticle [72]. Although this is a commonly used technique, one negative aspect is the wide variation in results depending on fixatives, fixation temperature, and fixation time [73]. Particles can also be viewed post-organ removal through microscopy techniques including light, fluorescent, or confocal microscopy. Fluorescent microscopy has the benefit of high sensitivity whereas confocal microscopy provides a three dimensional view of fluorescently labeled tissues [71].

8.2. Symptom reversal

Symptom reversal or phenotypic correction can be used to determine the effectiveness of an oral delivery system by looking for changes in enzyme or protein levels or observable changes in behavior or reaction to a stimulus. In in vivo models involving the oral delivery of insulin using nanoparticles, the change in glucose level in the plasma is used to determine the effectiveness of the delivery system. With this method, blood is drawn from the animal that has been orally administered the insulin loaded nanoparticles and the plasma glucose level is determined. At the same time, insulin concentration in the blood can be monitored by radioimmunoassay [43].

In the case of hemophilia, the symptom reversal is tested with the use of the tail clip assay, which determinates the time required for bleeding to stop after clipping the tail. This is done to prove phenotypic correction by transecting part of the tail and monitoring the survival and movement for 48 h, as well as the cessation of tail bleeding [74,75]. Another way to determine phenotypic correction for hemophilia involves the plasma clotting time, where samples and reference standards are mixed with equal parts of human FVIII-deficient plasma, veronal buffer, and aPTT reagent followed by addition of CaCl₂. The clotting time is determined and compared to the clotting time of non-hemophilic plasma [75]. Along the same lines, the behavior of an animal pre and post treatment can be representative of the therapeutic effect of the treatment. Parameters such as locomotion activity, exploratory patterns, and anxiety can be characterized to identify changes in neural symptoms [41].

8.3. Immune response

Anaphylactic reaction is a common response to allergens. This response can occur as an antibody response to the protein or an immune response, such as inflammation, and can be graded using a point system with: 0, no sign of reaction to 5, death [76]. Anaphylaxis is a poor indicator of the effectiveness of protein therapy; however, it is commonly used to determine the effectiveness of tolerance induction. Tolerance is important with protein replacement because it allows the patient to receive the protein or the gene that encodes the protein without the danger of an adverse immune response. Signs of an immune response in an animal model include hunched position, piloerection, slowing of movement, bronchospams, respiratory arrest, and anaphylaxis. The plasma can be obtained from the treated patient and measured for antibody titers of interest, which is another indicator of immune response [77].

Antibody analysis is used to determine if a foreign protein is present or if tolerance is developed [77]. For example, to determine the antibodies for cFVIII, plasma samples are mixed with equal volumes of buffered normal plasma and Bethesda units are determined using aPTT assay to compare test samples to human FVIII-deficient control plasma samples [75]. This provides a method for determining inhibitors to the cFVIII protein. An enzyme linked immunosorbent assay (ELISA) is amore commonly used method to determine antibody levels. There are multiple ways that an ELISA can be done; in one example, the plates are coated with the protein of interest, the Arah2 protein, serum samples were added, then biotinylated anti-mouse IgA, IgE, and IgG2a are added to determine the presence of those antibodies in the serum [76]. This yields results that determine which antibodies are specific for the protein of interest; if post-treatment levels are high, it could signify the presence of the foreign protein which is inducing an immune response, however if post-treatment levels are low, it could indicate that tolerance is being induced or that the protein is absent.

Another immune analysis technique involves measuring the amount of a specific antigen in the plasma. Although this antigen may not stimulate an immune response, the presence of the antigen signifies that the protein is being replaced, which is valid in protein and gene delivery applications. For example, in the case of oral delivery for hemophilia A, an aPTT assay determines the FVIII activity after administration of nanoparticles as compared to non-treated mice [75].

8.4. DNA/RNA extraction

In oral gene delivery studies, analysis of the genomic DNA in the tissue of the animal is one of the most sensitive methods of biodistribution analysis. This is done by lysing tissue samples, obtaining DNA from the tissue, purifying the DNA, and performing PCR [74]. In one study, the presence of the gene of interest is determined by PCR using primers that flank the gene segment; for example, to determine the presence of the cFVIII gene after oral gene delivery of FVIII with chitosan nanoparticles, forward and reverse primers that flank the cFVIII B-domain amplified the 152-bp fragment of interest and verified the presence of the gene [75]. The biodistribution of the nanoparticles is determined by observing the amount of amplified DNA in the extracted tissue samples.

8.5. Protein extraction

Protein analysis can be done by sacrificing the animal, removing the organ of interest, and homogenizing the organ to make the proteins determinable by Western blotting [41]. This analysis can determine relative amount of protein and type of protein based on molecular weight and standards.

8.6. Blood analysis

Blood is often used to determine the presence of a peptide or drug following oral delivery of a nanoparticle carrying drug, peptide, or nucleic acid. The whole blood or plasma can be used to perform assays. In one example, the Fluorogenic Thrombin Generation Test (FTGT), which is highly sensitive to low levels of FVIII uses plasma and fluorogenic thrombin substrate followed by use of a plate reader to quantify free thrombin as a measure of the transgene expression achieved by oral delivery of the cFVIII-chitosan nanoparticles [74]. In experiments performed by Sakuma et. al., electrical gradients in the blood determine the presence of the therapeautic effect. Blood is drawn from the rat after oral delivery of salmon calcitonin and the anionic nanoparticles of poly(methacrylic acid) and blood ionized calcium concentration is measured using calcium electrodes [78].

8.7. Radioactive labeling

Gamma scintigraphy is another in vivo analysis technique that is beneficial due to its ability to monitor real time transit of particles based on Technetium labeling. This method enables visualization and quantification of particles moving through the GI tract after oral delivery (Fig. 3). Although gamma scintigraphy is typically noninvasive, the GI organ of interest can be removed and analyzed for greater accuracy. This method requires the nanoparticles to be labeled with a gamma-emitting radioisotope [79–81]. The isotope is chosen based on several factors including radiation energy and half-life. Metal ion nuclides present the most suitable radioisotopes for gamma scintigraphy; the most commonly used radioisotope is technetium (^99mTc), which has a low radiation dose and a short half-life of 6 h. Labeling the particles with the radioisotopes can be done in multiple ways. One method, used to label particles with technetium, requires a non-absorbable chelate such as ^99mTc-DTPA [80]. Another way of labeling nanoparticles is the use of Na^99mTcO₄ solution, which involves ion exchange between the particle and the anionic ^99mTcO₄ to achieve a radiolabeled particle [82]. A gamma camera, with a 40 cm sodium iodide crystal activated with thallium [80], images the GI tract for the presence of technetium and the images are analyzed using software such as Siemens MicroDELTA Clinic software by manually delineating regions and counting the radioactivity in each region [79]. One of the primary drawbacks to the use of gamma scintigraphy is the radiation exposure [83], which is an important consideration in the use of this technique for in vivo human studies. Other disadvantages of this method are that it can be an invasive procedure and there is some instability of the labeling causing radioactive elements to leak into other parts of the body [71]. Despite these risks, gamma scintigraphy has been used in the medical field for imaging with chromium-51 labeling [84], showing its trustworthiness in clinical applications.

Alternatives to gamma scintigraphy that use radiolabeling are single-photon emission computed tomography/computed tomography (SPECT-CT) and positron emission tomography (PET). These two imaging techniques differ from scintigraphy because they result in 3D images with high resolution and correction of attenuation. The imaging techniques differ from each other because SPECT-CT instrumentation measures single photon release from radioisotopes while PET measures positron release [85]. In two studies using SPECT-CT to evaluated biodistribution, various gamma-emitting isotopes (such as ^99mTc, ¹²³I, ¹²⁵I, and ¹³³I) were used to track movement of multiple components at once. In one study, a nanoparticle containing chitosan and poly(γ-glutamic acid) was loaded with aspart-insulin and radiolabeled with ^99mTc while the drug alone was labeled with ¹²³I. The biodistribution after oral administration was observed using dual isotope dynamic SPECT-CT and can be seen in Fig. 4. The images showed the progression of the nanoparticles from the stomach to the small intestine to the large intestine, while the drug was shown in the kidneys and the urinary bladder, indicating a more rapid clearance of the drug [86]. In a later study, the same group showed that increased permeation in the intestines by chitosan did not increase the absorption of toxic lipopolysaccharides after oral administration of chitosan nanoparticles. SPECT-CT was used to show that ^99mTc-labeled lipopolysaccharides remained within the GI tract after administration of chitosan nanoparticles [87]. The imaging provided by SPECT-CT and PET provide clear, 3D images for cases in which scintigraphy may not be informative.

8.8. IVIS

Another technique for in vivo analysis after oral delivery is a form of optical imaging termed IVIS. This technique allows for in vivo imaging without radiation. IVIS is an optical imaging technique that uses fluorescence and results in a high signal-to-noise ratio due to the low bioluminescence of mammalian tissue. For application in oral delivery with nanoparticles, the nanoparticles can be labeled with fluorescent markers such as Sulforhodamine B or Cy5.5 [88,89]. In one study by Chen et. al., the biodistribution of modified nanostructured lipid carriers (NLCs) labeled with Sulforhodamine B was determined using IVIS. Results from IVIS showed that the NLCs with Myverol achieved higher retention due to the observance of fluorescently labeled nanoparticles in the abdominal area of the rat [88]. [88]. Lee et. al. used IVIS to study biodistribution of nano- and submicron-scaled zinc-oxide (ZnO) nanoparticles labeled with a monoreactive hydroxysuccinimide ester of Cy5.5 (Cy5.5-NHS). IVIS imaging showed a strong signal of Cy5.5-conjugated ZnO nanoparticles in the kidney and liver. By imaging the animal at different time points, the particle movement was tracked, and the results from this study showed that the nano-scaled ZnO nanoparticles moved through the GI tract faster than the submicron-scaled ZnO nanoparticles [89]. If further biodistribution studies are desired, the organs can be removed from the animal after IVIS and further imaging can be done on the individual organs (Fig. 5).