Food structure and gastric emptying

As an illustration of some of these effects, we undertook a study to determine the extent to which oat particle size in a porridge could alter glucose absorption, gastric emptying, gastrointestinal hormone response and subjective feelings of appetite and satiety^{(Reference Mackie, Bajka and Rigby12)}. In a crossover design, eight participants were fed porridge prepared from either oat flakes or oat flour with the same protein, fat, carbohydrate and mass. Subjective appetite ratings, gastric contents and plasma glucose, insulin and gastrointestinal hormones were determined over a period of 3 h post-consumption. The use of MRI provides direct visualisation of gastric content: changes in gastric emptying and also what is emptied. As an example, Fig. 1A and 1B shows MRI images for 5 and 25 min, respectively, post-consumption of oat flake porridge. The abdominal cross-sections show that most of the 175 ml liquid consumed with the 264 g porridge was emptied within 25 min of consumption. This highlights that structure is important in defining what is emptied. Regardless of the early emptying of gastric liquid from the oat flake porridge, its structure meant that after 3 h post-consumption the oat flake porridge had an average 25 % greater volume remaining in the stomach than the starch porridge. Despite the limited differences in the rate of gastric emptying, significant differences were seen in plasma GIP and insulin and minor differences in GLP-1. The peak in GIP and GLP-1 was at 20 min post-consumption for the starch porridge and 35 min for the flake porridge. The peak timing was reversed for insulin and no differences were seen in plasma glucose. Thus, highlighting that in healthy individuals blood glucose concentrations are tightly constrained even when very different amounts are being absorbed as suggested by the differences in GIP.

Fig. 1. Axial FIESTA (Fast Imaging Employing Steady-state Acquisition) MRI images of the stomach (outlined) taken 5 min (A) and 25 min (B) post-consumption of oat flake porridge. Image (A) shows a layer above the oat flake porridge that is not apparent after 25 min (B). Axial TrueFISP (True Fast Imaging with steady state precession) MRI images of the stomach (outlined) taken 5 min after consumption of a semi-solid (C) or liquid (D) version of the same meal.

Similarly, structural effects can be seen in the digestion and absorption of protein and lipids. Indeed food structures can be tailored to alter the timing of the delivery of specific macronutrients. Making use of density differences to drive creaming or sedimentation of components can be particularly effective^{(Reference Mackie, Rafiee and Malcolm13)}. When participants consumed a liquid or semi-solid meal with the same fat, protein, carbohydrate and energy as shown in Fig. 1C and 1D, differences in emptying behaviour were seen. The dark shapes in the stomach in Fig. 1C are boluses of high fat and protein cheese formed in the mouth and the swallowed. These remained visible for up to an hour before dispersing, trapping the protein and fat at the bottom of the stomach. In contrast, the liquid meal already showed evidence of creaming of the fat to the top of the stomach after 5 min. The food boluses were measured and compared with particle-size distributions from other meals, which highlighted the influence of the different meal structural properties on gastric chyme^{(Reference Hornby, Collado-González and Zhang14)}. Thus in the first hour the composition of the gastric chyme emptied into the small intestine would have been very different. This was confirmed rather circumstantially as the gastric-emptying rate for the liquid meal at 35 min post-consumption was significantly faster. The differences in emptying then led to differences in gastric volume and subjective appetite scores after 3 h. In another study using MRI to compare the effects of energy density and viscosity on gastric-emptying rate^{(Reference Camps, Mars and de Graaf15)}, the authors found that increasing the viscosity was less effective at slowing gastric emptying than increasing the energy density. However, the viscosity was more important in increasing the perception of fullness. The results highlighted the lack of satiation from ‘empty energy’ in quickly ingested drinks such as fizzy drinks. Although these examples highlight the role of food structure in digestion kinetics, it is also apparent that measuring the bioactive concentrations in peripheral blood cannot tell the whole story because we do not know what was being absorbed. In order to fill this gap, a number of simulations of digestion have been developed to mimic human physiology and enable the digestive fate of bioactives to be more closely followed.

In vitro simulations of the influence of food structure on digestive fate

The Infogest network has been a key player in developing physiologically relevant and widely usable in vitro simulations of food digestion. The network was originally assembled through a COST Action (FA1005) that had protein digestion as a core consideration in relation to food allergy^{(Reference Dupont, Bordoni and Brodkorb16)}. However, this was soon broadened to include all food and macronutrient types. The most widely cited simulation protocol is the static model that uses fixed conditions to simulate upper gastrointestinal tract digestion^{(Reference Minekus, Alminger and Alvito17,Reference Brodkorb, Egger and Alminger18)} . This simulation has been used in hundreds of studies and comprises three phases, oral, gastric and intestinal. It can provide valuable information on digestion endpoints for any bioactives where microbial fermentation does not play a role. The oral phase includes simulated salivary fluid, salivary amylase and recommendations for how the chewing of food should be simulated. The oral phase generates the food bolus that is passed into the stomach for further digestion. The gastric phase of the simulation includes simulated gastric fluid as well as the protease pepsin and a recommendation for the use of gastric lipase. The recommendation is for this phase to last for 2 h at pH 3, which is an estimate of mean pH values achieved in vivo when half of a meal has been emptied from the stomach into the duodenum. Although this is the recommendation, there is some evidence that the pH should be higher, perhaps closer to 5^{(Reference Sams, Paume and Giallo19)} depending on the nature of the meal. This is important because it determines which gastric enzymes are active. Salivary amylase is active in the stomach at higher pH, gastric lipase in the mid-range of pH and the protease pepsin at low pH. After 2 h, chyme from the gastric phase is emptied into the intestinal phase for digestion to be completed. The intestinal simulation includes simulated intestinal fluid, bile and pancreatic enzymes incubated at pH 7 for 2 h. This kind of approach allows a range of different endpoints to be determined, depending on the nutrients or bioactives of interest. These might include determination of peptide profiles or free amine groups for protein digestion, maltose concentrations for starch and NEFA profiles for lipid digestion. It should be highlighted that this is a static model using fixed parameters and thus is unlikely to give physiologically relevant kinetic data. However, it can be used for assessing digestibility of nutrients such as protein^{(Reference Sousa, Recio and Heimo20)} and is a strong alternative to replace animal models for determining protein nutritional quality. The key requirements of this simulation of the upper gastrointestinal tract are reproducibility and physiological relevance. The reproducibility has been confirmed in a number of ring trials^{(Reference Egger, Ménard and Delgado-Andrade21)} and the physiological relevance has been demonstrated in a number of different systems^{(Reference Egger, Schlegel and Baumann22–Reference Miralles, Sanchon and Sanchez-Rivera24)}. However, it should be noted that this type of simulation is only able to model luminal events and it lacks brush border enzymes and absorptive elements.

As highlighted earlier, it has become apparent that it is not just the extent of digestion that is important for disease but also the rate of digestion and nutrient release. Thus, it is important to have ways of determining digestion kinetics and this can be achieved using simple semi-dynamic simulations of digestion^{(Reference Mulet-Cabero, Egger and Portmann25)} or more sophisticated computer-controlled simulations^{(Reference Li and Kong26)}. The most important phase of digestion to simulate in relation to digestion kinetics is the gastric phase. Thus, kinetic simulations tend to concentrate on this phase and control factors such as gastric loading, gastric emptying, acidification rate, enzyme and simulated gastric fluid secretion rate and physical processing of the gastric chyme. The examples shown in this article draw on the Infogest semi-dynamic protocol that has been used to determine the impact of food structure on digestion kinetics but there are many other similar models in the literature.

In a study undertaken in 2012 and published in 2013, the effect of the structure of dairy products was initially investigated^{(Reference Mackie, Rafiee and Malcolm13)}. In this study, we were able show that the semi-solid structure persisted in the stomach and suppressed gastric emptying over the first hour compared to the liquid meal. This difference then led to persistent differences in volume of gastric contents and associated feelings of fullness for up to 3 h post-ingestion. Although this study hints at the role of food structure in digestion kinetics in support of previous work^{(Reference Marciani, Faulks and Wickham27)}, it does not directly show that the rate of appearance of the products of digestion varied. In order to determine that kind of information, studies must either use intubation^{(Reference Armand, Borel and Pasquier28)} or resort to animal or in vitro models. In a subsequent study the Infogest semi-dynamic model was used to follow the digestion of the same two meals in more detail^{(Reference Mulet-Cabero, Rigby and Brodkorb29)}. In those simulations, the detailed analysis of protein and lipid digestion was able to show that gastric behaviour was affected by the initial structure with creaming and sedimentation observed in the case of liquid and semi-solid samples, respectively. Lipid and protein digestion profiles showed clear differences in the amount of nutrients reaching the simulated small intestine and, consequently, the likely bioaccessibility after digestion. The semi-solid sample generated higher nutrient released into the small intestine at an early stage of digestion whereas nutrient accessibility from liquid sample was delayed due to the formation of a cream layer in the gastric phase. This shows the strong effect of the matrix on gastric behaviour, proteolysis and lipolysis.

In two similar studies, dairy processing (heating and/or homogenisation) was used to alter the microstructure of cows' milk prior to simulated digestion^{(Reference Mulet-Cabero, Mackie and Wilde30,Reference Ye, Cui and Dalgleish31)} . Both studies showed the typical clotting behaviour of whole milk in the gastric phase of digestion. They also showed differences in clot consistency depending on the processing applied to the milk. Unprocessed ‘raw’ milk had the firmest clot while homogenised and heated and homogenised presented weaker clots. In particular, the study by Mulet-Cabero et al. showed that the clot from raw or heated milk was dense enough to sediment, while the homogenised and heated and homogenised curd entrapped sufficient lipid to cream to the top of the gastric compartment. These structural changes occurring during the gastric phase resulted in different nutrient emptying, with significant differences between ‘raw’ and heated and homogenised, and more extensive digestion of milk proteins in the heat-treated samples due to the drastic denaturation of the proteins.

This research has shown that in vitro digestion can provide a platform for linking food structure to digestion^{(Reference Hiolle, Lechevalier and Floury32,Reference Bornhorst, Singh, Doyle and Klaenhammer33)} . The use of in vitro simulations of digestion driven by human study data provides a powerful tool to improve the nutritional quality of food. Most recently, both MRI and in vitro simulation have been combined to non-invasively follow the influence of food structure on digestion^{(Reference Deng, Janssen and Vergeldt34–Reference Deng, Seimys and Mars36)}. In all cases, it is the combination of a range of approaches linking human study data to physiologically relevant in vitro studies that has broadened understanding of the role of food structure in digestion kinetics.

Validation of simulations of digestion

It is clearly important for any model to be validated against data from the system that it is replicating. In the case of in vitro models of human digestion, this can often be problematic^{(Reference Dupont, Alric and Blanquet-Diot37)}. The rationale for using in vitro simulations of digestion is that they can provide direct information about the breakdown of food in the gastrointestinal tract that is often hard to gather in human studies. This can in turn improve interpretation of results from nutritional epidemiological studies that necessarily generate correlative outcomes. There are numerous examples of the use of in vitro digestion to determine the fate of specific macronutrients but much of the research focuses on protein and starch. The move to more plant-based diets has led to research on protein quality using the FAO-recommended digestible indispensable amino acid score system^{(Reference Rutherfurd, Fanning and Miller38,Reference Wolfe, Rutherfurd and Kim39)} . The Infogest simulation of digestion had also been used to provide the source data for digestible indispensable amino acid score^{(Reference Sousa, Recio and Heimo20)}, with the benefit no animals are involved. However, the in vitro results were validated against results from experiments in vivo. The validation was made by comparison of seven different protein sources with data collected both in pigs and in human subjects^{(Reference Hodgkinson, Stroebinger and van der Wielen40)}. The results showed good agreement between the results in vivo and the Infogest in vitro results but as the authors note, the comparison was only made with a limited set of food sources, so no general conclusions about the efficacy of the Infogest approach can be made for assessing all protein quality. The digestion of protein has been investigated in relation to food allergy, in particular because it has been suggested that stability to digestion may be a key parameter for a protein to be an allergen^{(Reference Astwood, Leach and Fuchs41)}. However, the outcomes are highly dependent on the nature of the in vitro digestion model being used and its relevance to specific allergens^{(Reference Torcello-Gómez, Dupont and Jardin42,Reference Torcello-Gómez, Dupont and Jardin43)} . In particular, the European Food Safety Authority currently uses a late phase gastric model with low pH and high protease activity but this is very different from an infant simulation^{(Reference Menard, Bourlieu and De Oliveira44,Reference Menard, Cattenoz and Guillemin45)} . As a result, there have been calls for European Food Safety Authority to review the methodology for novel protein risk assessment^{(Reference Verhoeckx, Bogh and Dupont46)}. The examples given here are for the Infogest-recommended simulations but there are many others in the literature with increasing levels of sophistication and focusing on different bioactives^{(Reference Duque-Soto, Quintriqueo-Cid and Rueda-Robles47,Reference Faubel, Cilla and Alegria48)} .

In addition to protein, there has been a lot of interest in understanding the role of processing and structure of starch in digestion relating to glycaemic response^{(Reference Pautong, Anonuevo and de Guzman49,Reference Fernandes, Madalena and Pinheiro50)} . Although comparison can be made with in vivo data, as highlighted by Phillips et al. earlier^{(Reference Phillips, Deane and Jones11)}, there are many individual factors that can make such a comparison problematic, not least the fact that glucose in peripheral blood is not a good indicator of bioaccessible glucose in the gut^{(Reference Priyadarshini, Moses and Anandharamakrishnan51)}. Because starch is the most important digestible polysaccharide in human nutrition usually accounting for 20–50 % of the total energy intake, it has been studied extensively^{(Reference Bohn, Carriere and Day52)}. The apparent health benefits of a low-GI diet led the Carbohydrate Quality Consortium to state ‘an urgent need to communicate information on GI and GL’^{(Reference Augustin, Kendall and Jenkins53)}. Consequently, it is important to note that a number of studies have shown the validity of the in vitro determination of the GI^{(Reference Fernandes, Madalena and Pinheiro50,Reference Monro and Mishra54,Reference Argyri, Athanasatou and Bouga55)} . The ability of resistant starch to pass through the upper gastrointestinal tract and into the colon has also highlighted the need for more sophisticated models of human digestion that include colonic fermentation.

What more is needed?

As the pressure to reduce animal experiments increases there a drive to find suitable models to replace them^{(Reference Langley, Evans and Holgate56,Reference Mak, Evaniew and Ghert57)} . As a result, there has been a proliferation of in vitro models of digestion, many of which are more sophisticated in the way that they mimic the physical and biochemical environment of the gut^{(Reference Hur, Lim and Decker58)}. These include simulations of the gastric phase^{(Reference Kong and Singh59,Reference Wickham, Faulks and Mann60)} where the biochemical and physical environment of the stomach are replicated, or the gastric and intestinal phases^{(Reference Minekus, Marteau and Havenaar61)} or the gastric, intestinal and colonic phases of digestion^{(Reference Chaikham, Apichartsrangkoon and Jirarattanarangsri62)}. However, these simulations lack the final stages of digestion and any absorption steps. Although some attempts have been made to include brush border enzymes^{(Reference Di Stasio, Picascia and Auricchio63)}, these have not been widely accepted because difficulties in defining activity and exposure time.

There are many in vitro models focusing on just the colonic phase of digestion often based on the early work of MacFarlane et al.^{(Reference Macfarlane, Macfarlane and Gibson64)}. These are becoming increasingly important in enabling research to understand the role of dietary fibre and plant bioactives in relation to gut microbiota and the gut–brain axis^{(Reference Silva, Bernardi and Frozza65)}. However, the issue with investigating the role of colonic fermentation in the digestion of complex foods using in vitro approaches is the carry-over of compounds from the small intestinal phase to the large intestinal fermentation phase. Transporters in the apical membrane of enterocytes often specifically control absorption in the small intestine. Thus, both highly nutritive molecules such as simple sugars, amino acids and peptides as well as fatty acids are largely removed from intestinal chyme. Additionally, potentially toxic bile acids are reabsorbed in the distal ileum. These and similar absorption mechanisms are not simulated well by passive dialysis so improvements need to be found in presenting realistic digesta to simulations of colonic fermentation.

In silico models of digestion are also becoming more widely available^{(Reference Le Feunteun, Verkempinck and Floury66,Reference Del Rio, Van der Wielen and Gerrits67)} but many of them are focused only on protein digestion using specific rules for modelling the action of proteases. However, more generally applicable in silico models of digestion will only become reliable when more human study data are made available to refine them^{(Reference Le Feunteun, Mackie and Dupont68)}. With the rise of machine-learning approaches, this is likely to become a more tractable approach in the future.