A standardised static in vitro digestion method suitable for food – an international consensus †

Simulated gastro-intestinal digestion is widely employed in many ﬁ elds of food and nutritional sciences, as conducting human trials are often costly, resource intensive, and ethically disputable. As a consequence, in vitro alternatives that determine endpoints such as the bioaccessibility of nutrients and non-nutrients or the digestibility of macronutrients (e.g. lipids, proteins and carbohydrates) are used for screening and building new hypotheses. Various digestion models have been proposed, often impeding the possibility to compare results across research teams. For example, a large variety of enzymes from di ﬀ erent sources such as of porcine, rabbit or human origin have been used, di ﬀ ering in their activity and characterization. Di ﬀ erences in pH, mineral type, ionic strength and digestion time, which alter enzyme activity and other phenomena, may also considerably alter results. Other parameters such as the presence of phospholipids, individual enzymes such as gastric lipase and digestive emulsi ﬁ ers vs. their mixtures ( e.g. pancreatin and bile salts), and the ratio of food bolus to digestive ﬂ uids, have also been discussed at length. In the present consensus paper, within the COST Infogest network, we propose a general standardised and practical static digestion method based on physiologically relevant conditions that can be applied for various endpoints, which may be amended to accommodate further speci ﬁ c requirements. A frameset of parameters including the oral, gastric and small intestinal digestion are outlined and their relevance discussed in relation to available in vivo data and enzymes. This consensus paper will give a detailed protocol and a line-by-line, guidance, recommendations and justi ﬁ cations but also limitation of the proposed model. This harmonised static, in vitro digestion method for food should aid the production of more comparable data in the future.


Introduction
In vitro methods simulating digestion processes are widely used to study the gastro-intestinal behaviour of food or pharmaceuticals. Although human nutritional studies are still being considered the "gold standard" for addressing diet related questions, in vitro methods have the advantage of being more rapid, less expensive, less labour intensive, and do not have ethical restrictions. This allows a relatively large number of samples to be measured in parallel for screening purposes. Reproducibility, choice of controlled conditions and easy sampling at the site of interest make in vitro models very suitable for mechanistic studies and hypothesis building.
Simulated digestion methods typically include the oral, gastric and small intestinal phases, and occasionally large intestinal fermentation. These methods try to mimic physiological conditions in vivo, taking into account the presence of digestive enzymes and their concentrations, pH, digestion time, and salt concentrations, among other factors. Some computerized sophisticated models such as the Dutch TNO gastrointestinal tract model, 1 the model by the English Institute of Food Research 2 or by the French INRA 3 allowing the simulation of dynamic aspects of digestion, such as transport of digested meals, variable enzyme concentrations and pH changes over time. However, the majority of models reported in literature are static ones, 4 i.e. models with constant ratios of meal to enzymes, salt, bile acids etc. at each step of digestion.
Static models of human digestion have been used to address such diverse scientic questions as the digestibility and bioaccessibility (i.e. the amount of a compound that is released from the matrix and is considered to be available for absorption through the gut wall) of pharmaceuticals, 5 mycotoxins, 6 and macronutrients such as proteins, 7,8 carbohydrates 9 and lipids. 10,11 They have also been used to study matrix release of micronutrients such as minerals and trace elements, 12 and secondary plant compounds including carotenoids 13 and polyphenols. 14,15 Some digestion methods are used to produce bioaccessible fractions that can be used to address further mechanistic questions, such as intestinal transport by employing Caco-2 cells. 16 Although many in vitro methods are derived from earlier reported methods, there is signicant variation in the use of in vitro digestion parameters between the individual models described in literature, [17][18][19] impeding the possibility to compare results across research-groups and to deduce general ndings. While altering some of these parameters may have a limited impact on the matrix release or digestibility of some compounds, there could be a large impact for other ingredients. Enzyme activity is also altered by pH and the concentration of salts such as calcium. The applied gastric pH may vary greatly between the models, i.e. from pH 2 20 to pH 4. 21 The COST action INFOGEST 22 is an international network joined by more than 200 scientists from 32 countries working in the eld of digestion. One aim of the network is to consolidate conditions for simulated digestion of food and nd a consensus, if possible, for a digestion model. The group is aware that no conditions outlined will be suited for all underlying research questions. However, the authors of this manuscript strived to describe a "smallest common denominator", i.e. a set of conditions that are close to the physiological situation, are practical, and can be seen as a basic suggestion to address various research questions. Further amendments of these suggested conditions may be needed, for example to simulate digestion in infants or the elderly, which may differ considerably in enzyme concentration. [23][24][25] For more accurate simulation of in vivo conditions, dynamic models should be used. In the next sections, we describe our recommendations for a standardised digestion method which is based on the current state of knowledge on in vivo digestion conditions, and employs widely available instrumentation and chemicals.

Experimental -in vitro digestion protocol
This section describes a detailed line-by-line protocol, which is also summarised in Fig. 1. Further information and justication on the choice and concentration of chemicals, inclusion or omission of certain steps are discussed in greater detail in the following section "In vitro digestion parametersrecommendation and justication".

Materials
All materials are standard analytical grade. Sodium bicarbonate (0.5 M) should be ltered through a 0.22 mm lter under vacuum. It can be stored at 2-5 C for approximately one month. The type of enzyme products, mostly provided by Sigma Aldrich (St Louis, Mo), is only a recommendation and similar products of comparable quality from other providers can be used. Enzyme activities are based on commonly used assays. Detailed protocols of the enzyme assays are outlined in the ESI. † ar Institute of Food Research, Norwich, NR4 7UA Norfolk, UK. E-mail: Alan.Mackie@ ifr.ac.uk Amylase (EC 3.2.1.1) activity is based on soluble potato starch: one unit liberates 1.0 mg of maltose from starch in 3 minutes at pH 6.9 at 20 C. Porcine Pepsin (EC 3.4.23.1) activity is based on bovine blood haemoglobin as a substrate: one unit will produce a DA 280 of 0.001 per minute at pH 2.0 and 37 C, measured as TCA-soluble products. Porcine trypsin (EC 3.4.21.4) activity is based on p-toluene-sulfonyl-L-arginine methyl ester (TAME): one unit hydrolyses 1 mmol of TAME per minute at 25

Simulated digestion uids
Simulated Salivary Fluid (SSF), Simulated Gastric Fluid (SGF) and Simulated Intestinal Fluid (SIF) are made up of the corresponding electrolyte stock solutions (Tables 1 and 2), enzymes, CaCl 2 and water. The electrolyte stock solutions are 1.25Â concentrated i.e. 4 parts of electrolyte stock solution + 1 part water give the correct ionic composition in the simulated digestion uids. For example 3.8 mL SGF electrolyte stock solution + 0.2 mL pepsin (made up in SGF electrolyte stock solution) + 25 mL 0.3 M CaCl 2 + 975 mL water ¼ 5 mL SGF. Enzyme activities are in units per mL of nal digestion mixture rather than secretion activity, unless stated otherwise.

Oral phase
Mastication of solid food is simulated by mincing an appropriate amount of food using a commercially available manual or electric mincer, such as the Eddingtons Mincer Pro (Product Code 86002, Berkshire, UK) or similar, commonly used in kitchens to mince meat. SSF electrolyte stock solution is added to create a thin paste-like consistency, similar to that of tomato paste or mustard. If necessary, the electrolyte stock solution can also be added during mincing. A nal ratio of food to SSF of 50 : 50 (w/v) is targeted. For liquid food an oral phase can be included, especially if the meal contains starch. In this case a nal ratio of 50 : 50 (v/v) is targeted. Human salivary a-amylase (EC 3.2.1.1) is added to achieve 75 U mL À1 in the nal mixture, followed by CaCl 2 to achieve 0.75 mM in the nal mixture and the necessary amount of water to dilute the stock solution of SSF. The recommended time of contact with the enzyme is 2 minutes at 37 C, which requires pre-warming of all reagents to 37 C.
In a typical example: 5 g of solid or 5 mL of liquid food is mixed with 3.5 mL of SSF electrolyte stock solution and minced together. 0.5 mL salivary a-amylase solution of 1500 U mL À1 made up in SSF electrolyte stock solution (a-amylase from human saliva Type IX-A, 1000-3000 U mg À1 protein, Sigma) is added followed by 25 mL of 0.3 M CaCl 2 and 975 mL of water and thoroughly mixed.

Gastric phase
Liquid food can be exposed to the oral phase (optional) or directly to gastric phase, as further discussed in the main section of the manuscript. Five parts of liquid food or oral bolus, is mixed with 4 parts of SGF stock electrolyte solution to obtain a nal ratio of food to SGF of 50 : 50 (v/v) aer addition of other recipients and water. Porcine pepsin (EC 3.4.23.1) is added to achieve 2000 U mL À1 in the nal digestion mixture,  followed by CaCl 2 to achieve 0.075 mM in the nal digestion mixture. 1 M HCl is added to reduce the pH to 3.0; it is recommended to determine the amount of required acid in a test experiment prior to digestion, hence acid can be added more rapidly and followed by verication of the pH. Finally, the necessary amount of water is added to the mixture to dilute the stock solution of SGF. The use of gastric lipase is not recommended at this time because it is not commercially available (further discussed in the main text). The recommended time of digestion is 2 hours at 37 C. The pH may have to be re-adjusted with 1 M HCl during digestion. Sufficient mixing during digestion is recommended, for example by placing the reaction vessel into a shaking incubator, water bath with integrated shaker or a rotator in a 37 C room. In a typical example: 10 mL of liquid sample or oral bolus is mixed with 7.5 mL of SGF electrolyte stock solution, 1.6 mL porcine pepsin stock solution of 25 000 U mL À1 made up in SGF electrolyte stock solution (pepsin from porcine gastric mucosa 3200-4500 U mg À1 protein, Sigma), 5 mL of 0.3 M CaCl 2 , 0.2 mL of 1 M HCl to reach pH 3.0 and 0.695 mL of water.

Non-standard gastric condition
In the absence of phospholipids or other low molecular weight surfactants in the tested food, it is recommended to include phospholipids (0.17 mM in the nal digestion mixture) in the gastric step. In this case freshly prepared SGF containing phospholipids is used instead of SGF. All other steps are as outlined above. For non-standard gastric condition using phospholipids, the following procedure is recommended. Prepare a stock solution of 50 mg mL À1 egg lecithin (Lipid Products, Redhill UK, 500 mg egg lecithin, approx. 63.5 mM, assuming mean M w : 787 g mol À1 ) by adding 1 vial containing 500 mL egg lecithin into a 10 mL volumetric ask wrapped in aluminium foil and lling with chloroform-methanol (1 : 1) solution up to the 10 mL mark; mix until dissolved. This can be stored for a several days at À20 C until required. Gastric liposomes (phospholipids) are prepared the day of usage: a 1 mL aliquot of the 50 mg mL À1 phospholipid stock solution is dried using a rotary evaporator until solvent is removed or dry remaining solvent under inert gas if no rotary evaporator is available, leaving 50 mg of dry phospholipids. Add 5 mL warm SGF to reach nal concentration of 10 mg mL À1 phospholipids in SGF. Incubate at 37 C, shaking at 170 rpm, for 10 min. Sonicate the solution in an ice bath until clear to the eye. Filter the sample through a 0.22 mm nylon syringe lter (Thermo Scientic™ Nalgene Syringe Filters or similar products) to remove any debris deposited by sonicator. The solution should be stored at 4 C and used the same day.

Intestinal phase
Five parts of gastric chyme is mixed with 4 parts of SIF electrolyte stock solution to obtain a nal ratio of gastric chyme to SIF of 50 : 50 (v/v) aer additions of other recipients and water. The gastric samples-chyme is mixed with SIF electrolyte stock electrolyte solution. Addition of base (1 M NaOH) will be required to neutralise the mixture to pH 7.0. Digestive enzymes Table 2 Preparation of stock solutions of simulated digestion fluids. The volumes are calculated for a final volume of 500 mL for each simulated fluid. We recommend to make up the stock solution with distilled water to 400 mL instead, i.e. 1.25Â concentrate, for storage at À20 C. In the Experimental section, these 1.25Â concentrates are referred to as Simulated Salivary Fluid (SSF), Simulated Gastric Fluid (SGF) and Simulated Intestinal Fluid (SIF) electrolyte stock solutions. The addition of enzymes, bile salts, Ca 2+ solution etc. and water will result in the correct electrolyte concentration in the final digestion mixture. CaCl 2 (H 2 O) 2 is not added to the electrolyte stock solutions as precipitation may occur. Instead, it is added to the final mixture of simulated digestion fluid and food a a * in brackets is the corresponding Ca 2+ concentration in the nal digestion mixture.
can be added as either pancreatin from porcine pancreas or individual enzymes. In the case of pancreatin, proteolytic, lipolytic and amylolytic activity of the extract should be determined using the assays outlined in Enzyme assays section. The amount of pancreatin added is based on the trypsin activity (100 U mL À1 in the nal mixture). If the food contains high amounts of fat or the fat digestion is at the centre of the study, pancreatin concentration should be either based on the lipase activity or additional porcine pancreatic lipase and colipase should be added to achieve 2000 U mL À1 lipase activity in the nal mixture. This is further discussed in the main section of the paper. Alternatively, individual enzymes can be added to the digestion mixture to achieve the following activities in the nal mixture: porcine trypsin ( ) and porcine pancreatic colipase (2 : 1 colipase to lipase molar excess, equivalent to a mass ratio of roughly 1 : 2 co-lipase to lipase as the mass of human pancreatic lipase is 51.2 kDa and the mass of human co-lipase is 10 kDa). Bile salts are added to give a nal concentration of 10 mM in the nal mixture. There are two options for bile; in both cases the concentration of bile salts needs to be determined, (see assay in Enzyme assays section): bile extract such as B8631 (porcine) or B3883 (bovine) from Sigma-Aldrich or fresh (frozen) porcine bile. CaCl 2 is added to reach 0.3 mM in the nal digestion mixture. It is recommended to assay the Ca 2+ content in pancreatin, if used, and take this into account when adding Ca 2+ to the digestive mixture. The pH may need re-adjustment before nally adding water to the mixture to dilute the stock solution of SIF. The recommended time of intestinal digestion is 2 hours at 37 C. The pH may need re-adjustment during digestion. This can be achieved either manually or by automated laboratory titrator.
In a typical example, 20 mL of gastric chyme is mixed with 11 mL of SIF electrolyte stock solution, 5.0 mL of a pancreatin solution 800 U mL À1 made up in SIF electrolyte stock solution based on trypsin activity (pancreatin from porcine pancreas, Sigma), 2.5 mL fresh bile (160 mM in fresh bile), 40 mL of 0.3 M CaCl 2 , 0.15 mL of 1 M NaOH to reach pH 7.0 and 1.31 mL of water. It is recommended to verify the pH and determine the amount of NaOH-HCl required in a test experiment prior digestion, hence base-acid can be added more rapidly and followed by verication of the pH.

Sampling during digestion
Sample conservation depends on the study focus (food structure, bioaccessibility, enzymatic digestion product etc.), and should be carefully considered for each study. It may be advisable to have individual sample tubes for each time point rather than withdrawing samples from the reaction vessel. Here are some recommendations to inhibit or slow down further enzymatic action on the food sample: (i) Snap freezing of samples is recommended in liquid nitrogen immediately aer the reaction for further analysis. (ii) If samples are sent to other labs, i.e. by courier or by post, the digestion has to be stopped completely; the following procedures are recommended: (a) neutralization of pH in the gastric phase by adding 0.5 M sodium bicarbonate before snap-freezing in liquid nitrogen and subsequent freeze drying of the samples or (b) addition of protease inhibitor (e.g. 1 mM 4-(2-aminoethyl) benzenesulfonyl uoride hydrochloride [AEBSF], Roche or similar), snap freezing in liquid nitrogen and subsequent freeze drying of samples.

In vitro digestion parametersrecommendation and justification
Oral phase Mastication and duration of oral phase. In the oral cavity the texture of solid food is signicantly altered by mastication and salivation. The food is wetted and lubricated by salivary secretion resulting in a cohesive bolus that is ready for swallowing. Mastication is a complex process that is inuenced by a number of factors like food composition, food volume, number of chewing cycles, bite force, teeth condition, degree of hunger and habits. [31][32][33][34] This all affects size, surface area and shape of food particles. 35 A prediction of particle sizes and particle numbers resulting from mastication that is based on toughness and Young's modulus of food particles can be obtained using the Food Fragmentation Index. 35 The particle-size distribution of the bolus depends largely on food type. Peyron et al. 36 and others compared the boluses produced aer mastication of raw vegetables (carrot, radish, and cauliower) and nuts (peanut, almond, and pistachio). Raw vegetables were transformed into similar boluses made up of particles larger than 2 mm, and nuts gave similar boluses containing 90% of particles smaller than 2 mm. In general, particle sizes of less than 2 mm are accepted for swallowing unless larger food particles are so enough to be swallowed. 36,37 In consideration of the highly individual chewing time and the complex situation of food breakdown during mastication we recommend standardizing the size of solid food particles by using a commercial mincer commonly used in kitchens. Versantvoort et al. 6 recommended an oral digestion time of 5 minutes in order to ensure proper mechanical action for static models. However, chewing time in vivo is generally much shorter. Therefore, a simulated oral phase of 2 min, i.e. the contact time with enzyme, is recommended in this model. This is somewhat longer than in vivo, however, accuracy and reproducibility in a lab situation may be compromised if using any shorter digestion time. In case of liquid food the simulation of an oral phase may be included, especially if the meal contains starch. However, most liquids do not require an oral phase, mainly due to the very short residence times in the oral cavity.
Volume of salivary secretions. Salivary secretion is also of inuence on parameters of the liquid phase of food like pH value, surface tension and viscosity. Human saliva is a watery complex uid, which is mainly produced by the parotid, the sublingual and the submandibular glands. The total amount of saliva produced per day is in the range of 1 to 1.5 L. 38,39 Saliva is excreted at different rates in the stimulated and unstimulated states. The stimulated salivary ow that is contributing to food digestion is a hypo-osmotic (110-220 mOsmol kg À1 ) uid. [40][41][42] In addition to 99.5% of water, human saliva contains 0.3% of proteins as well as various electrolytes like sodium, potassium, calcium, magnesium, phosphate and bicarbonate. Further components are glucose and nitrogenous products as urea. The main proteins are immunoglobulin A (IgA), a-amylase (ptyalin), lysozyme, lactoferrin, as well as mucosal glycoproteins (mucins). 38 In order to simulate the wetting and lubrication of solid food masses by salivation we recommend that at least 50% (w/v) of simulated salivary uid (SSF) is added to solid food masses prior to the homogenization process. The amount of SSF should be high enough to achieve a paste-like consistency, similar to that of tomato paste or mustard. The best lubricating components of saliva are mucins that are excreted from minor salivary glands. Mucins have the properties of low solubility, high viscosity, high elasticity, and strong adhesiveness. Mastication, speech, and swallowing all are aided by the lubricating effects of mucins. 43 Versantvoort et al. 6 and Sarkar et al. 44 both referred the use of 0.005 and 3% (w/v) mucin, respectively. The main objective of the simulation is to help the formation of the bolus that is largely held together by capillary force and allow a solution for the addition of amylase; mucin is not required for either of these. Besides, mucin is only a minor component of saliva thus it was not used in this standardized digestion method.
Amylase activity and pH. Salivary a-amylase (ptyalin) has a pH optimum at pH 6.8. 42 Its activity is generally limited to the mouth cavity and early gastric digestion when the pH can be high enough due to the buffering capacity of food. a-Amylase is inactivated by the acid milieu and the proteolytic activity in the stomach. It is therefore oen regarded to be of lesser signicance compared to the pancreatic a-amylase. 45 However, it has also been reported that even during 20 to 30 s of oral food processing, 50% of the starch in bread and 25% of the starch in spaghetti can already be hydrolysed. 46 Recent studies have shown that a-amylase plays an important role in the in vitro breakdown kinetics of bread boluses, 47 and between 25 and 50% of the starch in bread and pasta boluses was hydrolysed by salivary a-amylase in vivo. 46 While a small portion of starch is hydrolysed by the enzyme a-amylase due to the short retention time, almost no protein or fat digestion occurs in the mouth. Therefore we recommend a 2 min incubation, which may include mastication at the same time, see above, with a nal concentration of 75 U mL À1 of a-amylase in the mixture of food and SSF in case of the presence of carbohydrates that are digestible by a-amylase. Aer oral processing solid food is emptied from the oral cavity typically in at least two swallows for each bite. 48 Oesophageal passage is a short process with transit times of a few seconds. 49 No effect on food digestion has been reported.

Gastric phase
The primary purpose of the stomach is to deliver digesta to the duodenum in a regulated manner to optimize intestinal digestion. In the lower part of the stomach (antrum), the meal is mixed and digested with secreted enzymes and hydrochloric acid, ground by antral movements, and gradually emptied into the duodenum. The signicance of physiological and simulated gastric parameters will be discussed.
Duration of gastric step. Gastric emptying of a western type solid meal is usually completed between 3 and 4 h. [50][51][52] An initial lag phase may be observed before the linear decrease in gastric content. 53,54 Homogenization of the solid meal usually leads to a one-hour reduction of the length of gastric digestion. 55 By contrast, liquid meal digestion is characterized by an exponential emptying course with rapid onset of emptying. Emptying of 300 mL of water requires 1 h (ref. 56) whereas other studies on liquids with a low protein concentration has shown even shorter transit time (0.5 h). 56 The addition of nutrients (proteins, lipids or carbohydrates) to a liquid meal also affects the transit time. 57,58 In addition, the inter-and even intra-individual day to day variations in gastric secretion affects pH and the rate of gastric emptying. 59 A simplied static model cannot reproduce the dynamic and transient nature of the in vivo digestion process and the food is exposed to gastric conditions reached at approximately half-gastric emptying time. The pH is relatively low from the start of the digestion process, without the initial buffering effect of the food. Similarly, the food is exposed to an enzyme-substrate ratio, which is normally only reached at half-gastric emptying time. The conditions of the digestion protocol we recommend need to be applicable to a broad range of meals, therefore we recommend a time of two hours for gastric digestion. This time represents the half emptying of a moderately nutritious and semi-solid meal.
Volume of gastric secretions. The total volume of gastric secretion depends on fasted and fed state of humans and the consistency of the meal. A liquid meal will usually require from half to one volume of gastric secretion for digestion. 60,61 By contrast, two volumes of gastric juices are secreted for a solid meal. 53,60 The secretion during the rst hour represents half of the total secreted volume for both the liquid and solid meals, even though a continuous emptying will occur that is not possible to simulate by in vitro static digestion. It is thus advised to use one volume of simulated gastric juice for one volume of oral content whatever the meal physical state.
pH. Aer food intake, pH usually increases to 5 and above because of the buffering capacity of a typical western-type diet, 51,62 enriched in vegetable purees 52 or a cocoa beverage. 63 The secretion of hydrochloric acid lowers the pH to the values required for optimal enzyme activities. Consequently, pH slowly returns to fasted pH which is commonly found below 2. 64 Slightly acidic conditions with pH ranging from 4 to 6 are required for optimal gastric lipase activity 64 while pepsin will be mainly active between pH 2 and 4. In order to match the 2 hour recommendation for the length of the gastric simulation the pH we recommend must represent a mean value for a general meal as described above over the two hours suggested. Thus we recommend the use of a static value of pH 3.
Pepsin activity. Pepsin is the only proteolytic enzyme in the human stomach, however, many isoforms exist. The pepsin content in the stomach varies with individuals, however, mainly increases upon digestion from 0.26 (30 min) to 0.58 mg mL À1 (180 min). 62 Large variations in pepsin activities are reported in the literature, partly due to the use of different assays and calculations. [65][66][67] Our aim is to produce a standardised procedure and for this purpose pepsin activity is assayed using haemoglobin (Hb) as a substrate, see ESI, † where one unit will produce a DA 280 of 0.001 per minute at pH 2.0 and 37 C, measured as TCA-soluble products, also referred to as "Sigma" or "Anson" pepsin units. 68,69 A high homology between human and porcine pepsins (84%) and the low cost of porcine pepsin from gastric mucosa support a regular use of porcine pepsin in static in vitro digestion models. 70 Based on an evaluation of values given in the literature [65][66][67] we suggest that porcine pepsin is used at 2000 U mL À1 nal digestion mixture (equivalent to 4000 U mL À1 in secretion).
Gastric lipase activity. Lipase activity is markedly lower in the gastric compartment (10-120 U mL À1 ) compared to that in the duodenal tract (80-7000 U mL À1 ). 71,72 Gastric lipolysis is only partial (5-40%) mainly because of the lower amounts of enzyme present and its pH activity prole. 71 In the absence of triacylglycerols or when the digestion of proteins and polysaccharides is the main focus of the study, the addition of lipase in the gastric step of digestion can be omitted. Human gastric lipase or alternatives with similar characteristics are commercially unavailable at this time and alternatives such as fungal lipases 73,74 exhibit different activities and specicities. 75 For these reasons, gastric lipase is not included in the protocol at this time.
Bile salts. Low concentrations of bile acids (0.2 mM) may be found in the human fasting gastric uid 28 although not in all individuals. The detection of a concomitant pancreatic lipase activity suggested possible duodeno-gastric reux. 76 Thus, bile acids in the gastric phase will not be further considered in this protocol.
Phospholipids. Low concentrations of phospholipids are found in the gastric compartment 77 and these have been shown to affect the rate of protein digestion in the gastric and small intestinal environments. 78,79 The presence of surface active components such as phospholipids also has a marked effect on the extent of re-emulsication of lipids as it passes through the high shear regions of the pylorus. Therefore in the absence of phospholipids or other low molecular weight surfactants present in the food, 0.17 mM phospholipids in the nal gastric solution is recommended to be included in this static model as optional, non-standard gastric conditions.

Small intestine
Once the food has been through the simulated gastric phase of digestion it is transferred to a simulation of the digestion that occurs in the small intestine. It is reasonable to assume that this part of the simulation should be well mixed. Once again we suggest that the gastric contents should be diluted 50 : 50 v/v with simulated intestinal uid (SIF) as given in Table 2. There are many variables that have an impact on transit time through the small intestine but we suggest the time of simulated intestinal digestion should be 2 hours. 6,7,80 Aer emptying from the gastric compartment chyme is normally neutralised by the secretion of carbonate. Consequently the duodenal pH is around pH 6.5 depending on such factors as meal type and gastric emptying rate. The pH then increases slightly over its length to a value of around 7.5 in the distal ileum. Thus, in order to mimic the pH in the entire passage through the small intestinal phase in static conditions, we recommend using an average value of 7.0 6,7,62 through the addition of SIF and sodium hydroxide.
The most important components of the simulated small intestinal digestion are the pancreatic enzymes and the bile. In both cases we suggest that there are essentially two options offering differences in specicity, ease of use and cost of each experiment. For the enzymes we recommend either individual enzymes or porcine pancreatin and for bile we recommend the use of either bile extract or frozen porcine bile. There are some guiding principles that should be considered when deciding what approach to use. In the case of the bile, if the proposed experiment involves accurately following lipid hydrolysis in detail then frozen porcine bile should be used, otherwise the bile extract should suffice. The same argument could be used for the use of individual enzymes in that for a detailed analysis of lipid hydrolysis individual enzymes should be used or indeed if the system contains only protein, lipid or starch then the use of only proteases, lipases or amylase respectively may be appropriate. However, the cost and availability of enzymes should also be considered. In both cases the selection of the amount to use in a static model is difficult to assess as physiological measurements refer more to secretion rates rather than specic amounts.
Pancreatin. Porcine pancreatin is readily available and contains all the important pancreatic enzymes in differing amounts. However, as our aim is to produce a standardised procedure we must base the amount added on the activity of a specic enzyme and for this purpose, trypsin is the most appropriate. Thus we recommend that either 4Â USP (U.S. Pharmacopeia) or 8Â USP pancreatin is used and its trypsin activity assayed using the p-toluene-sulfonyl-L-arginine methyl ester (TAME) assay. 81 The amount of pancreatin added should then be based on the trypsin assay and should be sufficient to provide 100 TAME units per mL of intestinal phase content. The proteolytic, lipolytic and amylolytic activity of the pancreatin should also be determined. In addition, pancreatin also contains signicant amounts of various salts and given the importance of the Ca 2+ concentration in SIF we suggest that this is also assayed and taken into account when adding calcium to the SIF. It is important to recognise that the concentration of lipase and amylase in the pancreatin will differ from those recommended below and thus for high fat foods this approach may not be appropriate.
Individual enzymes. The alternative to using pancreatin is to use individual enzymes but then which enzymes should be included and how much activity should be used? The primary proteolytic enzymes in the lumen of the small intestine are trypsin and chymotrypsin. Based on an evaluation of values given in the literature we suggest that porcine trypsin is used at 100 U mL À1 nal concentration. 80 The activity is in TAME units where one unit hydrolyses 1 mmol of p-toluene-sulfonyl-L-arginine methyl ester (TAME) per minute at 25 C, pH 8.1, in the presence of 10 mM calcium ions. 82 The equivalence between TAME and BAEE units is: 1 TAME mM unit ¼ 55 BAEE A 253 units or 100 TAME U ¼ 5560 BAEE units. Chymotrypsin should be used at 25 U mL À1 nal concentration. 83 The chymotrypsin activity is in N-benzoyl-L-tyrosine ethyl ester (BTEE) units where one unit will hydrolyse 1.0 mmol of BTEE per minute at pH 7.8 at 25 C. 84 The ratio of chymotrypsin to trypsin is based on the work of Goldberg et al., who showed that the mass ratio of the two enzymes in human duodenal aspirates averaged about 2 : 1 trypsin to chymotrypsin and this corresponds to a 4 : 1 activity ratio. This is based on the activity of trypsin being 135 TAME U mg À1 and the activity of chymotrypsin being 64 BTEE U mg À1 . The main carbohydrate hydrolysing agent is pancreatic amylase that should be added at 200 U mL À1 nal concentration 85 where one unit will liberate 1.0 mg of maltose from corn starch in 3 min at pH 6.9 at 20 C. 86 The most difficult enzymes to accurately quantify in terms of activity are the lipases. We recommend using porcine pancreatic lipase at 2000 U mL À1 (ref. 87) where 1 unit will release 1 mmol of free fatty acid per minute from a substrate of tributyrin at 37 C, pH 8.0, in the presence of 2 mM calcium ions and 4 mM sodium taurodeoxycholate and excess colipase. This amount is based on the mean detection of 0.25 mg mL À1 pancreatic lipase in duodenal aspirates and the activity of the pure enzyme being 8000 U mg À1 . The assay should be conducted according to the recommendations of Carrière and coworkers, 88,89 also available at the CNRS website (http://eipl.cnrsmrs.fr/assay.php?module¼voir&id¼1). In the presence of bile, the rate of lipolysis is signicantly improved by the presence of co-lipase, which facilitates the binding of the lipase to the substrate. The co-lipase binds to the lipase in order to enable the lipase to adsorb to the oil-water interface. Thus, it is important to ensure that the co-lipase is added in a 2 : 1 molar excess with the lipase. This is equivalent to a mass ratio of roughly 1 : 2 co-lipase-lipase as the mass of the similar human pancreatic lipase is 51.2 kDa and the mass of human co-lipase is 10 kDa. Commercially available pancreatin usually contains enough colipase for maximum lipase activity, which can be veried, if necessary, by adding additional colipase in the course of the lipase assay and record changes in lipase activity.
Bile. Bile is important for the transport of the products of lipolysis and in the adult intestine the typical concentration in the fed state is 10 mM nal concentration in total uid. 62 As discussed above, we suggest two options for sources of bile for the intestinal stage, which are either to use a porcine or bovine bile extract 90 or frozen porcine bile, which is easily extracted from the porcine gall bladder. In either case the concentration of bile salts will need to be determined so that in the SIF the concentration is made up to 20 mM, resulting in a concentration of 10 mM in the nal digestion mixture. There are a number of different commercial kits available for the determination of bile that are mostly based on similar methods 91 (e.g. the bile assay kit 1 2212 99 90 313 from Diagnostic Systems GmbH in Germany) that can give a bile concentration in mM.

Suitability of static digestion models
In vitro digestion studies are widely used with the aim of predicting the behaviour of food components in the digestive tract. Most of these studies are performed in static models where gastric and small intestinal digestion is mimicked in two consecutive steps. During each step, the substrate is incubated for a specic time with simulated gastric and small intestinal digestive uids, respectively. The pH is generally maintained at a xed value by using a pH-stat or a buffer. This approach allows methods that are relatively simple to perform and permit high throughput testing. However, the simplicity of static models narrows the range of applicability, which drives the need for adapting a previously described method for a specic research question. This, and the lack of consensus on relevant physiological conditions, has led to a proliferation of different methods. In our consortium we aim to harmonize in vitro systems that simulate digestive processes by dening key parameters and conditions that need to be included to study a specic food or substrate and to measure a specic endpoint. As a starting point, we present in this paper a protocol with a set of standard conditions to be used in a simple static model. These standard conditions are based on relevant in vivo data and supported by rationale and source of data. This discussion focuses on the use and limitation of such a simple static model in relation to mimicking in vivo conditions.

Static models in relation to in vivo conditions
General aspects. The digestive tract is a complex system that aims to provide the body with optimal nutrition and energy. Therefore, feedback systems regulate every step of digestion. The feedback response may differ individually e.g. based on age, physical constitution (status of the body) and habits. This results in both food and individual dependent variation of aspects such as chewing, gastric emptying, secretion of digestive uids and gastro-intestinal transit times. An in vitro digestive system does not include the complex interaction between food and body, which is oen regarded as the major drawback of in vitro simulations. Whether or not this really is a drawback depends on the research question. Control over individual parameter in mechanistic studies for product optimization allows the effects of variation in product composition to be studied under the same conditions. In addition, accurately controlled conditions do not give the high variability oen encountered with in vivo studies, thus limiting the need for large numbers of replicates to obtain sufficient statistical power.
Oral step. Chewing and the consequent particle size reduction is a major determinant of the digestion of solid food. However, the consistency of chewed food, both in terms of particle size and hydration-lubrication with saliva varies widely depending on the type of food and the individual. The use of a food mincer standardizes the particle size and homogeneity of the food bolus but does not include the interaction between food and chewing on the digestion. Static models are not able to mix highly viscous food-saliva boluses as might be swallowed in vivo. Thus, the food needs to be mixed with articial saliva to obtain a sufficient liquid input for mixing in the gastric step.
Gastric step. The function of the stomach is to prepare and gradually deliver the meal to the small intestine in order to optimize further digestion in the small intestine. The meal is stored in the upper part of the stomach and gradually transferred to the lower part where the chyme is mixed and ground until particles are small enough to pass the pylorus. Homogeneity of the gastric content in vivo is generally low. The low level of motility in the upper part of the stomach causes solid ingested boluses to stack on top of each other and more liquid gastric content to phase separate. [92][93][94] Gastric emptying occurs gradually, strongly determined by the caloric value of the nutrients that enter the duodenum. 95 During the gastric phase, the meal is diluted by gastric juice containing enzymes and hydrochloric acid. Pepsin, gastric lipase and swallowed salivary amylase, have their optimum pH at 2.5, 5.4 and 6.8, respectively. 42,96,97 In the fasting state, the pH in the stomach is around 2 or below. During ingestion of the meal, the pH increases depending on the buffer capacity of the meal. Aer that, the pH is gradually decreasing due to hydrochloric acid secretion. The slow penetration of acid in a solid food bolus results in a prolonged high pH in the interior of the bolus. This all implies that during gastric emptying different fractions of the meal are exposed to different pH values and enzyme activities.
Static models use a relatively dilute digestive mixture that is well homogenised using a stirrer, shaker or impeller. Although this does not reect the mixing of gastric content in vivo, it exposes all substrates to the set point pH and related enzyme activities, and allows representative samples to be taken.
The complete meal with simulated gastric digestive uid is exposed to a xed pH during a xed period. Generally the gastric pH is maintained around 2, which may be the right value for the fasting state but does not reect the pH aer intake of a meal. Whether or not a static gastric digestion is adequate depends on the effect of each physiological parameter on the digestion and intended endpoint. In some cases a gastric step could even be omitted since the gastric digestion is completely overruled by the small intestinal digestion. On the other hand, the omission of gastric lipase during the gastric step, as chosen here, might not be fully adequate for mimicking the complete process of gastrointestinal lipolysis as for example preliminary digestion of dietary triglycerides by gastric lipase is known to further trigger pancreatic lipase activity on lecithin-stabilized emulsions in vitro. 98 In other cases incubation at pH 2 during 1 hour might lead to a complete peptic digestion, while this is not the case during a much milder exposure in vivo.
Small intestinal step. In the duodenum, the chyme that is gradually emptied from the stomach is neutralized with bicarbonate and mixed with bile and pancreatic juice. Bile is primarily important to emulsify fat and to form mixed micelles that solubilise and transport lipophilic products to the gut wall for absorption. During transit of approximately 3 hours through the small intestine, substrates and enzyme to substrate ratios are changing due to the digestion and absorption of digestive products and water. The major drawback of small intestinal static models is that they do not include removal of digestive products during the digestion process, which may cause product inhibition of enzymes. This is generally overcome by using non physiological low substrate concentrations in a dilute system.

Use and validation of static models
As with all models, digestive models are a simplication of reality and should be as simple as possible. However, as Albert Einstein stated, "we should make things as simple as possible, but not simpler". This also applies to designing model systems to study the behaviour of compounds in the gastro-intestinal tract. A digestion model should include all relevant parameters to predict the endpoint intended. The more relevant the parameters included are, the wider the applicability but also the higher the complexity.
An accurate prediction of the in vivo bio-accessibility (availability for absorption through the gut wall) is limited since static models lack the simulation of realistic enzyme substrate ratios, pH proles, transit times and removal of digested products, in time and place. Ranking of the digestion of different products is more feasible, provided that the set conditions are adequate for the variation in characteristics of the products. Static models might also be appropriate for mechanistic studies, where the digestion of a substrate under specic conditions is aimed for. The matrix composition of the different products should not differ too much and should be limited in complexity. In other words, static models are useful to study the digestion of single substrates or simple meals under specic conditions.
In addition to the limitations caused by the applied conditions, the assessment of digestion is strongly affected by the analysis of the digested fraction. The fraction of product released should be adequately separated from the undigested fraction. A centrifugation step will only separate insoluble undigested material with sufficient density. Undigested compounds might also be colloidal dispersions. Therefore, ultra-ltration or dialysis may be the better choice. Analysing free glucose, amino acids or fatty acids to determine the digestibility of macro-nutrients is not appropriate, since the pancreatic digestion is not complete. Therefore an additional step with brush border enzymes such as amylo-glucosidase or peptidase is required to complete starch and protein digestion, respectively. Analysis of lipid digestion in a static model is generally performed in a pH stat where the produced fatty acids are assumed to be equivalent to the amount of neutralizing alkali. Product inhibition can be overcome by continuous addition of Ca 2+ ions to precipitate free fatty acids as calcium soaps. 99 In contrast to the more holistic dynamic models that should be validated for their ability to reproduce the conditions in the gut, a static model should be validated against their intended use. In this paper we have described a protocol with conditions and composition of digestive uids that have a broad consensus in terms of physiological relevance. This protocol will be tested and validated by different research groups for a variety of applications to determine its use and limitations. This process will lead to the establishment of key parameters and settings for specic applications and endpoints. This allows model systems to be adapted and validated for specic applications and endpoints by choosing the physiological relevant parameters that have consensus in a big scientic community. This might lead to also using more complex systems but we should "make things as simple as possible, but not simpler".