Over the last 40 years, there has been interest in simulating human digestion in vitro, and specifically protein digestibility, since this knowledge is crucial in different areas, going from the nutritional assessment of novel foods and ingredients to the evaluation of protein allergenicity. However, human digestion and absorption are complex, multistage and adaptable processes in which several factors are involved (12). Thus, in vitro simulation of the digestion and absorption is a technically difficult, if not impossible, task. In this sense, although there are conditions that can be reproduced in vitro, with more or less success, such as, gastrointestinal enzymes, coenzymes and cofactors, pH, or temperature, there are other variables, such as, mechanical forces, regulation by gastrointestinal hormones, action of the intestinal microbiota or the participation of other organs, that are difficult to reproduce (13, 14). Furthermore, several studies have shown that the digestive capacity is adaptable, and for instance, the enzyme release at different levels of the gastrointestinal tract is regulated by the amount of ingested food (15, 16). This aspect, together with the ability of the products of digestion to modulate the intestinal function, adds extra complexity to the digestive process (17, 18). Although gut microbiota plays a critical role in digestion, its effects on protein quality evaluation could be overlooked in an in vitro approximation, since the goal of these methods is to simulate gastrointestinal digestion up to the ileum.

Another important challenge of the in vitro methods to evaluate protein digestibility is the definition of the digestible-, bioavailable- or absorbable-fraction. Because the calculation of in vivo protein digestibility is based on the difference between the amount of each amino acid ingested and that non-absorbed, many in vitro methods have tried to reproduce or approximate these digestible and non-digestible fractions through dialysis, filtration or protein precipitation (14). However, other methods have estimated protein digestibility in the whole digest, like in the pH-drop or pH-stat methods, as will be described later on. However, as it will be commented, these methods based on pH measurement or monitoring were found to be susceptible to the buffering capacity of components of some food materials (19, 20).

In order to improve the evaluation of protein and amino acid digestibility, the resulting digestible and non-digestible fractions can be analyzed by using the same analytical approaches as the ones used in the in vivo assays, i.e., total nitrogen by Kjeldahl or Dumas, and determination of total amino acids by gas chromatography (GC) or HPLC. The different methodologies used for protein quantification can generate discrepancies in the data. The Kjeldahl method is considered the standard method for the estimation of nitrogen in food, because of its universality and good reproducibility in liquid and solid samples. Kjeldahl is a chemical method that determines the nitrogen concentration released during digestion with strong acids (21). Dumas is a high temperature combustion method, and the elemental nitrogen is detected by a thermal conductivity detector. However, because not all of the food nitrogen comes from proteins, these two methods do not give a measure of the true protein. Results obtained with Dumas are usually a little bit higher than those with Kjeldahl due to the detection of nitrogen compounds like nitrates, nitrites and heterocyclic compounds that are not completely quantified by Kjeldahl. In this sense, both methods might need determination of the non-protein nitrogen fraction depending on the food evaluated, and more importantly, an accurate nitrogen to protein conversion factor (NPCF) to convert the nitrogen values into protein. For many foods and ingredients, an overestimation of protein content can result from the use of the standard nitrogen correction factor 6.25 (22), and this will be translated into an underestimated protein digestibility value.

In both, the in vivo and in vitro assays, an acidic hydrolysis with 6 N HCl at 110°C has to be performed for 18–24 h prior to total amino acid determination by GC or HPLC. The effect of this acidic hydrolysis on amino acid analysis was questioned by Darragh and Moughan (23), who showed that the hydrolysis process can generate an erroneous estimate of the amino acid composition. This is due to the presence of amino acids that require times greater than 24 h to cleave the peptide bond (isoleucine, leucine, and valine) and others, considered labile amino acids, which can be partially destroyed before measurement (serine and threonine). In this context, proteins produce different rates of release and loss of amino acids, depending on their amino acid composition, causing an inaccurate quantification. Furthermore, the effect of acid hydrolysis on the chemical integrity of amino acids has been a topic of great scientific interest for years. It is well known that during thermal processing and storage, some amino acids such as methionine, cysteine, threonine, and tryptophan become unavailable, decreasing their bioavailability between 1 and 10%, as in the case of histidine (24). This aspect is especially notable for lysine amino acids, which are easily damaged during the food processing. The ɛ-amino group of lysine can react with many compounds such as reducing sugar, vitamins, fats, polyphenols, generating reactions that produce isopeptides and causing a degradation of lysine. Of these reactions, the most important occurs when, during thermal processing, the amino group reacts with the reducing sugar forming early or late Maillard compounds, which generates a decrease in the availability of lysine. When the Maillard reaction is advanced, the lysine is completely destroyed and cannot be recovered. However, during the early stages of the Maillard reaction, a portion of the structurally altered lysine (Amadori products) is partially hydrolysed in the presence of strong acids that reverse to lysine. However, such reversion does not occur during gastrointestinal digestion (25, 26). This fact leads to an overestimation of lysine quantification in processed foods caused by the acid hydrolysis step that takes place during conventional amino acid analysis. For this reason, the ultimate measure of available lysine is considered the absorbed reactive lysine and new methods, such as the isotope method or the oxidation of an indicator amino acid, have been described (27).

Other methods widely used to measure protein hydrolysis degree are those based on the reaction of primary amino groups, such as the trinitrobenzenesulfonic acid (TNBS) or the o-phthaldialdehyde (OPA) procedure. However, the precision of these methods may depend on the method and the protein substrate being hydrolysed. For instance, several studies have shown that OPA and cysteine react weakly due to the sulfhydryl group of cysteine, generating an unstable product (28, 29). This aspect makes the OPA method unsuitable for quantifying the degree of hydrolysis in cysteine-rich substrates. Furthermore, TNBS or OPA do not react with secondary amino acids such as proline or hydroxyproline (30).

One of the critical points in the in vitro digestion protocols is the selection of the enzymes and conditions (pH, digestion times, and salt concentration) to mimic physiological digestion. To study protein digestibility, proteases of porcine origin that are commercially available, have been widely used, specifically, porcine pepsin for the gastric phase and porcine pancreatic extracts containing proteolytic, lipolytic and amylolytic activities or individual proteolytic enzymes (31). The physiological protease concentrations at different segments of the gastrointestinal tract have been revised to fix conditions in some in vitro protocols (32, 33). In addition, other methods included peptidases of bacterial origin to simulate the carboxy- and amino-peptidase activity of intestinal brush border enzymes (34). It is true that when the food contains a high fat or starch content, an insufficient digestibility of these macronutrients can affect protein digestibility, and amylolytic and lipolytic enzymes are less accessible or are available at high prices. Starch digestion starts in the oral phase and, and although it is inactivated by the low pH in the stomach, some activity may persist within the food bolus. However, oral amylase is not included in most protocols due to its high price. Similarly, lipid digestion starts in the stomach by the action of gastric lipase that reaches activities of ca 120 U/mL in gastric fluid (35). Due to the limited accessibility of human gastric lipase, and taking into account the triglycerol stereospecificity and pH stability, dog or rabbit gastric lipases have been proposed as closest substitutes (36, 37), however gastric lipase is not used in most of the in vitro methods to evaluate protein nutritional quality. Bile salts have also been reported to improve the protein hydrolysis by the action of pancreatic proteases (38), however, not all in vitro protocols include conjugated bile acids in the intestinal phase. More importantly, as detailed later on (Section 4), the key to ensure batch-to-batch and inter-laboratory reproducibility is the standardization of enzymatic activity. Most of the in vitro digestion protocols add a given amount of enzyme or fix an enzyme/substrate ratio (E/S) on weight basis. This adds an important source of variability since the enzymatic activity of commercial enzymes varies enormously from batch to batch and during prolonged or inappropriate storage.

During the simulated gastrointestinal digestion, the use of enzymes, pancreatic extracts or mucins adds a significant amount of protein that depends on the E/S. In a similar way as occurs in in vivo trials, the amount of “added nitrogen” needs to be subtracted in the final calculations of protein and amino acid digestibility. In those methods that separate the digestible and non-digestible protein fractions, this deduction should be done in one or in both fractions. Due to the high degree of autolysis of the digestive enzymes, in particular in absence or in low dietary protein concentrations (39), the use of water or simulated physiological fluids as blank will affect the calculation, underestimating digestibility.

The presence of antinutritional factors (ANF), i.e., lectins, saponins, polyphenols, or trypsin inhibitors, or the fiber content can also influence in vivo protein digestibility. ANF could compromise the protein digestibility by inhibiting the accessibility of the digestive enzymes to the protein or inhibiting enzyme activity (40). Trypsin inhibitors, found in field pea, peanut, wheat, lupin, and soybean, have been demonstrated to be capable of reducing the ability to bind the active site of the enzyme, while lectins may interfere with the digestion and absorption of nutrients (41). Furthermore, polyphenols have been shown to generate a complex with the digestive enzymes, inactivating them and therefore reducing the digestibility of proteins (13, 42). Moreover, fiber consumption has shown several effects on the gastrointestinal tract, from reducing enzymatic activity in the lumen and transit time to protecting the enzymes against degradation or stimulating the microbiome activity in the digestive tract. However, the effect of these compounds in the in vitro assays is still unknown and will depend on the conditions, especially the E/S used, which makes it difficult to simulate in vivo digestion with in vitro assays (43).

During protein hydrolysis, release of protons and amino acids from the cleaved peptide bonds results in changes in pH. These methods lie on the correlation between the rate of hydrolysis degree and protein digestibility. Hsu et al. developed a multi-enzyme method (three-enzyme method, trypsin + chymotrypsin + peptidase) for the evaluation of protein digestibility. Specifically, they showed that the pH-drop after 10 min of digestion with the three-enzyme solution of 23 human diets, mainly vegetables and dairy foods, was highly correlated (r = 0.90) with in vivo PER values in rats. However, substances with high buffering capacities could affect the results. Despite this, the pH-drop methods was able to predict the apparent digestibility of proteins. In addition, the trypsin inhibitory activities and the effect of heat processing on digestion could be detected (19). Two years later, Satterlee et al. slightly modified the Hsu et al. protocol by including an extra 10 min of digestion with a Streptomyces griseus protease (four-enzyme method). Numerous authors have evaluated the in vitro digestion process of several food sources through the pH-drop method. In 1981, Petersen and Eggum studied the applicability of the three-enzyme (19) and four- enzyme combinations (65) on 61 samples of food and feed. Their results demonstrated a high correlation (r = 0.89–0.90) between the pH-drop results and fecal protein digestibility in rats, especially for plant proteins and for mixtures of plant and animal proteins. However, the predicted in vitro digestibility value for animal proteins significantly differed from the in vivo results (46). In 1983 the same researchers evaluated the in vitro protein digestibility of 18 protein sources using the three-enzyme method of Hsu et al. and the four-enzyme method of Satterlee et al. The results showed a greater in vitro-in vivo correlation for the three-enzyme method (r = 0.78) than for the four-enzyme method (r = 0.56). Despite the good correlations obtained, the estimations were significantly affected by the different buffering capacities of some food substances, which was considered a major drawback of the pH-drop method (49). This aspect was further demonstrated by Moughan et al. who compared the in vitro digestibility of 20 meat and bone meal samples by the pH-drop method with the values of true ileal digestibility in rats. It was concluded that pH estimation methods may be influenced or affected by the strong buffering capacity of the ash content, mainly mineral content, of food (20). For this reason, it was recommended to determine the pH-drop after a dialysis treatment to eliminate salts with buffering capacities (20). Other authors such as Kim et al. and Wolzak et al. showed in vitro-in vivo correlation values of r = 0.95 and r = 0.421 for soy protein concentrate, and 33 vegetable proteins, respectively (47, 48). However, the difference in the response for different types of food proteins made it necessary to use different regression equations to obtain realistic estimates of digestibility. This task presents a major challenge due to the complexity involved in categorizing foods in each class of food.

Pedersen and Eggum revised the pH-drop method and modified it slightly in order to avoid the effect of substances present in the protein that could influence the drop in pH. In the pH-stat procedure, the pH is kept constant at pH 8 by automatic titration (0.10 M-NaOH titrant) during the incubation with enzymes. At the end of the incubation period, the amount of alkali added is recorded and the value is used as an indirect measure of protein digestibility (43, 46). Using pH-stat, Pedersen & Eggum showed an improvement in the prediction of protein digestibility of 30 samples, compared to the pH-drop. A high correlation coefficient (r > 90) with fecal digestibility in rats was obtained in pH-stat method, improving the one obtained by the pH-drop method (0.56–0.78). However, the digestibility of some foods, such as egg powder, was underestimated, because of the content of trypsin and chymotrypsin inhibitors of egg. A pretreatment with alkali to improve the correlation coefficients was then recommended (49). The pH-stat method has been widely used to evaluate the digestibility of different protein sources. Eggum et al. showed a good agreement in vitro vs. in vivo (measured by fecal protein digestibility in rats) in 17 foods, with the exception of two legumes, beans and chickpeas. The authors discussed that the discrepancies obtained for these foods, suggesting that it could be due to the high bacterial growth with the consumption of certain legumes in the diet, which caused an increase in the excretion of nitrogen in the feces. Excluding these legumes the obtained in vitro and in vivo digestibility percentages were similar (86.3–100.0% in vitro and 73.1–96.8% in vivo), although the correlation coefficient was only acceptable (R² = 0.66) (16). Better correlation coefficients were obtained by comparing the in vitro protein digestibility of maize, whole sorghum and pearled sorghum maize (r = 0.95) and 7 specific foods (R² ≈ 0.75) with their in vivo apparent fecal protein digestibility (50, 52). In the same way, good and significant correlations (R² = 0.82 and 0.64) were obtained when the protein digestibility of 10 salmonid diets were estimated by two the pH-stat in vitro assay methods and compared with in vivo digestibility in fish (51). Linder et al. measured the protein digestibility of an industrial veal protein hydrolysate, used as a gelatin-replacing ingredient for human consumption. The results showed a high correlation between fecal protein digestibility measured in rats and the pH-stat method (R² = 0.99), although already published regression equations were used (53). Recently, high correlation coefficients were obtained between in vitro protein digestibility of processed soybean meal and rapeseed meal through the pH-stat method and the results obtained from standardized ileal digestibility in growing pigs (r = 0.95) (54).

In summary, the methods based on pH measurement were shown to be suitable for predicting digestibility in many foods, with high correlations in plant substrates. The method was found to be highly reproducible across 6 laboratories that estimated protein digestibility of 17 protein sources by using the 3-enzyme method in a pH-stat (66). However, by using these methods, the results of the entire complex digestion process were evaluated based on a mere measurement of pH or pH-change. In other words, the crucial information on protein digestion that could be extracted from the use of gastrointestinal enzymes was neglected. In addition, the significant differences found for animal proteins, the use of different correlation curves for different samples, and the fact that certain physical and chemical characteristics, such as calcium content or buffering capacity, may prevent an accurate estimation of digestibility, and are important drawbacks for the use of these methods.

In the methods described in this section, enzyme incubations are followed by measurements of the insolubilized material collected after filtration, although in some cases measurements on the filtrate, or alternatively on one of the separated fractions after centrifugation are conducted. Digestibility is then related to in vitro solubility or the definition of an absorbable or bioaccessible fraction and a residue or non-absorbable fraction.

Early methods included one-step incubations giving lower digestible protein values than those obtained in vivo, and were rapidly replaced by two-step digestion. In the pepsin-jejunal fluid, a two-step incubation with pepsin digestion for 4 h followed by a further 4 h digestion with pig jejunal fluid was used (67). In vitro digestibility of protein was calculated by the determination of dry matter and crude protein on the residue after centrifugation for 10 min at 1,250 × g at 5° C, on the basis of the original protein content of the diet. A two-stage incubation with pepsin for 6 h at pH 2 followed by an incubation with pancreatin at pH 6.8 for 18 h in borate buffer was further developed (68) thus providing an animal-independent method. To calculate the digestibility, 1% TCA final concentration was used, followed by centrifugation for 1 h at 2,000 × g. This method was applied for routine analysis in quality control of feeds and feed ingredients for poultry. By reducing the particle size of the test material, passing through a 0.4 mm sieve, the accuracy of predicting in vivo digestibility was increased for all the tested diets, that included corn, barley, oats, soybean, corn gluten and wheat bran as protein sources. The correlations between ileal digestibility in broilers and in vitro estimates were high (r = 0.93 for crude protein) (55). A modification of this method was presented by Babinszky et al., where pepsin incubation was performed at pH 1 on fat-extracted feed samples, and the residue after pancreatin + amylase incubation and 1% TCA precipitation was centrifuged at 3,500 × g for 15 min, after decanting over a nylon cloth (particle size 40 μm). This method found an improved correlation with fecal digestible protein in pigs by reaching regression values for feedstuff and diets of 0.99 and 0.95, respectively (56). The additional determination of nitrogen content on the filtrate gave a similar correlation but was abandoned as it was considered to be too laborious.

With the aim to recover solubilized but not fully degraded proteins, precipitation with sulphosalicylic acid was introduced, while undigested materials were submitted to the standardized filtration equipment for measuring dietary fiber (43). Magnetic stirring was included during the enzymatic incubations in order to assure effective starch degradation. By the use of this method, prediction of individual amino acids in eight common feedstuffs showed that the in vitro digestibility of the individual amino acids was close to the in vitro digestibility of nitrogen. Hervera et al. (58) adapted the last in vitro method for estimation of digestible energy of dog foods and compared it with the pH-drop methodology. The results showed a correlation of r = 0.78 between the pH-drop with the three-enzyme method and the apparent fecal digestibility in dogs but higher accuracy, r = 0.81, was shown with the in vitro method using precipitation with sulphosalicylic acid (58). Wada and Lönnerdal determined digestibility in infant formulas by using total and non-protein nitrogen (NPN), i.e., soluble fraction in 12% final TCA concentration (69). They investigated the effect of industrial processing with in vivo digestibility using a suckling rat pup model in terms of chemical modifications and endurance of intact α-lactalbumin and β-lactoglobulin, but no direct in vivo-in vitro comparison was shown.

The role of nitrogen added in the form of enzymes was considered in further developments. When the in vitro digestibility of protein was calculated from the difference between nitrogen in the sample and the undigested residue after correction for nitrogen in the blank, it was shown that the resulting amino acid composition of the blank-derived protein was very close to reported values in the literature based on direct measurements of endogenous protein in vivo. Apparent ileal digestibility of individual amino acids was predicted in a similar way as for protein. The relationship was generally higher for essential amino acids, and generally lower for non-essential amino acids, than for protein (57). This procedure used precipitation with sulphosalicylic acid (2% final concentration) for 30 min at room temperature followed by rinsing with 1% sulphosalicylic acid of the filtered residues. A close relationship was found for the 17 single feedstuffs but meat and bone meal, and barley hull had to be excluded. The above conditions have been widely used to compare protein digestibility of different products, mainly using sulphosalicylic acid (62, 70) or TCA (71, 72) as precipitating agent.

In summary, the in vitro methods based on precipitation of a non-digestible fraction by using different agents such as TCA or sulphosalicylic acid have demonstrated good comparability with ileal digestibility in broilers and in pigs. Precipitation with sulphosalicylic acid after a 3-enzyme digestion protocol has shown higher accuracy than pH-drop when compared with dog fecal digestibility. The main advantage of these methods is the reproducibility of the precipitation step for the definition of a digestible and non-digestible fraction. However, the digestion conditions used by different authors would still require additional optimization and harmonization.

These methods are based on the continuous removal of low-molecular-weight products from digested material by ultrafiltration or dialysis to prevent enzyme inhibition by end products.

A two step-digestion method in which the intestinal digestion products (free amino acids and low molecular weight peptides) were removed through a dialysis membrane was proposed in order to reduce enzyme inhibition by hydrolysis products (73). After a 30 min digestion step with pepsin enzyme: substrate of 1:250 (pepsin activity 3,152 units/mg protein), intestinal digestion took place with pancreatin for 6 h, at an E/S of 1:25, in a dialysis cell of a 1,000 Da molecular weight cut-off, for the continuous elimination of digested products with 10 mM sodium phosphate buffer, pH 7.5, as circulating dialysis buffer. The essential amino acids released during the intestinal phase from beef, casein, rapeseed, soybean and gluten correlated with plasma levels found in portal and aortic blood in rats fed with the same substrates (59). A good correlation was found for plant sources (r = 0.92–0.93), although lower values were reported for beef or casein (r = 0.70). This protocol was also applied to protein mixtures (74) and to 19 selected foods, showing differences between in vitro-in vivo amino acid digestibility depending on the protein source. These variations were not related to the amino acid concentration in the protein and it was proposed that the amino acid sequence as the factor leading overall protein and amino acid digestibility (75). The in vitro protein digestibility by use of a dialysis cell method and pH stat was compared with the in vivo PER and true digestibility of heated rapeseed meal, soybean and lupine proteins (60). In vivo, PDCAAS correlated with pH stat and dialysis cell values with r = 0.92 and 0.98, respectively, although PER was poorly correlated with the in vitro protein digestibility. Similar strategies but using dialysis tubes in the intestinal phase have been used to predict ileal protein digestibility of pig feedstuffs (61), obtaining linear regression equations between in vitro digestibilities and porcine ileal apparent digestibilities. Dialysis cells have been more recently used in the estimation of the protein digestibility of novel food protein sources, such as seaweeds, where the high fiber content affected protein digestibility, likely by reducing the accessibility of the proteolytic enzymes (76, 77).

In some works, the use of chromatography or ultrafiltration with different cut off membranes has been used to characterize the digestible fraction. Besides, the characterization of the non-dialyzed digest has been conducted by ion-exchange or size exclusion chromatography, and ultrafiltration. The undigested residues were separated by ion-exchange chromatography into basic-neutral, lightly acidic and acidic fractions further resolved by sequential ultrafiltration (cut-off 10 and 1 kDa). Interestingly, large proportions of leucine, lysine, arginine, phenylalanine and tyrosine were found as part of peptides smaller than 1 kDa, both in the dialysates and retentates, while glutamine, threonine, serine and asparagine appeared mostly in fractions >1 kDa, while after 6 h with pancreatin, most of the proline appeared in the basic-neutral fraction >1 kDa (78). When this procedure was applied in the comparison of casein, cod, soy and gluten proteins, animal proteins were digested at a greater rate than plant proteins, and more resistant peptides were largely rich in proline and glutamic acid (79).

The impact of cooking on animal and plant protein digestion has been evidenced by the use of this strategy. The increase in protein digestibility of white and brown beans (Phaseolus vulgaris) after cooking was found to be related to a higher extent of proteolysis, as monitored by SDS-PAGE and recovery of low molecular weight peptides (< 30 kDa) after ultrafiltration of the digests (80). On the contrary, meat protein digestion in a microreactor fitted with a 10 kDa cut-off membrane in the gastric compartment and 1 kDa cut-off dialysis membrane in the intestinal compartment showed a decrease of protein digestibility with meat cooking (81). This study showed superior precision with the use of a semi-automatic flow procedure in comparison with the test tube method. Analytical size exclusion chromatography has been used to determine digestibility of casein vs modified casein with the glycation product pyrraline. The size pattern was used to show that the digestibility decreased with increasing pyrraline concentration of the peptide mixtures. Moreover, further ultrafiltration of digests using 1 kDa cut-off indicated that 50–60% of pyrraline was included in peptides (82).

The methods based on in vitro digestion and dialysis or ultrafiltration have shown good correlation in the prediction of ileal protein digestibility of food and feed. Some of these approaches have been used to characterize the gastrointestinal digests in combination with chromatographic methods. Main weaknesses of these methods would derive from the limited reproducibility of the use of ultrafiltration devices and the unspecific bound of protein material to the membrane material.

Dynamic systems have been proposed as in vitro alternatives for human or animal studies as physiologically relevant, including peristaltic mixing of food, computer- controlled pH values and realistic gastrointestinal transit times. Moreover, small molecules are removed from the digesta with hollow fiber membranes. The TNO-developed TIM^® system was tested to predict the true ileal digestibility of proteins including dairy, meat, wheat, faba bean or barley, and a linear relationship versus pig or calf data was obtained (83). A standard corn-based diet was compared with the same diet with coarse ground corn, 8% sugar beet pulp, 10% wheat bran, or 8% sugar beet pulp and 10% wheat bran. The dynamic model yielded digestibility coefficients comparable with in vivo ileal digestibility in growing pigs for the standard and coarse ground corn but the values were considerably affected by the incorporation of the fibrous ingredients. The linear fitting between the in vitro and the in vivo results for crude protein digestibility was not significant but resulted in R² = 0.99 when the coarse ground corn diet was excluded from the regression (63).

Using the tiny-TIM, digestibilities of ovalbumin, cooked and raw chicken egg white, and casein showed similar values to values reported in humans (R² = 0.96). The true ileal protein and amino acid digestibilities were used by the authors to estimate the DIAAS for immature herring egg proteins (84). More recently, cumulative true ileal digestibility of nitrogen data has been reported during 5 h tiny-TIM, expressed as the percentage of the exogenous nitrogen intake, correcting for nitrogen in gastric residue (85). These values served to calculate DIAAS for different protein ingredients, alongside the corresponding limiting amino acid. The DIAAS values for rice, whey, and pea-based proteins were in agreement with those collected from literature, using pig ileal data. However, for soy and a second source of pea protein with different processing, the values were significantly lower than those previously described in literature. This was ascribed to treatments applied to these specific ingredients during processing, including alkaline or heat treatment, leading to protein aggregation or structural changes. An alternative source of discrepancy was related to differences in the innate protein features due to cultivar of growing conditions. A low (under 50%) bioavailability of the majority of amino acids and low N digestibility was found for the last two products. Isolates with lower DIAAS also showed lower protein solubility and increased protein aggregation, which was identified as a potential cause inhibiting digestion. Indeed, DIAAS positively correlated to protein solubility and N-bioaccessibility. The dynamic system developed at INRAE, DIDGI^® was set up to mimic infant digestion upon an extensive analysis of literature on infant physiology and validated with piglet digestion (86). This system provided comparable results in vitro/in vivo for a reference dairy infant formula in terms of limiting essential amino acid, PDCAAS-like score and in vitro apparent digestibility. The last parameter was determined based on the soluble N lower than 10 kDa, as measured in the peptides by size exclusion chromatography and cumulated to the free amino acid nitrogen (64).

The use of dynamic systems to determine protein digestibility is still limited. Although these systems allow monitoring the progress and digestion kinetics, the calculation of the nitrogen mass balance could be more complex than in static systems. In addition to the difficulties to harmonize conditions in different apparatus, the availability of this sophisticated equipment could be an additional limitation to the extensive use of these methods.

The original INFOGEST protocol proposed for each digestion step a 1:1 ratio (w:w) between food and simulated fluid, ending up with a final ratio of 1:8 of food in digesta. No recommendation of nutrient normalization was proposed. However, in order to compare protein digestibility of different foods, a normalization may be needed. In four of the listed publications, protein input was normalized between 4 and 16% in the foods subjected to digestion. Increasing the amount of protein entering into the system reduced digestibility, as was observed in the case of different amounts of TCA soluble casein after size exclusion chromatography (91) and for casein and gluten digestibility, testing 4, 8, and 16% of protein input (94). Another approach to increase the food to enzyme ratio is the adaptation of digestive enzymes as proposed by Ariëns et al. (90) to reduce the background of enzymes. The authors reduced the addition of pancreatin from 100 U/mL of digesta by a factor of 10 to 10 U/mL and observed no impact on released NH₂ during digestion of whey protein isolate, which represents a highly digestible substrate. It would be interesting to test if this observation is also correct for substrates with lower digestibility. Alternatively, a reduction in enzyme background was achieved by Sousa et al., by using the supernatant of the pancreatin suspension after solubilization with ultrasound and subsequent centrifugation (92). This procedure did not reduce the trypsin activity in the pancreatin supernatant.

At the end of the intestinal phase of the original INFOGEST protocol, all products are in the same container. In order to assess protein digestibility, digestible and non-digestible fractions need to be separated. Different approaches were used by various authors, such as centrifugation (95), or ultrafiltration at different cut-off sizes, such as 5 or 10 kDa (90, 94), corresponding to peptides of 45–90 amino acids in length, assuming an average weight of 110 Da per amino acid. The choice of the rather high molecular weight cut-off compared to in vivo (500 Da) was justified by the lack of brush border enzymes in the system (94). Moreover, the use of ultrafiltration could as well lead to loss of material, impacting the mass-balance and in consequence the digestibility of the tested substrates (90). A second approach applied in the different protocols was a precipitation step either with different concentrations of TCA (6–12%) (91, 93) or with MeOH (80%) (92). Depending on the downstream analysis, the precipitation agent could disturb the measurements and it was removed by extraction with diethyl ether (93) or simply be evaporated in the case of MeOH (92).

In both, in vivo and in vitro situations, the endogenous enzymes and background proteins need to be considered. Different solutions to this challenge have been suggested. Probably the most precise way of differentiating endogenous material from the food of interest represents the use of isotopically labeled food sources as has been used by Ménard et al. (94). In this study, ¹⁵N isotopically labeled casein and gluten were digested at different concentrations. Unfortunately, the generation of isotopically labeled substrates is not always possible, in addition to being time consuming and expensive, and therefore other solutions are requested. However, for experiments of proof of principle and validation, isotopic labeling would be the method of choice. As an alternative, the enzyme background was subtracted by performing a parallel digestion with H₂O (90, 93–95) or using a protein-free food (92). However, as explained in Section 2, in the absence of substrate, a higher enzyme autolysis may occur (39, 94), which would cause an underestimated digestibility value.

Comparisons between in vitro and in vivo data were performed by four of the above-mentioned publications (Table 2). A high correlation between in vitro and in vivo DIAAS, as well as an agreement in limiting amino acid (DIAA) was demonstrated for the investigated substrates, although the in vitro digestibility values were below in vivo digestibilities in these studies (93, 95). In the same direction, a lower true in vitro digestibility value compared to in vivo was found for the two investigated substrates, casein and gluten (94). A direct comparison between in vivo and in vitro digestibility was performed for seven substrates (Figure 2, WPI (A) and black bean (B), representing two of the investigated substrates). The results showed a comparable digestibility with an average bias of 1.2% for all essential amino acids of the assayed substrates (Figure 2C). The DIAAS values were comparable with an average bias of 0.1% and a correlation of r = 0.96 between in vivo and in vitro results (92). It has to be highlighted that this latter study was carried out in vivo and in vitro by using identical substrates. In view of the published data, an increased number of substrates of different nature is needed to validate this in vitro model for digestibility.

Until recently, a major drawback of in vitro methods was the lack of comparability between different laboratories. In consequence, one of the major achievements of the INFOGEST network was to establish harmonized digestion protocols with satisfactory inter-laboratory reproducibility. Within the same INFOGEST network, protein digestibility is currently tested with several dairy products (SMP, whole milk powder, whey protein isolate, yogurt, and gruyere cheese) and with two plant sources (soy protein isolate, chickpea), applying the analytical workflow published by Sousa et al. (92). In parallel to these collaborative studies, a standardization of the method within the International Dairy Federation (IDF) and International Standardization Organization (ISO) was launched. The precision data (repeatability, reproducibility) obtained in the inter-laboratory trials will be included in the future IDF/ISO standard method, with the final goal to obtain a robust and validated protocol allowing the analysis of protein digestibility.