A feed is still only as good as its ingredients: An update on the nutritional research strategies for the optimal evaluation of ingredients for aquaculture feeds

A variety of strategies can be used to assess the nutritional quality of ingredients; however, the choice of strategies used can have a strong impact on the interpretation of that information. A landmark review paper published over a decade ago has been considered the benchmark approach by which to structure research assessing the quality of ingredients (Glencross, Booth, & Allan, 2007). In that review, a series of five steps (and the order in which they should be done) to develop a comprehensive data set on which to base judgements about ingredient quality were proposed; 1. Characterization, 2. Palatability, 3. Digestibility, 4. Utilization and 5. Functionality (processability). Once these five steps had been achieved, a formulator could make the judicious choice as to whether to use, and with what constraints to impose, any ingredient that they were presented with. Without any one of these steps, the risk exposure substantially increased as the formulator needed to make assumptions, and this significantly increased the risk of a feed failing in one or more specifications (Glencross et al., 2019). Typically, many studies in this domain have skipped many of these early steps and gone straight to the assessment of utilization (step 4). However, in doing so, many of these studies have ended up with erroneous outcomes and/ or misleading assessments of the ingredients that they are testing, not Received: 28 February 2020 | Revised: 29 May 2020 | Accepted: 15 June 2020 DOI: 10.1111/anu.13138


| INTRODUC TI ON
A variety of strategies can be used to assess the nutritional quality of ingredients; however, the choice of strategies used can have a strong impact on the interpretation of that information. A landmark review paper published over a decade ago has been considered the benchmark approach by which to structure research assessing the quality of ingredients (Glencross, Booth, & Allan, 2007). In that review, a series of five steps (and the order in which they should be done) to develop a comprehensive data set on which to base judgements about ingredient quality were proposed; 1. Characterization, 2. Palatability, 3. Digestibility, 4. Utilization and 5. Functionality (processability). Once these five steps had been achieved, a formulator could make the judicious choice as to whether to use, and with what constraints to impose, any ingredient that they were presented with. Without any one of these steps, the risk exposure substantially increased as the formulator needed to make assumptions, and this significantly increased the risk of a feed failing in one or more specifications . Typically, many studies in this domain have skipped many of these early steps and gone straight to the assessment of utilization (step 4). However, in doing so, many of these studies have ended up with erroneous outcomes and/ or misleading assessments of the ingredients that they are testing, not due to any limitations of the ingredient per se, but rather failure of the researcher to observe critical formulation constraints that allow the ingredient to be assessed on a basis commensurate with its potential to supply nutrients and energy. The remainder of this review will provide an oversight and some guidance on what steps should be undertaken to assess quality of ingredients and why those steps should be taken in that order and critically, what new steps should be considered for the assessment process in light of recent advances in the science.

| Step 1-Characterization
The characterization of ingredients is an often overlooked, but critical step in the evaluation process. As such, this initial step in the ingredient evaluation process remains as important as ever. For formulators to make use of technical documentation on ingredients, the users of that information must be able to relate the data to a particular type of ingredient. However, there is substantial variability in both the composition and nutritional values (as defined by the digestibility of nutrients from the ingredient) of most ingredients (Anderson, Lall, Anderson, & McNiven, 1995;Glencross et al., 2008a;b). Details on the species, origins, processing and/or storage history, let alone a chemical characterization, are often absent in a much of the scientific literature. Any chemical characterization needs to include, as a minimum, the basic parameters used to formulate feeds and/ or allow clear assessment of the ingredient. An example of what a standard characterization should include, in this case of five different fishmeals, is presented in Table 1. Clearly for other types of ingredients, such as plant-derived materials, there would also be some merit in including data on parameters such as starch, non-starch polysaccharides, acid-detergent and neutral detergent fibre, and lignin (Petterson, Harris, Rayner, Blakeney, & Choct, 1999). The methods for analysis should follow standardized methods such as those recommended by analytical associations such as AOAC, UKAS or similar. Without some form of characterization, the value of the work is somewhat diminished as it becomes difficult for the reader/user of the data to effectively relate the work to their materials. By providing a comprehensive characterization, it becomes much easier to relate the assessment to other materials and/or obtain the same material.
Another element to characterization that is gaining importance, and over time may warrant its own step, is an assessment of the sustainability of an ingredient. Various strategies have been examined to define sustainability of ingredients, but life cycle assessment is perhaps the approach gaining most favour (Boissy et al., 2011;Malcorps et al., 2019;Silva, Valente, Matos, Brandão, & Neto, 2018).

| Step 2-Palatability
Before the impact of the nutrients within a feed can be measured on animal performance, the animal clearly must ingest that feed. A decision hierarchy framework has been suggested as a means of defining the nature of the responses to feeds to aid defining how the feed is specifically impacting the response (Figure 1). In this figure, the influence of a feed (and by inference its ingredients) on an animal's response can be assessed to define the specific aspect of palatability that is affected (and note that this response might be positive or negative). It is clear that any factor(s) that negatively impact this hierarchy are going to limit the potential of the ingredient. Therefore, one of the qualities of any ingredient that is critically important to a feed is its effect on palatability of the feed. Clearly, if an ingredient reduces feed intake due to negative effects on palatability, it has some limitations as a potential feed ingredient. Conversely, those ingredients that can stimulate intake, and thereby improve palatability, have added value as ingredients. It can be seen from some studies how this variation in palatability can affect greater than 80% of the variability in growth response to diets trialling alternative ingredients ( Figure 2) Kousoulaki et al., 2018). In studies where fish are fed a fixed ration, it becomes impossible to assess the impact of diet (and by extension the test ingredient) on palatability responses. It should also be noted that effects on palatability of diets can often be detected within days of introduction and are usually at their most sensitive point of assessment within the first 10 days (minus days 1 to 3) of an animal being fed that new diet ( Figure 3). After this period, the animal may begin to adapt to the diet and the ability to discriminate palatability effects accordingly becomes diminished (Arndt, Hardy, Sugiura, & Dong, 1999;Glencross et al., 2006Glencross et al., , 2016Kousoulaki et al., 2018;Suresh & Nates, 2011, Nunes et al., 2019.

| Step 3-Digestibility
The assessment of the digestibility of nutrients and energy from diets and ingredients provides one of the clearest ways of unambiguously defining the nutritional value of an ingredient to an animal (Aslaksen et al., 2007;Glencross et al., 2004;Sugiura, Dong, Rathbone, & Hardy, 1998). By measuring the relative disappearance of nutrients between the feed and faeces, those nutrients available to an animal can be effectively measured. It is these nutrients that are used to drive feed intake (in response to digestible energy demands) and underpin growth. Therefore, variability in this digestibility is one of the critical points of variability in feed quality (Aslaksen et al., 2007;Bureau, Harris, & Cho, 1999;Glencross, Blyth, Wade, & Arnold, 2018;Glencross et al., 2008b). There are a range of methods that can be used to assess diet digestibility, and these can all potentially influence the absolute values determined in any given assessment. The important element to the reliable and consistent assessment of digestibility is to minimize non-nutritional effects (i.e. minimize effects of environment or sampling method).
For a comprehensive review of the range of strategies and their limitations, see Glencross et al. (2007).

| Step 4-Utilization
The assessment of nutrient and/or energy utilization is mostly commonly undertaken by a growth/feeding trial approach (Glencross TA B L E 1 An example of the characterization of different types of fishmeal. All values g/kg as received unless otherwise stated  Glencross et al. (2007).
While these trials are the most commonly encountered among the literature, their outcomes are largely predictable subject to diet specification choices, digestibility of ingredients used and the palatability of the diets produced. In those situations where the diets are formulated to the same specifications and on a digestible nutrient basis, the growth responses that result from the feeding study are usually largely just a reflection of feed intake (palatability) variation. Figure 2 shows a typical such response from the evaluation of a suite of alternative ingredients fed to Asian seabass (Lates calcarifer), where the feed intake alone accounts for over 80% of the variation in the growth response. Occasionally, unexpected responses do occur, and this is arguably where the true value of the utilization Each of the diets focussed on the assessment of a particular ingredient when included in diets balanced for digestible protein and digestible energy. Clear effects of each diet on palatability were noted. This effect of palatability resulted in greater than 80% of the variation in weight gain response being explained by vagaries in feed intake. Data derived from F I G U R E 3 Variation in daily feed intake in different treatments of fish (rainbow trout, Oncorhynchus mykiss) fed diets with a 30% inclusion of each of two different cultivars of yellow lupin kernel meal compared to a diet using fishmeal as its only protein source. All diets were formulated to the same digestible protein and energy specifications. Data derived from Glencross et al. (2006) scenario. In the crude specification scenario, it would be reasonable to suggest that the alternative ingredient causes a decline in performance at all inclusion levels and therefore should be considered a low-grade ingredient. By contrast, the digestible specification scenario shows no response in performance to the use of the alternative, and therefore, it would be reasonable to suggest that the alternative ingredient is well utilized and tolerated at all inclusion levels and therefore should be considered a high-grade ingredient. Of course, between each scenario the ingredient is the same, only the formulation strategy has changed.
When diets are formulated to non-constraining specifications it can be clearly seen that as the diet protein level increases that the differences between the responses to the different fishmeal begin to diminish. This difference between the three fishmeals continues to become negligible as the diet protein level increases up to a point at ~550 g/kg protein content where it would not be possible to discriminate among the three fishmeals in terms using a growth study approach. Important to note is that with the size of fish used in that study (7.5 25 g), the typical commercial diet specification is between 500 g/kg and 550 g/kg protein. As such, the use of a standard commercial diet formulation approach to assess quality in this example would have resulted in a null-hypothesis outcome and no ability to discriminate between the three fishmeals. While the relevance of a commercial style specification might be argued, the lack of sensitivity in using such a specification also needs to be considered.
Sometimes a commercial analogy might not be the most appropriate strategy to answer a nutritional question.

F I G U R E 4
Theoretical outcomes of two different series of diets formulated on either a crude specification or digestible specification basis F I G U R E 5 Effects of fishmeal quality and diet protein level on relative performance (weight gain) by Atlantic salmon. Data derived from Anderson et al. (1993)

| Step 5-Immunological and health allied assessments
A growing priority in the assessment of feeds and ingredients in animal diets is their impact on the immune response and general robustness (health) of the animal. There are a variety of strategies that are being used to examine these parameters, including specific pathogen (disease) challenges, which can be undertaken to assess the impacts of diets on the immunogenic function of fish against a particular patho-

| Step 6-Processing Effects (Functionality)
Physical constraints play an important role in the production of a viable feed with which to feed aquatic species. The fundamental logistics of handling ingredients play an often-overlooked role in the ingredient assessment process. Most modern aquaculture feeds are manufactured using extrusion processing, where the rheology of a given feed mix is managed to allow the plasticization of the components to produce a well bound and durable product that can be tailored to float or sink and contain a high-or low-oil content (Oterhals & Samuelsen, 2015;Samuelsen & Oterhals, 2016;Sørensen, 2012).
Some of those physical characteristics that are assessed are relatively simplistic but remain fundamental to managing the logistics of using ingredients. Parameters include those such as the flow-figure, which defines the ability of a powder/meal to flow and/or be conveyed. The bulk density, which defines storage demands and hygroscopicity which defines how much moisture the product absorbs from liquid or vapour exposure, is all critical to effective management of ingredients (Samuelsen, Mjøs, & Oterhals, 2013. Typically, assessment in this area involves the practical production of feeds using laboratory or pilot-scale extrusion systems (Draganovic, van der Goot, Boom, & Jonkers, 2011;Glencross, Hawkins, Maas, Karopoulos, & Hauler, 2010;Opstveldt et al., 2003;Samuelsen, Mjøs, & Oterhals, 2013. Variables such as the type of ingredients used and/or their inclusion levels are the main variables being tested in such science, and a range of parameters are subsequently assessed to provide information on the physicochemical properties of the ingredients on the extrusion process (e.g. specific mechanical energy input and expansion among others). The feed pellets produced from such studies usually then go on to other testing (e.g. density and oil absorption capacity) to assess a range of physical parameters important to the feed and feeding management process (Sørensen, 2012). There are other assessments which can now be used to provide alternative assessments of the extrusion process, including phase-transition assessment and rapid-viscosity assessment (Glencross et al., 2010;Oterhals, Ahmad, & Samuelsen, 2019;Samuelsen & Oterhals, 2016). An important feature of these physicochemical properties of ingredients is that they clearly play a role in impacting the physical qualities of feeds. Notably, these physical feed qualities can also directly impact feed intake and nutrient utilization and that this also should be taken into consideration when designing nutritional trials (Sørensen, 2012).

| Step 7-Product Quality Influences
In most cases, the animal species being fed are intended as a food product. Therefore, the sensory qualities (colour, taste and smell) are important criteria in determining quality of those products.
Accordingly, the use of sensory evaluation studies is sometimes included to evaluate the impact of different dietary treatments, including the use of ingredients (Rosenlund, Obach, Sandberg, Standal, & Tveit, 2001;Turchini et al., 2003;Liu et al., 2010). There are various elements to this evaluation, but principal among them is the evaluation of flavour and/or colour (Bjerkeng et al., 1997;Lie, 2001;Wade, Paulo, et al., 2014).
Flavour as a parameter has a large degree of complexity to it given that there are components defined by the tongue senses (taste: salt, sweet, acid, bitter and umami), and there are those components defined by the olfactory senses (smell). The evaluation of these sensory characteristics is typically undertaken using a range of strategies including the use of tasting panels, but increasingly the use of instrumentation is being used to assess colour, odour and texture (Chytiri, Chouliara, Savvaidis, & Kontominas, 2004;Liu, Zeng, & Sun, 2013;Olafsdottir, Högnadóttir, Martinsdottir, & Jonsdottir, 2000;Olafsdottir, Martinsdottir, & Jonsson, 1997;Olafsdottir et al., 2004). Sensory data are commonly considered qualitative in nature rather than quantitative. The advantage of using an instrumentation approach is that the data can be quantitative rather than qualitative. For some aspects of sensory evaluation, modern instrumentation analyses are even considered to be more sensitive than human senses and therefore provide a greater level of assay fidelity. Although in some instances, like with the detection of the muddy taint of geosmin, the advantages of instrumentation over human sensory perception have been questioned (Howgate, 2004).
There have been several studies devoted to examining the influ- Colour as a quality parameter, like flavour, also has many complexities to it subject to the objectives being sought, which are largely defined by the species being farmed. There are various methods that have been developed to assess colour including the use of colour-fans, colour-meters (L*a*b* meters) and computer-assisted imaging assessment techniques (Liu et al., 2013;Wade, Paulo, et al., 2014;Wathne, Bjerkeng, Storebakken, Vassvik, & Odland, 1998).
Most studies examining elements of colour end to focus on the application of carotenoids or carotenoid containing meals (Wade, Budd, Irvin, & Glencross, 2015;Wathne et al., 1998).

| OTHER N UTRITI ONAL A SS E SS MENTS
A range of other nutritional assessment methodologies and strategies exist that either value-add the primary in vivo work previously detailed and/or add new dimensions to that work by providing a greater mechanistic assessment of the functionality of the various nutritional responses seen. While many of these appear relatively academic in their intent, some have clear practical application in the ingredient assessment process.

| Rapid Analysis Technologies
The processes of in vivo assessment of diet and ingredient quality assessment are costly, laborious and time consuming. The development of technologies for the rapid analysis of nutritional value of raw materials, such as the use of in vitro assays and scanning technologies, like NIR, has been the subject of research since the 1980s (Bassompierre et al., 1997;Carter, Bransden, Van Barneveld, & Clarke, 1999;Dimes & Haard, 1994;Eid & Matty, 1989). This next section will examine some of the various assessment methods that have been attempted.

| In vitro Assessments
There are several options for the use of in vitro assessments of ingredient quality that can be applied. These extend from simple chemical analysis of specific parameters of the ingredient to provide a guide on various quality indices, to more complex in vitro assessments that aim to mimic the digestion process and provide an assessment of digestibility.
Perhaps the most commonly used in vitro assessment used as a quality criterion is those based on the level of certain indicators of quality/freshness in some ingredients, such as the TVN assay for fishmeals. Such indices have been used for other ingredients, like poultry and blood meals, though are less common, and new more rapid methods are being introduced (Johnson, Atkin, Lee, Sell, & Chandra, 2019; Lewis et al., 2019;Sheng et al., 2016). For fishmeals though, such assays have become so routine these days that they are used to provide demarcations between premium, high and average quality fishmeals (Jensen, Fiskeindustri, & Denmark, 1990). There are various methods that can be used to assess these parameters these days. In particular, the use of capillary electrophoresis (CE) provides a fast and sensitive procedure to simultaneously quantify volatile amines (TVN) and trimethylamine-oxide (TMAO) in samples (Bassompierre et al., 1998;Wu & Bechtel, 2008).
Various in vitro digestibility methods have been used to attempt to provide estimates of the nutritional (digestible) value of different raw materials for some time (Bassompierre et al., 1997;Eid & Matty, 1989;Lewis et al., 2019). Most methods use an enzyme-mediated process, but the key variable is often what enzymes are used (purified preparations or crude homogenates) and how the various products of the enzymatic process are interpreted that vary across the different methods. However, critical to the viable use of any rapid assessment method must be their validation against in vivo methods of assessment. Without these validations, the in vitro responses are merely academic. Although there has been much work done on developing and testing a range of in vitro assays, a range of problems surrounding their use persist. While they are clearly quicker than using in vivo testing, they are still time-consuming and have continued to have problems surrounding their reliability and inconsistencies in their predictive ability, though recent studies claim to have made significant gains in this area (Lewis et al., 2019;Yasumaru & Lemos, 2014

| Near-infraRed and Nuclear Magnetic Resonance Spectroscopy
In contrast to in vitro assays, technologies like near-infrared (NIR) spectroscopy and nuclear magnetic resonance (NMR) spectroscopy have allowed the assessment of the nutritional value of raw materials, on a near-real-time basis, and provide significant advancements in the responsiveness and cost savings in diet formulation by the feed industry (Conceição, Grasdalen, & Dinis, 2003;Cozzolino, Murray, & Scaife, 2002;Fontaine, Hörr, & Schirmer, 2001;Glencross et al., 2015). The use of NIR for determining the composition of raw materials is now relatively common in most modern feed production plants throughout the world. However, the use of NIR to assess the digestible value of protein and energy from raw materials is not as well established and reports on its successful application are scarce Glencross et al., 2015). To achieve a viable NIR calibration, it is critical that a wide range of samples is obtained from which to determine the nutritional (digestible protein and energy) values of the raw materials and to then correlate this with the NIR spectra of those same samples Glencross et al., 2015). The process of calibration development can be laborious and costly, though the potential gains in functionality through this method are enormous. Not only can NIR be used to determine compositional and nutritional (digestible) parameters of feeds and ingredients, it has also shown some prospect in being used as a discriminatory tool to determine the origins of raw materials to make certain ingredients (e.g. the type of fish used to make fishmeal) (Cozzolino et al., 2002).

| Nutrigenomic and allied assessments
A variety of "omics" (e.g. transcriptomics, proteomics, metabolomics) applications have emerged in aquaculture research with varying degrees of success in their application and capacity to deliver meaningful outcomes to nutritional science (Panserat & Kaushik, 2010).

| Transcriptomic analysis
Transcriptomic analysis is an assessment of the gene expression changes that occur in response to some stimulus (Panserat & Kaushik, 2010). In nutrition, this is usually with the use of a nutritional regime involving diets with different nutrients and/or ingredients and the assessment of the expression of specific genes relevant to various metabolic pathways (Panserat et al., 2009;Qian, Ba, Zhuang, & Zhong, 2014). It has become quite routine practice to include some level of transcriptomic analysis in modern nutritional research. However, it can be argued that in many cases, the practical value of using transcriptomics is questionable at best (Tacchi, Bickerdike, Douglas, Secombes, & Martin, 2011). Responses of the transcriptome (all those genes being expressed at any given time-point) to diet change dramatically postprandially, with tissue samples collected 24h after the animal has been fed arguably having little relevance to the total level of transcriptomic activity for any particular gene (Wade, Skiba-Cassy, Dias, & Glencross, 2014).
Perhaps a more pertinent approach would be to identify the postprandial period more responsive to diet and compare treatments from samples collected at that point. However, this requires an assessment of the variability postprandially and then often substantial changes to sampling regimes to allow the comparative sampling of experiments in time frames such as 2h after feeding (Poppi, Moore, Wade, & Glencross, 2019;Wade, Skiba-Cassy, et al., 2014).

| Proteomic analysis
Proteomic analysis is an assessment of the change protein expression that occurs with the use of a particular nutritional regime (Rodrigues, Silva, Dias, & Jessen, 2012). This "omics" analysis is arguably the next progression from transcriptomics in that it assesses the translation of the RNA produced from the differential gene expression in transcriptomics and allows variables of translational and posttranslational effects to be examined and therefore allows a closer look at the functional impact of a gene expression cascade (Seiliez et al., 2008;Wade, Skiba-Cassy, et al., 2014). Additionally, a range of tools can be applied at this level to examine post-translational modifications, like phosphorylation or methylation of proteins to identify whether their functional activity is also being affected (Zhou, Ding, & Wang, 2012).

| Metabolomic analysis
Metabolomic analysis is an assessment of the changes that occur in the abundance of metabolites within either a whole animal, tissues or cells. In this regard, metabolomics aims to use metabolite profiles to identify biomarkers indicative of physiological responses of organisms to nutritional or other conditions (Bankefors et al., 2011).
One of the benefits of the metabolomics approach is that it uses a broad scan of biological conditions to identify often unexpected problem or risk areas upon which to focus attention (Alfaro & Young, 2018). However, to date there has been little application of metabolomics in aquaculture nutrition and even less in studies devoted to ingredient assessment (Viegas et al., 2019).

| Endnote
Since the review by Glencross et al., (2007), not only has there been considerable progress in the science of ingredient evaluation for aquaculture feeds, but expectations from the users of those ingredients have also evolved. The original five steps proposed by the original authors still remain highly relevant, though emphasis on other parameters has now come to the fore as we increasingly adopt new ingredients into feeds for aquaculture species. Despite this, it still remains critically important that scientists evaluating ingredients do not skip the foundation steps in the rush to give quick answers.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available in peer-reviewed literature as detailed in the following references section. Where a specific data set was used, the reference for that source is linked to the analysis. All data were derived from resources available in the public domain. The data for