A comparison of regression models for

defining EPA + DHA requirements

using the gilthead seabream ( Sparus
aurata ) as a model species Sam J.S.
Houston
Visit to download the full and correct content document:
https://ebookmass.com/product/a-comparison-of-regression-models-for-defining-epa-
dha-requirements-using-the-gilthead-seabream-sparus-aurata-as-a-model-species-sa
m-j-s-houston/
 Aquaculture 556 (2022) 738308

Contents lists available at ScienceDirect

Aquaculture
journal homepage: www.elsevier.com/locate/aquaculture

A comparison of regression models for defining EPA + DHA requirements

using the gilthead seabream (Sparus aurata) as a model species
Sam J.S. Houston a, Vasileios Karalazos b, John Tinsley b, Douglas R. Tocher a, c,
Brett D. Glencross a, Óscar Monroig a, d, *
a
Institute of Aquaculture, Faculty of Natural Sciences, University of Stirling, Stirling FK9 4LA, UK
b
BioMar Ltd., North Shore Road, Grangemouth FK3 8UL, UK
c
Guangdong Provincial Key Laboratory of Marine Biotechnology, Shantou University, Shantou 515063, China
d
Instituto de Acuicultura Torre de la Sal (IATS), CSIC, 12595 Ribera de Cabanes, Castellón, Spain

A R T I C L E I N F O A B S T R A C T

Keywords: Carnivorous marine fish species such as gilthead seabream (Sparus aurata) require dietary eicosapentaenoic
Gilthead seabream nutrition (EPA) and docosahexaenoic (DHA) acids for optimal growth and wellbeing. The rapid growth of global aqua­
Essential fatty acids culture, along with increased proportions of dietary oil to increase growth rate of farmed fish, has meant that the
Long-chain polyunsaturated fatty acids
supply of marine oils used in aquafeeds has become limited. The shortfall has been satisfied by using vegetable
Nutrient requirements
Non-linear modelling
oils that lack EPA and DHA and, therefore, EFA (essential fatty acid) requirements of juvenile marine fish require
reassessment. A dietary trial was carried out with gilthead seabream (~25 g) that were fed diets with six EPA +
DHA levels ranging from 0.2% - 3.2% diet as fed. For each pellet size, the biometric data (weight gain, daily
growth index and feed conversion ratio) were analysed by four different regression strategies, namely split linear,
quadratic, the Gompertz function, and the four-parameter logistic function. Over the whole experimental period
(two pellet sizes) data suggested the current published requirement (1% of diet) was low and should be increased
to at least 1.2%. However, when the first pellet size for fish of 25–80 g was considered, the apparent requirement
was at least 1.4% of diet. This demonstrated in a single trial that EFA requirement was a function of fish mass,
decreasing as the fish grows. If FCR is considered, the requirement may be as high as 2%. The suitability of
different regression models varied, as the data for the first pellet was best fit by curves but, over both pellet sizes,
the split linear fit the data best. For asymptotic models (Gompertz and four-parameter logistic functions), a novel
way of defining the requirement was presented, the “elbow” calculation as a method to bisect an asymptotic
function. Therefore, using the raw data, we illustrate how a range of regression approaches could be explored
when determining nutrient requirement estimates as no single model was an ideal fit for all response curves.

1. Introduction
For marine carnivorous fish species, eicosapentaenoic acid (EPA;
20:5n-3) and docosahexaenoic acid (DHA; 22:6n-3) are regarded as
essential fatty acids (EFA) due to the limited ability that these species
possess for their biosynthesis from the C18 precursor, α-linolenic acid
(ALA; 18:3n-3) (Castro et al., 2016). Consequently, marine carnivorous
finfish have a dietary requirement for EPA + DHA to guarantee survival,
health and normal growth (Oliva-Teles, 2012; Tocher and Glencross,
2015; Xie et al., 2021). However, EPA and DHA are primarily found in
marine oils and, currently, fish oil (FO) is the primary source of these
EFA in aquafeeds.
An essential nutrient must be present in the animal’s diet because it
cannot be synthesised endogenously, or at a sufficient rate, to fulfil
physiological demands (Sargent et al., 2003). Animals respond to
nutrient levels and these can be plotted over a gradient of dietary supply

Abbreviations: AGR, absolute growth rate (g day− 1); DE, dietary energy; DGI, daily growth index; DHA, docosahexaenoic acid (22:6n-3); EFA, essential fatty acids;
EPA, eicosapentaenoic acid (20:5n-3); FCR, feed conversion ratio; FO, fish oil; FPL, four parameter logistic; LC-PUFA, long-chain polyunsaturated fatty acids; NRC,
National Research Council; VO, vegetable oil; WG, weight gain (g).
* Corresponding author at: Instituto de Acuicultura Torre de la Sal (IATS), CSIC, 12595 Ribera de Cabanes, Castellón, Spain.
E-mail addresses: karal@biomar.com (V. Karalazos), jtinsley@biomar.com (J. Tinsley), d.r.tocher@stir.ac.uk (D.R. Tocher), bglencross@iffo.com (B.D. Glencross),
oscar.monroig@csic.es (Ó. Monroig).

https://doi.org/10.1016/j.aquaculture.2022.738308
Received 5 October 2021; Received in revised form 25 April 2022; Accepted 26 April 2022
Available online 1 May 2022
0044-8486/© 2022 Elsevier B.V. All rights reserved.
S.J.S. Houston et al.
to give a response curve. Depending on the nutrient under study the
response curve will have different regions. The response shows negative
values when a nutrient is deficient, and is approximately zero when the
nutrient supply just balances the maintenance level (Lassiter and
Edwards, 1982). When the nutrient is in sufficient supply the response
will be positive (Bailleul et al., 2009; Hauschild et al., 2010) and this is
the level that is usually considered a minimum requirement in terms of
animal production. Further increases in the nutrient supply only elicit
small gains in the response variable until maximum performance
(plateau or asymptote) is approached (Fig. 1). These levels of response
are important when considering which models may apply best to the
nutrient and lifestage in question (Glencross, 2009). The most
commonly used response is growth, but not all nutrients will arrest
growth and, consequently, the deficient, and maintenance part of the
response may not be realised. This situation applies to EFA for juvenile
marine fish (Kalogeropoulos et al., 1992; Skalli and Robin, 2004), where
low supply leads to slower growth but does not stop it completely (Jin
et al., 2017), although other symptoms of deficiency may become
apparent (Bou et al., 2017).
It should be noted that determination of a nutrient requirement de­
fines a minimum level of nutrient supply to attain optimum/maximum
animal responses. This is different from a specification which is the di­
etary concentration chosen by a formulator to achieve the desired pro­
duction outcomes (not limited to essential nutrients) associated with the
feed (Glencross, 2009). The exact characteristics of the nutrient response
curve depend on the precise nutrient, the species, its life stage, and the
selected response (Rodehutscord and Pack, 1999). The National
Research Council (NRC) Nutrient Requirements of Fish and Shrimp
Fig. 1. The effect of nutrient level upon an animal response (a dose response
curve). Low nutrient supply indicates deficiency (Def), prevention of deficiency
indicates maintenance (Main), at the “elbow” or corner of the curve the
nutrient level can be described as optimal (Opt) where additional supply does
not elicit significant gains in the response and maximum (Max) performance,
whereby further increases in nutrient level cannot increase the response level
any further (further increases may become detrimental to the response level,
toxicity). This publication proposes that, after an experiment demonstrating
deficiency and surplus, the optimum requirement ( ) can be determined using
an “elbow” calculation. 1) A linear model is taken from the y minima to y
maxima ( ). 2) The maximum perpendicular distance between this line and
the fitted values of the relevant none-linear function is found. 3) The require­
ment ( ) is taken to be the nutrient level at this point. This represents a “point
of diminished returns”, whereby further gain in the response is small relative to
additional nutrient level. However, the “elbow” calculation is still sensitive to
experimental design and may not be appropriate for all response curves. The
figure has no units or axes and the response curve can take different forms to
what is depicted.
2
Aquaculture 556 (2022) 738308
(2011) reports that the EPA + DHA requirements for gilthead seabream
(Sparus aurata) for EPA + DHA is 0.9% of diet (dry weight) based on
studies by Kalogeropoulos et al. (1992) and Ibeas et al. (1994a, 1994b,
1997, 1996). Moreover, Ibeas et al. (1997) suggested that, in addition to
the EPA + DHA dietary level, the DHA:EPA ratio was also an important
factor, with a value of 0.5 regarded as appropriate at 0.9% EPA + DHA
dietary supply.
There are some drawbacks associated with previous studies in gilt­
head seabream that warrant further investigations into the EFA re­
quirements of this species. The studies by Ibeas et al. (1994a) and
Kalogeropoulos et al. (1992) used experimental fish less than 75 g in
weight and purified diets rather than commercial formulations. In
addition, fish growth and feed conversion achieved in the studies were
not as high as could be achieved, which is likely to cause underestima­
tion of requirements. Furthermore, the fastest growth rates in juvenile
sea bream can be realised at 24 ◦ C, which is warmer than the 18–21 ◦ C
used in previous studies. In fact, many species of fish such as Atlantic
salmon (Salmo salar), Asian seabass (Lates calcarifer), turbot (Psetta
maxima), European seabass (Dicentrarchus labrax), starry flounder
(Platichthys stellatus), Korean rockfish (Sebastes schlegeli) and red drum
(Sciaenops ocellatus), have been reported to have EPA + DHA (often
referred to as n-3 long-chain (≥C20) polyunsaturated fatty acids, LC-
PUFA) requirements of ~1% of diet (Coutteau et al., 1996; Gatesoupe
et al., 1977; Glencross et al., 2014; Lee et al., 2003; Lochmann and
Gatlin, 1993). However, other species such as yellowtail flounder
(Pleuronectes ferrugineus), Japanese flounder (Paralicthys olivaceus), sil­
ver bream (Rhabdosargus sarba) and striped jack (Pseudocaranx dentex)
have been shown to require EPA + DHA at a higher level (1.3–2.5%)
(Leu et al., 1994; Takeuchi, 1997; Takeuchi et al., 1992; Whalen et al.,
1998). Thus, EFA requirements vary among fish species, but also among
developmental stages, with larvae known to require relatively higher
levels of LC-PUFA, particularly DHA, to satisfy the high demands
required in rapidly growing neural tissues where DHA is required
(Takeuchi, 1997; Tocher, 2010).
It is appropriate to revisit the question of EPA + DHA requirements
for two key reasons. First, in the last two decades there has been a
substantial reduction in the use of fish meal (FM) and FO (sources of EPA
and DHA) in aquafeeds. Second, the oil content of modern aquafeeds has
increased to supply high energy to promote growth while sparing pro­
tein (Glencross, 2009; Sargent et al., 2003), and it has been suggested
that EFA requirement may vary with the lipid content of the diet
(Glencross and Smith, 2001; Watanabe, 1982). Therefore, the feed in­
gredients that contain EPA and DHA are being reduced while dietary
energy is increased by the addition of vegetable oils (VO) that lack LC-
PUFA, which may affect the apparent requirement. The present study
investigated the EFA requirements of a commercially important marine
species, the gilthead seabream, fed a dietary gradient of EPA + DHA
achieved by blending FO and VO. We herein discuss the benefits and
drawbacks of a range of regression models used to determine nutrient
requirements in fish (NRC, 2011), and provide estimates of EFA re­
quirements in juveniles calculated under the different regression stra­
tegies involving different response variables, including weight gain
(WG), daily growth index (DGI), and feed conversion ratio (FCR).
2. Materials and methods
2.1. Diets, fish husbandry and sampling
Gilthead seabream were maintained under the current European
legislation on handling experimental animals (2010/63/EU). In addi­
tion, all research performed by the Institute of Aquaculture, University
of Stirling (UoS) is subject to thorough ethical review carried out by the
UoS Animal Welfare and Ethical Review Board (AWERB) prior to any
work being approved. All projects, irrespective of where they are carried
out, are submitted to AWERB for approval using detailed Ethical
Approval forms that require all aspects of the experimentation to be
S.J.S. Houston et al. Aquaculture 556 (2022) 738308

described including all animal health and welfare issues as well as other Table 2
ethical considerations. The present research was assessed by the UoS Fatty acid composition of the experimental diets given as percentage of total
AWERB and passed the ethical review process. fatty acids.
Six experimental diets were formulated to deliver graded levels of Fatty acid (%) D1 D2 D3 D4 D5 D6
EPA + DHA to juvenile gilthead seabream. The diet formulations and 14:0 0.55 1.07 1.47 2.10 3.34 5.73
measured proximate compositions are detailed in Table 1 and fatty acid 16:0 15.44 15.62 15.71 16.05 16.69 17.52
compositions in Table 2. Commercially available oils including FO, 18:0 2.53 2.62 2.67 2.80 3.04 3.43
rapeseed oil and palm oil were blended to achieve specific levels of EPA 20:0 0.46 0.38 0.43 0.41 0.36 0.26
22:0 0.23 0.23 0.22 0.21 0.19 0.16
+ DHA. Other ingredients were selected to meet the known nutrient
ΣSFAa 19.35 20.04 20.64 21.70 23.75 27.22
requirements of this species (NRC, 2011; Oliva-Teles et al., 2011). Diets 16:1n-7 0.51 1.12 1.48 2.20 3.65 6.71
were produced by extrusion at the BioMar Tech-Centre (Brande, 18:1n-9 47.13 43.91 41.89 37.57 29.37 12.38
Denmark) and referred to as D1 – D6, with D1 containing only VO and 18:1n-7 2.47 2.55 2.58 2.69 2.77 2.99
D6 only FO as added oil. 20:1n 0.93 0.97 0.98 1.04 1.11 1.29
22:1n-11 0.08 0.14 0.19 0.28 0.43 0.75
The nutritional trial was carried out at the BioMar Aquaculture 24:1n-9 0.14 0.17 0.19 0.23 0.29 0.47
Technology Centre (Hirtshals, Denmark) as described previously ΣMUFAa 51.32 48.92 47.37 44.10 37.77 24.69
(Houston et al., 2017). Briefly, seabream juveniles of approximately 3 g 18:2n-6 21.42 20.42 19.69 18.56 16.16 11.22
were purchased from a commercial hatchery (Les Poissons du Soleil, 18:3n-6 0.00 0.03 0.04 0.07 0.12 0.25
20:2n-6 0.06 0.07 0.07 0.08 0.10 0.14
Balaruc-les-Bains, France) and initially fed a commercial diet containing
20:4n-6 0.05 0.11 0.15 0.23 0.40 0.78
FM and FO from first feeding until they reached ~24 g. At the start of the 22:5n-6 0.00 0.04 0.05 0.09 0.16 0.31
trial, 150 fish were randomly distributed in each of 18 × 1 m3 tanks with Σ n-6 PUFAa 21.52 20.67 20.01 19.03 17.01 12.90
each treatment being fed in triplicate (n = 3 per diet). The tanks were 18:3n-3 6.05 5.69 5.52 5.03 3.99 1.76
part of a recirculation aquaculture system with photoperiod, tempera­ 18:4n-3 0.08 0.30 0.44 0.70 1.23 2.40
20:4n-3 0.03 0.08 0.11 0.18 0.32 0.62
ture and salinity maintained at 12 L:12D, 24 ◦ C and 32 g kg − 1,
20:5n-3 0.70 1.96 2.73 4.32 7.48 14.34
respectively. The total duration of the feeding trial was 128 days, during 22:5n-3 0.11 0.26 0.35 0.53 0.91 1.71
which time the fish were fed two pellet sizes, firstly a 3 mm pellet (P1) 22:6n-3 0.63 1.47 1.98 3.06 5.23 9.89
for 56 days and, secondly a 4.5 mm pellet (P2) for 72 days. Therefore, Σ n-3 PUFAa 7.61 9.76 11.13 13.83 19.16 30.77
EPA:DHA 1.11 1.33 1.34 1.41 1.43 1.45
three experimental periods were considered, namely P1, P2 and over the
EPA + DHA (%)b 0.22 0.57 0.78 1.10 1.91 3.22
whole trial (OV), with the tank biomass being measured at 0, 56 and
a
128 days. At the end of P1 and P2, feeding was stopped 24 h before SFA, saturated fatty acids; MUFA, monounsaturated fatty acids; PUFA,
polyunsaturated fatty acids; EPA, eicosapentaenoic acid (20:5n-3); DHA, doco­
sahexaenoic acid (22:6n-3).
b
These values are the predictor variable for all analyses in this publication,
Table 1
expressed in this table as a percentage of diet.
Diet formulations and proximate analyses of the six experimental diets.
DIET D1 D2 D3 D4 D5 D6
sampling and, prior to weighing, fish were anaesthetized with benzo­
Ingredients (% of diet as caine (Centrovet, Kalagin, Santiago, Brazil), recovered from the anaes­
fed) thetic bath to a table to allow excess water to drain away, fish weighed
Fishmeal 12.50 12.50 12.50 12.50 12.50 12.50 and biomass from each of the tanks recorded and fish counted. The fish
Soy protein concentrate 21.90 21.90 21.90 21.90 21.90 21.90 were fed twice daily ad libitum and the delivered and wasted feed
Rape seed meal 10.00 10.00 10.00 10.00 10.00 10.00
recorded for an accurate determination of biological FCR.
Wheat gluten 4.00 4.00 4.00 4.00 4.00 4.00
Corn gluten 25.00 25.00 25.00 25.00 25.00 25.00
Wheat 7.10 7.10 7.10 7.10 7.10 7.10
2.2. Requirement analysis
Lysine 0.70 0.70 0.70 0.70 0.70 0.70
Methionine 0.07 0.07 0.07 0.07 0.07 0.07
Vitamin and mineral Although EPA and DHA are individual nutrients, they are supplied
0.45 0.45 0.45 0.45 0.45 0.45
premixa together in marine oils used by the aquafeed sector and, thus, in the
Microingredientsb 2.67 2.75 2.79 2.88 3.07 3.40 present study, were analysed as though they represent a single nutrient.
Yttrium 0.03 0.03 0.03 0.03 0.03 0.03
Therefore, the predictor variable in the following analysis was EPA +
DHA% of diet (as fed). Three related response variables were selected for
Oils (%)
analysis, weight gain (WG = final weight – initial weight), daily growth
Fish oil (SA) 0.00 1.75 2.64 4.43 8.00 14.85
Rapeseed oil 10.40 9.20 8.60 7.30 4.80 0.00 index (DGI = (final weight0.333 – initial weight0.333) / days ⋅ 100) and
Palm oil 5.19 4.56 4.22 3.64 2.39 0.00 food conversion ratio (FCR = (final weight – initial weight) / feed
consumed). To estimate the requirements of gilthead seabream juveniles
Proximate composition (% of diet as fed) for EPA + DHA% of diet, a series of regression analyses were performed
Crude protein 41.8 43.0 42.9 41.7 42.4 42.5 on the three responses WG (g), DGI and FCR, for the P1 (3 mm), P2 (4.5
Crude lipid 22.2 21.7 21.7 21.5 21.8 20.8 mm) and OV (both pellets) response data. All regression analyses were
Ash 6.1 6.1 5.9 5.9 6.1 6.2
undertaken using the statistical package R (R Core Team, 2013, version
Moisture 10.3 9.7 9.4 10.7 8.9 8.9
Crude energy (MJ/kg) c
22.0 22.1 22.2 21.8 22.2 22.0
3.4.0; Vienna, Austria). The model parameters were obtained using the
a
function nls(), and the following “packages” to analyse the data: nlstools
Vitamin and mineral premix including (as IU or ppm of premix): Vitamins A (Baty et al., 2015), propagate (Speiss, 2013) and ggplot2 (Wickham,
(666,666 IU/kg), E (40,000 ppm), K3 (4002 ppm), B1 (4000 ppm), B2 (4800
2016). The package nlstools contains a range of functions for performing
ppm), B3 (12,000 ppm), B5 (16,000 ppm), B6 (4000 ppm), B12 (4 ppm), Biotin
regression diagnostics, while propagate’s function (predictNLS()) was
(120 ppm), Folic acid (4000 ppm), C (28,063 ppm), Copper (222 ppm), Cobalt
(222 ppm), Iodine (133 ppm), Manganese (1784 ppm), Zinc (22,223 ppm). used to calculate the prediction intervals for a model’s fitted values (by
b
Monocalcium-phosphate (MCP) (1.62% diet), cholesterol (0.08–0.17 diet), Monte-Carlo simulation and second order Taylor expansion), with
Emulthin G35 (0.6–1.4% diet), antioxidants (0.03% diet). graphics produced in ggplot2. There are many types of non-linear
c
Estimated by using the mean values of gross energy for proteins, lipids and functions that can model these experimental data and the approach
carbohydrates 23.6, 39.5 and 17.2 kJ/g, respectively. presented was empirical, as opposed to mechanistic. Preference was

3
S.J.S. Houston et al. Aquaculture 556 (2022) 738308

given to models with low error (mean squared error, MSE), and that
were a strong fit to the requirement region of the response. Suitable
models were assessed diagnostically according to techniques described
by Ritz and Streibig (2008). The four regression models used were: split
linear model, split quadratic model, Gompertz model and four-
parameter logistic (FPL) model. For the latter two models (asymptotic
models) two requirement estimates are presented: 1) 95% of the
asymptote (NRC, 2011), which is referred to as the “NRC criterion” and
plotted as ▴in the relevant figures and, 2) the “elbow” calculation,
plotted as ● in the relevant figures. The elbow calculation is a graphical
method that takes the maximum and minimum response values of a
model, joins these with a straight line, then defines the point of the
model’s curve which has the furthest perpendicular distance from this
line (see Fig. 1). The elbow’s usual application is in K-means clustering
to select an appropriate number of clusters within unstructured datasets
(Syakur et al., 2018). The R-code used was adapted from a stackerflow
post (Eickhart, 2017).
Unless otherwise stated n = 18, which in regression is sufficient to
detect large (f2 = 0.5) effects in the response.
2.2.1. Split linear model
The first model applied to the responses was the so-called “broken
line” or split regression model, in which the first segment is a linear
model usually ascending (descending for FCR) to a maximum response,
i.e. a horizontal segment (a mean). The requirement is defined as the
intersection of these two segments. Split linear regression can be
expressed as follows:
f(x) = m ⋅ x + c for x < req
f(x) = m ⋅ req + c for x > req
where x is the percentage of EPA + DHA in the experimental diet, c is
the model’s intercept and m is the gradient (of the first segment) and req
is x at the breakpoint (the requirement).
2.2.2. Split quadratic model
The second model applied to the data was the split quadratic model,
in which the first segment is a quadratic function ascending to a
maximum level of response, whereas the second segment is horizontal.
The requirement is defined as the intersection of these two segments.
Split quadratic regression can be expressed as follows:
f(x) = m ⋅ x + (− 0.5 ⋅ m / req) ⋅ x2 + c for x < req
f(x) = m ⋅ req + (− 0.5⋅m / req) ⋅ req2 + c for x > req
where x is the percentage of EPA + DHA in the experimental diet, c is
the model’s intercept, m is the rate constant and req is x at the breakpoint
(the requirement).
2.2.3. Gompertz model
The first asymptotic function fitted to the response data was the
Gompertz that can be defined as:
f(x) = a ⋅ exp. (b ⋅ c x)
where x is the percentage of EPA + DHA in the experimental diet, a is
the asymptote or Y maxima, b sets the displacement on the x-axis and c is
the scaling parameter, setting the rate of growth towards the asymptote.
2.2.4. Four parameter logistic model
The second asymptotic function fitted to the response was the FPL
model that is defined as:
f(x) = b + (a-b) / (1 + exp.((xmid-x)/c))
where x is the percentage of EPA + DHA in the experimental diet, a is
the upper asymptote, b is the lower asymptote, xmid is the value of x at
the inflection point and c is the scaling parameter.
3. Results
3.1. Growth performance
Table 3 shows the mean initial weights, final weights, DGI and FCR.
During the feeding trial, weight of fish in all dietary groups increased
(Table 3). However, the increase in dietary EPA + DHA (particularly
diets D5-D6) resulted in higher weight gain in gilthead seabream
juveniles.
3.2. Requirement analysis
A series of regression analyses were conducted to explore different
ways to determine the EPA + DHA requirement of juvenile gilthead
seabream. All functions and their respective parameters are given in
Table 4. The requirement estimates derived from these models are given
in Table 5, along with measures of the model fit, the Akaike information
criterion (AIC) and the MSE. Below we present the EFA requirement esti­
mations resulting from applying the four models to growth parameters (WG
and DGI) and FCR.
3.2.1. Weight gain and daily growth index
Using the split linear model to fit the WG data, the requirements were
estimated as 1.21, 1.47 and 1.13% EPA + DHA as a percentage of diet for
OV, P1 and P2, respectively (Fig. 2A). Using the split linear model to fit
the DGI data, the requirements were 1.20, 1.46 and 1.09% EPA + DHA
for OV, P1 and P2, respectively (Fig. 2B). The requirement values agree
between the two metrics of growth, WG and DGI. The distribution of
error for P1, WG and DGI exhibited a curve. When using the split
quadratic model for the WG data, the EFA requirements were estimated
as 1.70, 2.25 and 1.57% EPA + DHA for OV, P1 and P2, respectively
(Fig. 2C). The error distribution in the P1 data was better using the
quadratic model. Using the split quadratic model for the DGI data, the
requirement levels were 1.64 and 2.15% EPA + DHA for OV and P1,
respectively. This model failed to fit the DGI data of P2 due to the high
DGI values attained with fish fed diet D4 (Fig. 2D). The OV and P2 data
was best modelled by the split linear method, but the P1 data was best
described by the split quadratic model (Table 5).
The asymptotic non-linear models, namely the Gompertz and the
four-parameter logistic (FPL) models, applied to the WG and DGI data
are presented in Fig. 3. Using the Gompertz model and elbow calculation
(Fig. 1) to fit the WG data, the EFA requirements were estimated as 1.22,
1.34 and 1.18% EPA + DHA for OV, P1 and P2, respectively (Fig. 3A).
Considering the NRC recommendation of 95% of the maximum
response, the EFA requirements were calculated as 1.13, 1.40 and 1.07%
EPA + DHA for OV, P1 and P2, respectively. With regards to the DGI

Table 3
Mean values (± SD) for the dietary treatments for the biometric data analysed in this study (n = 3).
D1 SD D2 SD D3 SD D4 SD D5 SD D6 SD

Initial weight (g) 24.4 ±0.3 24.3 ±0.1 24.0 ±0.1 24.6 ±0.3 24.3 ±0.1 24.5 ±0.3
Final weight (g) 198.6 ±1.3 219.4 ±6.6 224.8 ±5.0 241.3 ±3.1 245.0 ±4.0 248.3 ±1.5
Weight gain (g) 174.2 ±1.3 195.1 ±6.6 200.8 ±4.9 216.7 ±2.8 220.7 ±3.9 223.8 ±1.4
DGIa 2.29 ±0.01 2.44 ±0.05 2.49 ±0.04 2.58 ±0.01 2.62 ±0.02 2.63 ±0.01
FCRb 1.33 ±0.02 1.33 ±0.02 1.31 ±0.02 1.27 ±0.01 1.22 ±0.01 1.21 ±0.01
a
Daily Growth Index = Final weight (g) 0.333 – Initial weight (g) 0.333/Days feeding.
b
Food conversion ratio = (Feed fed (kg) – feed waste (kg))/Δ Biomass (kg).

4
S.J.S. Houston et al. Aquaculture 556 (2022) 738308

Table 4
Model parameters fitted to the biometric data of S. aurata juveniles in the present trial to derive estimates for the EPA + DHA requirement of this species. The four
models presented are the split linear, split quadratic, Gompertz and the four parameter logistic (FPL) function. The models were applied to the overall data (OV, both
pellet sizes), the 3 mm pellet size (P1, in the text) and the 4.5 mm pellet size (P2 in the text). The name of the parameter and the notation used in the text is indicated in
the column headers. n = 18 except for the Gompertz model applied to FCR data, where n = 15.
Model type Parameters

Metric Period Plateau Gradient Intercept Requirement

m c
Split regression WG (g) OV 222.22 47.13 165.25 1.21
3 mm 78.81 11.02 57.61 1.47
4.5 mm 148.38 36.03 107.62 1.13
DGI OV 2.63 0.33 2.23 1.20
3 mm 3.05 0.32 2.59 1.46
4.5 mm 2.30 0.32 1.96 1.09
FCR OV 1.21 − 0.07 1.36 2.03
3 mm 1.09 − 0.06 1.23 2.50
4.5 mm 1.25 − 0.10 1.41 1.62

Plateau Rate constant Intercept Requirement

m c req
Ascending polynomial WG (g) OV 222.43 75.26 158.52 1.70
3 mm 74.04 15.62 56.43 2.25
4.5 mm 148.76 59.18 102.31 1.57
DGI OV 2.63 0.55 2.17 1.64
3 mm 3.06 0.48 2.54 2.15
FCR OV 1.21 − 0.11 1.37 3.03
3 mm 1.02 − 0.06 1.23 6.94
4.5 mm 1.24 − 0.13 1.42 2.71

Plateau Displacement Scale parameter

a b c
Gompertz WG (g) OV 224.42 − 0.38 0.17
3 mm 72.84 − 0.31 0.28
4.5 mm 149.99 − 0.42 0.14
DGI OV 2.64 − 0.22 0.16
3 mm 3.08 − 0.21 0.26
4.5 mm 2.31 − 0.23 0.10
FCR OV 1.20 0.23 0.26
3 mm 1.06 0.21 0.47
4.5 mm 1.24 0.26 0.20

Plateau Bottom Inflection Scale parameter

a b xmid c
FPL WG (g) OV 223.00 150.65 0.45 0.32
3 mm 74.62 24.75 − 0.32 0.66
4.5 mm 149.00 106.37 0.55 0.25
DGI OV 2.63 2.00 0.29 0.36
3 mm 3.08 − 6.10 − 2.03 0.76
4.5 mm 2.31 1.98 0.55 0.19
FCR OV 1.21 1.34 1.04 0.10
3 mm 1.09 1.21 1.41 0.29
4.5 mm 1.25 1.38 0.95 0.19

data, the Gompertz function and the elbow calculation estimated EFA
requirements of 1.18, 1.30 and 1.08% EPA + DHA for OV, P1 and P2,
respectively. When the NRC criterion (95% of the maximum response)
was applied to estimate EFA requirements from DGI data, much lower
values were obtained, namely 0.79, 1.04 and 0.64% EPA + DHA for OV,
P1 and P2, respectively (Fig. 3B). Using the FPL model to fit the WG
data, EFA requirements were estimated as 1.25, 1.37 and 1.19% EPA +
DHA (elbow calculation) and 1.01, 1.34 and 0.94% of dietary EPA +
DHA (NRC criterion) for OV, P1 and P2, respectively (Fig. 3C). For the
DGI data, the EFA requirements were estimated as 1.21, 1.3 and 1.09%
(elbow calculation) and 0.77, 1.05 and 0.68% of dietary EPA + DHA
(NRC criterion) for OV, P1 and P2, respectively (Fig. 3D). The fourth
parameter allowed the model to fit the growth data well in OV and P1.
However, parameter estimates for b and, or xmid had large standard
errors, due to the absence of data for the lower part of the curve. It
should be noted that for P1 the model parameters were not significant
leading to wide prediction intervals (Fig. 3D).
3.2.2. Feed conversion ratio
The split linear and quadratic models provided requirement esti­
mations that were high. This was due to the inadequacy of these models
to fit the unusual distribution of the experimental FCR data. Therefore,
the values are not discussed in detail here, but are reported in Table 5
and Fig. 4 A and B for completeness and to demonstrate the lack of fit. To
fit the Gompertz models to the FCR data it was necessary to remove data
of diet D1 (N = 15). Using the Gompertz curve to fit the FCR data, the
EFA requirements were estimated as 1.53, 1.67 and 1.47% (elbow
calculation) and 1.15, 1.92 and 1.03% EPA + DHA (NRC criterion) for
OV, P1 and P2, respectively (Fig. 4C). The removal of diet D1 data
removed error and, therefore, the model is not directly comparable to
the FPL model. Despite the unusual shape of the FCR data, the FPL model
was flexible enough to fit these data without the removal of diet D1 data.
With FCR as response, the EFA requirements were 1.55, 2.03 and 1.45%
EPA + DHA (elbow calculation) and 1.05, 1.45 and 0.99% EPA + DHA
(NRC criterion) for OV, P1 and P2, respectively (Fig. 4D). The estimated
requirement values attained for FCR were higher than for growth, and

5
S.J.S. Houston et al. Aquaculture 556 (2022) 738308

the FPL model was the only model to offer a satisfactory fit to the

size (P1, in the text) and the 4.5 mm pellet size (P2 in the text). In the Gompertz and FPL models, the EPA + DHA estimate derived from the “elbow” calculation is given. The mean squared error (MSE) and the Akaike
Requirement estimates for EPA + DHA (% of diet as fed) of juvenile gilthead seabream, Sparus aurata, from the models listed in Table 4. The models were applied to the overall data (OV, both pellet sizes), the 3 mm pellet

− 56.72
− 52.53
− 81.76
− 91.85
− 68.64
experimental data.

114.14

107.98
− 67.6
71.92
AIC
4. Discussion

0.00036
0.00021
0.00072
0.0008
0.0014
0.0018
The present study explored four different models and three perfor­

19.06

13.56
MSE

1.83
mance responses that can be used to advance estimates of EPA + DHA
requirement. The experimental design involved diets formulated with
FPL requirement
commercially relevant oils and energy densities, and strong growth
performance was observed in all dietary groups. There were two key
conclusions from this study. Firstly, it showed, within a single study that
the requirement for EPA + DHA is a function of fish mass (decreasing
1.25
1.37
1.19
1.21
1.30
1.09
1.55
2.03
1.45
with increasing weight) and could therefore be modelled with data from
more time points. Secondly, the current published requirement for EPA
+ DHA of 0.9% (dry diet) appeared too low for juvenile gilthead seab­
− 68.75
− 58.72

− 70.09
− 82.84
108.07

− 52.2

− 57.7
113.4
70.00

ream (Kalogeropoulos et al., 1992; NRC, 2011). We report the results

AIC

from different analysis methods and acknowledge that these data are
open to interpretation. However, we advance tentative recommenda­
0.00083

0.00032
0.00014
0.00073
0.0014
0.0021

tions for the two pellet sizes used in this study. For juveniles of 24–80 g,
20.44

15.22
MSE

1.84

the diet (3 mm pellet) may contain at least 1.4% EPA + DHA as a

minimum requirement but, to achieve maximum performance (highest
growth and feed conversion), the level is 2%. This level may be reduced
Gompertz requirement

in larger fish (80–250 g; 4.5 mm pellet) to 1.2% to satisfy minimum

requirement, but 1.5% to achieve maximum performance. The data
suggested that the current published guidance (~ 1%) would become
adequate at some body weight during this growth phase (P2). Further­
more, the data presented here support the reduction in the use of FM
1.22
1.34
1.18
1.18
1.30
1.08
1.53
1.67
1.47

and, or FO, provided that EFA requirements are covered in this species,
this is in line with studies using very low FM and FO levels and achieving
− 69.86
− 57.86

− 78.25

− 67.28
112.02

105.76

good growth performance, in particular in the later stages of production

70.24

− 82
AIC

(Benedito-Palos et al., 2016; Simó-Mirabet et al., 2018).

The previous requirement estimates for juvenile gilthead seabream
employed smaller fish than this study (1.2 g) and used split linear
0.00078

0.00049
0.00039
0.00089
0.0015
18.94

13.39
information criterion (AIC) are also given as an indication of how well the models fit the experimental data.

regression to analyse weight gain (g). The juvenile seabream used in the
MSE

1.86

present study attained growth rates and feed efficiency superior to the
fish than the fish used to determine the current known requirement
Split quadratic requirement

(Kalogeropoulos et al., 1992; National Research Council, 2011). As fish

performance, in particular feed efficiency increases, the requirement for
nutrients will increase to support increased growth rates (Shearer,
1995). A further study in European seabass also employed split linear
regression with DGI as the response, and reported a requirement of 0.7%
LC-PUFA (of diet dry diet) for fish of 14 g initial weight (DGI: 0.84–1.0)
(Skalli and Robin, 2004). The different approaches presented in the
2.25
1.57
1.64
2.15

3.03
6.94
2.71
1.7

present study also showed that there is no one model that suits every
–

parameter or each of the experimental periods, as suggested previously

− 67.27
− 56.22

− 80.19
− 85.29
− 70.53
109.75

101.33

(Baker, 1986; Rodehutscord and Pack, 1999).

− 57.9
70.61
AIC

The higher EFA requirements, especially for DHA, of larval fish is

relatively well understood (Izquierdo, 1996; Sargent et al., 1997). A
recent study of micronutrients and taurine in seabream showed that
0.00089

0.00044
0.00034
0.00083
0.0018
0.0015
18.67

11.67

requirements in larvae were several times higher than in fish in the grow
MSE

2.12

out phase (Izquierdo et al., 2019). The present study has shown a change
in EFA requirement between two fish (pellet) sizes and, therefore,
Split linear requirement

demonstrated the existence of a relationship between fish mass and EFA

requirement. It is likely that this relationship takes on a similar form to
digestible protein [requirement = − a ⋅ fish mass (Kg)b] (Lupatsch et al.,
2003), but to define this relationship the precise requirement for EPA +
DHA would need to be known over a range of fish masses, ideally in a
single experiment achieving optimal growth. Here, we only provide
1.21
1.47
1.13
1.20
1.46
1.09
1.97
2.50
1.62

requirement estimates for fish of around 25–80 g and 100–250 g, and so

only the negative slope between these fish weights can be estimated,
4.5 mm

4.5 mm

4.5 mm

with the value being approximately − 1.8. Alternatively, if the require­

3 mm

3 mm

3 mm
Pellet

ment is viewed per unit energy (Glencross, 2009) or lipid (Watanabe,

OV

OV

OV

1982) the decreasing requirement (% of diet) may be interpreted as a

consequence of the rising energy (DE) demand for maintenance as the
WG (g)
Table 5

Metric

fish grows (Lupatsch, 2005). Regardless of how the requirement is

FCR
DGI

numerically expressed, or even understood, its fraction in the diet (% or

6
S.J.S. Houston et al.
g kg− 1) still needs to be known to inform diet specification. The dis­
cussion below focuses on the strengths and weaknesses of the models.
The first two models applied to the data were linear and quadratic
split regression. Conceptually, these models derive the maximum
requirement because the equation describes a model approaching a
horizontal plateau, at which point the response is equivalent to the mean
of the highest performing dietary groups. For this reason, it is critical for
these techniques that the maximum response is known with at least two
dietary treatments. The exception to this is where an effect of toxicity is
observed at higher levels of nutrient supply. An example of this can be
found in black seabream (Acanthopagrus schlegelii) where high levels of
n-3 LC-PUFA led to lower growth and, in this situation, the authors used
quadratic regression without the horizontal segment (Jin et al., 2017). In
the present study, we did not observe a detrimental effect of higher EPA
+ DHA levels, so quadratic regression was not appropriate. The key
advantage with linear and quadratic split models is the provision of the
requirement by the fitting process (Glencross, 2009). Linear split
regression is the most frequently applied model in fish nutrient
requirement studies (Benedito-Palos et al., 2014; Luzzana et al., 1998;
Mai et al., 2006; Murillo-Gurrea et al., 2001; Skalli et al., 2006).
Quadratic split regression, to our knowledge, has only been applied in a
meta-analyses of fish data (Hernandez-Llamas, 2009), but there are
more examples in terrestrial animals (Hauschild et al., 2010). Interest­
ingly, the linear split model had the best fit for WG response in OV and
P2 data, and it produced EPA + DHA requirement estimates marginally
higher than the two asymptotic models (Gompertz and FPL), this being
somewhat unexpected as this method had been found to underestimate
nutrient requirements (Hernandez-Llamas, 2009; Shearer, 2000). In the
present study, the quadratic split regression gave higher estimates than
the linear split regression. Neither of these methods seemed applicable
to the FCR data due to the unusual shape of this response. As a result, the
7
Aquaculture 556 (2022) 738308
Fig. 2. The split linear (A and B) and split quadratic (C and D)
models applied to the weight gain (A and C) and daily growth
index (B and D) data against EPA + DHA % of diet (as fed, x-
axis). The raw data are marked with coloured spots, the
relevant model fit with an unbroken line, 95% prediction in­
tervals (calculated by Monte-Carlo simulation) are shaded in
colour. Colours for the different periods are as follows: the
whole trial, OV = ; the 3 mm pellet size, P1 = and the 4.5
mm pellet size, P2 = . Note, in panel D the quadratic model
would not fit the P2 data due to the high performance of D4.
Estimated requirement values are given in Table 5.
EPA + DHA requirements estimated with FCR as response variable were
considerably higher than published requirement estimates for S. aurata
(Kalogeropoulos et al., 1992) when any of the split regression models
were used. Feed efficiency/feed conversion ratio is usually given in
requirement papers, but rarely modelled as a parameter. An example
with lysine requirements of the Japanese seabass (Lateolabrax japonicus)
used split regression with feed efficiency (Mai et al., 2006).
Asymptotic functions (such as Gompertz and FPL) can often be a
strong fit to experimental data (Mercer et al., 1989) since they
adequately address the concept of diminishing returns where the in­
crease in the response slows as the plateau is reached. A limitation of the
continuous asymptotic models is that a method is required to derive a
requirement estimate, for example 0.95* asymptote (Cowey, 1992; NRC,
2011). The present results suggested that this method was highly sen­
sitive to parameter selection and might lead to erroneous requirement
estimates, at least in the context of EFA. Thus, the requirement estimates
during the OV period for both growth responses (DGI and WG) disagree
when either the FPL or Gompertz models are applied. An alternative to
the 0.95* asymptote criterion, the proposed “elbow” calculation, was
developed as a means of defining an “elbow” or optimal requirement on
a response curve. In nutrient requirement studies, it is important that
both the deficient and excess part of the nutrient response curve are
well-defined (Glencross, 2009; Shearer, 2000). It was anticipated that
the elbow calculation would be less sensitive to the selected dietary
levels of nutrient supply. Unfortunately, this is not the case and the
elbow calculation still relies on the nutrient levels determined at the
experiment’s conception. However, this calculation did seem useful in
analysing the data obtained in the present study as it gave similar
requirement estimates for both metrics of growth and both models
(albeit, marginally higher for the DGI parameter). This is due to its
insensitivity to the scale of the response. However, it is advocated that
S.J.S. Houston et al.
the elbow calculation is only applied to asymptotic functions and only to
experiments that have adequately described the deficient and excess
regions of the response. Only the elbow method has been employed
here, and we are not advocating it as a replacement method for selecting
a requirement, and there are apparent nuances with this method.
The two asymptotic models used in the present study (Gompertz and
FPL) have been used previously to fit nutrient responses in terrestrial
animals (Gahl et al., 1991; Pesti et al., 2009) but, to our knowledge, have
not been applied in fish. Other asymptotic functions (four and five-
parameter saturation kinetics models) were used in a study investi­
gating the tryptophan requirements of hybrid striped bass (Morone
chrysops x M. saxatilis) finding requirements just over 2 g kg− 1 when the
asymptote was multiplied by 0.95 (Gaylord et al., 2005). Two meta-
analyses in fish have employed a variety of non-linear models to pre­
viously published data and found that often the curves fit responses
better, but that no one model was appropriate for all data sets (Rode­
hutscord and Pack, 1999; Shearer, 2000). Therefore, we explored a
range of non-linear models and the two attaining the best fit were re­
ported here. In the present study, the Gompertz model was a good fit for
most of the growth data, especially for P1. The FPL model was also able
to fit these data, but this model carried high uncertainty because it has
parameters that are not covered by the data (b and xmid). To apply the
Gompertz model to the FCR data it was necessary to disregard data for
diet D1 and, therefore, it does not capture reality well, which created
problems comparing fits as error associated with diet D1 was discarded.
The only model that could describe the FCR response to the diets was the
FPL model and, as such, requirements for FCR differed between the two
asymptotic models and it is difficult to advance an estimate, although it
can be said that the requirement to achieve the best FCR was higher than
for the growth metrics.
It is evident from these empirical data that the current published
8
Aquaculture 556 (2022) 738308
Fig. 3. The non-linear Gompertz model (A and B) and four
parameter logistic model (FPL in the text, C and D) applied to
the weight gain (A and C) and daily growth index (B and D)
data against EPA + DHA % of diet (as fed, x-axis). The raw data
are marked with coloured spots, the relevant model fit with an
unbroken line, requirements with black symbols and the 95%
prediction intervals (calculated with 2nd order Taylor expan­
sion) are shaded in colour. Symbols for the requirement esti­
mates are as follows: Published = ■, National Research
Council = ▴ (NRC criterion in text) and Elbow = ●. Colours for
the different periods are as follows: the whole trial, OV = ;
the 3 mm pellet size, P1 = and the 4.5 mm pellet size, P2 =
. Note in panel D in P1, the error was not plotted for clarity
because it was very wide, an indication that the curve for this
data was over-parameterised. Estimated requirement values
are given in Table 5.
requirement of 0.9 (dry weight) / 1% EPA + DHA is too low for juveniles
of this species, especially when considering fish of 25–80 g. The present
study has demonstrated that no model is a fit for each situation and the
NRC definition of a requirement at 95% of a response maximum may not
be entirely appropriate for EPA + DHA requirements. One reason for this
is that the models suggest that n-3 LC-PUFA (i.e. EPA and DHA) are not
completely essential to growth (all models have positive intercepts,
implying that there is no maintenance demand and no weight loss), at
least over the timescale and EFA levels used for this study. Interestingly,
results modelling the turnover of individual fatty acids in Asian seabass
(L. calcarifer) suggested zero maintenance requirement for EPA and DHA
(Salini et al., 2016). The elbow-calculation was applied to determine a
point on an asymptotic response curve that was less sensitive to the
design of the experiment although, unfortunately, it is still not insensi­
tive to experimental design. The calculation, however, does bisect the
curve well and the data demonstrated it returned values that agreed with
each other across models and metrics. However, its application was not
appropriate unless the maximum response and deficient responses were
well defined and we only advocate its use for asymptotic models. The
elbow calculation defines a point on the curve that further increases in
dietary EPA + DHA levels lead to little gain in the response and,
therefore, denotes an optimum requirement (Fig. 1).
In summary, the key conclusions of this work are that 1) S. aurata
juveniles of 25 g – 80 g (3 mm pellet) have a higher requirement for EPA
+ DHA of at least 1.4% of diet for optimal/maximum growth, but
perhaps as high as 2% if FCR is considered, 2) the requirements for EPA
+ DHA are a function of fish mass, the implication of this being that
requirements could be modelled but the trial would need a larger
number of sampling points than the two reported here and, 3) the exact
definition of EPA + DHA requirement is a question of context; mainte­
nance, optimum production, maximum performance, health, well-being
S.J.S. Houston et al.
or product quality could all ultimately reflect different end-points and,
therefore, give different estimated requirements.
Authorship
VK and JT designed and executed the nutritional trial and all authors
contributed to planning the analyses. SJSH, VK, and JT carried out the
sampling. OM and DRT supervised the lead author and BG provided
technical expertise during manuscript preparation. SJSH analysed the
data and prepared the draft manuscript, which was then shared between
all authors who made amendments, contributions and
recommendations.
CRediT authorship contribution statement
Sam J.S. Houston: Investigation, Methodology, Formal analysis,
Writing – original draft, Visualization. Vasileios Karalazos: Concep­
tualization, Investigation, Methodology, Supervision, Writing – review
& editing, Funding acquisition. John Tinsley: Conceptualization,
Investigation, Methodology, Supervision, Writing – review & editing,
Funding acquisition. Douglas R. Tocher: Conceptualization, Investi­
gation, Supervision, Writing – review & editing, Funding acquisition.
Brett D. Glencross: Writing – review & editing. Óscar Monroig:
Conceptualization, Investigation, Supervision, Writing – review & edit­
ing, Funding acquisition.
Declaration of Competing Interest
None.
9
Aquaculture 556 (2022) 738308
Fig. 4. The linear (A) and quadratic (B) split regression models
applied to the FCR data against EPA + DHA % of diet (as fed, x-
axis), the 95% prediction intervals (shaded colour) were
calculated by Monte-Carlo simulation. The non-linear Gom­
pertz (C) and four parameter logistic (FPL in the text, D)
applied the FCR data against EPA + DHA % of diet (as fed, x-
axis), the 95% prediction intervals (shaded colour) were
calculated by second-order Taylor expansion. The raw data are
marked with coloured spots, the relevant model fit with an
unbroken line and requirements with black symbols. Symbols
for the requirement estimates are as follows: Published = ■,
National Research Council = ▴ (NRC criterion in text) and
Elbow = ●. Colours for the different periods are as follows: the
whole trial, OV = ; the 3 mm pellet size, P1 = and the 4.5
mm pellet size, P2 = . Note in panel B, the error for P1 was
not plotted, but clearly it is a poor fit to the data. Estimated
requirement values are given in Table 5.
Acknowledgements
This work was co-funded by BioMar AS and the Marine Alliance for
Science and Technology Scotland (MASTS). BioMar provided the
experimental feeds, trial facilities and fish, and covered travel expenses.
The authorsexpress thanks to the technical staff at the BioMar Feed Trial
Unit, Hirtshals, Denmark for expert care of the fish during this trial.
References
Bailleul, P.J. dit, Bernier, J.F., van Milgen, J., Sauvant, D., Pomar, C., 2009. The
utilization of prediction models to optimize farm animal production systems: the
case of a growing pig model. In: Modelling Nutrient Utilization in Farm Animals.
CABI, pp. 379–392. https://doi.org/10.1079/9780851994499.0379.
Baker, D.H., 1986. Problems and pitfalls in animal experiments designed to establish
dietary requirements for essential nutrients. J. Nutr. 116, 2339–2349. https://doi.
org/10.1093/jn/116.12.2339.
Baty, F., Ritz, C., Charles, S., Brutsche, M., Flandrois, J.-P., Delignette-Muller, M.-L.,
2015. A toolbox for nonlinear regression inR: the Packagenlstools. J. Stat. Softw. 66
https://doi.org/10.18637/jss.v066.i05.
Benedito-Palos, L., Ballester-Lozano, G., Pérez-Sánchez, J., 2014. Wide-gene expression
analysis of lipid-relevant genes in nutritionally challenged gilthead sea bream
(Sparus aurata). Gene 547, 34–42. https://doi.org/10.1016/j.gene.2014.05.073.
Benedito-Palos, L., Ballester-Lozano, G.F., Simó, P., Karalazos, V., Ortiz, Á., Calduch-
Giner, J., Pérez-Sánchez, J., 2016. Lasting effects of butyrate and low FM/FO diets
on growth performance, blood haematology/biochemistry and molecular growth-
related markers in gilthead sea bream (Sparus aurata). Aquaculture 454, 8–18.
Bou, M., Berge, G.M., Baeverfjord, G., Sigholt, T., Østbye, T.-K., Romarheim, O.H.,
Ruyter, B., 2017. Requirements ofn-3 very long-chain PUFA in Atlantic salmon
(Salmo salar L): effects of different dietary levels of EPA and DHA on fish
performance and tissue composition and integrity. Br. J. Nutr. 117, 30–47. https://
doi.org/10.1017/s0007114516004396.
Castro, L.F.C., Tocher, D.R., Monroig, O., Castro, L.F.C., Tocher, D.R., Monroig, O., 2016.
Long-chain polyunsaturated fatty acid biosynthesis in chordates: insights into the
evolution of Fads and Elovl gene repertoire. Prog. Lipid Res. 62, 25–40. https://doi.
org/10.1016/j.plipres.2016.01.001.
Coutteau, P., Van Stappen, G., Sorgeloos, P., 1996. A standard experimental diet for the
study of fatty acid requirements of weaning and first Ongrowing stages of the
S.J.S. Houston et al.
European sea bassdicentrarchus labraxL.:Comparison of extruded and extruded/
coated diets. Arch. für Tierernaehrung 49, 49–59. https://doi.org/10.1080/
17450399609381863.
Cowey, C.B., 1992. Nutrition: estimating requirements of rainbow trout. Aquaculture
100, 177–189. https://doi.org/10.1016/0044-8486(92)90370-z.
Eickhart, E., 2017. Finding the best trade-off point on a curve [WWW document]. Stack
Overflow. URL https://stackoverflow.com/questions/2018178/finding-the-be
st-trade-off-point-on-a-curve. (Accessed 17 July 2017).
Gahl, M.J., Finke, M.D., Crenshaw, T.D., Benevenga, N.J., 1991. Use of a four-parameter
logistic equation to evaluate the response of growing rats to ten levels of each
indispensable amino acid. J. Nutr. 121, 1720–1729. https://doi.org/10.1093/jn/
121.11.1720.
Gatesoupe, F.J., Leger, C., Boudon, M.-C., Luquet, P., 1977. Alimentation lipidique du
turbot (Scophthalmus maximus L.). II. Influence de la supplementation en esters
methyliques de l’acide linolenique et de la complementation en acides gras de la
serie w 9 sur la croissance. Ann. Hydrobiol. 8, 247–254.
Gaylord, T.G., Rawles, S.D., Davis, K.B., 2005. Dietary tryptophan requirement of hybrid
striped bass (Morone chrysops x M. saxatilis). Aquac. Nutr. 11, 367–374. https://doi.
org/10.1111/j.1365-2095.2005.00360.x.
Glencross, B.D., 2009. Exploring the nutritional demand for essential fatty acids by
aquaculture species. Rev. Aquac. 1, 71–124. https://doi.org/10.1111/j.1753-
5131.2009.01006.x.
Glencross, B.D., Smith, D.M., 2001. Optimizing the essential fatty acids, eicosapentaenoic
and docosahexaenoic acid, in the diet of the prawn, Penaeus monodon. Aquac. Nutr.
7, 101–112. https://doi.org/10.1046/j.1365-2095.2001.00158.x.
Glencross, B.D., Tocher, D.R., Matthew, C., Gordon Bell, J., 2014. Interactions between
dietary docosahexaenoic acid and other long-chain polyunsaturated fatty acids on
performance and fatty acid retention in post-smolt Atlantic salmon (Salmo salar).
Fish Physiol. Biochem. https://doi.org/10.1007/s10695-014-9917-8.
Hauschild, L., Pomar, C., Lovatto, P.A., 2010. Systematic comparison of the empirical
and factorial methods used to estimate the nutrient requirements of growing pigs.
Animal 4, 714–723. https://doi.org/10.1017/s1751731109991546.
Hernandez-Llamas, A., 2009. Conventional and alternative dose–response models to
estimate nutrient requirements of aquaculture species. Aquaculture 292, 207–213.
https://doi.org/10.1016/j.aquaculture.2009.04.014.
Houston, S.J.S., Karalazos, V., Tinsley, J., Betancor, M.B., Martin, S.A.M., Tocher, D.R.,
Monroig, O., 2017. The compositional and metabolic responses of gilthead seabream
(Sparus aurata) to a gradient of dietary fish oil and associated n-3 long-chain PUFA
content. Br. J. Nutr. 118, 1010–1022.
Ibeas, C., Izquierdo, M., Lorenzo-Herández, A., 1994a. Effect of different levels of highly
unsaturated fatty acids on growth and fatty acid composition of juvenile gilthead
bream (Sparus aurata). Aquaculture 124, 285. https://doi.org/10.1016/0044-8486
(94)90393-x.
Ibeas, C., Izquierdo, M.S., Lorenzo, A., 1994b. Effect of different levels of n− 3 highly
unsaturated fatty acids on growth and fatty acid composition of juvenile gilthead
seabream (Sparus aurata). Aquaculture 127, 177–188. https://doi.org/10.1016/
0044-8486(94)90424-3.
Ibeas, C., Cejas, J., Gómez, T., Jerez, S., Lorenzo, A., 1996. Influence of dietary n − 3
highly unsaturated fatty acids levels on juvenile gilthead seabream (Sparus aurata)
growth and tissue fatty acid composition. Aquaculture 142, 221–235. https://doi.
org/10.1016/0044-8486(96)01251-3.
Ibeas, C., Cejas, J.R., Fores, R., Badía, P., Gómez, T., Hernández, A.L., 1997. Influence of
eicosapentaenoic to docosahexaenoic acid ratio (EPADHA) of dietary lipids on
growth and fatty acid composition of gilthead seabream (Sparus aurata) juveniles.
Aquaculture 150, 91–102. https://doi.org/10.1016/s0044-8486(96)01473-1.
Izquierdo, M.S., 1996. Essential fatty acid requirements of cultured marine fish larvae.
Aquac. Nutr. 2, 183–191. https://doi.org/10.1111/j.1365-2095.1996.tb00058.x.
Izquierdo, M., Domínguez, D., Jiménez, J.I., Saleh, R., Hernández-Cruz, C.M.,
Zamorano, M.J., Hamre, K., 2019. Interaction between taurine, vitamin E and
vitamin C in microdiets for gilthead seabream (Sparus aurata) larvae. Aquaculture
498, 246–253. https://doi.org/10.1016/j.aquaculture.2018.07.010.
Jin, M., Lu, Y., Yuan, Y., Li, Y., Qiu, H., Sun, P., Zhou, Q.-C., 2017. Regulation of growth,
antioxidant capacity, fatty acid profiles, hematological characteristics and
expression of lipid related genes by different dietary n-3 highly unsaturated fatty
acids in juvenile black seabream (Acanthopagrus schlegelii). Aquaculture 471, 55–65.
https://doi.org/10.1016/j.aquaculture.2017.01.004.
Kalogeropoulos, N., Alexis, M.N., Henderson, R.J., 1992. Effects of dietary soybean and
cod-liver oil levels on growth and body composition of gilthead bream (Sparus
aurata). Aquaculture 104, 293–308. https://doi.org/10.1016/0044-8486(92)90211-
3.
Lassiter, J.W., Edwards, H.M., 1982. Animal Nutrition. Prentice-Hall International,
Hemel Hempstead. https://doi.org/10.5962/bhl.title.155283.
Lee, S.-M., Lee, J.H., Kim, K.-D., 2003. Effect of dietary essential fatty acids on growth,
body composition and blood chemistry of juvenile starry flounder (Platichthys
stellatus). Aquaculture 225, 269–281. https://doi.org/10.1016/s0044-8486(03)
00295-3.
Leu, M.Y., Yang, S.D., Wu, C.H., Liou, C., 1994. Effect of dietary n-3 highly unsaturated
fatty acids on growth, feed efficiency and fatty acid composition of juvenile silver
bream. Asian Fish. Sci 7, 233–240.
Lochmann, R.T., Gatlin, D.M., 1993. Essential fatty acid requirement of juvenile red
drum (Sciaenops ocellatus). Fish Physiol. Biochem. 12, 221–235. https://doi.org/
10.1007/bf00004370.
Lupatsch, I., 2005. Protein and energy requirements in Mediterranean species. Cah.
Options Méditerranéennes 63, 9–18.
10
Aquaculture 556 (2022) 738308
Lupatsch, I., Kissil, G.W., Sklan, D., 2003. Defining energy and protein requirements of
gilthead seabream (Sparus Aurata) to optimize feeds and feeding regimes. Isr. J.
Aquac. - Bamidgeh. https://doi.org/10.46989/001c.20354.
Luzzana, U., Hardy, R.W., Halver, J.E., 1998. Dietary arginine requirement of fingerling
coho salmon (Oncorhynchus kisutch). Aquaculture 163, 137–150. https://doi.org/
10.1016/s0044-8486(98)00228-2.
Mai, K., Zhang, L., Ai, Q., Duan, Q., Zhang, C., Li, H., Liufu, Z., 2006. Dietary lysine
requirement of juvenile Japanese seabass, Lateolabrax japonicus. Aquaculture 258,
535–542. https://doi.org/10.1016/j.aquaculture.2006.04.043.
Mercer, L.P., May, H.E., Dodds, S.J., 1989. The determination of nutritional requirements
in rats: mathematical modeling of sigmoidal, inhibited nutrient-response curves.
J. Nutr. 119, 1465–1471. https://doi.org/10.1093/jn/119.10.1465.
Murillo-Gurrea, Coloso, Borlongan, Serrano, 2001. Lysine and arginine requirements of
juvenile Asian sea bass (Lates calcarifer). J. Appl. Ichthyol. 17, 49–53. https://doi.
org/10.1046/j.1439-0426.2001.00242.x.
National Research Council, Council, N.R., others, 2011. Nutrient Requirements of Fish
and Shrimp, Aquaculture International. Springer Science and Business Media LLC.
https://doi.org/10.1007/s10499-011-9480-6.
Oliva-Teles, A., 2012. Nutrition and health of aquaculture fish. J. Fish Dis. 35, 83–108.
https://doi.org/10.1111/j.1365-2761.2011.01333.x.
Oliva-Teles, A., Lupatsch, I., Nengas, I., 2011. Nutrition and feeding of Sparidae.
Sparidae. https://doi.org/10.1002/9781444392210.ch7.
Pesti, G.M., Vedenov, D., Cason, J.A., Billard, L., 2009. A comparison of methods to
estimate nutritional requirements from experimental data. Br. Poult. Sci. 50, 16–32.
https://doi.org/10.1080/00071660802530639.
R Core Team, 2013. R: A Language and Environment for Statistical Computing.
Ritz, C., Streibig, J.C., 2008. Nonlinear Regression with R. Springer Science & Business
Media.
Rodehutscord, M., Pack, M., 1999. Estimates of essential amino acid requirements from
dose-response studies with rainbow trout and broiler chicken: effect of mathematical
model. Arch. für Tierernaehrung 52, 223–244. https://doi.org/10.1080/
17450399909386164.
Salini, M.J., Poppi, D., Turchini, G.M., Glencross, B.D., 2016. Defining the allometric
relationship between size and individual fatty acid turnover in barramundi Lates
calcarifer. Comp. Biochem. Physiol. Part A Mol. Integr. Physiol. 201, 79–86. https://
doi.org/10.1016/j.cbpa.2016.06.028.
Sargent, J.R., McEvoy, L.A., Bell, J.G., 1997. Requirements, presentation and sources of
polyunsaturated fatty acids in marine fish larval feeds. Aquaculture 155, 117–127.
https://doi.org/10.1016/s0044-8486(97)00122-1.
Sargent, J.R., Tocher, D.R., Bell, J.G., 2003. The lipids. Fish Nutr. https://doi.org/
10.1016/b978-012319652-1/50005-7.
Shearer, K.D., 1995. The use of factorial modeling to determine the dietary requirements
for essential elements in fishes. Aquaculture 133, 57–72.
Shearer, 2000. Experimental design, statistical analysis and modelling of dietary nutrient
requirement studies for fish: a critical review. Aquac. Nutr. 6, 91–102. https://doi.
org/10.1046/j.1365-2095.2000.00134.x.
Simó-Mirabet, P., Felip, A., Estensoro, I., Martos-Sitcha, J.A., de las Heras, V., Calduch-
Giner, J., Pérez-Sánchez, J., 2018. Impact of low fish meal and fish oil diets on the
performance, sex steroid profile and male-female sex reversal of gilthead sea bream
(Sparus aurata) over a three-year production cycle. Aquaculture 490, 64–74.
Skalli, A., Robin, J.H., 2004. Requirement of n-3 long chain polyunsaturated fatty acids
for European sea bass (Dicentrarchus labrax) juveniles: growth and fatty acid
composition. Aquaculture 240, 399–415. https://doi.org/10.1016/j.
aquaculture.2004.06.036.
Skalli, A., Robin, J.H., Le Bayon, N., Le Delliou, H., Person-Le Ruyet, J., 2006. Impact of
essential fatty acid deficiency and temperature on tissues’ fatty acid composition of
European sea bass (Dicentrarchus labrax). Aquaculture 255, 223–232. https://doi.
org/10.1016/j.aquaculture.2005.12.006.
Speiss, A.N., 2013. Introducing ‘propagate’ [WWW Document]. RMazing. URL https
://rmazing.wordpress.com/2013/08/31/introducing-propagate/. (Accessed 27 May
2017).
Syakur, M.A., Khotimah, B.K., Rochman, E.M.S., Satoto, B.D., 2018. Integration k-means
clustering method and elbow method for identification of the best customer profile
cluster. In: IOP Conference Series: Materials Science and Engineering, p. 12017.
Takeuchi, T., 1997. Essential fatty acid requirements of aquatic animals with emphasis
on fish larvae and fingerlings. Rev. Fish. Sci. 5, 1–25. https://doi.org/10.1080/
10641269709388592.
Takeuchi, T., Shiina, Y., Watanabe, T., 1992. Suitable levels of n-3 highly unsaturated
fatty acids in diet for fingerlings of Red Sea bream. Nippon Suisan Gakkaishi 58,
509–514. https://doi.org/10.2331/suisan.58.509.
Tocher, D.R., 2010. Fatty acid requirements in ontogeny of marine and freshwater fish.
Aquac. Res. 41, 717–732. https://doi.org/10.1111/j.1365-2109.2008.02150.x.
Tocher, D.R., Glencross, B.D., 2015. Lipids and fatty acids. Diet. Nutr. Addit. Fish Heal.
https://doi.org/10.1002/9781119005568.ch3.
Watanabe, T., 1982. Lipid nutrition in fish. Comp. Biochem. Physiol. Part B Comp.
Biochem. 73, 3–15. https://doi.org/10.1016/0305-0491(82)90196-1.
Whalen, K.S., Brown, J.A., Parrish, C.C., Lall, S.P., Goddard, J.S., others, 1998. Effect of
dietary n-3 HUFA on growth and body composition of juvenile yellowtail flounder
(Pleuronectes ferrugineus [Limanda ferruginea]). Bull. Aquac. Assoc. Canada 21–22.
Wickham, H., 2016. ggplot2-Elegant Graphics for Data Analysis. Springer International
Publishing, Cham, Switz.
Xie, D., Chen, C., Dong, Y., You, C., Wang, S., Monroig, Ó., Li, Y., 2021. Regulation of
long-chain polyunsaturated fatty acid biosynthesis in teleost fish. Prog. Lipid Res.
101095.
Another random document with
no related content on Scribd:
 J. C. Hoadley

I was able to get from our landlord and purchasers of our tools the
necessary extension of time, and made the engine for him. It and the
loom were each a complete success. Mr. Waters told me long after
that he never observed a single variation from exact uniformity of
motion, without which his loom would have had to be abandoned.
I had one day the pleasure of meeting there the president of the
Lancaster mills, the only other great industry of Clinton, who had
come over expressly to examine the running of our engine. Before
he left he said to me that the engine certainly presented a
remarkable advance in steam engineering.
I saw there one thing that interested me greatly. That was, the
method of painting wire cloth. This was carried on in a large tower
high enough to enable a twenty-yard length of the “cloth” to be
suspended in it. This was taken through a tub of paint, and drawn
slowly upward between three successive pairs of rollers, the last pair
of india-rubber, held firmly together. By these the paint was
squeezed into every corner, both sides were thoroughly painted, and
the surplus paint removed, so that every mesh was clear, a uniform
perfection unattainable by hand painting, and two boys would paint
in ten minutes as much as a painter could paint in a day. I think this
was an invention by Mr. Waters.
With the completion of the engine for the Clinton Wire Cloth
Company, the manufacture of the high-speed engine was closed for
three years, from the spring of 1873 to the spring of 1876.
This long rest proved to be most valuable. Looking back upon it, I
have always been impressed with its importance at that very time to
the development of the high-speed system.
The design of the engine needed to be revised, and this revision
involved study, to which time and leisure were essential.
I had also an order from Elliott Brothers of London, to prepare a
new and enlarged edition of the pamphlet descriptive of the Richards
Indicator. I determined to make this a comprehensive book,
embracing new information required by the steam engineer, so far as
I knew it. This was published simultaneously in London and New
York in the summer of 1874.
 I was enabled also to turn to account the report of the experiments
of M. Regnault, which I had been at so much trouble to get, and with
the help of English authorities to prepare and embody in this book
Tables of the Properties of Saturated Steam, which the American
Society of Mechanical Engineers honored me by adopting as its
standard.
I felt warranted in giving to this edition an amended title, as
follows: “A treatise on the Richards Steam Engine Indicator, and the
Development and Application of Force in the Steam Engine.”
This also was a job requiring much time and undivided application.
It is needless to say that without this long and entire rest from
business neither of these tasks could have been undertaken.
I found in the Astor Library a remarkable old book, entitled “Canon
triangulorum,” published at Frankfurt in 1612, containing a Table of
Natural Trigonometrical Functions, computed for every minute of arc,
and extended to the fifteenth place of decimals. The column of
versed sines enabled me to prepare tables exhibiting the rates of
acceleration and retardation of the motion of a piston controlled by a
crank, neglecting the effect of the angular vibration of the
connecting-rod. This effect was afterwards shown separately. For my
treatment of this subject, I must refer the reader to the book itself.
A little incident in connection with this work, which made a deep
impression on my mind, and has since afforded me some food for
reflection, seems worth relating. The printing was done in London,
and I did not see the proof, so I had to take especial pains with the
copy, having no opportunity to revise it. I was living in Harlem, and at
one time having no suitable envelope for mailing, and none being
obtainable there, I took a Third Avenue horse-car for an eight-mile
ride down to the New York post office, intending to get some
envelopes at a stationery store on Beekman Street, and mail the
portion of the copy which I then had ready at the general post office.
I had hardly taken my seat when Mr. Allen got into the car. He was
living in Mott Haven, and I had not seen him for a long time. Besides
ourselves the car was nearly if not quite empty. He came and sat
down by me, and I opened my copy and read to him something in
which I knew he would be interested. He said to me, in his gentle
way, “You would not express it exactly that way, would you?” On the
instant it flashed on my mind that I had made a stupid blunder, and I
replied, “I guess I wouldn’t,” and, thanking him for calling my
attention to it, I left the car, and returned home and corrected it. I
have quite forgotten what the point was, and if I remembered it, I
would not tell. But I have often asked myself who sent Mr. Allen
there, saving me from publishing a mortifying blunder. I expect some
sweet spirit will tell me before long.

The Prototype of the Modern High-speed Engine, Fly-wheel Side.

The Prototype of the Modern High-speed Engine, Crank Side.

 As soon as this book was off my hands, I devoted myself to the
revision and standardizing of the engine.
As made up to that time, it was not reversible, and the valves
could not be handled. It could not therefore be used in rolling-mills,
the field to which I felt already that it was especially adapted.
Moreover, every engine should be capable of being backed in
starting, as otherwise whenever it had stopped with the piston at a
point later than the latest point of cut-off, or say in the last half of the
stroke, which it would do half the time, it would need to be pulled
around by hand to a position in which one of the admission ports
would be open. This in a large engine, or one connected with
extensive lines of shafting, would be a serious matter, so much so
that in some engines little starting cylinders are required.

Longitudinal Section of Cylinder and Valves.

Cross-section of Cylinder and Valves.
 Elevation and Plan of Valve Connections.

I had also determined to use the equilibrium admission valves with

adjustable pressure plates, according to the drawings sent to me by
Mr. Allen in 1863, and to abandon the separate steam chest, and put
the exhaust valves on the opposite side of the cylinder.
Then the engine needed to be standardized, so as to cover the
field with the fewest number of sizes, symmetrically distributed. The
existing practice with all makers of engines had been to let the
purchaser dictate the size and speed of the engine he wanted, a
practice which resulted in a lot of patterns and drawings not adapted
to other people’s requirements, and not properly distributed. For an
organized manufacturing business, this habit must be entirely broken
up.
Mr. Allen had in his shop in Mott Haven an unoccupied second
story, in which I had stored our patterns and drawings and drawing
implements. Here I established my quarters, and spent my working
hours until this second job was finished.
 The two perspective views of opposite sides of the engine, show
these changes as they appear externally, and the remaining views
show some constructive details.
These latter show the exhaust valves transferred to the front side
of the engine, and located so as to drain the cylinder, and the
admission valves set at different elevations, to accommodate the
differential connection, the abandonment of the separate steam-
chest, and this chest with the exhaust chambers cast with the
cylinder, with openings over the valves; the levers by which the
differential movements are given to the admission valves; and the
single-link rod, and the gab by which this rod is unhooked, with the
method of moving the admission valves by hand.
In place of the levers on the steam rock shaft, I at that time drew
cast-iron disks, which being polished and vibrating in place I thought
very handsome. They gave me lots of trouble, till I learned enough to
get rid of them, the story of which I will tell by and by. The front view
shows the admission valve stems balanced by being extended
through at the back end, a feature which helped the governor action
when high steam pressures were employed, but which was
abandoned as unnecessary after I abandoned the disks on the
rocker shaft.
 First Arrangement of Exhaust Valves.

The first two figures show the valves in section and the adjustable
pressure plate and mode of its adjustment. The closeness of the
piston to the head may be observed. I never allowed more than one-
eighth inch clearance, and never had a piston touch the head. This
was because the connecting-rod maintained a constant length, the
wear of the boxes being taken up in the same direction.
These illustrations show the exhaust valves after alteration made
several years later in Philadelphia. As first designed by me, these
are shown in the foregoing sectional views. As will be seen, the
exhaust valves lay with their backs towards the cylinder, worked
under the pressure of the steam in the cylinder, made four openings
for release and exhausted through the cover.
I consented to the change in Philadelphia because this
arrangement involved too much waste room, but the change was not
satisfactory after all. I had become possessed with the idea that the
engine running at high speed needed 50 per cent. more room for
exhausting than for admission. This was not the case. I have always
regretted that I did not retain this design, and content myself with
reducing the exhaust area.
 The lightness of the piston in this view will be observed. This was
a special design for adapting the engine to be run at 200 revolutions,
giving 1200 feet piston travel per minute. The stuffing-box was a
freak which was abandoned.
The next figures show the valve-stem guides, rocking-levers,
coupling-rods and gab, which latter when thrown over unhooks the
link-rod, as is done on steamboat engines.
The following figures show the construction of the main bearing
with adjustments on opposite sides, by which the shaft is kept in
exact line, and shows also the solid support of the shaft quite out to
the hub of the crank. This view contains one error. The cap is not
made a binder. I relied on the strength of the thick continuous web of
the bed under the boxes in addition to the depth of the bed. But we
once had a bed break right here under enormous strain, and since
then the caps have been made binders. It will be observed that the
wedges are drawn upward to tighten the boxes. It is not necessary to
explain why.
Main Bearing.
 Front View of Wiper
Section on the Line a-b
Center Line of Shaft

Eccentric and Crosshead Pin Crank-pin Lubricator.

Lubricator.

The above left-hand cut shows the automatic lubricator of the

eccentric and the cross-head pin. The stud A, on the eccentric strap
and on the strap of the connecting-rod, carries a curved blade, a,
which at the beginning of each forward stroke rises to take the drop
of oil from the stem of the sight-feed lubricator. This is set on an arm
fixed on the cap of the main-bearing and on a bridge between the
upper guide-bars. The drop is made sure to come to this central
point by a wire B filling the brass tube, grooved on opposite sides
and terminating in a point.
The automatic lubrication of the crank-pin presented a still more
serious problem. It was solved by the construction shown, in the right
hand view, which will be understood without further description. It will
be seen that here the oil tube is inclined, and the drop follows it to a
point on its lower side. Both these lubricators proved absolutely
reliable. The last one is equally applicable on double-crank engines.
Fly-wheels.
Made
Dimensions when Practicable
of to Serve also
Cylinders. Travel Indicated Powers. as Belt-Drums.
Revolutions of Piston Without With
of Crank in Feet Conden- Conden- Weight
Bore. Stroke. per Minute. per Minute. sation. sation. Diameter. of Rim. V

Horse Horse
Inches. Inches. Powers. Powers. Feet. Inches. Lbs.
6 12 350 700 25 3 350
7 12 350 700 35 3 6 400
8 16 280 746 45 60 4 650
9 16 280 746 60 75 4 6 700
10 20 230 766 75 100 5 1300
11.5 20 230 766 100 125 5 6 1450
13 24 200 800 130 160 6 6 2100
14.5 24 200 800 160 200 7 2350
16 30 165 825 200 260 8 4000
18 30 165 825 250 330 9 4000
20 36 140 840 320 400 10 6000
22 36 140 840 400 500 11 6000
24 42 125 875 480 620 12
26 42 125 875 560 730 13
28 48 112.5 900 670 870 16
 32 48 112.5 900 870 1140
36 48 112.5 900 1100 1430
40 48 112.5 900 1360 1750
44 48 112.5 900 1600 2100

The powers are those given by an initial pressure of 85 lbs. on the square inch, cut
one quarter of the stroke. For the best economy steam should not be cut off earli
unless a higher pressure is carried. At the latest point of cut off, the powers de
double those given in the above Table. The engines can be worked under
pressures, with corresponding increase of power.

After considerable study I finally adopted the above table of

standard sizes and speeds, covering the ground from 25 horse-
power up with nineteen sizes.
As the bed could not be reversed, I needed both a right-hand and
a left-hand bed for each size. I avoided half of this expense in
patterns by planning two diameters of cylinders with the same
stroke, and making one bed answer for both.
Until I found something else to do, I employed myself in preparing
complete drawings for three or four smaller sizes of engines; a work
which afterwards proved exceedingly useful.
 CHAPTER XXI

Production of an Original Surface Plate.

will introduce here a description of the method of

producing an original surface plate.
The production of mechanically true planes by the
method of scraping was first suggested by Mr.
Whitworth, and was brought to perfection in his works.
Having had and having improved the opportunity there
to study this system, and having employed it largely in the
manufacture of high-speed engines, it seems to me that an account
of it should find a place in these reminiscences, especially as the
importance of mechanical truth is coming to be more and more
realized in this country. I will therefore describe the process of
producing an original standard surface plate.
The first point, of course, is the design. The square form, 30
inches square, has been found most suitable. I could not, however,
use this form myself, a long surface plate being required for the
guide-bars and steam-chest joints of my engine.
The plate must be incapable of deflection. To insure this it is ribbed
on the under side with ribs seven inches deep, and is supported at
three points, equidistant from each other and from the center, so that
its equal support cannot vary, whatever may be the surface on which
it stands. It is provided on two opposite sides with handles, by which
it can be lifted and rotated. The arrangement of the ribs and feet is
here shown.
It must be cast of hard and close-grained iron, which will possess
the most durable qualities, in a baked mold without a cope, so that
the gas shall escape most freely. As cast, the plate should be one
inch thick. About three eighths of an inch is planed off, removing all
dirt, and leaving a perfectly sound surface, with a thickness of about
five eighths of an inch. Three of these plates are made. After these
have been planed, their edges squared and steel handles put in they
are delivered to the fitter.

Surface Plate for Producing a True Plane.

 I will first describe the tool used in scraping. Originally this was a
hooked tool, and the scraping was done by a drawing motion. Two of
these tools were employed, one for the roughing work, in which the
hook projected downward about three quarters of an inch, and the
other for fine scraping. In the latter the hook projected downward
only about one quarter of an inch, and absolute freedom from
vibration was aimed at. These tools were used for a number of
years, but afterwards a radical change was made. The modern
method is to employ a single straight tool, like a carpenter’s chisel,
about an inch and a quarter wide and an eighth of an inch thick, with
a square end. This end is slightly curved, and its corners are
rounded to avoid scratching the plate. The scraping is done by a
pushing motion.
This tool has been found preferable on all accounts. Projections
needing to be removed are in front of the tool, instead of being
covered by it. The tool is perfectly rigid, and can be inclined to the
surface at any desired angle. The cutting edge is a square angle,
and being well supported keeps sharp for a considerably longer time
than when it is an acute angle, and when ground or honed two
edges are formed. Moreover, the pushing motion is preferred.
Two of the plates only are first brought together. For disclosing the
high points, one of these is covered with a raddle made of finely
sifted red lead and oil. This is made quite stiff, and all of it that can
be removed by the palm of the hand is rubbed off, leaving only a
very thin uniform film on the surface. Any dust having been carefully
removed from both surfaces by a soft brush, one of these plates is
inverted on the other, and at one corner each plate is marked in the
edge with a prick-punch. The upper plate is then rubbed about on
the lower one for, say, half a minute. When lifted off, the high
portions of the surfaces are shown on one plate by the raddle put on,
and on the other by that rubbed off. The workman then gives to
these parts of the surfaces a general scraping, giving to his tool a
long sweep, say from four to six inches. This is repeated two or three
times, the stroke being shortened each time, and the upper plate
being placed in a position at right angles with its last one, which can
be determined by the prick-punch marks. This change of position is
necessary to avoid a cross-wind or spiral form. The scraping should
now extend over the entire surfaces, and these should have a
general uniform bearing on each other, with the points of contact
uniformly distributed and equally distinct. The work should be
continued in the same way until all these requirements are fulfilled.
Now appears the use of the third plate. The two surfaces thus
formed are sure to be, one of them convex and the other concave, in
some corresponding degree. The workman now numbers the plates,
by numbers stamped in the edges, these being marked Nos. 1 and
2, and the third plate No. 3. No. 2 is now set aside, and No. 3 is
scraped to fit No. 1. It is thus made a duplicate of No. 2. Next, No. 1
is set aside and Nos. 2 and 3 are brought together. Supposing these
to be convex, they will bear together at the middle point, on which
the upper plate will rock, and the degree of their convexity will thus
be shown. The workman then in the same manner scrapes these
plates equally to the best of his judgment, until their entire surfaces
are brought together, with equal distribution of the points of contact.
These two surfaces will now again be, one convex and the other
concave, though in a much less degree. The next step is to apply
No. 1, which is concave, to either No. 2 or No. 3, and scrape it to fit.
It is then applied to the other, of which it has now been made a
duplicate, and the same process is repeated, until the three plates
can be interchanged in any way, and will have a uniform general
bearing on each other, with equal distribution and distinctness of the
points of contact. We have thus, in a general way, produced three
demonstrated true planes, but the surfaces are yet far from the
desired approximation to absolute truth.
Now follows the fine scraping, which is not attempted until general
truth has thus been established. The object of this is to multiply the
points of contact and perfect their equal distribution and prominence.
For this operation no raddle is used, but the surfaces are rubbed
together dry. When the plates are separated, the points of contact
shine like stars. Here skill and care are pre-eminently required. The
scraping takes off only a dust. If too strong depressions may be
made deeper than before, and requiring the reduction of the entire
surface. The superiority of the modern tool is now especially shown.
By lowering the angle of the tool, the workman presents the slightly
curved edge to the surface in a position as nearly parallel with it as
he desires. Interchanges similar to the former ones are now
repeated, until the bright points are brought as close together as is
desired, with uniform distribution and distinctness. The tedious
operation is now finished, and these bright points remain as
witnesses.
The three plates were necessary to the production of one. They
have also a permanent use. They are indispensable to the
preservation of the true plane, which it has cost so much patient
labor to produce. The date of their completion is stamped on their
edges. Then plates 1 and 2 are put away in the store-room, their
surfaces carefully protected from rust or injury, which last is best
avoided by inverting one on the other, and No. 3 is put into use. A
prominent use is for the production of smaller plates or straight-
edges adapted to special purposes. After a while, perhaps in a little
while, this plate loses its truth by unequal wear. Indeed, speaking
with absolute truth, it may be said that the first time this plate is used
it is ruined. But by taking pains to use different parts of its surface as
equally as possible, it may be kept in fair condition for some time. It
can at any time be restored to its original condition by scraping it to
No. 2, taking the same pains to turn it one quarter way around at
every rub. In the course of time No. 2 will itself become worn
unequally, when its truth can be restored by rubbing it on No. 1.
Finally the three plates can all be restored to their original condition
by rubbing them together interchangeably as at first. Thus the true
plane can be absolutely perpetuated.
The importance of this work can only be realized when we
consider that the true plane affords the only means by which true
cylindrical work also can be either produced or verified. It is thus
seen to be fundamental to all mechanical truth.
 CHAPTER XXII

Efforts to Resume the Manufacture. I Exhibit the Engine to Mr. Holley. Contract
with Mr. Phillips. Sale of Engine to Mr. Peters.

n the years ’74 and ’75 I was filled with eagerness to

get the engine on its legs again, and tried a number of
schemes in vain. One morning I read in the paper that
Alexander L. Holley had just returned from Europe,
where he had been making a tour of the steel-making
establishments, studying both the Bessemer and the
open hearth or Siemens-Martin processes, on a scheme of
interchanging improvements in manufacture between American and
foreign licensees under both these systems.
It occurred to me that Mr. Holley might be the very man I wanted. If
he could be got to recommend the engine to the steel-makers, they
might take it up for their own use. I had not applied the engine in
rolling-mill work, but felt sure that it would prove especially adapted
to that service. So I called on Mr. Holley at his home in Brooklyn. I
had never before met him, but I found that he knew something about
the engine from its exhibition in Paris, and from his brother-in-law,
Frederick J. Slade, then an officer of the New Jersey Steel Company,
and who was one of the engine’s warm admirers. I have already
mentioned Mr. Slade and the help he gave me while in Paris in
solving the problem of piston acceleration.
Alexander Lyman Holley