Processing and protein-fractionation characteristics of different polymeric membranes during filtration of skim milk at refrigeration temperatures

Accepted Manuscript Processing and protein-fractionation characteristics of different polymeric membranes during filtration of skim milk at refrigeration temperatures Shane V. Crowley, Veronica Caldeo, Noel A. McCarthy, Mark A. Fenelon, Alan L. Kelly, James A. O’Mahony PII:

S0958-6946(15)00016-3

DOI:

10.1016/j.idairyj.2015.01.005

Reference:

INDA 3792

To appear in:

International Dairy Journal

Received Date: 9 October 2014 Revised Date:

8 January 2015

Accepted Date: 9 January 2015

Please cite this article as: Crowley, S.V., Caldeo, V., McCarthy, N.A., Fenelon, M.A., Kelly, A.L., O’Mahony, J.A., Processing and protein-fractionation characteristics of different polymeric membranes during filtration of skim milk at refrigeration temperatures, International Dairy Journal (2015), doi: 10.1016/j.idairyj.2015.01.005. This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

ACCEPTED MANUSCRIPT 1

Processing and protein-fractionation characteristics of different polymeric membranes

2

during filtration of skim milk at refrigeration temperatures

3 4

RI PT

5 6

Shane V. Crowleya, Veronica Caldeoa, Noel A. McCarthyb, Mark A. Fenelonb, Alan L.

8

Kellya, James A. O’Mahonya,*

SC

7

9

M AN U

10 11 12 a

School of Food and Nutritional Sciences, University College Cork, Cork, Ireland

14

b

Teagasc Food Research Centre, Moorepark, Fermoy, Co. Cork, Ireland

19

*

Corresponding author. Tel.: +353 21 4903625

20

E-mail address: [email protected] (J. A. O’Mahony)

16 17 18

21

AC C

15

EP

TE D

13

22 23 24 25

1

ACCEPTED MANUSCRIPT 26

___________________________________________________________________________

27

Abstract

28 29

Serum protein concentrates (SPCs) were generated from reconstituted skim milk (3.2% protein) using lab-scale tangential-flow filtration at 3–4 °C. The influence of

31

membrane type on process performance (e.g., permeate flux) and protein-enrichment was

32

assessed with polyvinylidene-difluoride membranes (0.1 µm and 0.45 µm pore-size), and a

33

polyethersulfone membrane (1000 kDa cut-off). The 1000 kDa membrane exhibited the

34

highest starting flux (6.7 L m-2 h-1), followed by the 0.1 µm (5.4 L m-2 h-1) and 0.45 µm (4.8

35

L m-2 h-1) membranes. Flux decreased by >40% during filtration with the 1000 kDa and 0.1

36

µm membranes, while the decrease was lower (<20%) with the 0.45 µm membrane. β-Casein

37

comprised >97% of casein in SPCs from the 0.1 µm and 1000 kDa membranes. SPCs from

38

the 0.45 µm membrane had higher β-casein:αS-casein ratios than the feed and higher levels of

39

minor whey proteins (e.g., lactoferrin) relative to the other SPCs.

40

___________________________________________________________________________

SC

M AN U

TE D

EP AC C

41

RI PT

30

2

ACCEPTED MANUSCRIPT 42

1.

Introduction

43 44

In the manufacture of bovine milk-based first age infant milk formula (IMF) products, the protein component is formulated by reconstituting skim milk and whey powders in

46

combination to yield a casein:whey ratio of approximately 40:60, which is equivalent to that

47

of human milk (de Wit, 1998), and markedly different from that of bovine milk (80:20).

48

Efforts to extend the humanisation of IMF typically involve increasing the levels of β-casein,

49

α-lactalbumin and lactoferrin, while reducing levels of α-casein and β-lactoglobulin (Guo,

50

2014). Ingredients such as α-lactalbumin-enriched whey protein concentrates are already

51

being incorporated into IMFs (Kuhlman, Lien, Weaber, & O’Callaghan, 2005). Similarly, the

52

application of β-casein-enriched ingredients is expected to increase as manufacturers attempt

53

to further bridge the gap between the protein profile of IMFs and human milk. However,

54

industrial processes for the production of β-casein-enriched ingredients remain in the early

55

stages of development commercially.

SC

M AN U

TE D

56

RI PT

45

One of the most promising processes for the manufacture of β-casein-enriched serum protein concentrates (SPCs) is pressure-driven membrane filtration. Traditionally, filtration

58

processes (e.g., ultrafiltration, microfiltration), are performed at high temperatures of 45–50

59

°C, where there is the advantage of a high flux (short processing time for a given volume of

60

product) and limited growth of mesophilic bacteria (Gésan-Guiziou, 2013; Hurt, Zulewska,

61

Newbold, & Barbano, 2010; O’Mahony & Tuohy, 2013; Zulewska, Newbold, & Barbano,

62

2009). For the enrichment of β-casein, filtration is performed at cold (typically 4–10 °C)

63

temperatures (Christensen & Holst, 2014; Coppola Molitor, Rankin, & Lucey, 2014; Holland,

64

Corredig, & Alexander, 2011; Le Berre & Daufin, 1994; O’Mahony, Smith, & Lucey, 2014;

65

Seibel, Molitor, & Lucey, 2014; Woychik et al., 1992), under which conditions β-casein

66

exists in its monomeric state in the serum phase (Payens & van Markwijk, 1963; Rose, 1968).

AC C

EP

57

3

ACCEPTED MANUSCRIPT 67

In the dairy industry, the term “cold”, as applied to membrane separation processes, currently encompasses a relatively broad temperature range and can include any filtration

69

process temperatures <20 °C. As low filtration temperatures may have additional benefits

70

distinct from β-casein separation (e.g., reduction in the denaturation of whey proteins,

71

reduced fouling of membranes by calcium phosphate and reduced growth of thermophiles),

72

many manufacturers are currently transitioning from traditional “warm” (~40–50 °C)

73

processes to cold processes (Lawrence, Kentish, O’Connor, Barber, & Stevens, 2008).

74

However, the lower temperatures reduce diffusivity, and therefore the mass transfer

75

coefficient, with a concomitant marked decrease in permeate flux (Le Berre & Daufin, 1994).

76

This issue is exacerbated when polymeric membranes (e.g., polyvinylidene-difluoride,

77

PVDF, and polyethersulfone, PES) are used instead of ceramic membranes, as the latter

78

generally have considerably higher flux values (Beckman, Zulewska, Newbold, & Barbano,

79

2010; Zulewska et al., 2009). For many years, ceramic membranes were the most widely used

80

material for membrane separation processes in the dairy industry, due to their excellent pH-,

81

temperature- and cleaning-tolerance, in addition to high flux performance; however, in recent

82

years, polymeric membranes have become increasingly popular due to their cost-

83

effectiveness (O’Mahony & Tuohy, 2013).

SC

M AN U

TE D

EP

The materials and conditions that have been described for the enrichment of β-casein

AC C

84

RI PT

68

85

using membrane filtration vary widely, and few researchers have provided a detailed analysis

86

of processing performance. Woychik (1992) reported filtration of milk at temperatures of 2–8

87

°C using microfiltration membranes with pore sizes of 0.1 and 0.2 µm or an ultrafiltration

88

membrane with a molecular mass cut-off of 100 kDa in a plate-and-frame configuration. Le

89

Berre and Daufin (1994) separated β-casein from sodium caseinate at 4 °C using tubular

90

ceramic membranes (ZrO2 filtering layer on carbon supporting layer) with pore sizes ranging

91

between 0.02 and 0.08 µm. Microfiltration of skim milk at 2–5 °C using 0.55 µm spiral-

4

ACCEPTED MANUSCRIPT wound polymeric (PVDF) membranes has also been reported (O’Mahony et al., 2014).

93

Holland et al. (2011) removed β-casein from skim milk using 80 kDa cut-off polymeric

94

ultrafiltration (PES) membranes in a plate-and-frame configuration at 7 °C. More recently,

95

Glas, te Biesebeke, Kromkamp, and Klarenbeek (2013) used different membranes with pore

96

sizes of between 0.3 and 0.5 µm in a spiral-wound configuration for enrichment of β-casein

97

from skim milk at much higher temperatures (10–20 °C), while Christensen and Holst (2014)

98

reported using a 800 kDa cut-off membrane in a spiral-wound configuration to generate a β-

99

casein-enriched permeate from micellar casein isolate. Most of these researchers noted that

SC

101

the separation of β-casein had potential application in infant formula products. As the above examples illustrate, studies on the enrichment of β-casein have involved

M AN U

100

RI PT

92

the use of a multitude of membrane materials, pore sizes and cut-offs, configurations and

103

processing temperatures. Presently, the optimal materials and conditions for the filtration of

104

skim milk at refrigeration temperatures are unclear. Le Berre and Daufin (1994) performed a

105

thorough analysis of processing characteristics during their β-casein enrichment experiments;

106

however, the researchers used a 1% (w/w) sodium caseinate solution as the feed and ceramic

107

membranes for the separation. In industrial applications for the development of infant

108

formula ingredients, skim milk is more likely to be used as the feed material for β-casein

109

enrichment, as it facilitates co-enrichment with whey proteins and the generation of a

110

separate “native” functional retentate stream enriched in casein micelles (O’Mahony et al.,

111

2014; Seibel et al., 2014).

EP

AC C

112

TE D

102

In this study, membrane filtration at refrigeration temperatures (≤4 °C) was performed

113

using polymeric microfiltration membranes (0.1 µm and 0.45 µm pore-size PVDF) and a

114

polymeric ultrafiltration membrane (1000 kDa cut-off PES), with detailed analysis

115

undertaken of the filtration process itself, in addition to the composition, physicochemical

5

ACCEPTED MANUSCRIPT 116

properties and protein profile of the process streams generated, with a view to developing a

117

protein base for IMF manufacture.

118 119

2.

Materials and methods

2.1.

Materials

121

RI PT

120

122

Low-heat skim milk powder (SMP) was supplied by the Irish Dairy Board (Fermoy,

SC

123

Ireland). Reconstituted SMP was prepared in deionised water under constant magnetic

125

stirring at 22 °C to attain a 3.2% (w/w) true protein suspension. Reconstituted SMP was then

126

stored at 4 °C for ~23 h to ensure complete rehydration.

M AN U

124

127 128

2.2.

Membrane filtration: processing parameters, process analysis and cleaning

130

TE D

129

Filtration experiments were performed at lab-scale using a pressure-driven, tangentialflow filtration device (Pellicon 2 mini-holder; Merck-Millipore, Tullagreen, Carrigtwohill,

132

Ireland), as described by Crowley et al. (2015), who reported using the unit for ultrafiltration

133

of reconstituted SMP and IMF, with the following modifications: in the present study, both

134

PES (Biomax, Merck-Millipore) and PVDF (Durapore, Merck-Millipore) membranes were

135

used in individual trials (Table 1); the heat-exchanger (plate and frame) was used to control

136

temperature at between 3–4 °C; an equilibration time of at least 30 min was allowed to ensure

137

stable conditions (e.g., temperature and flux). Reconstituted SMP was concentrated to a

138

volume concentration factor (VCF) of 3 by removing permeate, with the retentate being

139

continuously recirculated back to the feed. Samples of the retentate and permeate were taken

140

for analysis when VCF = 3, while the feed was sampled before concentration (VCF = 1).

AC C

EP

131

6

ACCEPTED MANUSCRIPT 141

Critical flux was determined as described by Crowley, O’Callaghan, Kelly, Fenelon, and

142

O’Mahony (2015). The filtration system was cleaned fully after each concentration process and critical

143

flux determination using flushing and recirculation steps with water, sodium hypochlorite and

145

phosphoric acid for PVDF membranes and water and NaOH for the PES membrane. The

146

contribution of the membrane, reversible fouling and irreversible fouling to hydraulic

147

resistance were determined for each membrane by resistance-in-series modelling of water

148

flux data at 25 °C on clean and fouled membranes, according to Darcy’s law (Beckman &

149

Barbano, 2013): viscosity of water was taken as 8.91 × 10-4 Pa s, water flux was determined

150

at a TMP of 0.1 bar, and a cross-flow velocity of 0.16 m s-1 was used. The degree of fouling

151

was determined as described by Beckman and Barbano (2010).

M AN U

SC

RI PT

144

152 153

2.3.

Composition and physicochemical analyses

TE D

154

Total nitrogen (TN) and non-protein nitrogen (NPN) were determined as per ISO

156

8968-IDF 20 (ISO/IDF, 2001), using the Kjeldahl method and a nitrogen-to-milk protein

157

conversion factor of 6.38 for calculation of crude protein (CP) and true protein (TP). A

158

HB43-S Halogen rapid moisture analyser (Mettler Toledo GmbH, Schwerzenbach,

159

Switzerland) was used to measure total solids (TS) content. Conductivity was measured at 5

160

°C with a five-ring conductivity measuring cell (Metrohm Ireland Ltd., Carlow, Ireland).

161

Apparent viscosity was measured with a Thermo Scientific HAAKE RotoVisco 1 viscometer

162

(Karlsruhe, Germany) at a shear rate of 300 s-1, using a cup and bob geometry cooled with

163

water at 5 °C. A portable refractometer (N1-α, 0-32% Brix, Atago USA Inc., Bellevue, WA,

164

USA) was used to measure the Brix value of the retentate during concentration, to estimate

165

changes in solids content.

AC C

EP

155

7

ACCEPTED MANUSCRIPT 166 167

2.4.

Electrophoresis

168

Qualitative assessment of the protein profile of feed, retentate and permeate streams

169

was performed using electrophoresis. Reducing sodium dodecyl sulphate-polyacrylamide gel

171

electrophoresis (SDS-PAGE) was performed as described by Laemmli (1970), with minor

172

modifications. Prepared samples were run on pre-cast 4-20% acrylamide Tris-glycine gels

173

(Pierce Protein Research Products, Scoresby, Australia) using an AcquaTank mini gel unit

174

(Acquascience, Bellbrook Industrial Estate, Uckfield, UK). SDS-PAGE gels were scanned

175

using a desktop scanner (HP Scanjet G4010, HP, Leixlip, Ireland). All gels were Coomassie-

176

stained.

M AN U

SC

RI PT

170

177 178

2.5.

Reverse-phase high-performance liquid chromatography

180

TE D

179

The protein profile of feed, retentate and permeate streams was measured quantitatively using a Varian reverse-phase high-performance liquid chromatography (RP-

182

HPLC) system (Varian Associates Inc., Walnut Creek, CA, USA) comprising an

183

autosampler, a ProStar solvent delivery system with three pumps, and a ProStar

184

programmable multi-wavelength spectrophotometer. The RP-HPLC system was interfaced

185

with a PC with Varian Star Workstation v.6 software package for system control and data

186

acquisition

187

AC C

EP

181

A Phenomenex Widepore XB-C18 reverse-phase column with 3.6 µm particle size

188

and 250 × 4.6 mm SecurityGuard Ultra UPLC Widepore guard column for 4.6 mm ID

189

(Phenomenex, Macclesfield, UK) was used for protein separation. The column temperature

190

was maintained at 45 °C. Elution was monitored at 214 nm at a flow rate of 1 mL min-1 and a

8

ACCEPTED MANUSCRIPT mobile phase of two solvents, A, containing 10% acetonitrile (HPLC far UV grade; Labscan

192

Ltd, Blackrock, Ireland) and 0.1% trifluoroacetic acid (TFA, sequencing grade; Sigma-

193

Aldrich Ireland Ltd, Arklow, Ireland) in deionised water and B, containing 90% HPLC-grade

194

acetonitrile, 9.93% deionised water and 0.07% TFA. The gradient was as follows: linear

195

gradient from 5% to 25% B in 5 min, increasing from 25% to 45% B in 25 min, increasing

196

from 45% to 95% B in 1 min, holding at 95% B for 9 min, and reducing from from 95% B to

197

5% B in 1 min, followed by 5% B for 40 min.

RI PT

191

A solution (pH 7) containing 0.1 M bis-Tris buffer, 6 M guanidine hydrochloride, 5.37

SC

198

mM sodium citrate, and 19.5 mM dithiothreitol was added to samples (volume such as to

200

contain 10 mg protein) in a 1:1 ratio (v:v) at room temperature. Each sample was incubated

201

for 1 h at room temperature, and diluted 1:1 (v:v) with solvent A. Before injection, samples

202

were filtered through a 0.20 µm filter (MINIsart RC-15, Sartorius AG, Göttingen, Germany)

203

and 40 µL was injected. αS2-Casein was omitted from HPLC analysis due to previously

204

reported difficulties associated with its quantification (Hinz, Huppertz, & Kelly, 2012).

205 206

2.6.

Statistical analysis

EP

207

TE D

M AN U

199

Results reported are the means of data from three independent trials on separate days,

209

unless stated otherwise. Analysis of variance (one-way ANOVA) of protein profile data from

210

protein peak areas identified by RP-HPLC were conducted using SPSS Version 20.0 for Mac

211

OS X (SPSS Inc., Chicago, IL, USA). When differences were significant (P < 0.05), the

212

means were analysed using Tukey’s HSD test.

AC C

208

213 214

3.

Results

215

9

ACCEPTED MANUSCRIPT 216

3.1.

Membrane characteristics and process design

217 218

Key geometrical parameters (e.g., membrane area and hydraulic diameter) were identical for the three membranes used, with the membranes differing only in their material

220

of construction and pore-size/cut-off (Table 1). A low cross-flow velocity (0.14 m s-1) was

221

selected to allow operation at minimum TMP to limit the effects of membrane and/or deposit

222

layer compaction (Persson, Gekas, & Trägårdh, 1995), which could alter membrane

223

selectivity.

Measurement of the impact of TMP on permeate flux during recirculation of skim

225

milk at <4 °C at a cross-flow velocity of 0.14 m s-1 showed that PES1000 had the highest

226

critical flux (7.8 L m-2 h-1 at a TMP of 0.027 bar), followed by PVDF0.1 (5.2 L m-2 h-1 at a

227

TMP of 0.031 bar) and PVDF0.45 (3.5 L m-2 h-1 at a TMP of 0.032 bar). Critical flux values

228

for all membranes, particularly the PVDF membranes, were substantially lower than the

229

starting flux values that have been reported for microfiltration at 10 °C (17 L m-2 h-1;

230

Lawrence et al., 2008), 21 °C (14 L m-2 h-1; O’Mahony et al., 2014) or 50 °C (14 L m-2 h-1;

231

Beckman & Barbano, 2010) using polymeric membranes, and particularly for microfiltration

232

of skim milk at 50 °C using ceramic membranes (54 L m-2 h-1, Hurt et al., 2010; 64–72 L m-2

233

h-1, Zulewska et al., 2009).

234

EP

TE D

M AN U

224

AC C

SC

RI PT

219

Operation at sub-critical flux was selected for subsequent generation of MF

235

permeates, as this is known to limit the effects of fouling (Bacchin, Aimar, & Field, 2006).

236

Critical flux data for the membranes did not correlate well with results for water permeability

237

measurements (Table 1), with the lowest water permeability value measured for the PES1000

238

membrane. This membrane exhibited the highest flux with skim milk as the feed, but exerted

239

the greatest hydraulic resistance against a fluid that did not contain foulants (water). This

240

result suggested that reduced interactions between proteins with the PES1000 membrane

10

ACCEPTED MANUSCRIPT compared with the PVDF membranes (possibly due to a lower hydrophobicity of the former)

242

were responsible for the higher flux; indeed, adhesion of proteins to membrane surfaces is

243

known to affect polymeric membranes more than ceramic membranes (O’Mahony & Tuohy,

244

2013).

245 246

3.2.

Process performance of membranes during filtration

247

Across the individual trials with the three membranes, temperature was maintained at

SC

248

RI PT

241

3.4 ± 0.3 °C during concentration, while pH was 6.68 ± 0.05. Temperature decreased by 0.4 ±

250

0.2 °C when the process was switched from full recirculation to concentration mode, due to

251

improved efficiency of the heat-exchanger in cooling the reduced volume of feed/retentate on

252

removal of permeate.

253

M AN U

249

Permeate flux declined during concentration with all three membranes (Fig. 2a), which is likely due to a combination of rapid concentration polarisation and gradual fouling

255

(Beckman & Barbano, 2013). This decrease in flux was approximately linear, with R2 values

256

of 0.92, 0.98 and 0.81 for PES1000, PVDF0.1 and PVDF0.45, respectively. Flux decline

257

(expressed as a percentage of the starting flux) was >40% for PES1000 and PVDF0.1, but

258

<20% for PVDF0.45 membranes. The decrease in flux for the PES1000 and PVDF0.1

259

membranes was comparable with data obtained by O’Mahony et al. (2014), who reported

260

decreases of 40-60% during microfiltration/diafiltration of skim milk with 0.55 µm spiral-

261

wound PVDF membranes. In addition, strong exponential increases in TMP (R2 ≥ 0.99; Fig.

262

2b) and brix values (R2 ≥ 0.98; Fig. 2c) were recorded during processing with all three

263

membranes.

AC C

EP

TE D

254

264

The higher starting flux obtained with the PES1000 membrane resulted in a

265

processing time of 80 ± 5 min to achieve a VCF of 3, which was more rapid than both the

11

ACCEPTED MANUSCRIPT PVDF0.1 (95 ± 0 min) and PVDF0.45 (98 ± 8 min) membranes. Resistance-in-series

267

modelling (Table 2) confirmed that the formation of a reversible fouling layer (i.e., which

268

could be removed with cleaning agents but not water) was the largest single contributor to

269

overall resistance, in agreement with the results of Le Berre and Daufin (1994). Resistance

270

due to the membrane itself and irreversible fouling (i.e., that not removable by cleaning

271

agents) were comparatively minimal, and their influence decreased further as reversible

272

fouling increased during the latter stages of processing (Table 2). Measurement of degree of

273

fouling from differences in clean and fouled water permeability for two independent trials,

274

where 100% indicates total blockage, indicated that the fouling decreased in the order

275

PVDF0.1 (40.7 ± 9.4%), PES1000 (19.7 ± 0.4%) and PVDF0.45 (3.5 ± 2.7 %). Despite such

276

large differences in fouling, cleaning of the PES and PVDF membranes was effective in

277

maintaining the water permeability at ±20% of the original value of the membrane at first use

278

in all cases.

M AN U

SC

RI PT

266

281 282

3.3.

Gross composition and physicochemical properties of process streams

Membrane filtration of skim milk at refrigeration temperatures to a VCF of 3 resulted

EP

280

TE D

279

in increased levels of true protein and solids in retentate fractions and the permeation of a

284

limited quantity of true protein into the permeate (Table 3). As the PVDF0.45 membrane was

285

more porous than the PES1000 and PVDF0.1 membranes, it allowed the permeation of

286

additional protein into the permeate. More than 20% of the total nitrogen in the permeate

287

from the PES1000 and PVDF0.1 membranes comprised NPN, while the value was 13% for

288

the higher protein permeate of the PVDF0.45.

AC C

283

289

After processing, the pH of retentate and permeate fractions were slightly higher than

290

that of the feed (Table 3). Conductivity values of all retentates decreased relative to the feed,

12

ACCEPTED MANUSCRIPT while the conductivity of the permeate increased compared with that of the feed, due to

292

depletion and enrichment of minerals in the serum phase of skim milk, respectively. The

293

retentate of the PVDF0.45 membrane had a lower viscosity than the other retentates, due to

294

loss of micellar casein and a lower solids content (Table 3). The presence of micellar material

295

was evidenced by the high turbidity of the PVDF0.45 permeate, as assessed both visually and

296

using analytical centrifugation (results not shown); conversely, the permeates from the

297

PES1000 and PVDF0.1 membranes were clear.

299

3.4.

Protein profile of process streams

M AN U

300 301

SC

298

RI PT

291

SDS-PAGE analysis of the process streams showed that β-lactoglobulin and αlactalbumin, in addition to β-casein, were enriched in the permeate of the PVDF0.1 and

303

PES1000 membranes (Fig. 3). RP-HPLC was also used to quantify individual casein and

304

whey proteins in the membrane streams (Fig. 4); β-casein comprised >97% of the casein

305

protein fraction in the permeates from the PVDF0.1 and PES1000 membranes (Table 4). In

306

comparison, β-casein was reported to have accounted for ~50% of the casein fraction in the

307

permeates generated by Woychik (1992) using milk at temperatures between 2–8 °C and

308

microfiltration membrane pore-sizes of 0.1–0.2 µm.

EP

AC C

309

TE D

302

Compared with permeates from PVDF0.1 and PES1000 membranes, significantly (P

310

< 0.05) higher levels of αS1- and κ-caseins were identified in the permeates of the PVDF0.45

311

membrane (Table 4), due to permeation of small casein micelles; however, separation using

312

the PVDF0.45 membrane still enriched for β-casein relative to the other caseins (Table 4),

313

due to the high levels of serum-phase β-casein at ≤4 °C. Based on true protein (Table 3) and

314

protein profile (Table 4) data, it was calculated that the β-casein content of permeate from the

315

PVDF0.1, PVDF0.45 and PES1000 membranes was 37, 31 and 35 g 100 g-1 total protein,

13

ACCEPTED MANUSCRIPT respectively. The permeates from the PVDF0.1 and PES1000 membranes were β-casein-

317

enriched SPCs, with markedly higher whey protein:casein ratios than the feed (Table 4). The

318

permeate from the PVDF0.45 membrane also had a higher whey protein:casein ratio than the

319

feed, but the increase was less marked than in the permeates from the other membranes. The

320

generation of a more whey-dominant system is important for first-age IMF applications,

321

where the final product should have a high whey protein:casein ratio similar to that of human

322

milk (Guo, 2014).

RI PT

316

The loss of minor whey protein (~80 kDa) was observed in permeates generated from

SC

323

the PVDF0.1 and PES1000 membranes, but not for the PVDF0.45 membrane, in which the

325

minor whey protein was enriched relative to the feed (Fig. 3). Based on data from Anema

326

(2009) and the use of internal standards in SDS-PAGE (data not shown), the main high-

327

molecular mass whey proteins were identified as the heavy chain of IgG, BSA and lactoferrin

328

(LF) in order of increasing mass. The results in Fig. 3 indicated that LF may be present in

329

substantial quantities in the permeate of the PVDF0.45 membrane, the larger pore size of

330

which facilitated its permeation during membrane filtration, while it was absent or present at

331

only trace levels in permeates from the PES1000 and PVDF0.1 membranes; however,

332

differentiation between LF and lactoperoxidase (LPO) using SDS-PAGE was not possible

333

due to their approximately equivalent molecular masses (Veith & Reynolds, 2004).

334 335 336

4.

AC C

EP

TE D

M AN U

324

Discussion

337

Selection of an appropriate membrane is one of the key considerations when

338

designing separation processes for the enrichment of milk proteins, including β-casein. Both

339

PVDF and PES membranes have been used in studies on the enrichment of β-casein using

14

ACCEPTED MANUSCRIPT 340

cold membrane filtration, but their performance in terms of both processing and protein-

341

enrichment characteristics has not been compared in substantial detail previously.

342

In terms of the composition and physicochemical properties (Table 3), and protein profile (Figs. 3 and 4; Table 4) of the process streams, the PES1000 and PVDF0.1

344

membranes are essentially equivalent; however, where these membranes diverge is primarily

345

in their influence on the filtration process itself. The higher flux afforded by the PES1000

346

membrane facilitated a more rapid separation process (Figs. 1 and 2). The former factor

347

cannot be underestimated, as the low flux afforded by cold filtration processes is one of the

348

single biggest factors preventing their uptake at commercial level. Moreover, the PES1000

349

membrane had the additional benefit of being easier to clean, with most of the clean water

350

permeability being restored by a simple clean with warm water. As the selectivity of the

351

PES1000 and PVDF0.1 was essentially equivalent, the differences in process performance

352

may be due to factors such as differences in membrane hydrophobicity (O’Mahony & Tuohy,

353

2013). Both the PES1000 and PVDF0.1 membranes were suitable for the manufacture of β-

354

casein-enriched SPCs, where 97-100% of total casein consisted of β-casein. Approximately

355

20% of the total nitrogen in these permeates was comprised of NPN (Table 3), which is

356

similar to the proportions found in human milk; in addition, the permeates had increased

357

whey protein:casein ratios compared with the feed (Table 4). However, β-lactoglobulin was

358

present in all of the SPCs, although it is absent from human milk (Guo, 2014). Thus, levels of

359

β-lactoglobulin would need to be reduced in sufficient quantities in the SPCs produced using

360

the PES1000 and PVDF0.1 membranes for them to be promising ingredients in humanised

361

IMFs (Woychik, 1992). More generally, these SPCs could find use in other food application

362

as, for example, increasing levels of β-casein in whey protein-based ingredients has been

363

shown to improve functional properties such as foaming (Coppola et al., 2014).

AC C

EP

TE D

M AN U

SC

RI PT

343

15

ACCEPTED MANUSCRIPT In contrast to the other membranes studied, the PVDF0.45 membrane allowed some

365

micellar casein to pass into the permeate (Table 4), resulting in a casein fraction in which β-

366

casein comprised only 53%. Interestingly, this membrane displayed certain desirable

367

processing characteristics, such as low flux decline (Fig. 2a) and minimal fouling. The good

368

processing characteristics of this membrane may be due to the loss of small casein micelles

369

into the permeate, as evidenced by high turbidity (results not shown), higher protein and

370

solids content (Table 3) and the presence of high levels of αS- and κ-caseins (Figs. 3 and 4;

371

Table 4). Reduced numbers of small micelles would result in a reduced polydispersity of the

372

casein micelle population in the retentate, with a concomitant decrease in the propensity for

373

tight packing of micellar material at the membrane surface. Pre-treatment of feeds to lower

374

levels of small casein micelles may be a useful way of reducing fouling and improving

375

membrane performance.

SC

M AN U

376

RI PT

364

For the first time, the effect of membrane processing on the minor, high molecular mass whey proteins during the generation of β-casein-enriched SPCs was also investigated.

378

PVDF0.45 was the only membrane studied in which LF/LPO appeared to be present in

379

substantial quantities in the permeate after filtration of skim milk at refrigeration

380

temperatures, with markedly higher levels compared with permeates from the other two

381

membranes studied (Fig. 3). As previously stated, LF and LPO cannot be differentiated based

382

on mass by SDS-PAGE; however, as levels of LF are several fold higher than that of LPO in

383

bovine milk (de Wit, 1998; Korhonen, 2011; Mills, Ross, Hill, Fitzgerald, & Stanton, 2011),

384

much of the intensity of the LF/LPO band in Fig. 3 can be attributed to LF. The effect of

385

membrane type on LF levels during the enrichment of β-casein may have important

386

implications for infant formulae ingredient development. An iron-binding protein, LF is

387

considered to be highly bioactive, with purported antimicrobial, antioxidative, and

388

immunomodulatory functions among others (Korhonen, 2011; Mills et al., 2011); in addition,

AC C

EP

TE D

377

16

ACCEPTED MANUSCRIPT LF is considered an important constituent of the protein profile of human milk, where it is

390

present at much higher levels than bovine milk (Guo, 2014). These factors have led to

391

increasing interest in the use of LF as a fortificant in IMFs (Korhonen, 2011; McCarthy,

392

Kelly, O’Mahony, & Fenelon, 2014). Despite an associated decrease in the purity of β-casein

393

in the casein fraction (Table 4), the finding that LF levels may not be depleted relative to the

394

feed (Fig. 3) supports the potential application of more porous membranes such as PVDF0.45

395

in the development of β-casein-enriched ingredients for infant formulae. Future studies on

396

SPCs generated using membrane filtration would benefit from additional analysis of the

397

minor whey protein fraction using other powerful techniques for protein profiling (i.e., mass

398

spectrometry).

M AN U

SC

RI PT

389

399 400

5.

Conclusion

401

The selection of different membrane materials can have a marked influence on

TE D

402

filtration performance at refrigeration temperatures during the enrichment of β-casein from

404

skim milk. A PES membrane with a 1000 kDa cut off and a PVDF membrane with a 0.1 µm

405

pore-size were both used successfully to enrich β-casein at high casein purity, producing

406

SPCs of equivalent composition, physicochemical properties and protein profile. However,

407

although the selectivity of these membranes was essentially identical, the PES membrane

408

displayed improved flux and low fouling. The 0.45 µm PVDF membrane was not as selective

409

for β-casein, but it yielded a permeate with a unique protein profile (i.e., increased β-

410

casein:α-casein ratio, LF not depleted); in addition, it was the least susceptible to fouling of

411

all the membranes tested, which was hypothesised to be due to a reduced packing ability of

412

the fouling layer caused by loss of small casein micelles to the permeate. This study will aid

AC C

EP

403

17

ACCEPTED MANUSCRIPT 413

in the selection of parameters for refrigerated filtration processes for the development of β-

414

casein-enriched ingredients for nutritional applications such as infant formulae.

415 416

Acknowledgements

418

RI PT

417

The authors would like to acknowledge the financial support of the Food Institutional Research Measure (FIRM) initiative of the Irish Department of Agriculture, Food and the

420

Marine.

SC

419

422

References

423 424

M AN U

421

Anema, S. G. (2009). The use of “lab-on-a-chip” microfluidic SDS electrophoresis technology for the separation and quantification of milk proteins. International Dairy

426

Journal, 19, 198-204.

427

TE D

425

Bacchin, P., Aimar, P., & Field, R. W. (2006). Critical and sustainable fluxes: Theory,

429

experiments and applications. Journal of Membrane Science, 28, 42-69.

AC C

430

EP

428

431

Beckman, S. L., & Barbano, D. M. (2013). Effect of microfiltration concentration factor on

432

serum protein removal from skim milk using spiral-wound polymeric membranes.

433

Journal of Dairy Science, 93, 4506-4517.

434 435

Beckman, S. L., Zulewska, J., Newbold, M., & Barbano, D. M. (2010). Production efficiency

436

of micellar casein concentrate using polymeric spiral-wound microfiltration

437

membranes. Journal of Dairy Science, 96, 6199-6212.

18

ACCEPTED MANUSCRIPT 438 439 440

Christensen, J., & Holst, H. H. (2014). Method of producing beta-casein compositions and related products. Patent application 051315 (WO 2014/114709 A2).

441

Coppola, L. E., Molitor, M. S., Rankin, S. A., & Lucey, J. A. (2014). Comparison of milk-

RI PT

442 443

derived whey protein concentrates containing various levels of casein. International

444

Journal of Dairy Technology, 67, 467-473.

446

SC

445

Crowley, S. V., O’Callaghan, T. F., Kelly, A. L. Fenelon, M. A., & O’Mahony, J. A. (2015). Use of ultrafiltration to prepare a novel permeate for application in the functionality

448

testing of infant formula ingredients. Journal of Separation and Purification

449

Technology [in press].

450

454 455 456 457 458

TE D

453

products. Journal of Dairy Science, 81, 597-608.

Gésan-Guiziou, G. (2013). Concentrated milk and powders. In A. Y. Tamine (Ed.),

EP

452

de Wit, J. N. (1998). Nutritional and functional characteristics of whey proteins in food

Membrane processing: Dairy and beverage applications (pp. 208-225). Oxford, UK:

AC C

451

M AN U

447

Blackwell Publishing Ltd.

Glas, C., te Biesebeke, R., Kromkamp, J., & Klarenbeek, G. (2013). Functional serum

459

protein product for use in infant food and therapeutic compositions and methods for

460

the preparation thereof. Patent 13/185,353 (US 2013/8349386 B2).

461

19

ACCEPTED MANUSCRIPT 462

Guo, M. (2014). Chemical composition of human milk. In M. Guo (Ed.), Human milk

463

biochemistry and infant formula manufacturing technology (pp. 19-32). Cambridge,

464

UK: Woodhead Publishing Ltd.

465

Hinz, K., Huppertz, T., & Kelly, A. L. (2012). Susceptibility of the individual caseins in

467

reconstituted skim milk to cross-linking by transglutaminase: influence of

468

temperature, pH and mineral equilibria. Journal of Dairy Research, 79, 414-421.

RI PT

466

SC

469

Holland, B., Corredig, M., & Alexander, M. (2011). Gelation of casein micelles in β-casein

471

reduced milk prepared using membrane filtration. Food Research International, 44,

472

667-671.

M AN U

470

473 474

Hurt, E. E., Zulewska, J., Newbold, M. W., & Barbano D. M. (2010). Production of micellar casein concentrate with a 3-stage UTP ceramic membrane process at 50 °C. Journal

476

of Dairy Science, 93, 5588-5600.

479 480 481 482

ISO/IDF. (2001). Milk - Determination of nitrogen content. ISO 8968-5/IDF 20-5. Geneva,

EP

478

Switzerland: International Organisation for Standardisation/Brussels, Belgium:

AC C

477

TE D

475

International Dairy Federation.

Korhonen, H. J. (2011). Bioactive milk proteins, peptides and lipids and other functional

483

components derived from milk and bovine colostrum. In M. Saarela (Ed.), Functional

484

foods: Concept to product (2nd ed., pp. 471-551). Oxford, UK: Woodhead publishing.

485

20

ACCEPTED MANUSCRIPT 486

Kuhlman, C. F., Lien, E., Weaber, J., & O’Callaghan, D. (2005). Infant formula

487

compositions comprising increased amounts of alpha-lactalbumin. Patent 10/318,977

488

(US 2005/6913778 B2).

489

491

Laemmli, U. K. (1970). Cleavage of structural proteins during the assembly of the head of

RI PT

490

bacteriophage T4. Nature, 227, 680-685.

492

Lawrence, N. D., Kentish, S. E., O’Connor, A. J., Barber, A. R., & Stevens G. W. (2008).

SC

493

Microfiltration of skim milk using polymeric membranes for casein concentrate

495

manufacture. Separation and Purification Technology, 60, 237-244.

M AN U

494

496 497 498

Le Berre, O., & Daufin, G. (1994). Fouling and selectivity of membranes during separation of β-casein. Journal of Membrane Science, 88, 263-270.

500

TE D

499

McCarthy, N. A., Kelly, A. L., O’Mahony, J. A., & Fenelon, M. A. (2014). Sensitivity of emulsions stabilised by bovine β-casein and lactoferrin to heat and CaCl2. Food

502

Hydrocolloids, 35, 420-428.

504 505 506

AC C

503

EP

501

Mills, S., Ross, R. P., Hill, C., Fitzgerald, G. F., & Stanton, C. (2011). Milk intelligence: Mining milk for bioactive substances associated with human health. International Dairy Journal, 21, 377-401.

507 508 509

O’Mahony, J. A., Smith, K. E., & Lucey, J. A. (2014). Purification of beta casein from milk. Patent 11/272,331 (US 2014/8889208 B2).

510

21

ACCEPTED MANUSCRIPT 511

O’Mahony, J. A., & Tuohy, J. J. (2013). Further applications of membrane filtration in dairy

512

processing. In A. Y. Tamine (Ed.), Membrane processing: Dairy and beverage

513

applications (pp. 225-262). Oxford, UK: Blackwell Publishing Ltd.

514

516

Payens, T. A. J., & van Markwijk, B. W. (1963). Some features of the association of βcasein. Biochimica et Biophysica Acta, 71, 517-530.

517

RI PT

515

Persson, K. M., Gekas, V., & Trägårdh, G. (1995). Study of membrane compaction and its

519

influence on ultrafiltration water permeability. Journal of Membrane Science, 100,

520

155-162.

521 522 523

M AN U

SC

518

Rose, D. (1968). Relation between micellar and serum casein in bovine milk. Journal of Dairy Science, 51, 1897-1902.

TE D

524

Seibel, J., Molitor, M. S., & Lucey, J. A. (2014). Properties of casein concentrates containing

526

various levels of beta-casein. International Journal of Dairy Technology, 67. In press.

528 529 530 531 532

Veith, P. D., & Reynolds, E. C. (2004). Production of a high gel strength whey protein

AC C

527

EP

525

concentrate from cheese whey. Journal of Dairy Science, 87, 831-840.

Woychik, J. H. (1992). Preparation of simulated human milk protein by low temperature microfiltration. Patent 07/791,691 (1992/5169666 A).

533

22

ACCEPTED MANUSCRIPT Zulewska, J., Newbold, M., & Barbano, D. M. (2009). Efficiency of serum protein removal

535

from skim milk with ceramic and polymeric membranes at 50 °C. Journal of Dairy

536

Science, 92, 1361–1377.

AC C

EP

TE D

M AN U

SC

RI PT

534

23

ACCEPTED MANUSCRIPT 1

Figure legends

2

Fig. 1. Permeate flux as a function of trans-membrane pressure during the filtration of

4

reconstituted skim milk at 3.4 ± 0.3 °C in full recirculation mode with a 0.1 µm

5

polyvinylidene-difluoride (
), 0.45 µm polyvinylidene-difluoride (
) or 1000 kDa

6

polyethersulfone () membrane. Critical flux values are indicated by open symbols. Results

7

are from a single trial for each membrane.

9

Fig. 2. Permeate flux (a), trans-membrane pressure (b) and brix (c) as a function of time

SC

8

RI PT

3

during the filtration of reconstituted skim milk at 3.4 ± 0.3 °C with a 0.1 µm polyvinylidene-

11

difluoride (
), 0.45 µm polyvinylidene-difluoride (
) or 1000 kDa polyethersulfone ()

12

membrane. Permeate was continuously removed during filtration to achieve a volume

13

concentration factor of 3. Results are the means ± standard deviations of data from three

14

independent trials.

TE D

15

M AN U

10

Fig. 3. Sodium dodecyl sulphate-polyacrylamide gel electrophoretogram of molecular mass

17

marker (lane 1), and feeds (lanes 3, 4, 5), permeates (lanes 6, 7, 8) and retentates (lanes 9, 10,

18

11) from filtration of reconstituted skim milk at 3.4 ± 0.3 °C with 0.1 µm polyvinylidene-

19

difluoride, 0.45 µm polyvinylidene-difluoride or 1000 kDa polyethersulfone membrane

20

(samples from membranes are shown in this order from left to right for individual streams).

AC C

21

EP

16

22

Fig. 4. Chromatograms from reverse-phase high-performance liquid chromatography for (a)

23

feed and (b) permeate and (c) retentate streams from membrane separation of skim milk at

24

3.4 ± 0.3 °C using a 0.1 µm pore-size polyvinylidene-difluoride membrane, showing the

25

identification of peaks associated with proteins of interest.

1

ACCEPTED MANUSCRIPT

Table 1

a

SC

PVDF0.45 n.a. 0.45 Yes 0.16 0.001 0.04 0.625 0.1 12 13 0.021 2006 3 0.14

M AN U

PVDF0.1 n.a. 0.1 Yes 0.16 0.001 0.04 0.625 0.1 12 13 0.021 2057 3 0.14

TE D

PES1000 1000 n.a. Yes 0.16 0.001 0.04 0.625 0.1 12 13 0.021 1666 3 0.14

EP

Membrane code Molecular mass cut-off (kDa)a Pore-size (µm)a Spacer presenta Channel length (m)a Channel height (m)a Channel width (m)a Total active membrane width (m)a Membrane area (m2)a Number of feed channelsa Number of permeate channelsa Total hydraulic diameter (m)b Water permeability (L m-2 min-1 bar-1)b Volumetric flow rate (L m-2 min-1)b Cross-flow velocity (m s-1)b

RI PT

Geometric and hydrodynamic properties of membranes used for filtration of reconstituted skim milk.

b

AC C

Values taken from documentation received from, or personal communication with, Merck-Millipore technical representatives; n.a. = not applicable. Values calculated as described by Crowley et al. (2015).

ACCEPTED MANUSCRIPT

Table 2

Rr (1012 Pa)

Ri (1010 Pa)

Rt/Rm -

Rm/Rt (%)

t~0 PVDF0.1 PES1000 PVDF0.45

9.7 7.4 8.5

2.0 2.4 2.0

9.4 7.1 8.3

9.8 7.5 0.2

49.5 30.4 41.5

2.04 3.35 2.45

t = 60 PVDF0.1 PES1000 PVDF0.45

45 58 27

2.0 2.4 2.0

44 58 27

9.8 7.5 0.2

228 239 133

a

Ri/Rt (%)

0.45 0.42 0.76

Rr/Rt (%)

SC

Rm (1011 Pa)

0.99 1.03 0.03

97.0 95.6 97.5

0.21 0.13 0.01

99.3 99.4 99.2

M AN U

Rt (1012 Pa)

TE D

Membrane

RI PT

Resistance-in-series modelling of fouling of membranes at 0 and 60 min processing time. a

AC C

EP

Results are the means of data from three independent trials. Abbreviations are: Rt, total resistance; Rm, membrane resistance; Rr, resistance due to reversible fouling; Ri, resistance due to irreversible fouling.

ACCEPTED MANUSCRIPT

Table 3

Feed

PVDF0.1 PVDF0.1

Retentate Permeate

7.76 ± 0.22 0.56 ± 0.05

0.027 ± 0.001 0.026 ± 0.001

PVDF0.45 PVDF0.45

Retentate Permeate

6.36 ± 0.86 1.22 ± 0.42

0.026 ± 0.001 0.026 ± 0.001

PES1000 PES1000

Retentate Permeate

7.56 ± 0.06 0.56 ± 0.01

0.026 ± 0.001 0.025 ± 0.001

Total solids (%, w/w) 11.5 ± 0.25

M AN U

Non-protein nitrogen (%, w/w) 0.027 ± 0.000

Physicochemical properties pH Conductivity Viscosity (mS cm-1) (mPa s)

SC

All

Composition True protein (%, w/w) 3.21 ± 0.04

3.05 ± 0.02

3.20 ± 0.31

17.7 ± 0.19 7.30 ± 0.15

6.71 ± 0.04 6.71 ± 0.03

2.54 ± 0.01 3.14 ± 0.02

11.5 ± 0.75 1.95 ± 0.16

15.9 ± 1.34 8.35 ± 0.61

6.70 ± 0.00 6.65 ± 0.06

2.67 ± 0.07 3.13 ± 0.04

9.00 ± 2.19 2.57 ± 0.38

18.0 ± 0.33 7.28 ± 0.29

6.70 ± 0.00 6.70 ± 0.04

2.52 ± 0.06 3.12 ± 0.06

13.3 ± 0.81 2.08 ± 0.19

TE D

6.65 ± 0.01

Physicochemical properties were measured at ~5 °C; results are the means ± standard deviations of data from three independent trials.

AC C

a

Stream

EP

Membrane

RI PT

Composition and physicochemical properties of individual process streams after separation of reconstituted skim milk (feed) at refrigeration temperatures using different polymeric membranes. a

ACCEPTED MANUSCRIPT

Table 4

RI PT

Protein profile of individual process streams after separation of reconstituted skim milk (feed) at refrigeration temperatures using different polymeric membranes determined using reverse-phase high-performance liquid chromatography. a

Feed

PVDF0.1 PVDF0.45 PES1000

Permeate Permeate Permeate

0.00a 36.6b 0.72a

PVDF0.1 PVDF0.45 PES1000

Retentate 48.9 Retentate 48.0 Retentate 47.5

0.00a 10.2b 2.74a

38.0 39.0 37.4

13.1 13.0 15.1

% of total peak area Casein Whey protein 90.2 9.8

69.6a 70.5a 68.0a

30.4a 29.5a 32.0a

49.2a 77.7b 48.3a

50.8b 22.3a 51.7b

69.0 66.9 71.4

31.0 33.1 28.6

94.2 94.1 94.4

5.82 5.86 5.64

TE D

100b 53.2a 96.5b

% of total whey protein peak area β-lactoglobulin α-lactalbumin 67.6 32.4

SC

All

% of total casein peak area αS1-casein β-casein κ-casein 46.3 41.4 12.3

M AN U

Stream

EP

a

Membrane

Total whey protein was considered to consist of only β-lactoglobulin and α-lactalbumin. Results are the means ± standard deviations of data

AC C

from three independent trials; values for permeates within a column with different superscripts were significantly different (P < 0.05), values for retentates within a column were not significantly different.

ACCEPTED MANUSCRIPT

9 8

RI PT

6 5 4 3

SC

Permeate flux (L m-2 h-1)

7

1 0 0.00

0.03

0.06

0.09

M AN U

2

0.12 0.15 0.18 0.21 0.24 Trans-membrane pressure (bar)

AC C

EP

TE D

Fig. 1.

0.27

0.30

0.33

ACCEPTED MANUSCRIPT 8.0

a

6.0 5.0 4.0 3.0 2.0 1.0 0.0 10

20

30

40

50

60

70

90

100

110

0.25 0.20 0.15 0.10 0.05 0.00 0

10

30

40

50

60

70

80

90

100

110

17

Brix (%)

AC C

16

120

c

EP

18

20

120

b

M AN U

0.30

TE D

Trans-membrane pressure (bar)

0.35

80

SC

0

RI PT

Permeate flux (L m-2 h-1)

7.0

15 14 13 12 11 10

0

Fig. 2.

10

20

30

40

50 60 70 Process time (min)

80

90

100

110

120

ACCEPTED MANUSCRIPT Feeds

Mass marker kDa

1

2

3

4

Permeates

5

6

7

8

Retentates

9

10

11

250

RI PT

130 100

LF/LPO

BSA

70

IgG - Heavy chain

SC

55

M AN U

35 25

15

AC C

EP

Fig. 3.

TE D

10

αS2-casein αS1-casein β-casein κ-casein

β-lactoglobulin α-lactalbumin

AC C

EP

TE D

M AN U

SC

RI PT

ACCEPTED MANUSCRIPT

Fig. 4.