Production of Milk Phospholipid-Enriched Dairy Ingredients

Milk phospholipids (MPLs) have been used as ingredients for food fortification, such as bakery products, yogurt, and infant formula, because of their technical and nutritional functionalities. Starting from either buttermilk or beta serum as the original source, this review assessed four typical extraction processes and estimated that the life-cycle carbon footprints (CFs) of MPLs were 87.40, 170.59, 159.07, and 101.05 kg CO2/kg MPLs for membrane separation process, supercritical fluid extraction (SFE) by CO2 and dimethyl ether (DME), SFE by DME, and organic solvent extraction, respectively. Regardless of the MPL content of the final products, membrane separation remains the most efficient way to concentrate MPLs, yielding an 11.1–20.0% dry matter purity. Both SFE and solvent extraction processes are effective at purifying MPLs to relatively higher purity (76.8–88.0% w/w).


Commercialized Milk Phospholipid Products and Concentrate
Phosphoric 500/600/700 and Gangolac 600 (products manufactured by Fonterra) comprise 34%, 75%, 62%, and 15% MPLs, respectively, representing one source of highly-purified MPLs [52,53]. Arla Foods Amba have developed phospholipid-rich, concentrated dairy milk commodities for infant milk formulas and skin care. It has been claimed that Lacprodan ® MFGM 10 supports physiological development of the infant gut and provides infants with similar health benefits to breast milk because of their similarities in fatty acid profile [54]. Arla dairy products PL 20/75 consist of 20% and 75% MPLs, respectively [55].

Laboratory Extraction of Milk Phospholipids
Intact MFGM makes up 2-6% of the total mass of MFG [26]. However, MFGM represents 60%-70% of the total milk phospholipids [69]. Raw bovine milk comprises 0.2-0.4 g MPLs/kg, and raw milk is generally a laboratory source of MPLs [5,70]. Intact MFGs can be isolated with low-speed centrifugation. The cream layer from raw milk skimming can be washed with phosphate buffered saline (PBS; pH 6.8, 0.1 M, 1:10, v/v) and centrifuged at 390 g for 10 min at 10 • C. The final cream layer after three washes is the large MFG fraction [71]. Different from isolating intact MFGM, Sanches-Juanes et al. [72] ruptured MFGM and recrystallized milk lipids, and starting from raw milk, they washed cream with a 0.15 M NaCl solution and precipitated casein using centrifugation at 5000× g.
Cream washing is a step used to remove casein and other non-MFGM materials from cream [44]. After centrifugation, casein will precipitate, with lipid stratification at the top layer [73]. Also, calcium, naturally present in casein micelles, can form a complex between MFGM and the casein micelles through its binding to the phospho-casein and phospholipids of MFGM, leaving impurities with MPLs [74]. In addition, washing causes a severe loss of phospholipids, almost 60% per wash [32]. Hence, washing facilities for separating MPLs are costly and energy-intensive [44], thereby they are mainly only used for laboratory purposes [5,73,75].
In addition to washing and centrifugation, the microfiltration of raw milk has been applied to produce MFGM material. It has been found that a 1.4-µm ceramic membrane was superior to 0.8 µm, yielding a high-purity MFGM material, which was composed of 7% phospholipids and 30% protein [76].
For analysis purposes, MPL samples are usually prepared using solvent extraction. The Folch [77] and Bligh [78] methods use chloroform and methanol to dissolve lipids. Other lipophilic extraction formulas include the Mojonnier solvents [79], dichloromethane [80], and the ammoniacal ethanolic solution of lipids with dimethyl ether and light petroleum in the Röse-Gottlieb extraction [81,82]. The total lipid content in samples can be determined with a gravimetric assay, Gerber-van Gulik butyrometer, infrared spectroscopy according to an International Dairy Federation (IDF) method [81], or gas chromatography equipped with a flame ionization detector [83].
To determine the MPLs and their subclasses, solid-phase extraction can fractionate polar lipids from non-polar lipids. Silica-gel-bonded cartridges or silica gel plates can be used for such a purpose [84]. The obtained MPLs can be solvent dried using a vacuum and stored at −20 • C before using [85]. In addition, chloroform and methanol are also valid elution solvents [86]. Total MPLs can be measured using the IDF molybdate assay [87], Fourier transform infrared spectroscopy [88], or a fluorescence assay on cleaved choline group [89]. Both nuclear magnetic resonance of 31 P and chromatography can quantify MPLs and their subclasses [90,91]. High-performance liquid chromatography coupling with detectors as a charged aerosol detector, evaporative light-scattering detector, and mass spectroscopy is more acceptable than thin layer chromatography [92].

Solvent Extraction
Many kinds of polar solvents have been used to extract MPLs, such as ethanol and alkanes [21,66]. To separate casein from MPLs, proteins can also be thermally denatured or in an acid solution (pH 4.6) [48,81], the aggregated particles are subsequently filtrated. Regarding fractionation of MPL from WPPC, ethanol (70% v/v) at 60-80 • C denatures proteins; after filtration the MPL concentration is ≈45.8% in the filtrate in Figure 2a [48]. This notable method uses no toxic solvent. However, the incompleteness of the phospholipid recovery may be a concern [48]. solution of lipids with dimethyl ether and light petroleum in the Röse-Gottlieb extraction [81,82]. The total lipid content in samples can be determined with a gravimetric assay, Gerber-van Gulik butyrometer, infrared spectroscopy according to an International Dairy Federation (IDF) method [81], or gas chromatography equipped with a flame ionization detector [83].
To determine the MPLs and their subclasses, solid-phase extraction can fractionate polar lipids from non-polar lipids. Silica-gel-bonded cartridges or silica gel plates can be used for such a purpose [84]. The obtained MPLs can be solvent dried using a vacuum and stored at −20 °C before using [85]. In addition, chloroform and methanol are also valid elution solvents [86]. Total MPLs can be measured using the IDF molybdate assay [87], Fourier transform infrared spectroscopy [88], or a fluorescence assay on cleaved choline group [89]. Both nuclear magnetic resonance of 31 P and chromatography can quantify MPLs and their subclasses [90,91]. High-performance liquid chromatography coupling with detectors as a charged aerosol detector, evaporative light-scattering detector, and mass spectroscopy is more acceptable than thin layer chromatography [92].

Solvent Extraction
Many kinds of polar solvents have been used to extract MPLs, such as ethanol and alkanes [21,66]. To separate casein from MPLs, proteins can also be thermally denatured or in an acid solution (pH 4.6) [48,81], the aggregated particles are subsequently filtrated. Regarding fractionation of MPL from WPPC, ethanol (70% v/v) at 60-80 °C denatures proteins; after filtration the MPL concentration is ≈45.8% in the filtrate in Figure 2a [48]. This notable method uses no toxic solvent. However, the incompleteness of the phospholipid recovery may be a concern [48].  al. [93], and (c) Shulman et al. [21]. BMC, buttermilk concentrate; MPL, milk phospholipid; AI, acetone insoluble; WPPC, whey protein phospholipid concentrate (liquid, reconstituted from powder).
Compared to 58.1% recovery by ethanol, the tertiary amine CyNMe2 (N,Ndimethylcycloexylamine) yielded a 99.96% recovery rate of MPLs. At various solvent-sample weight ratios, the lipid extraction was conducted at ambient temperature. The dissolved MPLs in the amine were released by bubbling CO2 at atmospheric pressure, which converts CyNMe2 into the carbonate salt in Figure 2b. By injecting nitrogen and removing CO2, the carbonate salt regenerated into the amine  [93], and (c) Shulman et al. [21]. BMC, buttermilk concentrate; MPL, milk phospholipid; AI, acetone insoluble; WPPC, whey protein phospholipid concentrate (liquid, reconstituted from powder).
Compared to 58.1% recovery by ethanol, the tertiary amine CyNMe2 (N,N-dimethylcycloexylamine) yielded a 99.96% recovery rate of MPLs. At various solvent-sample weight ratios, the lipid extraction was conducted at ambient temperature. The dissolved MPLs in the amine were released by bubbling CO 2 at atmospheric pressure, which converts CyNMe2 into the carbonate salt in Figure 2b. By injecting nitrogen and removing CO 2 , the carbonate salt regenerated into the amine form for reuse ( Figure 2b). Though the recovery rate for BM was as high as 99.96 ± 1.2%, the extraction rates for BS and concentrated BM were only 7.57 ± 0.59% and 77.27 ± 4.51%, respectively. Aside from the input sensitivity, the amine may interact with dairy components and cause toxic consequences [93], and the chemical facilities required may be incompatible in a dairy factory setting.
MPLs can be dissolved in ethanol and alkanes [21,67,68], and may not dissolve in acetone, ethyl acetate, and 2-pentanone [21,67,68]. Lipid BMP (100 g) dissolved in ethanolic hexane (1:4 v/v, 800 mL) under constant agitation at 45 • C for 2 h will produce an extract. The permeate of vacuum filtration (repeated twice) can then be vacuum-dried at 1 kPa (Figure 2c). The residue (≈20 g) is then defatted twice with 120 mL acetone, and the resulting acetone is insoluble (AI, ≈7 g), composed of mainly polar lipids, and in the final step vacuum, is dried again at 1 kPa [21]. However, acetone poses a degree of toxicity, as acetone residue in defatted MPLs may reach 5-10 ppm. Further, acetone can form a mesityl oxide via a condensation reaction, causing an off flavor [94]. Hence, toxic residues in acetone-insoluble fractions need analysis and monitoring.

Supercritical Fluid Extraction
Supercritical CO 2 with ethanol as a co-solvent can be used to extract MPLs effectively, yielding purities of 26.26% and 16.88% from WPPC and BMP extractions, respectively ( Figure 3a). The SFE operation can be conducted at 50-60 • C [95] and 350-550 bar [49]. The SFE co-solvent (CO 2 and 20% ethanol) allowed for complete extraction of PE, PC, and SM. However, neither PS (i.e., the vital compounds for cognitive function) nor PI were extracted [61,96]. Therefore, the co-solvent method may be an invalid industrial process due to the incompleteness of PS/PI recovery. In addition to co-solvents, dimethyl ether near the critical point (DME, 20%-30% solubility, 333 K, 40 bar) and supercritical CO 2 are able to dissolve polar and neutral lipids, respectively [59].
form for reuse (Figure 2b). Though the recovery rate for BM was as high as 99.96 ± 1.2%, the extraction rates for BS and concentrated BM were only 7.57 ± 0.59% and 77.27 ± 4.51%, respectively. Aside from the input sensitivity, the amine may interact with dairy components and cause toxic consequences [93], and the chemical facilities required may be incompatible in a dairy factory setting.
MPLs can be dissolved in ethanol and alkanes [21,67,68], and may not dissolve in acetone, ethyl acetate, and 2-pentanone [21,67,68]. Lipid BMP (100 g) dissolved in ethanolic hexane (1:4 v/v, 800 mL) under constant agitation at 45 °C for 2 h will produce an extract. The permeate of vacuum filtration (repeated twice) can then be vacuum-dried at 1 kPa (Figure 2c). The residue (≈20 g) is then defatted twice with 120 mL acetone, and the resulting acetone is insoluble (AI, ≈7 g), composed of mainly polar lipids, and in the final step vacuum, is dried again at 1 kPa [21]. However, acetone poses a degree of toxicity, as acetone residue in defatted MPLs may reach 5-10 ppm. Further, acetone can form a mesityl oxide via a condensation reaction, causing an off flavor [94]. Hence, toxic residues in acetoneinsoluble fractions need analysis and monitoring.

Supercritical Fluid Extraction
Supercritical CO2 with ethanol as a co-solvent can be used to extract MPLs effectively, yielding purities of 26.26% and 16.88% from WPPC and BMP extractions, respectively (Figure 3a). The SFE operation can be conducted at 50-60 °C [95] and 350-550 bar [49]. The SFE co-solvent (CO2 and 20% ethanol) allowed for complete extraction of PE, PC, and SM. However, neither PS (i.e., the vital compounds for cognitive function) nor PI were extracted [61,96]. Therefore, the co-solvent method may be an invalid industrial process due to the incompleteness of PS/PI recovery. In addition to cosolvents, dimethyl ether near the critical point (DME, 20%-30% solubility, 333 K, 40 bar) and supercritical CO2 are able to dissolve polar and neutral lipids, respectively [59].
Supercritical fluid DME has been used to extract polar lipids, resulting in a yield of 69.1-77.8%. The SFE process shown in Figure 3b can accept both liquid and powder inputs [59,97]. This unit can work with CO2 and DME in two-stages, extracting neutral and polar lipids, separately. In addition to a two-step operation, this unit can also operate a single extraction with DME. Near-critical DME dissolves both polar and non-polar lipids in the SFE chamber. Through a two-stage de-pressurization, lipids are separated from the protein fraction, whereas vaporized DME is compressed and condensed for reuse (Figure 3b). This method features properties such as non-toxicity, a compact skid, feeding flexibility, and a high content of MPLs (65.7-75.5 g MPLs per 100 g DM). However, the MPL recovery rate (69.1-77.8%) needs further improvement.   Supercritical fluid DME has been used to extract polar lipids, resulting in a yield of 69.1-77.8%. The SFE process shown in Figure 3b can accept both liquid and powder inputs [59,97]. This unit can work with CO 2 and DME in two-stages, extracting neutral and polar lipids, separately. In addition to a two-step operation, this unit can also operate a single extraction with DME. Near-critical DME dissolves both polar and non-polar lipids in the SFE chamber. Through a two-stage de-pressurization, lipids are separated from the protein fraction, whereas vaporized DME is compressed and condensed for reuse (Figure 3b). This method features properties such as non-toxicity, a compact skid, feeding flexibility, and a high content of MPLs (65.7-75.5 g MPLs per 100 g DM). However, the MPL recovery rate (69.1-77.8%) needs further improvement.

Enrichment of Milk Phospholipids via Filtration
BM or BS is composed of milk fat, casein and whey protein, lactose, and ash. The particle sizes range from 0.4-4 µm for MFGM fragments or phospholipid micelles [98], 0.02-0.3 µm for casein, 0.03-0.06 µm for whey protein, and 0.002 µm for lactose and ash, respectively [99]. The size of MFG is around 0.2-15 µm [3]. As illustrated in Figure 4, the MF unit removes lactose and whey protein, and UF separates the smaller casein proteins from MPLs. Due to the size overlap of casein micelles and MPL particles, their separation is usually incomplete. Casein micelles disintegrate into peptides and amino acids in the proteolysis unit [34,42], and hydrolysates enter into the permeate stream during the subsequent UF operation [42,96]. Alcalase (E.E. 3.4.21.62), a serine-type endoprotease with esterase activity, catalyzes amino esters at pH 7.5 and 35-60 • C [96], while tryptic and peptic hydrolysis may be carried out at 42 • C for 2-16 h, at a pH of 7.7 and 2.0, respectively [42].

Enrichment of Milk Phospholipids via Filtration
BM or BS is composed of milk fat, casein and whey protein, lactose, and ash. The particle sizes range from 0.4-4 µ m for MFGM fragments or phospholipid micelles [98], 0.02-0.3 µ m for casein, 0.03-0.06 µ m for whey protein, and 0.002 µ m for lactose and ash, respectively [99]. The size of MFG is around 0.2-15 µ m [3]. As illustrated in Figure 4, the MF unit removes lactose and whey protein, and UF separates the smaller casein proteins from MPLs. Due to the size overlap of casein micelles and MPL particles, their separation is usually incomplete. Casein micelles disintegrate into peptides and amino acids in the proteolysis unit [34,42], and hydrolysates enter into the permeate stream during the subsequent UF operation [42,96]. Alcalase (E.E. 3.4.21.62), a serine-type endoprotease with esterase activity, catalyzes amino esters at pH 7.5 and 35-60 °C [96], while tryptic and peptic hydrolysis may be carried out at 42 °C for 2-16 h, at a pH of 7.7 and 2.0, respectively [42].
Membrane filtration is a typical process for enriching BM (Figure 4a). Proteolytic treatment plus UF, as illustrated in Figure 4b, successfully differentiates MFGM from protein particles and yields product purities of 14 ± 3.4% (DM) [42] and 11.05 ± 0.02% (DM) [96]. The combined process of proteolysis and membrane separation can yield a 100% recovery rate of MPLs from BM, as illustrated in Table 3. Considering membrane units exist in most dairy factories [99,100], the process remains the most effective method for concentrating MPLs, requiring less investment than the other processes [101]. As illustrated in Table 3

Available Processes for Extracting Milk Phospholipids
In brief, there are three options for the large-scale manufacturing of MPLs, including solvent extraction [21,68], SFE [59,97], and proteolysis plus membrane concentration [34,42,82,96]. The membrane concentration of MPLs have yielded a 20% (w/w, DM basis) purity, as achieved by Lecico [58] and Arla [10]. Tatua [56] and Westland and Synlait [44] have extracted MPLs from BS powder (2.28%, w/w, DM basis), achieving approximately 12.8% (w/w, DM basis) purity using membrane filtration. The proteolysis and UF unit recovers MPLs completely [34,82,96] and cost-effectively [44]. This process is more efficient than SFE and solvent extraction, whereas SFE and solvent extractions are effective steps for manufacturing high purity MPLs. Therefore, the three processes have features of a high recovery rate, facility availability, and food compatibility, representing current industrial practices (in Table 3). Membrane filtration is a typical process for enriching BM (Figure 4a). Proteolytic treatment plus UF, as illustrated in Figure 4b, successfully differentiates MFGM from protein particles and yields product purities of 14 ± 3.4% (DM) [42] and 11.05 ± 0.02% (DM) [96]. The combined process of proteolysis and membrane separation can yield a 100% recovery rate of MPLs from BM, as illustrated in Table 3. Considering membrane units exist in most dairy factories [99,100], the process remains the most effective method for concentrating MPLs, requiring less investment than the other processes [101]. As illustrated in Table 3, this method [96] recovered more MPLs than the other processes.

Available Processes for Extracting Milk Phospholipids
In brief, there are three options for the large-scale manufacturing of MPLs, including solvent extraction [21,68], SFE [59,97], and proteolysis plus membrane concentration [34,42,82,96]. The membrane concentration of MPLs have yielded a 20% (w/w, DM basis) purity, as achieved by Lecico [58] and Arla [10]. Tatua [56] and Westland and Synlait [44] have extracted MPLs from BS powder (2.28%, w/w, DM basis), achieving approximately 12.8% (w/w, DM basis) purity using membrane filtration. The proteolysis and UF unit recovers MPLs completely [34,82,96] and cost-effectively [44]. This process is more efficient than SFE and solvent extraction, whereas SFE and solvent extractions are effective steps for manufacturing high purity MPLs. Therefore, the three processes have features of a high recovery rate, facility availability, and food compatibility, representing current industrial practices (in Table 3).

Life-Cycle Accessment Method of Carbon Footprint
The ISO 14,040 life-cycle assessment (LCA) is an internationally accepted methodology used to calculate a product's environmental footprint [103]. Life-cycle carbon footprints (CFs) of dairy products cover the direct emission from the dairy factory (scope 1); the energy carrier footprint for factory operations (natural gas, steam, power, nitrogen, and compressed air in scope 2); and the raw material, packaging, and logistics in scope 3. In addition, the life-cycle CF comprises the emissions from the dairy farm (upstream) and distribution center (downstream) [104]. The boundaries are set as shown in Figure 5a. The CFs of MPL products were reported as equivalent CO 2 emission for one kg of MPLs, according to the ISO 14,067 reporting standards [105]. equation: CFBMC = CFBM + CFMS, where CFBMC, CFBM, and CFMS were the CF of BMC, BM, and MS, respectively. The CF of MPL products using SFE or solvent extraction was calculated using the equation CFMPLs = CFBMC + CFSFE or CFMPLs = CFBMC + CFSol where CFMPLs, CFBMC, CFSFE, and CFSol were the CF for the MPL product, BMC, SFE, and solvent extraction process, respectively, as illustrated in Figure 5b. The CF for BMP (1.5% purity) was 1.10 kg CO2/kg MPL, and the CF of BMC (11.0% purity) was 87.40 kg CO2/kg MPL, as calculated in Table 5. Starting from BMC, the CFs of MPL products were 170.59, 159.07, and 101.05 kg CO2/kg MPL for processes of CO2-DME supercritical fluid extraction, DME SFE, and organic solvent extraction, respectively.  Figure 5. (a) Boundary definition of the life-cycle carbon footprints (CFs) of dairy products and exemplary emissions from scope 1 (direction emission), scope 2 (energy carriers), and scope 3 (raw material procured, packaging material, and transportation). (b) Cascade of CFs of BMP, BMC, and MPLs using the following processes: "CO2-DEM" (supercritical CO2 and DME), "DME" (supercritical DME), and "Solvent" (hexane and ethanol extraction and acetone de-fatting, kg equivalent CO2/products). Scopes of BM CFs: adapted from References [45,[106][107][108][109]. MPLs, milk phospholipids; BM, buttermilk; BMC, BM concentrate; DME, dimethyl ether; SFE, supercritical fluid extraction. * Ultra High Temperature processing.

Carbon Footprint Estimation
In Table 4, four MPL enrichment processes were used as references for estimating and comparing the total CFs. The membrane separation process was used to concentrate MPL from the original BM. The resulting product was a BM concentrate (BMC), which may be further processed to yield MPL products by either using an SFE technique or a solvent extraction method. The CF of "utility" consumed for the three individual MPL enrichment methods was obtained by multiplying the utility amount and CF conversion factor, which represents the amount of carbon emission for a unit weight of utility. Normalized CF: CFNormalized = CF/CMPLs, where CMPLs was the MPL purity (g MPLs per 100 g product).
The normalized CF of the product uisng membrane separation was as high as 87.4 kg CO2/kg BMC since the BMC comprised of only 11.05% MPLs. The CFs for products using SFE and solvent extraction were much higher than their baseline (CFBMC) because of the intensive process during purification. As shown in Table 4, the CFs of fractions using SFE were 170.59 and 159.07 kg CO2/kg MPLs for CO2/DME co-extraction and DME extraction, respectively. CO2/DME co-SFE exhibited a BMC, and MPLs using the following processes: "CO 2 -DEM" (supercritical CO 2 and DME), "DME" (supercritical DME), and "Solvent" (hexane and ethanol extraction and acetone de-fatting, kg equivalent CO 2 /products). Scopes of BM CFs: adapted from References [45,[106][107][108][109]. MPLs, milk phospholipids; BM, buttermilk; BMC, BM concentrate; DME, dimethyl ether; SFE, supercritical fluid extraction. * Ultra High Temperature processing.
The CF of BM (baseline CF, 1.10 kg CO 2 /kg BM powder) was cited directly from data derived from the Unified Livestock Industry and Crop Emissions Estimation System (ULICEES) model in Canada [45]. The data abides by the Intergovernmental Panel on Climate Change (IPCC) methodology [106]; it covers emissions like methane [45], nitrous oxide [107], and carbon dioxide using the F4E2 model [108]; and uses an allocation matrix to partition six inventory flows (i.e., fuel, power, raw milk transportation, alkaline/acid, water, and waste water) into 11 major dairy products [109].
In this study, BM was assumed as the starting material for producing MPLs. Therefore, the CF for producing BM was set as the baseline. The CF of MPLs in Figure 5b and Table 4 is a sum of the CF of BM (as the baseline) and the CF for extracting MPLs from BM at dairy factories. The starting amount of BM was assumed to be 100 kg (1.3%, w/w, DM basis). Since MPLs are considered as the target products, CF of protein in the MPL fractions was not included in the estimations.
The CF of BM concentrate (BMC) via membrane separation (MS) was calculated using the equation: CF BMC = CF BM + CF MS , where CF BMC , CF BM , and CF MS were the CF of BMC, BM, and MS, respectively. The CF of MPL products using SFE or solvent extraction was calculated using the equation CF MPLs = CF BMC + CF SFE or CF MPLs = CF BMC + CF Sol where CF MPLs , CF BMC , CF SFE , and CF Sol were the CF for the MPL product, BMC, SFE, and solvent extraction process, respectively, as illustrated in Figure 5b. The CF for BMP (1.5% purity) was 1.10 kg CO 2 /kg MPL, and the CF of BMC (11.0% purity) was 87.40 kg CO 2 /kg MPL, as calculated in Table 5. Starting from BMC, the CFs of MPL products were 170.59, 159.07, and 101.05 kg CO 2 /kg MPL for processes of CO 2 -DME supercritical fluid extraction, DME SFE, and organic solvent extraction, respectively.

Carbon Footprint Estimation
In Table 4, four MPL enrichment processes were used as references for estimating and comparing the total CFs. The membrane separation process was used to concentrate MPL from the original BM. The resulting product was a BM concentrate (BMC), which may be further processed to yield MPL products by either using an SFE technique or a solvent extraction method. The CF of "utility" consumed for the three individual MPL enrichment methods was obtained by multiplying the utility amount and CF conversion factor, which represents the amount of carbon emission for a unit weight of utility. Normalized CF: CF Normalized = CF/C MPLs , where CMPLs was the MPL purity (g MPLs per 100 g product).
The normalized CF of the product uisng membrane separation was as high as 87.4 kg CO 2 /kg BMC since the BMC comprised of only 11.05% MPLs. The CFs for products using SFE and solvent extraction were much higher than their baseline (CF BMC ) because of the intensive process during purification. As shown in Table 4, the CFs of fractions using SFE were 170.59 and 159.07 kg CO 2 /kg MPLs for CO 2 /DME co-extraction and DME extraction, respectively. CO 2 /DME co-SFE exhibited a higher environmental impact compared to supercritical DME extraction due to direct emissions from co-SFE. Solvent extraction demonstrated a lower environmental impact and a higher MPL recovery rate than SFE. However, the products obtained using solvent extraction were less food-compatible than SFE unit-extracted products.
MPLs from proteolysis and filtration processes carry 87.40 kg equivalent CO 2 /kg product, much higher than all the milk fat products (Table 5). With less CF than SFE and solvent extraction, membrane separation is the most efficient process in terms of process intensity, energy consumption, and environmental impact. In addition, this process is compatible with most dairy factories. Membrane separation is a necessary step for concentrating BM into BMC. BMC can then be purified using SFE (DME). The relevant processes with a significant MPL CF include membrane filtration, evaporation and spray drying, SFE, and solvent recovery, the improvement of which offer opportunities to reduce the CF of the final products. For example, 0.1-µm polymeric spiral-wound MF membranes have been used to separate casein from milk, exhibiting a higher energy efficiency at 0.024 (MF) and 0.015 (DF) kWh/kg permeate than that of graded permeability membrane (0.143 and 0.077 kWh/kg permeate for MF and DF, respectively [110]. Furthermore, permeate flux, volume concentration ratio, transmembrane pressure, and temperature all had an impact on the energy efficiency of membrane UF, ranging from 0.26-0.33 kWh/kg retentate [116]. Another approach toward reducing the environmental impact is to improve the purity of MPLs during filtration by differentiating the particle size of casein micelles (i.e., hydrolysis) from the fragmented MFGM and subsequent application of membrane filtration.

Conclusions
This paper identified three dairy streams for milk phospholipid (MPL) manufacturing at an industrial scale: buttermilk, beta serum, and whey protein phospholipid concentrate. The life-cycle CFs of the MPLs were 87.40, 170.59, 159.07, and 101.05 kg CO 2 /kg MPLs for the membrane separation process, CO 2 /DME supercritical fluid extraction, SFE by DME, and organic solvent extraction, respectively. The extracted products comprised 11.1, 76.8, 69.9, and 88.0% MPLs, with recovery rate of 100, 69.1, 67.4, and 100%, respectively. In conclusion, to improve the efficiency of an MPL concentration process, casein in BM needs to be proteolyzed before running UF/DF processes. By doing so, it is possible to achieve full recovery of MPLs from BM; moreover, this method may result in a relatively low CF. SFE using dimethyl ether is the most effective method for the production of high-purity (≈66.8%) MPL products, albeit at the cost of a high CF. This study provided insights into the best available industrial practices for extracting MPLs and estimating their life-cycle CFs. Funding: This research received no external funding.

Conflicts of Interest:
The authors declare no conflicts of interest.