The whole tomato fruits were fractionated into juice, pomace (pulp and peel), and seeds (Fig. 1). It was impossible to separate the peel and pulp because at this stage the peel appeared as fragmented pieces. The juice formed the largest proportion of material in tomatoes in terms of dry biomass (60%) and seeds (12%) were least. However, a large proportion of these tomatoes was comprised of water and only 49 g of tomato seeds were recovered from 4.5 kg fresh tomatoes. The processing of a larger quantity of round tomatoes (100 kg) resulted in a lower proportion of seeds being recovered (250 g) which would still be insufficient to perform pilot scale studies [Baker et al., unpublished]. Therefore, a different strategy using sedimentation of seeds was implemented to recover a larger quantity of material from a tomato juice processing plant . It was impossible to separate the pulp and seeds from processed pomace using sieving because the gelatinous layer surrounding the seeds was removed during heat processing that would enable the seeds to pass through the sieve. The remaining pomace after processing resulted in the recovery of 44% of the dry biomass as seeds. However, the development of a strategy to recover crude protein from tomatoes that would be otherwise discarded is important process that should be independently investigated.
Compositional analysis of the pomace and seeds indicated that non-fibre was the major component in both fractions (Table 2), and a relatively high lignin content in the pomace (25.3 ± 1.5%). Whether the remaining material after fibre analysis was lignin is unclear and one possibility is that cutin, a macromolecule with properties similar to lignin, and that can form a structural component in mosses  may show acid resistant properties that are similar to lignin. In contrast, the lignin content in the seeds was considerably lower (9.51 ± 0.43%), with cellulose and lignin contents of the seeds having similar values to those reported previously . There were significant quantities of cellulose in the pomace and of both hemicellulose and cellulose in the seeds, indicating that the application of an enzymatic processing step using carbohydrases could have an effect in releasing proteins from both fractions.
The protein content was determined for each of the tomato fractions, with the milled tomato seeds containing the highest amounts (Table 2). The protein content in un-milled seeds was 0.17 ± 0.02% and so it was necessary to mill the seeds to reduce particle size and facilitate optimised, downstream protein recovery . Kjeldahl analysis revealed that the protein concentrations in milled seeds (Table 2) were similar to those reported previously [11, 12], but there are no reported values for the protein content of tomato pulp. It is possible that some of the protein remained inaccessible even under the alkaline treatment, resulting in lower values determined using the Bradford assay. Similar findings were revealed in another study, where protein determination of tomato seeds under weak alkaline conditions, using the Bradford assay, was lower than values determined using Kjeldahl analysis . The Bradford assay measures soluble protein although the majority of soluble proteins will be determined at specific pH values . In contrast, Kjeldahl analysis will also include insoluble proteins such as globular proteins although very low concentrations of other nitrogenous compounds besides amino acids may also be included.
Three quarters of phosphorus in plants is stored as phytates, although it cannot be absorbed by ruminants and the presence of phytates in food can lead to the deficiency in other minerals such as iron and zinc . Our results showed that 76% of the total phosphorus was in the form of phytic acid and the amount was within the expected range for most cereal grains (Megazyme, Ireland). A previous study has shown that the carryover of phytates into the protein extract is much less compared with the quantity found in the original seeds .
Enzyme Mediated Protein Extraction from Seeds
A range of different enzyme mixtures were assessed in order to determine any improvements in protein extraction, each with minor differences in optimum temperature and pH conditions (Table 1). Enzyme mediated protein extraction from seeds indicated that the only carbohydrase which resulted in an increase in protein yield, following treatment, was Filta 02L. This was 10% higher and significantly different compared with the control (Fig. 2A) without enzyme (P = 0.008). Filta 02L possessed β-glucanase, xylanase and cellulase activities that would enable these carbohydrases to hydrolyse both the hemicellulose and cellulose present in the matrix. In contrast, Cellux 01L possessed β-glucanase and cellulase activities and did not show increased protein recovery. The other carbohydrases evaluated as part of this study, Tail 157 and Tail 01, did not possess cellulase activity and subsequently yielded lower quantities of protein compared with Filta 02L. Tail 113 which had similar carbohydrases to Filta 02L did not show increased levels compared with the control, perhaps due to lower levels of carbohydrase activity or possessed activity that targeted different carbohydrates. Therefore, carbohydrases that could completely degrade both hemicellulose and cellulose, such as xylanase, cellulase and β-glucanase appear to be necessary in order to increase the recovery of protein from tomato seeds. Some of the carbohydrases, Tail 157, Tail 113 and Cellux 01L, appeared to show a decrease in protein yield in comparison to the control without carbohydrase. However, only Tail 157 was significantly lower (P = 0.008) and it unclear why this resulted in a lower yield. It was apparent that this enzyme contained higher quantities of pectinases than the other enzymes and the release of galacturonate may form a gel-like substance in the presence of metallic ions , that may result in protein binding.
Enzyme Mediated Protein Extraction from Tomato Pulp
Protein extraction studies from the pulp indicated that Tail 157 increased the yield by 210%, which was significantly higher (P = 0.005) (Fig. 2B). The other enzymes also appeared to increase the protein yield between 30 and 40%, although only Tail 01 and Filta 02L were significantly higher (P < 0.05). Tail 157 appeared to act on the tomato peels causing them to completely disintegrate when whole tomatoes were incubated in the presence of this enzyme (unpublished). Considering the low levels of proteins that were extracted, it is important to include controls that account for the quantity of protein associated with each carbohydrase and these concentrations were subtracted from protein yields obtained with each carbohydrase. Tail 157 contains mostly pectinases, and the presence of a significant pectin content in tomatoes ranging from 5 to 10% of the total dry biomass  may be a factor in the effectiveness of this carbohydrase. Furthermore, the cellulase activity of Tail 157 at 94 ± 18 units per ml (Table 2) was considerably higher than the other carbohydrases where cellulose was a major fibre component (Table 3). Therefore, higher cellulase activity may contribute to higher protein recovery although further experiments would be required. There could be other factors such as physical treatments that could improve protein yields, including the use of microwave extraction from freeze dried tomato pulp to recover 7.8% of protein . The remaining pulp could be valorised to recover carotenoids using solvent extraction .
Further Separation of the Milled Seeds
Another strategy was explored to separate the milled tomato seeds into two separate fractions, one containing predominately the seed hulls and the other fraction predominately containing the seed kernels. Tomato seeds are composed of ~ 50% non-fibrous material (Table 1), and it would appear, based on results from previous studies, that a significant amount of this soluble material could be tomato seed oil which has been found to comprise 31% of the total tomato seed weight . Most studies have described the formation of tomato meal where the oils are removed through a defatting process by hexane extraction [11, 23, 28,29,30,31] or by using pressing . However, the use of large quantities of organic solvent in pilot scale studies presents significant flammability issues, along with increasing CAPEX and OPEX and it is desirable to develop a green process that limits organic solvent use. Therefore, a process was investigated that would postpone the use of organic solvents until a later stage where lower volumes would be required due to the decrease in biomass.
The total protein yields after the first extraction and from the pooled sequential extraction mixtures was ~ 14% and 36% g based on total protein determined using Kjeldahl analysis (Fig. 3). Almost all of the alkali extractable protein was recovered as determined by the Bradford assay. This indicated that a significant amount of protein could be recovered from subsequent washes, although this does require the use of large volumes of water and ethanol. Rather than using large quantities of ethanol, protein could be precipitated using citric acid and up to 500 mg/l chitin as a coagulating agent  or different pH acidification ranges . The volume of water to weight of biomass ratio at 18:1 during the initial wash was similar to previously reported values, conducted under dilute alkaline conditions . However, a considerable amount of residual protein remained in the seed hull fraction following processing (Table 4). The quantity of protein in the crude protein extracts as determined by the different methods, Bradford and Kjeldahl analysis, were similar at ~ 30%, indicating that the majority of the protein was soluble and that a considerable proportion of other components were co-extracted with the protein. A hexane extraction revealed that 25% of the total crude protein was composed of oils and fats, indicating that the protein content would increase to ~ 40%. Further increases in protein yield could be obtained by ultrafiltration to collect proteins that remained in suspension even after ethanolic precipitation.
It was determined that the sequential aqueous extraction of proteins revealed that use of higher centrifugation speeds resulted in a lower proportion of material remaining suspended and more protein was associated with the centrifuged pellet which was assumed to contain seed hulls fragments based on appearance (Table 4). Higher levels of protein were recovered when lower speeds were used but the precipitates appeared slighter browner compared with precipitates obtained with higher centrifugation speeds due to the carryover of smaller seed hull fragments into this fraction. Nevertheless, there were minor differences between the crude fractions in terms of protein content determined using Kjeldahl analysis. The protein contents of the material left in the suspensions at different low speed centrifugation speeds were generally similar. Protein contents determined using Kjeldahl analysis were generally similar to those determined by the Bradford assay in protein extracts in contrast to differences between Kjeldahl and Bradford determined in the milled seeds. This might indicate that the proteins in this separated fraction were more accessible to the reagents in the Bradford assay whereas the majority of those in the seeds were inaccessible. Fibre analysis of the residual hulls indicated that the lignin content had increased more than three-fold compared with the original seeds, whereas the levels of hemicellulose and cellulose remained unchanged, and the non-fibre content decreased (Fig. 4). This clearly indicated a separation of the tomato seed components. The fraction containing the higher proportion of lignin contained a similar quantity of protein at 30% of the dry matter content to the protein extract, whereby this residual material could be evaluated as a potential ruminant by-product for animal or chicken feed (Fig. 5).
An investigation into alkali extracted tomato seed proteins indicated that 61% were salt soluble globulins and 37% were glutenin and gliadins that were soluble in acetic acid and ethanol, respectively . Similarly, analysis of the protein composition of the seed fraction containing mostly seed kernels using SDS-PAGE (Fig. 6) showed that the major proteins were globulins based on comparison with previously published data . Therefore, protein bands migrating in the reducing lane (+ DTT) at 49 kDa, 35 kDa and 20 kDa, represented globulins (Fig. 5). Another study showed that globulins precipitated at pH 3.8- 6.2, whereas those that were soluble proteins were precipitated at pH 3.5- 4.6 . Therefore, it is possible that a higher proportion of soluble globulins were extracted under neutral pH conditions.
The fraction enriched with protein and fibre recovered in this study remained as a colloidal suspension but showed low solubility when precipitated and dried. However, the use of ultrafiltration has been shown to improve functional properties of proteins and this process could be combined with other physical pre-treatment processes such as microwaving for further enhancement . One of the major challenges at this stage is upscaling from gram to kilogram scale quantities and understanding the caveats at larger scale to achieving the success determined at smaller scale.