Four tomato introgression lines (ILs), IL2-5, IL5-4, IL6-2 and IL12-4, originating from a cross between cv. M82 and the wild green-fruited species Solanum pennellii (accession LA 716) (Eshed and Zamir, 1994), were selected for the variety of fruit physical and chemical properties (Rousseaux et al., 2005). All ILs are almost isogenic to the original M82 genotype with a single homozygous chromosome segment (location indicated in the name) incorporated from S. pennellii (Liu et al., 2003). M82, the reference genome, has been previously described as “drought sensitive” (Liu et al., 2017). IL2-5 is described as a “drought tolerant” mutant compared to M82 (Liu et al., 2018). IL5-4 is described as improved in terms of °Brix, reduced firmness, and sensitivity to blossom end rot (Matsumoto et al., 2021). IL6-2 contains a Beta mutant gene (B) that alters carotenoid biosynthesis, which leads to an increase in fruit β-carotene content at the expense of lycopene (Ronen et al., 2000). IL12-4 fruits contain higher levels of antioxidants, ascorbic acid, and soluble solids compared to M82 (Sacco et al., 2013). In addition, tomato cv. H1311 was selected for its high lycopene content (Ozminkowski, 2018).

One goji (L. barbarum) cultivar, SWEET CAROLINE “FPW07” (FPW Développement, France) was selected. It displays beneficial agronomic traits such as high yield and powdery mildew resistance (Wang, 2019).

Tomato seeds were sown in plug trays in a climatic plant growth chamber with 70% relative air humidity, 120 µmol m⁻² s⁻¹ photosynthetic photon flux intensity, 14 h artificial daylight, and day–night temperatures set at 22–17°C. Following germination, seedlings were transferred to a glasshouse and transplanted individually into 0.3-L pots, then, at the third true leaf stage, into 7.5-L pots filled with compost substrate made of 90% organic matter (2/3 frozen black peat moss, 1/3 peat moss) with a 80% water retention capacity (Potgrond h70 047, Klasmann-Deilmann France). Six blocks of two plants per genotype (N = 12 plants per genotype) were randomly distributed over six rows surrounded by two border rows. The day–night temperature set point was 25–17°C. Flowers were pollinated by bumblebees. Two-year-old goji plants (N = 60 plants) were grown into 7.5-L pots in the same growing conditions as tomato plants. The goji plants were homogenized in terms of ramifications per plant, following the three-level model and were then arranged on a trellis (Jiao and Liu, 2020). Flowers were self-pollinated.

The nutrient solution (Liquoplant Rose, dilution 4/1000, Plantin, Courthézon, France) was supplied to plants by a drip irrigation system (drip flow of 2 L h⁻¹) for the whole growth period with an average electroconductivity of 1.8 mS cm⁻¹, and a pH of 6.0. During the vegetative period, the number of daily triggers was revised periodically to match 100% replacement of plant evapotranspiration. When flowers became visible in 75% of plants, half of the rows (N = 3 blocks of two plants for each tomato genotype, N = 30 for goji) were assigned to one or other of the two irrigation regimes, namely, control irrigation (CI) and DI. CI plants were irrigated in the same manner as during the vegetative period. For DI plants, the water supply was gradually decreased until a 50% reduction of irrigation volume that was maintained over the whole reproductive period up to final fruit harvest (Djidonou et al., 2016). Soil water content was tracked using probes (N = 5 random plants per species per treatment) inserted below the soil surface (EC-5 Soil Moisture Sensor, Decagon). The average volumetric water content for the CI and DI time periods was around 45% (soil water potential about −0.1 MPa) and 25% (soil water potential about −0.2 MPa), respectively ( Supplementary Figure 1 ). Over the whole fruit production period, water delivered to CI tomato plants ranged from 159 to 167 L plant⁻¹, falling to 69–76 L plant⁻¹ under DI. Total water delivery for goji was on average 190 L plant⁻¹ for CI and 105 L plant⁻¹ for DI.

Red-ripe fruits were collected twice a week during the harvest period, which lasted 48 days for tomato and 75 days for goji. Commercial fruit fresh yield (CFFY) was calculated by weighing all harvested mature fruits except for tomato fruits with blossom end rot. Average fruit weight was approximated by dividing CFFY by the total number of collected fruits. Calculation of commercial fruit dry yield (CFDY) was based on dry matter (DM) content assessed gravimetrically by drying a sub-sample at 75°C until weight stabilization. Water use efficiency was calculated on the basis of fresh (FW-WUE) or dry (DW-WUE) weight by dividing CFFY or CFDY by the total amount of water supplied in each treatment.

For the biochemical assessment of fruit quality, five (tomatoes) to six (goji berries) separate batches were formed for each treatment and genotype by mixing fruits (n > 30) harvested over the entire harvest period, in order to attenuate any seasonal changes in fruit quality (Poggioni et al., 2022). Each representative sample was grounded (IKA A11 basic; IKA-Werke GmbH, Germany) in liquid nitrogen and around 30 g of fresh powder was stored at −80°C in sealed containers. The biochemical analyses were performed on fresh powder (H-ORAC_FL and carotenoids) or on freeze-dried samples [alcohol-insoluble matter (AIM), sugars, acids, vitamin C, and phenolics].

AIM, as an estimation of fruit dietary fiber, was measured after extraction and removal of soluble molecules using 80% and 50% ethanol solutions, following starch hydrolysis by amyloglucosidase, as previously described (Gilbert et al., 2009).

Using the H-ORAC_FL assay, hydrophilic antioxidant capacity was measured. For this purpose, 0.5 g of thawed tomato or goji powder was diluted in 5 mL of pH 7.4 phosphate buffer and mixed in a sealed container at 160 rpm for 1 h. After centrifuging at 3,300 g for 10 min, 25 µL of the supernatant was added to a 96-well plate. One hundred fifty microliters of 0.5 nM fluorescein (FL) solution (Sigma Aldrich) was then added and incubated at 37°C for 10 min. Antioxidant capacity was gauged using fluorescence intensity at 485/520-nm wavelengths. After a 5-min stable phase, 25 µL of 150 mM AAPH [2,2’-azobis(2-amidinopropane)dihydrochloride] was added, starting the oxidation. Fluorescence was measured at 40-s intervals for 1 h, and values were compared to the initial stable phase reading. Results, derived from the fluorescence decay curve, were given in micromoles of Trolox equivalent (TE) per gram on a DM basis using a Trolox calibration curve (Huang et al., 2002). Three technical replicates were averaged for each genotype-treatment condition.

Soluble sugars (glucose, fructose, and sucrose) and organic acids (citric acid and malic acid) were extracted with a chloroform–methanol–water mixture (3:5:5 v:v:v) as described elsewhere (Gomez et al., 2002). Soluble sugars and organic acids were determined by HPLC, using a refractometer detector for soluble sugars (Gomez et al., 2002) and a UV detector for organic acids (Wu et al., 2002). In goji berry, with the chromatographic peak for citric acid being poorly resolved, this molecule was assayed using indirect enzymatic measurement of the disappearance of NADH by the action of citrate lyase, malate dehydrogenase, and lactate dehydrogenase with a microplate reader as described elsewhere (Wu et al., 2002).

Tomato carotenoids (β-carotene, lycopene, phytoene, and phytofluene) were extracted with a hexane-dichloromethane-ethyl acetate mixture (1:4:50 v:v:v) and then assayed by HPLC DAD as described elsewhere (Sérino et al., 2009). For goji berry, carotenoids (β-carotene, zeaxanthin dipalmitate, zeaxanthin, β-cryptoxanthin palmitate, and β-cryptoxanthin) were extracted with an ethanol–hexane mixture (2:1 v:v) then assayed by UPLC ESI TQ as described elsewhere (Dumont et al., 2020). No distinction was made between the E and Z forms of carotenes, and the mass of the peaks observed was integrated into a single molecule.

Tomato phenolic compounds (chlorogenic acid, cryptochlorogenic acid, rutin, and naringenin chalcone) were extracted with a methanol–water mixture (7:3 v:v) and then assayed by HPLC-DAD (Jordan et al., 2020). For goji berry (rutin, chlorogenic acid, and neochlorogenic acid), the protocol applied was a methanol–water–formic acid extraction (7.6:2:0.4 v:v:v) and an assay by UPLC ESI TQ as described elsewhere (Dumont et al., 2020).

Ascorbic acid (vitamin C) was measured using a microplate technique that reduces Fe³⁺ to Fe²⁺ using the reduced form of vitamin C (Stevens et al., 2006).

All these compounds were quantified using an external calibration based on commercially available pure standards (Sigma, Extrasynthese, or CaroteNature). In the absence of a pure standard, zeaxanthin palmitate could not be assayed with maximum confidence and it is therefore not studied in this article.

Carotenoid bioaccessibility was measured on CI and DI fruits for goji and for tomato H1311, M82, IL6-2, and IL12-4, as they displayed the greatest differences in fruit carotenoid content ( Supplementary Figure 2 ). Micellization efficiency, here described as bioaccessibility, was computed as the ratio between the compound quantity released from the food matrix and incorporated into mixed micelles and the compound quantity in the tested food. Carotenoid micellization was assessed using an in vitro digestion protocol including oral, gastric, and duodenal steps as previously described (Morand-Laffargue et al., 2023b) with the following modification: at the beginning of the protocol, 2 g of thawed tomato or goji fruit powder was placed in a 100-mL Erlenmeyer flask containing 32 mL of NaCl 0.9%. For goji, carboxyl ester lipase (CEL, Porcine pancreas, Sigma Aldrich, #26745) was added at the beginning of the duodenal digestion step, using 1 U mL⁻¹ digestate, as suggested by a previous study (Wen et al., 2018). The CEL enzyme hydrolyzes numerous ester compounds, including the xanthophyll esters present in goji, and is known to be biologically active in the human digestive system (Chitchumroonchokchai and Failla, 2006). The rest of the protocol remained unchanged and was followed by centrifugation and 0.22-µm filtration (Millex-GS, mixed cellulose esters, Millipore) of the micellar supernatant. Aliquots were stored at –80°C in sealed tubes. The carotenoids in the micellar fractions of the digestates were extracted using slightly different protocols to those used for fresh fruit. For tomato, 500 µL of the micellar fraction was extracted with an ethanol–hexane mixture (1:4 v:v). The hexane phase of the extract was evaporated under nitrogen flow and then solubilized in a MeOH–DCM mixture before injection into HPLC-DAD as described elsewhere (Borel et al.; Morand-Laffargue et al., 2023a). For goji berry, 450 µL of the fraction was contacted with 30 mg of CaCO₃ and 600 µL of hexane–ethanol (2:1 v:v) with butylated hydroxytoluene (BHT) (0.2%; w:v). After adding 300 µL of water, the two phases (hexane and aqueous) were separated by centrifugation. The hexane phase was placed in a microtube while 600 µL of 0.1% BHT hexane was added to the aqueous phase. A second centrifugation was used to recover the hexane phase and was added to the first. Between each step, a shaking step was performed. The two hexane phases were evaporated under nitrogen. The resulting dry extract was reconditioned immediately prior to analysis using UPLC-ESI-TQ by adding 150 µL of MTBE (methyl tert-butyl ether) followed by 300 µL of ethanol as described elsewhere (Dumont et al., 2020).

Micellar fractions from tomato digesta (100 µL) were injected into the HPLC-DAD as described elsewhere (Borel et al., 2021; Morand-Laffargue et al., 2023b). In our work, specific wavelengths were assigned for the quantification of all-trans-lycopene (472 nm), β-carotene and lutein (450 nm), phytoene (286 nm), and phytofluene (350 nm).

For tomato fruits, two-way analyses of variance (ANOVAs) were applied to test the impact of genotype (G), water treatment (T), and their interaction (G*T) (R software, version 4.2.2). Where at least one factor, or the interaction, was significant (p < 0.05), post-hoc pairwise comparisons were performed with Fisher’s least significant difference (LSD) test followed by a false discovery rate p-value adjustment using the Benjamini–Hochberg procedure (package “agricolae”, v 1.3-5). The impact of DI on goji fruits was evaluated independently using an unpaired two-tailed Student’s t-test. The standard deviations (SDs) of WUE and commercial fruit yield on a DW basis (CFDY) were calculated by considering error propagation formula for each product of the variables (Ku, 1966). For heatmaps, the Euclidean distance matrix was calculated and the complete-linkage clustering method (R, package “pheatmap”, v1.0.12) was applied, indicating significant comparisons with a “*” (p < 0.05). Plots were drawn using either R (package “ggplot2”, v3.4.0, and “ggprism”, v1.0.3) or GraphPad Prism (GraphPad Software, v9.0.0). Principal component analysis (PCA) of tomato carotenoids was carried out on centered and scaled data (R, package “ade4”, v.1.7-20) with individual values (95% confidence ellipse for each genotype-irrigation group) and variables plotted in the first two PC axes (R, package “factoextra”, v.1.0.7).

Overall, DI and genotype had a strong impact on tomato production in terms of CFFY and CFDY ( Figure 1 ). On average, the CFFY of CI plants ranged from 2,023 g plant⁻¹ for IL6-2 to 3,531 g plant⁻¹ for IL5-4, and for all six tomato genotypes, it fell by −64% under DI (from −47% for M82 to −84% for IL12-4). The average CFFY for goji plants was 433 g plant⁻¹ under CI, but the average reduction under DI (−34%) was less severe than in tomato. Fruit DM content was higher under DI compared to CI: +14% on average for goji and from +26% (IL6-2) to +45% (IL12-4) in the tomato genotypes ( Table 1 ). As a result, the average decrease of CFDY under DI ranged from −34% (M82) to −76% (IL12-4) for tomato and it was only −25% for goji.

Considering commercial fruits, tomato plants produced 18.4 g fresh mass L⁻¹ water under CI and 15.4 g L⁻¹ under DI. The relative variation under DI ranged from −61% (IL12-4) to +26% (M82) compared to the CI treatment. The FW-WUE was lower in goji than in tomatoes, with an average of 2.28 g fresh mass L⁻¹ water for CI and a +19% increase under DI. On a dry mass basis ( Figures 1C, D ), tomato plants produced 1.32 g dry mass L⁻¹ water under CI and the relative effect of DI ranged from −44% for IL12-4 to +57% for M82. Goji produced 0.54 g dry mass L⁻¹ water with a +36% increase under DI compared to CI.

Quality traits exhibited strong heterogeneity between tomato and goji, but also among tomato genotypes ( Table 1 ). In tomato grown under CI conditions, the largest variations were observed for total carotenoid content with a 3.2-fold difference between the poorest (IL6-2) and richest (H1311) genotype. The sugar:acid ratio varied by a factor of 2.4 between the lowest (IL12-4) and highest (IL6-2) values. A 2-fold increase was observed between the lowest (IL6-2) and highest (M82) total acid contents and a 2.4-fold increase was observed between the lowest (IL12-4) and highest (IL2-5) total phenolic content. The high-lycopene cultivar H1311 also had the highest contents in phytofluene, lutein, ascorbic acid, and rutin. Under comparable conditions (CI), goji fruits displayed higher dry matter content, lower AIM content, and higher sugar and lower acid contents on a dry mass basis than ripe tomatoes, resulting in a higher sugar:acid ratio (34.1%). Goji fruits also displayed lower ascorbic acid (121.9 mg per 100 g⁻¹ fruit DW) and lower hydrophilic antioxidant capacity (4.3 µmol TE per g⁻¹ fruit DW) than tomatoes. The total phenolic and carotenoid contents on a dry mass basis were one order of magnitude higher in tomato fruit than in goji but values reported in Table 1 are the sums of quantified compounds only, and thus they may be underestimated.

The relative effect of DI on fruit traits is presented on the DW ( Figure 2A ) and FW ( Figure 2B ) basis. On a DW basis, DI had a negative or non-significant impact on total ascorbic acid, malic and citric acid, AIM, and hydrophilic antioxidant capacity in both tomato and goji fruits. DI had a negative or non-significant impact on total carotenoid and β-carotene content in tomato fruits, but a positive significant impact on the same compounds in goji. DI had a positive or insignificant impact on rutin, chlorogenic acid, total polyphenols, sugar:acid ratio, total sugars, glucose, fructose, and sucrose for both tomato and goji fruits. On an FW basis, DI had a positive or non-significant impact on 13 out of 16 measured traits for both tomato and goji genotypes. The exceptions were sucrose, which positively increased or was stable under DI for tomato but not for goji, β-carotene, which significantly increased in goji under DI but was stable for tomato, and malic acid, which significantly decreased for goji but was rather stable for tomato.

For all of the 24 measured fruit traits on a DW basis ( Figure 3 ), univariate ANOVAs revealed a high proportion of variance explained by genotype (p < 0.001), ranging from 34.2% of total variance for ascorbic acid to 96.5% for β-carotene. The genotype factor accounted for 41.5% of total sugar variability, for 83.3% of total acid variability, for 91.1% of sugar:acid ratio, and for 87.3% of total carotenoid variability. Genotype variations in phenolic compounds ranged from 39.9% for rutin to 73.1% for cryptochlorogenic acid. The contribution of DI to the total variance of fruit metabolites ranged from 4.9% for fructose (p < 0.05) to 23.4% for glucose (p < 0.001) and 38% for sucrose (p < 0.001). It contributed to 32.4% of ascorbic acid variance (p < 0.001) and to less than 20% for all other compounds. Concerning carotenoids, DI significantly contributed (p < 0.001) from 1.4% for lycopene to 9.8% for lutein variance. With regard to the interactions between water treatment and genotypes, the highest contribution was observed for sugars, ranging from 12.3% for sucrose (p < 0.001) to 13.6% for fructose (p < 0.01) and 16.2% for glucose (p < 0.001) ( Figure 3 ).

The absolute quantities of carotenoids incorporated into mixed micelles after in vitro digestion are displayed in Figure 4 . In tomatoes, the effect of genotype on fruit carotenoid content was not systematically reflected in the amount recovered in micelles due to variations in bioaccessibility ( Table 2 ). For instance, more lycopene was observed in H1311 fresh fruits (+27.3%, p < 0.001) compared to other genotypes, but the total micellar lycopene recovered was independent of genotype (p = 0.26) as lycopene bioaccessibility was quite low ( Table 2 ). Regarding the treatment effects, the micellar amounts of all carotenoids were reduced by DI compared to CI. This effect was significant and independent of genotype for phytoene (−30.3% on average) and phytofluene (−41.3%) recovery. Similarly, DI reduced, though not significantly, the micellar amount of lycopene (−36.4%, p = 0.09), β-carotene (−39.8%, p = 0.06), and lutein (−11.5%, p = 0.09) for all genotypes. When considering total carotenoids, a significant genotype effect (p < 0.001) was observed on the micellar carotenoid content, with a consistent (p = 0.61 for interaction) but negative DI impact (−31.7% on average, p < 0.01).

These variations resulted from variations in bioaccessibility for each compound ( Table 2 ). Lycopene bioaccessibility was reduced from −34.1% (IL12-4) to −54.9% (IL6-2) under DI (p = 0.03 for all genotype). The bioaccessibility of phytoene and phytofluene was higher than that of lycopene, but still a negative effect of DI was observed for all genotypes, ranging from −16.6% (IL12-4) to −29.8% (IL6-2) for phytoene and from −39.3% (H1311) to −47% (M82) for phytofluene. Similarly, lutein bioaccessibility was reduced (p < 0.05) by DI for all genotypes, from −21% (H1311) to −30.2% (IL12-4) with the exception of IL6-2 (+3.7%). The β-carotene bioaccessibility was significantly impacted by genotype and treatment (p < 0.01) without interaction between the two factors. DI reduced β-carotene bioaccessibility from −14.6% (IL6-2) to −50.7% (M82). IL6-2 fruits displayed the highest β-carotene content on an FW basis ( Figure 4 ) but the lowest bioaccessibility value (4.1% for CI, Table 2 ). Regarding total carotenoid, the bioaccessibility, which averaged 8.2% for all CI fruit, significantly depended on genotype (p < 0.001) and it was reduced under DI (p < 0.001), from −27.5% (IL6-2) to −41.2% (M82). Across all tomato genotypes, carotenoid bioaccessibility under CI conditions can be ranked as follows: lutein (50.6%) > phytofluene (41.8%) > phytoene (19.9%) > β-carotene (11.4%) > lycopene (0.3%), while the ranking of the average relative impact of DI on carotenoid bioaccessibility is AS FOLLOWS: lutein (−19.3%) < phytoene (−23.5%) < β-carotene (−37.5%) < phytofluene (−41.9%) < lycopene (−44.2%).

As found by Hempel et al., the use of porcine CEL to cleave carotenoid esters in our protocol resulted in a relatively low percentage of free forms (<0.2%) in the digestate, raising questions over this enzyme’s ability to cleave carotenoid esters, although this is expected to occur at higher rates in vivo (Hui and Howles, 2002; Hempel et al., 2017 a). That said, a strong incorporation of free forms was observed in the micellar phase, averaging 100% of the free forms quantified at the end of the digestion phase for both zeaxanthin and β-cryptoxanthin (data not shown). Therefore, zeaxanthin and beta-cryptoxanthin in the fruit and micelles were expressed as the sum of the free and esterified forms. The fractions of free compound forms incorporated in the mixed micelles produced at the end of the in vitro digestion are shown in Figure 5 .

On an FW basis, DI significantly increased the fresh fruit content in zeaxanthin (+ 31.1%, p < 0.001), β-cryptoxanthin (+49.4%, p < 0.001), and β-carotene (+36.1%, p < 0.001) compared to CI fruits ( Figure 5A ). Fresh fruit β-cryptoxanthin mainly accumulated in palmitate form with trace amounts of free β-cryptoxanthin in both CI and DI fruits (<0.6 µg g⁻¹ fresh fruit weight) ( Figure 5 ). No differences in zeaxanthin, β-cryptoxanthin, or total carotenoids were observed between CI and DI micellized fractions ( Figure 5B ). This was probably due to the reduction of the bioaccessibility of zeaxanthin (−17.7%, p = 0.33) and β-cryptoxanthin (−32.76%, p < 0.001) under DI ( Table 2 ). Overall, while total carotenoids increased in fresh fruit under DI (+31.4%, p < 0.001), there was no difference in the total carotenoids recovered in mixed micelles between CI and DI samples because of a slight decrease in total carotenoid bioaccessibility under DI (−16.7%, p = 0.19) ( Figure 5B and Table 2 ). With regard to deposition forms in the micellar phase, most of the zeaxanthin was recovered as dipalmitate, with an average of just 6.4% of total zeaxanthin being recovered as free zeaxanthin without any significant differences between CI and DI treatments. Most β-cryptoxanthin was recovered as palmitate but, compared to zeaxanthin, a higher fraction was recovered in free compound form, comprising, on average, 43.7% of total β-cryptoxanthin in the micellar phase, with no significant differences between CI and DI treatments ( Figure 5C ).