Does Tomato Breeding for Improved Performance Under LED Supplemental Lighting Make Sense?

Differences in growth have been reported for tomato under LED compared to HPS light, however, it is not clear if breeding specic for LED supplemental light is worthwhile. Therefore, we derived four recombinant inbred line (RIL) tomato populations from parents with contrasting growth responses to different light spectra. These RIL populations were grown for four weeks under supplemental HPS or 95% red and 5% blue LED light in the greenhouse. For one population we also studied fruit production. Plant height and size of the side shoots of the young plants were strongly reduced under LED supplemental lighting compared to HPS in all populations. The adult plants showed shorter internode lengths, less trusses, less fruits, and lower yield of ripe fruits per plant under LED. However, when the unripe fruits at the last harvest day were included, the difference in yield between HPS and LED disappeared, indicating that the plants under LED light were compacter and slower in development, but in the end produced similar yield. We found numerous QTL, but hardly any of these QTL appeared to be signicantly LED-specic. Also, we found very signicant genetic effects of maternally inherited plastids and mitochondria, showing the importance of using a parental genotype as mother or as father. However, these effects were very similar between the two light conditions. We conclude that our study does not justify tomato breeding programs that are specically targeted at 95% red and 5% blue LED supplemental lighting.


Introduction
Year-round tomato production in temperate climates with long winter periods of low light and low temperature requires greenhouses with supplemental lighting. Those are not only costly for the grower but also leave a large ecologic footprint due to the energy usage. The supplemental light is often provided by High Pressure Sodium (HPS) lamps that emit a warm, relatively broad light spectrum that proved favourable for many plant species. However, HPS lamps consume considerable amounts of energy.
Narrow band Light-Emitting Diodes (LED) -lamps could be an alternative as these consume far less energy for the same light output (Gómez and Mitchell 2015;Kaiser et al. 2019). By providing the plants with a narrow spectrum of supplemental lighting, the lighting energy can be focussed on a more e ciently used light spectrum (Paradiso et al. 2011). Tomato plants grown in climate rooms under red and blue (RB) LED lighting had a higher CO 2 assimilation rate than under white light from uorescent tubes (Arena et al. 2016). However, the plants in the study of Arena et al. (2016) had, despite the higher CO 2 assimilation rate, in RB-light a lower biomass compared to the white light control.
Light composition not only in uences photosynthesis and biomass accumulation, but in uences plant performance in many other ways: The concentration of soluble sugars is in uenced by the quality of the light regime (Lanoue et al. 2017) which in turn is involved in oxidative stress signalling (Couée et al. 2006) and metabolic regulation (Smeekens et al. 2010). Photoreceptors, like phytochromes, in uence owering time (Strasser et al. 2010; Cao et al. 2015), and root and leaf growth (Rao et al. 2011;Costigan et al. 2011). The light regime under which plants grow also affects secondary metabolites (Carvalho and Folta 2014) and thereby could indirectly in uence quality traits and disease resistance.
The genotype by environment interaction is important in the selection of the optimal spectral composition for a crop. Next to species speci c responses (Bantis et al. 2018), natural variation for light responses was also identi ed between different tomato accessions (Ouzounis et al. 2016). In tomato, quantitative trait loci (QTL) for plant architecture that could in uence light absorption in the canopy were identi ed (Nakano et al. 2016;Fulop et al. 2016) and also QTL for photosynthetic traits (Oyanedel et al. 2000; de Oliveira Silva et al. 2018) but light spectrum differences were not taken into account. Velez-Ramirez et al. (2014) identi ed a QTL for continuous light tolerance in tomato and developed from this a genotype that showed up to 20% higher yield when grown under continuous light. This study from Velez-Ramirez et al. is thus an example for the possibilities of genetic adjustments for light use e ciency in tomato. Nonetheless, there are, as far as we know, no QTL studies in tomato looking at light spectrum effects. Therefore, we aimed at identi cation of QTL that are LED or HPS speci c.
Chloroplasts are critical in light perception and photosynthesis, and mitochondria are important for energy use. Chlorplasts and mitochondria are inherited from the mother, and contain their own genomes, and can thereby contribute to the genetic variation (Gray 1993). Therefore, we did not focus on genetic effects located in nuclei only (the QTL), but in addition we studied the effects of maternally inherited plastids and mitochondria. For this purpose, we made reciprocal crosses when developing F1 hybrids. Per combination of parents, half of the RILs were derived from F1 plants from Parent A x Parent B, whereas the other half of the RILs was derived from an F1 from the reciprocal cross Parent B x Parent A. As a result, the rst half of the RILs from this population harboured the plastids and mitochondria from Parent A, and the other half of the RILs contained the plastids from Parent B. We did this for all four RIL populations, leading to eight sub-populations.
We hypothesised that there is natural variation relevant for breeding that is speci c to LED light cultivation using tomato as a representative species that is often grown in protected conditions. The effect of red, blue and white LED light on biomass and physiological traits was determined for 40 genotypes in climate rooms. According to contrasting light reactions of those plants, crosses were made between selected genotypes and four different RIL populations were developed. Reciprocal crosses allowed the study of effects from plastids and mitochondria for each RIL population. We grew those populations in greenhouse conditions with supplemental HPS and LED light to identify physiological and morphological trait differences in the two light environments and investigated the underlying genetics by means of QTL mapping. We analysed four RIL populations in three replications in young plant stages and one RIL population of full-grown fruit yielding plants under HPS versus LED supplemental lightning, and genetically mapped plant growth and yield.

Materials And Methods
The initial 40

tomato accessions
In an initial experiment 40 tomato genotypes, including cultivars and wild accessions (Table S1) (Ji et al. 2020) with arti cial lighting. The plants were pre-grown in potting soil under uorescent light for seven to nine days after sowing and then transferred to sterile river sand and grown in one of the following LED conditions: white light with additional UVB (ca. 0.16 W m -2 ) and FR (ca 26 μmol m -2 s -1 ) or R/B LEDs at a 16h photoperiod (light intensities ca. 130 µmol m -2 s -1 ). The white light was created with white/red LEDs (GreenPower LED-TL-DR/W-MB-VISN, Philips, The Netherlands), and the additional UVB and FR was provided by Philips GreenPower LED production modules far-red (GreenPower LED -PM-FR-150), cool uorescent tubes (38W, Philips) and Philips Broadband TL 20W/12 RS Ultraviolet-B lamps. The R/B LEDs had 88% red light and 12% blue light (GreenPower LED-PM-DR/B-MB-150, Philips). The temperature and humidity of the climate room were approximately 22°C and 70%, respectively. Destructive measures for total dry biomass (roots, shoots and leaves) were done three weeks after transplanting.

Development of four RIL populations
From the 40 accessions, genotypes were selected that showed contrasting responses to the different light environments. The following four combinations of contrasting genotypes were selected for development of RIL populations: Ailsa Craig (LA2838A) and Kentucky Beefsteak (TR00021) hereafter abbreviated with "A ~ K"; LA1578 (EA00674; S. pimpinellifolium) and Rutgers (EA00465; S. lycopersicum) "L ~ R"; Moneymaker (CGN14330; S. lycopersicum) and Momotaro (Tough Boy; TR0003; S. lycopersicum) "Mm M o"; and Kentucky Beefsteak (TR00021; S. lycopersicum) and NunhemsFM001 (S. lycopersicum) "K ~ N". For each of these cross-combinations a reciprocal cross was performed, swapping the two parents, and therefore each RIL population consisted of two subgroups differing in maternally inherited plastids and mitochondria. All individuals of the populations were grown via single seed decent at least to the F5 generation (the Mm ~ Mo RIL population was an F6 Growth conditions for the four RIL populations analysed in young stage For the screening of the RIL populations, the plants were grown in glasshouses at Unifarm, Wageningen University and Research. In each experiment the same set of RILs was grown in parallel in two different compartments, one with HPS supplemental lighting and one with LED supplemental lighting. Plants for both supplemental light treatments were germinated under HPS-supplemental lighting. Of each genotype ca. 12 seeds were sown. The seeds were placed on stone-wool plugs (Gro Plug, Grodan Rockwool B.V., Roermond, The Netherlands), covered with Agra-Vermiculite (PULL Rhenen, Rhenen, The Netherlands) and a transparent foil for additional humidity for approximately the rst ve days. The plugs were soaked with nutrient solution: NH 4 1.2 mM, K 7.2 mM, Ca 4 mM, Mg 1.82 mM, NO 3 12.4 mM, SO 4 3.32 mM, P 1 mM, Fe 35 μM (as mixture of Fe-DTPA/ Fe-EDDHSA), Mn 8 μM, Zn 5 μM, B 20 μM, Cu 0.5 μM and Mo 0.5 μM. The macronutrients were provided as a mixture of fertilizers from Yara Benelux BV (Rotterdam-Vlaardingen, The Netherlands) and the micronutrients were from the Agrispoor product line of Horticoop BV (Bleiswijk, The Netherlands). The pH was between 5.5 and 5.8 (adjusted with KOH) and the EC was approximately 2.0 dS m -1 .
Seedlings with fully expanded cotyledons were selected for transfer after one week. The plants were transferred to 10 cm x 10 cm x 6 cm stone-wool blocks (Grodan Delta, Grodan Rockwool B.V., Roermond, The Netherlands). Upon transfer the plants were divided among the two compartments, one with supplemental HPS-lighting (Master green power, cgt 400W, Philips) and one with supplemental DR/LB LED-lighting (Green Power LED top lighting module DR/B LB, 190W, Philips).
The light spectra of the lamps were measured ( Figure S1) with a Field Spectroradiometer (SS-110, Apogee Instruments, Logan, UT, USA) and PSS values calculated according to the method of Sager et al. (1988). For the analysis, blue light over the wavelength range 400-500nm, green over 500-600nm, red over 600-700nm and FR over 700-800nm were considered (individual intensity measures per wavelength were used for the average if above 0.05 µmol s -1 m -2 ). Percentages of those wavelengths in the spectrum were compared to the overall intensity in the PAR range from 400 to 800 nm and rounded to 5% intervals. To shield the experiment from light of adjacent compartments, the walls of the compartments were covered with white horticultural foil. The supplemental light intensity was about 200 µmol s -1 m -2 (in a wavelength range from 400 to 700 nm) during 16 h, starting 16 h before sunset each day. The compartments were heated and set to an air temperature of 22°C during the supplemental lighting period and of 19°C outside the supplemental lighting time. Climate data inside the greenhouse were measured by a climate box (PT500 Temp./ RH sensor, Hoogendoorn Automatisering) every 20 s and averaged over 5 min intervals. The incoming solar light and other environmental factors were recorded and averaged over time (Table S2). The solar light was measured on top of the greenhouse. Eight light measurements (light meter LI-250A from LI-COR) inside the empty greenhouse were taken at the approximate plant height on one day. With this data a coe cient between incoming light and the light measured on top of the glasshouse was calculated to estimate the incoming solar light for all experiments. In case the solar radiation (in the third and fourth experiment) outside the greenhouse was above 200 W m -2 , an energy screen (Harmony 4215 O FR) was closed above the lamps in the greenhouse (leading to 42% less light at 100% screen-closure, the light-coe cient was adapted accordingly). For the mature plants the intensity of the solar light received by the plants was estimated using the measured solar radiation outside the greenhouse and a transmissivity factor of 0.62 (based on previous measurements). In addition, the measured solar radiation was reduced to 47%, as this is the approximate amount of photosynthetically active photon ux density in the solar radiation; and then assuming that 1 W m -2 solar light equals 4.57 µmol m -2 s -1 the conversion into µmol m -2 s -1 was done (McCree 1972; Britton and Dodd 1976).
The RILs of the populations L ~ R, A ~ K and Mm ~ Mo were grown together in three, repeated experiments between October 2018 and March 2019 with one replicate per RIL and experiment, giving three replicates in time. The population A ~ N was grown in one experiment in March/ April 2019 with three replicates per RIL simultaneously. The seven parents of those RIL populations were grown in each experiment with four replicates. The rst experiment was harvested 5 weeks after sowing and the consecutive three experiments were harvested in week 4 after sowing. The individual plants of each experiment were divided into four batches and sown on four consecutive days each on one ebb-and-ood table. Destructive measurements were performed during four consecutive days as well, so that the growth-period was the same for all plants. All plants were distributed in a randomized design over the four tables. For the rst three experiments the tables were divided into four areas (front, back and two middle parts of the rectangular tables) for the randomization pattern with similar numbers of plants from each RIL-population and reciprocal cross per area. From the parents one replicate was grown per table. The parents used for one population were situated close to each other on each table.
Plants were watered by ushing the stone-wool blocks for 5 minutes with nutrient solution (see above) on the ebb-and ood-tables. The watering schedule over the course of the experiment was irregular, it aimed at keeping the blocks always moist but not wet.
Phenotyping during the growth of the young plants

Destructive measurements of biomass related traits of the young plants
At the end of the growth period, four or ve weeks after sowing (depending on the experiment), the plant stems were cut at the height of the stone-wool blocks and leaves were separated from the stem. In the axil of the true leaves, side shoots developed in most genotypes. The numbers of true leaves (TL) and side shoots were counted separately from the second experiment onwards. The leaf/ shoot-structures were counted in case they were longer than ca. 5mm and the total respective fresh weights per plant determined. Final leaf area (of true leaves and side-shoots) was determined with an area meter (Li-cor, LI-3100C Area Meter, Li-cor Inc., Lincoln, Nebraska, USA). True leaves, side shoots and stems were put into separate paper bags and placed into a ventilated oven at 70°C for 24 hours followed by 105°C for 24 hours, after which the dry weight was measured (in the rst experiment the side shoot and true leaf dry weight was determined together, in all the following experiments separately). The speci c leaf area (SLA) was determined as the ratio between leaf area and leaf dry weight.
Growth conditions of the fruiting RIL population Stone-wool (Rockwool Grodan B.V., Roermond, The Netherlands) was used as planting medium, and the plants were drip-irrigated using the same nutrient solution as for the young plants. The EC of the nutrient solution in the stone wool slab was on average 3.8 ds m -1 under HPS versus 3.2 ds m -1 under LED. The compartments were heated if the temperature dropped below ca. 22°C (in a later stage of the experiment 24°C) during the supplemental lighting period or 19°C outside the supplemental lighting time. In compartment with LED supplemental lighting, heating pipes below the plants and above the lamps were used, to compensate for the heat irradiance from the HPS lamps. The HPS compartment was only heated by pipes below the plants. For pollination we used bumblebees (Koppert, biological systems). Harvesting and measurements started in week 17 after sowing and continued up until week 28. In each compartment the same set of 136 RILs was grown with one replicate each, as well as 11 replicates of each parent of the RIL population.

Growth abnormalities in the young RIL populations
The RIL populations were grown under two different supplemental lighting conditions in the greenhouse, under standard HPS lighting and R/B supplemental LED lighting. In some plants the formation of callus like structures starting at the veins on the abaxial side of leaves, called intumescence, was observed in both supplemental lighting conditions ( Figure S2). Although all populations had individual plants with light intumescence symptoms (Williams et al. 2016), in some populations the occurrence was more severe. Intumescence was scored from 0 (no intumescence) to 3 (most of the leaves affected) and because this invasive physiological trait could affect other measured traits, genotypes that had a score of 2 and above were excluded from further analysis (0.1 to 19.2% of all individual plants, depending on the population).

Measuring fruit traits
The number of fruit trusses was counted on week 17 and week 25 after sowing. The fruit trusses were labelled in order of development, starting at truss number 1. The number of fruits were counted on week 17 and week 25 after sowing as well. Internode length was measured once during the trial on week 17 after sowing. This length was measured in centimetres from the third truss to the fth truss. Fruit fresh weight was measured throughout the trial with intervals of three to four days between measurements, with the measurements starting in week 17 and ending in week 28 after sowing. The fresh fruits were measured from the breaker stage and onwards. This breaker stage is de ned as the phase at which the fruit starts developing orange/red colour on its surface skin. During the last harvest during week 28 after sowing, all fruits were collected, including the non-ripe ones. By adding the weight of all unripe fruits per plant to the yield of ripe fruits of that plant, a total fruit yield (both ripe and unripe) was calculated per plant. By taking the number of fruits per plant into account, the average fruit weight per plant was obtained. Dry weight was measured for fruits from the 3rd and the 5th truss. This was done by cutting the fruits and subsequent drying the cut pieces in a ventilated oven for the rst 12 hours at 70 °C and followed for at least 84 hours at 105 °C. Subsequently the weight of the dried fruit was determined.

Genotyping of the RIL populations
The four RIL populations were genotyped by Bejo Zaden B.V., Rijk Zwaan B.V, and Nunhems Nederlands B.V. The genotyping was done with 185, 219, 183 and 177 KASP markers for the populations A ~ K, L ~ R, Mm ~ Mo, and N ~ K, respectively, with approximately 2.9%, 12.5%, 5.2%, and 6.7 % of the marker calls scored as heterozygous.

Statistical Analysis
A Wilcoxon rank sum test was used to determine the signi cance of the light condition or reciprocal cross effect on the individual populations, using R (version 3.5.0; www.r-project.org). Descriptive statistics were done with the R package psych (version 1.8.4; Revelle 2018). All graphics were made with the R package ggplot2 (version 3.0.0; Wickham 2016). For the relative trait value in HPS and LED light, the trimmed mean under LED light, excluding the phenotypically upper 10% and lower 10% of individuals of each population (for A ~ K, L ~ R and Mm ~ Mo over three experiments), was expressed as percentage of the trimmed mean in HPS light (Mean LED *100/ Mean HPS ). For the relative trait range per population, the trimmed average (again with a fraction of 0.1 of extremes trimmed at each end) trait value of the top 10% and bottom 10% plants within each population was calculated, and the difference between those values in percent as (((top-bottom)/top)*100) was calculated. Equally the relative difference between the sub groups of the RIL population that were derived from the reciprocal crosses were calculated, by comparing the trimmed means per subgroup as ((High GroupMean -Low GroupMean )/ High GroupMean )*100. A Shapiro-Wilk normality test per condition, population and experiment showed that nearly 2/3 of the trait values were not normally distributed. Broad sense heritability was estimated by using the mean sum of squares values (Mean Sq) from a two-factor ANOVA (type III SS) on the rank-transformed data, that uses the RILgenotypes and the experiment as main factors or in case of the population "K ~ N" from a single ANOVA with genotype as factor. The ANOVA was done with the R package car (version 3.0-0; Fox and Weisberg 2019). For the estimation of the heritability, we used the equation H 2 =σ 2 g /(σ 2 g +σ 2 e ). The Mean Sq of the residuals was used as environmental variance (σ 2 e ) and ((Mean Sq RIL-genotype -Mean Sq residuals )/replicate number) gave the genotypic variation (σ 2 g ). To test for an interaction of the light treatments and the subgroups derived from the reciprocal crosses, a two-factor ANOVA on the rank transformed data was done as described above per population, using the subgroups and light-environment as interacting factors.
The QTL analysis was performed using the R package R/qtl (Broman et al. 2003;Broman and Sen 2009). The genetic map was based on the position information from the companies and re-calculated for each RIL population. A single QTL analysis was performed with the non-parametric mapping function for both the young and mature RIL populations, using individual values per experiment as input. However, in case of the young population N ~ K for which all replicates were grown within the same experiment an average value per genotype was used for the QTL analysis. In addition, a single QTL analysis with the nonparametric mapping function was performed for differences in response under LED and HPS, using as parameter (x LED -x HPS ) / x HPS. For the populations A ~ K, L ~ R and Mm ~ Mo we used the BLUE function from the R package polyqtlR (Bourke et al. 2021) in the following way: BLUE(data=data, model = pheno~geno, random = ~1|experiment, genotype.ID = "geno"). In case of the population N~K the block in which the replicates were grown was used as random effect.
Also, a multiple QTL analysis was performed for the young populations with the multiple imputation method of Sen and Churchill (Sen and Churchill 2001) using rank transformed data as input, and in case of the rst young three populations the experiment number was used as covariate. All permutation tests for LOD-thresholds/ penalties were done with 1000 permutations. In the single QTL analysis, the LOD threshold was determined speci cally per population, trait and condition. For the multiple QTL analysis, the automated stepwise procedure was used (assuming a maximum of 7 possible QTL) with a penalized LOD score criterion balancing heavy and light interaction penalties. For computational reasons the penalties were calculated based on one trait with a median level heritability under HPS light conditions in the population N ~ K. LED-speci c QTL were selected based on the multiple QTL analysis for the young population. For the mature population, these were selected based on the single QTL analysis. For the young population, QTL on the same chromosome were considered different if their approximate 99% Bayes credible intervals did not overlap. The signi cance of the QTL that were only identi ed under LED light was veri ed with an ANOVA for the effect of the allele distribution at the peak-marker on the phenotype under LED light. Also, the allele by light-condition interaction of those LED-speci c QTL was assessed with an ANOVA (using the peak marker and the light condition as interacting factors explaining the trait). For both ANOVAs the rank transformed data from the individual plants was used and for the young populations A ~ K, L ~ R and Mm ~ Mo, the experiment number was used as a block factor. The phenotypic mean difference of the two QTL with signi cant allele by light interaction was tested with a Tukey HSD-test with the package Agricolae (De Mendiburu 2020).

Selection of contrasting parental genotypes for the four RIL populations
In an initial study 40 cultivars and wild accessions (Table S1) were compared on the basis of their total dry biomass in different light conditions. Contrasts between genotypes in different light conditions were used to select parents for crosses to construct RIL populations. In the two light conditions, i.e. R/B LED light and white LED light with UVB and FR, Moneymaker, Momotaro, Ailsa Craig and Kentucky Beefsteak were among the genotypes with the largest difference in biomass comparing plants growing in white light with plants growing in R/B light. Moneymaker and Kentucky Beefsteak had a higher biomass in R/B while Momotaro and Ailsa Craig had a higher biomass under the white light ( Figure S3). In R/B light, Rutgers was among the genotypes with the highest biomass and LA1578 was among the genotypes with the lowest biomass. The genotype NunhemsFM001 had an approximately median biomass under both conditions and one of the smallest differences in biomass between the light treatments. It was in contrast to Kentucky Beefsteak which had one of the largest biomass in R/B light and the largest difference in biomass when compared between the two light environments. Those contrasting genotypes were used to create four RIL populations: Ailsa Craig (LA2838A) and Kentucky Beefsteak (TR00021) ), abbreviated as "A ~ K"; LA1578 (S. pimpinellifolium) and Rutgers "L ~ R"; Moneymaker and Momotaro "Mm ~ Mo"; and Kentucky Beefsteak and NunhemsFM001 "K ~ N". In view of possible effects from chloroplasts and mitochondria on the light use e ciency under LED and HPS, and considering that the plastids and mitochondria are maternally inherited, for each RIL population two subgroups were made, by swapping the parents when making the F1 hybrids.
Young plants were compacter, and had smaller side shoots under LED For each RIL population a relative trait value was calculated to compare LED to HPS, where the trimmed average trait value of the plants grown under HPS light represents 100% of the trait value (Figure 1, Figure  S4). Despite few traits that were most likely more affected by the seasonal changes than by the population, the overall trend between the populations' average light responses was similar. In all four populations, the plants grown under LED supplemental lighting had on average less biomass, were more compact, darker pigmented and had smaller side shoots with less biomass compared to plants grown under the standard HPS supplemental lighting. Among the different traits, the side shoot size and the height of the plants were lowest in LED light compared to HPS light. The leaf area of the side shoots under LED supplemental lighting was between 65% and 79% of that of plants under HPS lighting, depending on the RIL population. Likewise, the dry weight (DW) of the side shoots under LEDs was considerably lower, being 63% to 85% compared to those under HPS lighting. Moreover, the area and DW of the true leaves was lower under LED (between 84% to 90% of HPS). However, the reduction of the side shoots size and biomass was on average greater than that of the true leaves (TL), as the ratio of side shoots to true leaves was signi cantly smaller for plants grown under LED light compared to those under HPS (ratio side shoots/ TL area, Ratio side shoots/ TL DW; Figure 1). Also the total number of leaves and side shoots was slightly but signi cantly lower for LED than HPS grown plants. The anthocyanin and avonol content indices were in all populations between 12% and 50% higher in LED compared to HPS grown plants. Chlorophyll content was in most populations also slightly increased in LED grown plants but overall similar between the conditions (Figure 1). For all mentioned observations, we see a similar pattern for the parental genotypes ( Figure S5). The e ciency of operating electron transport of photosystem II (Φ PSII ) was 2% to 3% lower in LED light compared to HPS in the rst three experiments.
Side shoot size and plant height showed high heritability We showed signi cant differences in overall trait values between HPS and LED supplemental light grown young plants. The ranges of the trait values within a RIL population were calculated by taking the relative difference between the highest and the lowest trait value, after removing the 10% highest and 10% lowest values ( Figure 2). Traits related to side shoots (area, SLA, DW) and anthocyanin content index, had the highest relative trait range compared to all other traits. These trends were independent of the light spectrum.
The range of the traits within a population shows the variability of the trait but not how much of the trait is genetically or environmentally driven. This genetic component can be estimated by the heritability of the trait. Across all populations, the heritability estimate for side shoot size and plant height was the highest (up to 0.68 for side shoot area, 0.72 for the area ratio between side shoot and true leaves or 0.66 for plant height; Figure 2). The lowest heritability was identi ed in all populations for Ф PSII and SLA (side shoots as well as true leaves). In the case of Ф PSII it is possible that the light uctuations in the greenhouse caused uctuations in Ф PSII measurement that were confounding the analysis.
From these data we conclude that the largest variation and highest heritability were found for the size of the side shoots. This holds both for the absolute size and weight of the side shoots, and the relative size and relative weight compared to the main shoot. The height of the seedlings four weeks after sowing showed a reasonable variation, but a very high heritability. These conclusions hold for both light environments.
Genetic differences between maternally inherited plastids showed considerable impact on side shoots and plant size under both light conditions Each RIL population consisted of two sub populations from reciprocal crosses, which means that half of a RIL population was derived from a cross between parent A and parent B (A x B), using A as mother and B as father, whereas the other half of the population was derived from B x A, in which parent B was used as mother and A as father. The rationale behind this was that photosynthesis and energy related processes occur for a large part in chloroplasts and mitochondria. These plant cell organelles are maternally inherited. Figure 3 shows that there was a clear impact of the reciprocal cross group on the evaluated traits, especially on plant size and architecture. The population in which the most traits were different between the sub populations, meaning strong effects of maternally inherited plastids, was L R . The ratio of side shoots to true leaves was most affected in this population under LED-supplemental lighting. When the S. pimpinellifolium LA1578 was used as mother the side shoots were 35% larger compared to using Rutgers (EA00465; S. lycopersicum) as mother. Apparently, the formation of side shoots is signi cantly under control of maternally inherited plastids in the L ~ R population. For the K ~ N RIL population the height of four-week-old seedlings was very signi cantly in uenced by the maternally inherited organelles too: When N (S. lycopersicum NunhemsFM001) was the mother, the seedlings were 5 % longer compared to K (S. lycopersicum Kentucky Beefsteak) as maternal parent. The plants were not only 5 % longer, the dry weight and area of the true leaves were also 5% higher when the Nunhems accession was used as mother rather than as father. Apparently, the plant size was signi cantly in uenced by the origin of the plastids in this population.
Remarkably, the maternal effects were very similar between the two light conditions. An ANOVA did not show any signi cant interactions between the light treatments and the subgroups (except for PSII in the population L ~ R with a signi cant interaction at a p-value of 0.03).
Numerous reproducible QTL for young plant traits under both LED and HPS, but hardly any differential light impact Comparing the QTL detected in a single QTL analysis for all traits and populations, under HPS and LED light, showed that the distribution of QTL largely differed per population (Figure 4). Detailed information about trait values, raw marker data, and the LOD-scores per trait are provided in the Suppl. Files 1 and 2. Related traits like the area and dry weight of the true leaves or side shoots, or the DW and height of the stem were often co-localizing for each population (Figure 4). The QTL for true leaves and side shoots are at different locations. Although the location of most QTL differed between populations, the QTL for DW and area of the side shoots (and the related ratios of the side shoots to the true leaves) are co-locating in the populations that share Kentucky Beefsteak as a parent, i.e. A ~K and K ~ N. In both cases the presence of the Kentucky Beefsteak-allele at the QTL gave a lower trait value.
The QTL patterns within one population were overall similar for the two light conditions (Figure 4). However, some QTL were only detected in one of the light conditions. The QTL that emerged only in LED light treatment with a multiple QTL analysis, and not at HPS supplemental light, are graphically represented per chromosome, population and trait in Figure 5. More detailed information about the positions and LOD-scores can be found in the Suppl. File 2. Figure 5 shows in total 44 putative LEDspeci c QTL, with LOD scores between 1.4 and 11.3 and an explained variance between 1.8 and 17.2% ( Figure S6). We used an ANOVA to evaluate if the allelic effect at the QTL was signi cantly different for the two light conditions (interaction term in the ANOVA) beyond the overall difference in phenotype caused by the different lights. If the interaction would be signi cant, the allelic variation at the QTL can be useful for breeding for LED speci c conditions. For only two out of those 44 QTL a signi cant (p < 0.05) allele by light interaction was determined i.e., for leaf distance (= internode length) in the population Mm ~ Mo, and for anthocyanin content index in the population A ~K, although the differential impact was not very strong for these two QTL ( Table 1).
As an alternative approach, the relative difference in trait value per genotype for the two light environments was calculated, using the expression (x LED -x HPS ) / x HPS . Subsequently, we used this relative difference in light response in a QTL analysis. No QTL were identi ed for this response value for the populations A ~K, N ~ K and L~R for the selected traits. For population Mm~Mo the QTL for leaf distance on ch06 was identi ed again, with a peak LOD score of only 2.78 just above the threshold, con rming the weak LED speci c QTL for this trait ( Figure S7).
We conclude that we found many QTL in the four populations for the evaluated traits. Usually, these QTL appeared in both light environments, but 44 QTL could be LED speci c when looking at the QTL graphs. However, a more robust statistical evaluation of these 44 QTL showed that the allelic effects were not signi cantly different for the two light conditions, apart from two exceptions. This means that only two small effect QTL were really LED-speci c for the young plants.
LED reduced internode length, plant development, and number of fruits per plant, but eventually no yield reduction For one out of the four RIL populations we also analysed fruit traits, namely for the K ~ N population (Kentucky Beefsteak ~ NunhemsFM001). The reason for selecting this population was that one parent (NunhemsFM001) is a recent variety and its performance represents better modern varieties than old varieties like Moneymaker do. In addition, the parents had similar fruit sizes, i.e. NunhemsFM001 245.48 ± 7.25 g/ fruit and Kentucky Beefsteak 281.95 ± 15.18 g/ fruit under HPS light, which should prevent a very pronounced segregation of fruit size in the RIL population. This is relevant for analysing fruit yield.
We evaluated this fruiting RIL population under LED and HPS supplemental lighting. The population showed signi cantly (p = 0.02) shorter internode lengths under LED compared to HPS supplemental lighting in mature plants, 8% and 18% less trusses respectively in week 17 (p = 0.002) and week 25 ( p = 1.0e-10), 8% and 17% less fruits during these weeks (p = 0.018 and 1.2e-05, respectively), and 15% lower yield of ripe fruits per plant (p = 5.2e-07, Figure S8, Suppl. File 3). However, when the unripe fruits at the last harvest day were included, the difference in yield between HPS and LED disappeared, indicating that the plants under LED light were a bit compacter, slower in development, but in the end produced similar yield.
Interestingly, the fraction of dry matter partitioned to the fruits was similar for HPS and LED. Furthermore, the average weight per fruit was 8% higher for LED, which was a signi cant difference (p = 0.035).
Apparently, the lower number of fruits per plant under LED was compensated by a somewhat larger fruit size, giving a similar yield eventually.

High correlations between fruit traits
As expected, the number of trusses per plant in week 17 was highly correlated with the number of trusses in week 25, and highly correlated with the number of fruits per plant during these weeks ( Figure 6). The weight per fruit was very clearly negatively correlated with number of fruits per plant ( Figure 6). There was a weaker positive correlation between the number of fruits and number of trusses at the one hand, and the total fruit yield (in kg per plant) at the other hand. The table also shows that the number of fruits per plant showed a higher correlation with yield than weight per fruit did. Both, number of trusses and weight per fruit, contribute to yield, but apparently, the number of trusses was a more important driver for yield.

Several major QTL for fruit traits under LED and HPS, but no convincing LED speci c ones
To identify possible LED-speci c QTL for the measured traits, a QTL analysis was performed on data from both lighting systems individually. Signi cant QTL were found on ch01 and ch11 for the following traits: number of trusses in week 17 and 25, number of fruits in these two weeks, and fresh weight per fruit (Figure 7; Suppl. File 2). These ve traits are physiologically closely related, showed therefore a high correlation (Figure 7), and apparently are under control of the same QTL on ch01 and ch11. These two QTL emerged under both lighting conditions. For the fruit yield (kg per plant) we found weak QTL on ch11 for HPS, and on ch01 and ch10 for LED (Figure 7, Suppl. File 2). The presence of QTL under LED light, and absence of these QTL under HPS light, may indicate LED-speci c QTL. To study this in more detail, we mapped the relative differences for these traits between the two lighting conditions, using the expression (x LED -x HPS ) / x HPS . However, no light speci c QTL were found, neither on ch01, nor on ch11 ( Figure 7C). For none of the ve traits, LED-speci c QTL emerged, showing the absence of a LED-speci c QTL for number of trusses, number of fruits, and fresh weight per fruit in the K ~ N population.
Also, for internode length and the dry matter fraction of the fruits we detected QTL with low LOD scores that seemed to be different between the lightning conditions ( Figure 7A, B). However, again no QTL for the differential light effect was found ( Figure 7C). We conclude that we were able to detect QTL for the different fruit traits and internode length, but none of these QTL appeared to be signi cantly speci c to one light condition. Between the young populations the identi ed QTL were mainly different, re ecting the diverse genetic background used in this experiment. The parents of the RIL populations were selected according to their diverse light responses and the QTL identi ed were therefore largely speci c to the respective genotypes. Only the populations with one common parent, Kentucky Beefsteak, showed co-locating QTL (for side shoot area and DW). Overall, this suggests a complex regulation of light responses and a large diversity among the evaluated accessions. The reliability of the series of QTL described in this study was generally high, as for three of the four populations, the effect of three experimental replications was taken into account for the analysis. In addition, the similarity in QTL pattern (Figure 4) between the two light conditions for young plants shows not only the similar genetic control in both light setups but further also the reproducibility of the data, obtained from parallel experiments.

Discussion
The overall trait ranges and heritabilities were considerably consistent across populations and light conditions. There were a few differences, especially between the populations tested in the rst three experiments compared to the population K ~ N that was tested in the fourth experiment. These differences may have resulted from the increased solar light period and higher light intensity during the fourth experiment. Though there was a strong impact of the light spectrum on the overall plant performance, the QTL patterns for most traits were similar between the conditions within one population.
Only two QTL out of the 44 LED-speci c QTL were found to have a signi cant interaction with the light environment, which points to a general shift in phenotype in the different light conditions, but all tested genotypes responded in a similar way. Also Wolyn et al. (2004) studied an A. thaliana RIL population in ve different light qualities for hypocotyl length and out of eight QTL only one QTL, speci c for red light, was found to have a signi cant genotype by environment interaction. So, although the parental accessions of our RIL population were selected based on differences to light responses, the data on the RIL populations gave no strong indication for a differential interaction between the tested tomato genotypes and the two light conditions. The more uctuating environmental conditions and additional solar light in the greenhouse compared to the climate chamber experiment under which the parents of the RIL population were characterized may account for the difference in environmental response.
Next to the nuclear genome, the organelle genomes in the mitochondria and plastids carry genetic information (Gray 1993 ). This important role of the organelles in bud outgrowth could explain the differences we observed in side shoot formation between the two sub-groups from reciprocal crosses. Chloroplasts are critical in light perception and photosynthesis, and therefore genetic differences in chloroplasts might lead to differences in responses to HPS and LED supplemental light. However, we did not observe this in our experiments, as clear interactions between supplemental lighting and reciprocal crosses were not found. Higher levels of green light or lower blue to green ratios were implicated in e.g. shade avoidance (Zhang et al. 2011;Fraser et al. 2016) that also lead to hypocotyl and maybe (early) stem elongation (Folta 2004;Sellaro et al. 2010). Therefore, the green light in the HPS lamps (ca. 50%) may also play a role in the morphological differences compared to LED light. Another factor that could affect the plant growth was that the HPS emit radiative heat, the broad range HPS radiation has more infrared light and may lead to a higher meristem temperature at the top of the plants. This could favour leaf initiation and outgrowth. We have seen a slightly higher leaf number under HPS light, which could be a meristem temperature effect (Savvides et al. 2016).
Most striking for the young plants were the sizes of the side shoots compared to the true leaves, both regarding high heritability and difference between the light conditions. When looking at side shoot formation we should subdivide into axillary bud outgrowth that breaks its quiescence, and consecutive side shoot development. The rst could be indicated in our study by the number of side shoots and the second by the overall size and dry weight of these side shoots. The number of axillary buds was less affected than the size of the side shoots by the different supplemental lighting conditions. Tucker (1976) proposed that far red triggers abscisic acid formation near the auxiliary bud and thereby inhibits it outgrowth. He showed that a pulse of far red light stunts side shoot formation in young tomato plants. Generally, a low R/ FR ratio can favour apical growth instead of side shoot growth (Girault et al. 2008;Evers et al. 2011;Dierck et al. 2017). The R/ FR ratio is higher in the LED-compartment, so we would expect more side shoots there than under HPS supplemental lighting. These assumptions con ict with our experimental outcome of less and smaller side shoots in LED light. Blue light was also found to impact axillary bud outgrowth but the effect seems to vary between species and genotypes (Leduc et al. 2014; Huché-Thélier et al. 2016). However, the percentage of supplemental blue light is similar between our two conditions (ca. 5% in both). Therefore, also other light quality differences seem to impact side shoot outgrowth. In our experiments, this could be a lower blue/red ratio (0.03 in compared to 0.12) or the lack of green light in LED supplemental lighting.

Conclusions
The primary aim of this study was discovery of LED-speci c QTL for growth related traits and fruit yield. That could help plant breeders when developing modern cultivars that are optimal for greenhouses with supplemental LED-lighting, and developing other cultivars that are optimal for HPS supplemental light. Therefore, we developed four RIL populations, that were derived from parents that showed contrasting phenotypes under different light environments. We observed numerous QTL, both at HPS and LED supplemental lighting, at the seedling stage and at the fruiting stage. However, we detected hardly any light environment-speci c QTL. Therefore, our study indicates that breeding for optimal tomato cultivars for greenhouses with supplemental HPS simultaneously leads to optimal cultivars for supplemental LED lighting. And the other way around, cultivars that perform well under LED, very likely also perform well under HPS.
Moreover, we made reciprocal crosses for the four RIL populations, in view of possible effects of the maternally inherited plastids, such as chloroplasts, and the mitochondria, which are critical for photosynthesis and energy regulation. Again, we found very clear and signi cant effects from the maternally inherited plastids and mitochondria, showing the importance of using a parent either as mother or as a father. However, these differences did not interact signi cantly with the light environment. So, similarly as for the genetic factors from the nuclei (the QTL), the genetic factors from the maternally inherited plastids and mitochondria, are of paramount importance for plant architecture and fruit traits, but these genetic factors did not show a speci c interaction with either supplemental HPS or supplemental LED light.
Therefore, our data do not support tomato breeding for speci c light environments in greenhouses, as far as supplemental 95% red and 5% blue LED light is concerned.

Funding
The presented research was funded by the Netherlands Organisation of Scienti c Research (NWO), Applied and Engineering Sciences (TTW) as part of the "LED it be 50%" program, project number 14211. This study was also supported by Bejo Zaden B.V., Glastuinbouw Nederland, Nunhems Netherlands B.V.

Con icts of interest
The authors declare that there is no con ict of interest.

Availability of data and material
The allelic marker scores and raw data are provided in the supplemental material.

Code availability
The R scripts are added to the supplemental material S. initiated the project and contributed with discussions. All authors contributed to varying degrees in the writing of the manuscript.    Effects of maternally inherited plastids and mitochondria in the different tomato populations. For each combination of parents, two sub populations were derived, swapping the father and mother of the crossing. We refer to this as reciprocal crosses. The bars represent the trimmed mean difference between these subgroups of each RIL population. In the case that the bars are displayed as positive effect (light grey color), the sub-populations A x K, L x R, Mm (Moneymaker) x Mo (Momotaro) and K x N had the higher trait value (the parent that is named rst in the crosses is the maternal line). Bars with negative effect value (dark grey) represent that the other maternally inherited plastids (sub-populations K x A, R x L, Mo x Mm and N x K) gave the higher trait value. Signi cances of differences between the subgroups are indicated as follows: "." means no signi cant difference; * means p-value < 0.05, ** p-value < 0.01, and *** p-value < 0.001). The traits are sorted by their responsiveness to supplemental lighting quality, and are divided by the supplemental lighting conditions, HPS or R/ B LED light.