The primary genetic resource was IRRI IR64 mutant library (consisting of 66,891 lines) that was derived from a single rice plant (Oryza sativa cv. IR64), IR64-21 (Wu et al., 2005). The mutant library, as for all IRRI seed stock, follows all the international guidelines and legislation of the International Treaty on Plant Genetic Resources for Food and Agriculture. The original IR64-21 parent plant was selected for phenotypic uniformity from breeder seeds grown under field conditions. IR64 mutants used in the screening were either deletions (mainly dry seeds mutagenized with γ-radiation at 250 GY at the International Atomic Energy Agency) or point mutations (mainly pre-soaked seed mutagenized with EMS at 0.4, 0.6, 0.8, 1.0, and 1.6% concentrations at IRRI). Mutagenized lines were advanced every 4 months from the initial M₁ (first mutant generation) population by single seed descent until at least the M₃ generation. This was to ensure near homozygous inbred lines, which was essential for quantifying phenotypic characteristics in replicated trials.

Increased vein density lines were selected using a high throughput screen (Feldman et al., 2014). Rice plants from the deletion mutant populations had their veins counted on-site using a handheld microscope (Readiview handheld microscope, Meade Instruments Corporation, Irvine, CA, United States) at 80x magnification.

High vein density lines from the mutant populations were subjected to detailed microscopy analysis. Vein counts per 2 mm and cellular arrangements were viewed with a light microscope (Olympus BX51 Motorized Research Microscope, Tokyo, Japan, connected to an Olympus DP71 Microscope Digital Camera, Tokyo, Japan). Images were taken using Olympus CellˆP imaging software (from the Olympus Cell^∗ software family) and had their parameters quantified using ImageJ (Version 1.44).

Vein density viewings were taken of fresh leaf samples at 4x magnification. Samples were then fixed in FAA (formalin-acetic acid-alcohol solution) (500 ml absolute ethanol : 270 ml formaldehyde : 50 ml acetic acid : 180 ml water) for at least 12 h. These were cross-sectioned by hand using a razor blade. Leaf sections were placed in 85% ethanol for 24 h, after which the ethanol was removed and rinsed off with water. A fresh application of 85% ethanol was then applied for 24 h. The ethanol was removed and rinsed off with water after this period and replaced by lactic acid for 30–36 h before being washed with water. At 40x magnification, MCs and BSCs could be viewed. Measurements of leaf thickness, IVDs, numbers of MCs between veins and MC lengths were made for leaf 5–7 of three replicate plants per line. Figure 1 illustrates how these parameters could be taken from leaf section images.

Photosynthesis measurements were performed using the portable IRGAs, LI-COR LI-6400 and LI-6400XT machines (LI-COR Environmental, Lincoln, NE, United States). A/C_i [assimilation rate (μmol CO₂ m^-2 s^-1) over intercellular CO₂ concentration (μmol mol^-1)] curves were run according to the following program: CO₂R (CO₂ reference) steps (in order) of 1800, 1800, 1600, 1400, 1200, 1000, 800, 600, 400, 200, 100, 80, 60, 40, 20, and 1800 μmol mol^-1 with 3–5 min waiting time, 1500 μmol m^-2 s^-1 PAR (photosynthetically active radiation), 400 μmol s^-1 gas flow and 30°C block temperature. The program descended in CO₂R levels for each subsequent step (besides the second and last ones) as it was found from trial testing that it took the leaf longer to adjust to increasing, as opposed to decreasing, ambient CO₂ levels. The first step of 1800 μmol mol^-1 [CO₂R] was repeated to ensure that the leaf had enough time to stably react to the high [CO₂]. The last step of 1800 μmol mol^-1 [CO₂R] was included to prepare the photosynthesis machine for the next round of measurements.

A/PAR [assimilation rate over incident PAR (μmol photons m^-2 s^-1)] curves were run according to the following program: PAR steps (in order) of 2000, 1750, 1500, 1250, 1000, 750, 500, 250, 100, 50, 20, 0, and 2000 μmol m^-2 s^-1 with 3–5 min waiting time, 400 μmol mol^-1 CO₂R, 400 μmol s^-1 gas flow and 30°C block temperature. The program descended in PAR levels for each subsequent step (besides the last one) as it was found from trial testing, that it took the leaf longer to adjust to increasing, as opposed to decreasing, levels. The last step of 2000 μmol m^-2 s^-1 PAR was included to prepare the photosynthesis machine for the next round of measurements and was not used in data analysis.

Cuvette humidity levels were always kept between 67 and 70%. Plants were kept well-watered during the photosynthesis measurements to ensure that stomatal conductance was always high (stomatal conductance at ambient CO₂ level was ≥0.4 mol m^-2 s^-1).

Three replicate measurements of the same leaf (the youngest fully expanded leaf of the main tiller) of a plant at mid-vegetative stage were taken using the same IRGA machine for each photosynthesis curve of each plant measured. Controlling for variation with time, rather than sibling plants per line, was found to be the most important factor and was controlled by measuring the same leaf/same plant/same line/mutant without adding variation with intra-genetic-line variation. Measurements on control wild-type plants were made in a way that effectively accounted for time, environment and developmental variation. Measurements took place between 08:30 and 14:30 under greenhouse conditions. Plants were treated with artificial lighting that was controlled to give a light intensity of 200 μmol m^-2 s^-1 when days were particularly overcast.

Photosynthetic parameters were calculated both manually and through modeling and curve fitting software (Sharkey et al., 2007). From an A/C_i curve, the CP (μmol mol^-1) was the x-axis intercept of the regression line for photosynthesis rates when C_i was <200 μmol mol^-1 (six data points). The model used was based on the Farquhar and von Caemmerer biochemical model of leaf photosynthesis and calculated the remaining parameters, i.e., Vc_max (rate of RubisCO-catalyzed carboxylation) (μmol m^-2 s^-1), J_max [the regeneration of RuBP (ribulose-1,5-bisphosphate) controlled by the rate of TPU) (μmol m^-2 s^-1], TPU (capacity of the leaf to use the primary product of the chloroplast) (μmol m^-2 s^-1), g_m (mesophyll conductance) (μmol m^-2 s^-1 Pa^-1).

From an A/PAR curve, R_d (μmol CO₂ mol^-1) was the negative y-axis intercept of the regression line for photosynthesis rates when PAR was <200 μmol m^-2 s^-1 (four data points). The slope of the regression line was taken to be the maximum QE. The photosynthesis rate at saturating PAR was taken to be A_max. Instantaneous WUE (assimilation rate divided by transpiration rate; A/E) and intrinsic WUE (assimilation rate divided by stomatal conductance rate; A/g_s) were calculated at A_max.

Measurements took place at the IRRI greenhouses (14° 10′ 23.29″, 121° 15′ 32.44″). Seeds were incubated at 50°C for 72 h if their dormancy had to be broken, i.e., if seeds were more than 2 years old or if germination rates were particularly poor (<30%). They were then pre-germinated in petri dishes on filter paper treated with sterilized distilled water and incubated at 35°C until the radical and coleoptile was sufficiently swollen (usually by 24 h). Soil was sourced from the IRRI upland farm and was treated with 0.09–0.01–0.09 g NPK fertilizer per kg soil as a basal dressing and 0.09 g N fertilizer per kg soil every 2–3 weeks, depending on plant growth rate (determined qualitatively on a relative scale). Irrigation was applied generously to ensure that the soil moisture content was always high.

One way analyses of variance (ANOVAs) tested differences between categories (e.g., rice genotype) for dependent variables (e.g., vein density, leaf width and anatomical traits). Simple linear regressions tested whether two dependent variables were correlated. Regression analyses and ANOVAs were calculated using Genstat Twelfth Edition (Version 12.1.0.3278).

In a previous study, it became apparent that variants that had narrower leaf widths were more likely to have higher vein densities (Feldman et al., 2014). This led to a candidate list of narrow leaf mutants, which both segregated for the increased vein density trait close to an expected ratio (1:3 for a monogenic recessive gene) and transmitted the phenotype over successive generations.

It was clear that leaf width and vein density were also linked in BC₁F₂ (second filial generation after one backcross) segregants of the high vein density, narrow leaf width mutant lines (Figure 2). Segregants with the wild-type leaf width phenotype predominantly had wild-type vein density phenotypes. The opposite was the case for the remaining segregants, which most often had narrow leaves and high vein densities. It can be seen in Figure 2 that the wild-type leaf width segregants had almost the same frequency distribution of vein densities as for the IR64 wild-type. This was unlike the mutant leaf width segregants, which had a distinct frequency distribution curve for vein densities that peaked at 2 veins mm^-1 greater than the peak of the controls (wild-type leaf width segregants and IR64 wild-type).

Figure 3 illustrates the level of phenotypic distinction of the M₅ candidate lines (Figures 3B–F) from the IR64 wild-type (Figure 3A). This phenotypic distinction was equivalent to that between mutant and wild-type segregants in the BC₁F₂ mutant candidate lines. Though plants were qualitatively not much shorter and had similar numbers of tillers, mutants had distinctly narrow leaf sheaths in addition to narrow leaf blades. This also made their stems narrower. The degree to which mutant plants had narrower leaves is illustrated in Figure 4. Line E22097-1-3-1 (Figure 3C) had the widest leaves of the mutants: 9 mm for the widest section of the widest leaf compared to 5–6 mm for other mutant lines and 11 mm for the parental control.

The qualitative scoring of the narrow leaf trait in mutant plants was straightforward and reliable. In addition to having narrow stems and leaves, they had a darker green color (stems and leaves) and had less leaf ridges (which could be sensed both tactilely and visually) (Figure 4).

The vein density screen was partly set up to test the hypothesis that increases in vein density would lead to changes in leaf cell structure, fundamentally in terms of decreasing interveinal MC numbers (for a low MC to BSC ratio), which is especially important to C₄ photosynthesis as the two cell types perform separate functions in C₄ photosynthesis.

Increased vein densities in C₄ plants are the direct result of a leaf anatomy that has smaller IVDs and fewer MCs of approximately the same size, compared to C₃ plants. The link between vein density and cellular arrangement in the Oryza genus was confirmed in various rice lines (high vein density deletion mutants, non-high vein density deletion mutants, and ssp. indica and ssp. japonica wild-type cultivars) in a long-term greenhouse experiment (Supplementary Table S1). Vein density correlated negatively with IVD (p < 0.001) and negatively with MC number (p < 0.001) but did not correlate with MC length (p > 0.05). MC number was positively correlated with IVD (p < 0.001) and was negatively correlated with MC length (p < 0.005). Finally, MC length correlated positively with IVD (p < 0.001). Similar trends were found in a subsequent experiment of high vein density mutants (Table 1), except that rather than vein density correlating with MC number (p > 0.05), it correlated negatively with MC length (p < 0.001).

Mutant lines in the final candidate deletion mutant population (E11068-1-10, E19076-1-5, E22097-1-3, E26181-1-1 and G558-11-5) were not only distinguished by their higher vein densities and narrower leaf widths, but were also distinct in terms of cellular structure. They had smaller values of IVD and MC length compared to their parental control (p < 0.001) (Figure 5 and Table 1). MC counts were generally not different from the control with the exceptions of E216181-1-1-2 and G558-11-5-2, both of which had fewer MCs (p < 0.05).

Apart from the association with other C₄ metabolic adaptations, vein density has also been linked to photosynthetic performance possibly by controlling the capacity of the hydraulic system and photosynthate transport (Brodribb et al., 2007). To help test this, photosynthetic rates of high vein density candidate mutant plants were measured at varying CO₂ and light levels (Figure 6). In this way, candidate mutants could be characterized by key photosynthetic parameters that distinguish C₄ from C₃ species (Table 2). No C₄–specific properties were found in these plants.

Figure 6 gives an overview of the A/C_i and A/PAR responses and shows that the mutant plants had higher (p < 0.001) photosynthesis for higher levels of incident light (PAR) and C_i compared to the IR64 wild-type control.

Table 2 shows the results of photosynthetic analysis. From the A/C_i curve, Vc_max indicates the amount of active RuBisCO present in the leaf, J_max represents the capacity for thylakoid electron transport and TPU is an estimation of the ability to use the end-products of photosynthesis in the chloroplast. From the light response curve, A_max is the light saturated photosynthetic capacity and R_d is the rate of CO₂ loss with or without light energy (i.e., respiration rate). All five parameters were increased in mutant lines (p < 0.05 for Vc_max; p < 0.001 for J_max, TPU, R_d and A_max). Supplementary Table S2 illustrates variation between mutant lines for these parameters.

A plant’s leaf thickness can be correlated with photosynthetic capacity due to the accumulation of photosynthetic components per unit leaf area in C₃ plants (Peng, 2000; Marenco et al., 2009). Additionally, thinner leaves are characteristic of C₄ species where they ensure closer contact between MCs and BSCs (Dengler et al., 1994; Leegood, 2000) as well as a shorter vertical distance between veins and the stomatal epidermis. The leaf thickness of candidate lines was generally lower (p < 0.001, combining all lines) than the control, even though leaf thickness varied between mutant lines (p < 0.001). The leaf thickness of IR64-21 had an average of 66.46 ± 0.47 μm compared to 57.43 ± 0.66 μm for E26181-1-2, 59.71 ± 0.69 μm for E11068-1-10-1, 61.05 ± 0.78 μm for E19076-1-5-3, 62.93 ± 0.66 μm for G558-11-5-2 and 66.98 ± 1.07 μm for E22097-1-3-1.