Interactions between nitrogen nutrition, canopy architecture and photosynthesis in rice, assessed using high-resolution 3D reconstruction

Increasing nitrogen use efficiency is a key target for yield improvement programs. Here we identify features of rice canopy architecture during altered N availability and link them to photosynthetic productivity. Empirical mathematical modelling, high-resolution 3-dimensional (3D) reconstruction and gas exchange measurements were employed to investigate the effect of a mild N deficiency vs. surplus N application on canopy architecture, light and photosynthesis distribution throughout development. Three contrasting rice lines: two Malaysian rice varieties (MR219 and MR253) and a high-yielding indica cultivar (IR64) were cultivated. 3D reconstruction indicated key N-dependent differences in plant architecture and canopy light distribution including changes to leaf area index (LAI), tiller number, leaf angle and modelled light extinction coefficients. Measured leaf photosynthetic capacity did not differ substantially between the high and reduced N treatments; however, modelled canopy photosynthesis rate indicated a higher carbon gain per unit leaf area for the reduced N treatment but a higher carbon gain per unit ground area for the high N treatment. This is a result of altered canopy structure leading to increased light distribution under reduced N which partially offsets the reduced LAI. Within rice, altered N availability results in the development of full pho-tosynthetically functional leaves, but leads to altered canopy architecture, light distribution and overall productivity suggested that N availability can be fine-tuned to optimize biomass production. We propose wider use of 3D reconstruction to assess canopy architecture and productivity under differing N availabilities for a range of species.

Increased crop yield per hectare will be needed to sustain the growing global population. However, yield barriers are imposed by the decreasing availability of land and resources combined with a rapidly changing climate (Ray et al. 2012;Challinor et al. 2014). Nitrogen (N) is one of the most costly agricultural inputs, in terms of finance and environmental impact, despite being one of the most important mineral nutrients required to sustain yields. Field grown crops therefore require an external input of N as fertilizer but strategies for application vary substantially (Peng et al. 2006(Peng et al. , 2010. Large amounts of N fertilizers are used to increase yield and to prevent fluctuating resources from affecting production (Kant et al. 2011); however, growing concerns over the environmental consequences of mineral N use, and its potential contamination when not used efficiently, has led to the need for research in the interactions between availability and crop growth (Peng et al. 2010).
Rice is a staple food in many countries, accounting for more than 40 % of global food production. The impact of rice on health and livelihoods is even greater in South East Asia, where rice provides the main source of nutrition as well as income and employment (Makino 2011;GRiSP 2013). Evidence suggests that in recent years the average local rice yield in some rice growing countries is less than half (30-50 %) of achievable potential based on local verification trials (e.g. those performed by Malaysian Agricultural and Development Institute (MARDI); Omar and Perai 2008). The current average rice yield has been reported at 4.5-5 t ha −1 ; however, this average is mostly as a result of application of more than the recommended dose of N (Nori et al. 2008). Nevertheless, increases in yield by 50 % are estimated to be required in all rice growing countries to meet demand by 2050 (Sheehy and Mitchell 2013). The most productive systems are those which contain irrigated rice, accounting for approximately 45 % of rice cultivation area, where multiple harvests occur per year and yield is high (Redfern et al. 2012). The potential for expanding crop area under cultivation is limited within most countries, with a reduction in the rate of expansion in irrigated land, damage to current cultivated land (e.g. salinization and intensification-induced degradation of soil) plus transfer of cultivated land to other uses. Therefore, increases in rice yield must come with a concurrent reduction in the amount of land under cultivation.
There has been a general trend for increased use of N fertilizer consumption in SE Asia (FAOStat). However, the use of N fertilizers is not economical and increased N levels do not necessarily improve yield or the crop's tolerance to uncertain climatic conditions (Kropff et al. 1993;Murchie et al. 2009;Peng et al. 2010). Furthermore, studies indicate that at any given soil N content, significantly lower yields were achieved towards the end of the 21st century than the preceding three decades (Cassman 1999). Application of N fertilizer in excess of that required can even lead to negative effects including mutual shading, lodging and pest damage (Peng et al. 2006(Peng et al. , 2010. Thus, understanding the crop response to a change in N levels, and selecting varieties that are capable of outperforming others will be critical to reduce overreliance on fertilizers. As a primary constituent of essential proteins and enzymes that are involved in important plant metabolic processes, N is essential in the formation of the plant canopy and increasing photosynthetic leaf area. Photosynthetic components are a significant sink for leaf N: chloroplasts account for up to 80 % of total leaf N with Rubisco being the dominant enzyme (Makino and Osmond 1991). Leaf photosynthetic capacity and Rubisco content per unit leaf area is highly correlated with leaf N both within and between species (Evans and Seemann 1989;Theobald et al. 1998). N affects a number of developmental traits including plant height, panicle number, leaf size and spikelet number, all of which contribute to the yield potential of the crop. As a key requirement for cell division and expansion, N is integral for development, growth and final organ size (Wann and Raper 1979). During the vegetative growth stage, absorbed N primarily promotes early growth and increases the number of tillers (Mae 1997). For the formation of dense canopies, large concentrations of N are required (Connor et al. 2011). In N deficient conditions, the plant counterbalances the lack of N by producing a lower number of tillers; a compensations step that allows for fewer but fully functional leaves (Chen et al. 2003). Consequently, N deficiency generally reduces leaf area index, intercepted radiation, plant height and canopy photosynthesis rate (Connor et al. 2011). Even a mild N deficiency can make moderate changes to plant structure that will have a large impact on the light distribution and thus productivity of canopies but little is known about the changes in 3D structure (and hence light dynamics) in crop canopies with differing N content.
Assessing the productivity of crops is confounded by heterogeneous nature of plant and crop canopies; they commonly consist of multiple plants exhibiting different growth and developmental patterns (Kozłowska-Ptaszyńska 1993;Godin 2000). Therefore, understanding plant response to changes in N levels requires experimental data combined with high-resolution information on the physiological characteristics associated with a particular canopy architecture. This could be achieved through modelling approaches that can make more accurate predictions of the canopy light environment, and thus the influence of architecture, compared with manual measurements. Monitoring plant growth and estimating canopy photosynthesis rate and efficiency in the field on a large scale is a complex task. While some research has been carried out to study the effects of varying N treatments on crop systems (e.g. Harasim et al. 2016 for wheat) and on isolated rice varieties (Mae 1997;Herman et al. 2015), few studies exist to investigate how different varieties respond to varying N treatments in terms of changes to their canopy architecture. Using photosynthesis measurements alongside the three-dimensional (3D) modelling of crop canopies, we can explore plant structure and estimate crop productivity at the whole canopy scale, which would not be feasible using manual measurements (Song et al. 2013;Burgess et al. 2017).
Here we employ such methods (Pound et al. 2014) to investigate the effect of N availability on three rice lines. Because N is an integral component in photosynthetic machinery and in forming structural tissue, we hypothesize that the differences in both canopy architecture and photosynthesis from alteration of N content will influence not only the vertical light gradient and the spatio-temporal variation in light. Further we hypothesize that specific N-dependent changes in architecture such as leaf angle will influence this canopy light distribution. This work will identify whether the optimal canopy architecture for reduced N conditions differs to that of high N conditions.

Plant material and experimental design
Two Malaysian rice varieties, MR219 and MR253, both from MARDI, were selected for study due to their potential biotic and abiotic resistance (e.g. MR253 in resistant to leaf blast) and performance in marginal soils. A high-yielding IRRI cultivar, IR64, was also chosen as a control due to its high-yielding potential, tolerance to multiple diseases and pests plus wide adaptability, as well as previous studies on its response to varied N application (Morris et al. 1989;Dickman et al. 1996). Seeds were sown into module trays containing Levington Module compost with sand in the 'FutureCrop' Glasshouse facilities, University of Nottingham Sutton Bonington Campus, UK on 8 May 2014. The seedlings were transplanted into soil beds at the appearance of the third true leaf. These glasshouses are 'agronomy' glasshouses which permit the sowing of entire crop stands in sunken concrete pits under controlled glasshouse conditions, described in full in Hubbart et al. (2018). The three rice varieties were assigned in a completely randomized design. The experimental plot was divided into 18 microplots, with each microplot containing 42 plants of the same variety (7 × 6 plants). We imposed a mild or moderate nitrogen deficiency as follows. The high nitrogen plots at the start of the experiments contained 350 kg N ha −1 and the low nitrogen plots contained 250 kg N ha -1 . Additional fertilizer was not supplied throughout the duration of the experiment. Irrigation was supplied using drippers for 15 min, twice daily. Metal halide lamps provided supplementary lighting when an external light sensor detected intensity (Photosynthetic photon flux density, PPFD) below 300 μmol m −2 s −1 . A 12-h photoperiod (07:00 to 19:00) was maintained in the glasshouse using blackout blinds with a constant temperature of 30 °C and relative humidity (RH) of 50-60 %.

Composition and morphology
Five replicate measurements per plot for plant heights and soil-plant analyses development (SPAD) measurements were obtained weekly, from 20 days after transplanting (DAT) until the start of the flowering stage (100 DAT). Five replicate measurements per plot were also taken for tiller numbers between 14 and 35 DAT. SPAD measurements were taken in situ using the Minolta 502 Plus Chlorophyll Meter (Spectrum Technologies, USA), to obtain the total available chlorophyll within a specified leaf area. Chlorophyll a and b content were determined spectrophotometrically. Frozen leaf samples of known area were ground in 80 % acetone. The samples were then centrifuged for 5 min at 300 rpm and the absorbance (at 663 and 645 nm) of the supernatant was measured using a spectrophotometer. Chlorophyll a and b content were calculated using the protocol of Porra et al. (1989).
Leaf thickness was measured at the major and minor veins using leaf sectioning. Sections of the penultimate leaf on the main stem were cut from the widest part of the leaf using a sharp razor blade, for mounting on microscope slides. After mounting, the leaf sections were cleared using 85 % (w/v) lactic acid saturated with chloral hydrate. The slides were heated in a hot water bath (70 °C) for an hour. After clearing, the leaf sections were washed with distilled water and stained using 1 % toluidine blue dye in 1 % (w/v) disodium tetraborate. A few drops of glycerol were added to the leaf sections to preserve the samples before being viewed under a calibrated light microscope and images captured using a digital camera (Nikon DXM 1200). Stomatal density and length were determined using leaf impressions of both the adaxial and abaxial surfaces on the widest part of the flag leaf. Impressions were made using Coltène® PRESIDENT Plus silicone-based impression putty. Clear nail varnish was then applied to the hardened putty and later peeled and mounted on microscope glass slides for view under a ×40 magnification confocal light microscope. Images of six fields of view were taken for each variety under each treatment for analysis (Hubbart et al. 2012). All images for leaf thickness and stomata were analysed using the analytical software ImageJ.

Leaf nitrogen
Leaf nitrogen analysis was carried out by Lancrop Laboratories, York, UK. Three plants per plot were chosen at random for this analysis. For each plant, leaves were clipped at the top portion of the canopy (including the flag leaf, at 10 cm height) to make up 200 g of fresh plant material. Samples were labelled and couriered to the laboratory on the same day.

Gas exchange
Data were taken from the glasshouse grown rice in plots in the same weeks as the imaging for reconstruction (below). Leaves were not dark-adapted prior to measurements. Light-response curves (LRCs) and ACi curves were taken with a LI-COR 6400XT infra-red gasexchange analyser (LI-COR, Nebraska). The block temperature was maintained at 30 °C using a flow rate of 500 mL min −1 and light was provided by a combination of in-built red and blue LEDs. For LRCs, illumination occurred over a series of 7 PPFD values, prior to flowering and a series of 12 values post flowering, between 0 and 2000 μmol m −2 s −1 , with a minimum of 2 min at each PPFD. The LRCs were taken at two different canopy heights; designated top and bottom, where the top layer refers to the last fully expanded leaf and the bottom layer refers to a fully expanded leaf in the bottom half of the canopy that did not show signs of senescence. An additional middle canopy layer was included at full canopy development (GS5) to better capture any spatial differences in large, fully grown, plants. For the ACi curves, leaves were exposed to 1000 μmol m −2 s −1 throughout. They were placed in the chamber at 400 ppm CO 2 for a maximum of 2 min and then CO 2 was reduced stepwise to 40 ppm. CO 2 was then increased to 1500 ppm, again in a stepwise manner. At least one replicate was taken per treatment plot but with 5 replicates taken for each of the 6 treatments.
Analysis of variance (ANOVA) and Tukey's multiple comparison tests were carried out using GenStat for Windows, 17th Edition (VSN International Ltd). All individual and interaction terms were considered in the model. Data were checked to see if it met the assumption of constant variance and normal distribution of residuals.

3D reconstruction and ray tracing
3D analysis of plants was made according to the protocol of Pound et al. (2014) and Burgess et al. (2015). Every 2 weeks and following photosynthesis measurements, the rice plants (roots and shoots) were carefully removed from the plots, placed into pots and moved to the imaging studio located next to the glasshouse to prevent excessive movement and damage to leaves. Roots were supplied with water to prevent wilting. It was found that this process did not alter the key architectural features of the plants. They were imaged within 10 min using three fixed Canon 650D cameras, with a minimum of 40 images per plant. Images were captured using a revolving turntable, including a calibration target of set width (397 mm). An initial point cloud was obtained using the PMVS software (Furukawa and Ponce 2010;Wu 2011). The PMVS photometric-consistency threshold (Furukawa and Ponce 2010: Equation 2) was set at 0.45 to optimize the amount of plant material recognized in the point cloud. Default parameters were used within the Reconstructor software, except for maximum cluster size and boundary sample rate that were changed to 120 and 15, respectively. One plant per plot was removed at each growth stage leading to three replicates per line; at least two of these were used to form the final canopies. As only one plant was removed per plot, per growth stage, removal was expected to have minimal effect on the remaining plants however, to ensure this; care was taken to leave a buffer plant (i.e. the edge plant) next to removal sites. Previous work has validated the reconstruction process, indicating the in silico plants represent differ between 1 and 4 % in area compared with that of measured plants and accurately capture of leaf angles (Pound et al. 2014;Burgess et al. 2015). Duplicating and randomly rotating the individual reconstructed plants into a 3 × 3 grid with 10 cm within and between rows formed reconstructed canopies.
Reconstructed canopies consist of a number of 2D triangles within a mesh. Total light per unit leaf area for each triangle at a given time point was predicted using a forward ray-tracing algorithm implemented in fastTracer (fastTracer version 3; PICB, Shanghai, China; Song et al. 2013). Latitude was set at 3 (for Kuala Lumpur, Malaysia), atmospheric transmittance 0.5, light reflectance 7.5 %, light transmittance 7.5 %, day set at the day of the imaging. The diurnal course of light intensities over a whole canopy was recorded in 30-min intervals. The ray tracing boundaries were positioned within the outside plants so as to reduce boundary effects. The software fires rays through a box with defined boundaries: when they exit one boundary (i.e. the side) they enter again from the opposite side.

Modelling
All modelling was carried out using Mathematica (Wolfram).
Cumulative leaf area index (cLAI; leaf area per unit ground area as a function of depth) was calculated from each of the canopy reconstructions. For each depth (d; distance from the highest point of the canopy, i.e. the highest point on the z axis), we found all triangles with centres lying above d (Equation 1).
where d is also used as a reference to dived canopies into layers, with all triangles above the midpoint, d mid assigned the upper layer, and those below the lower layer. Two reference points were used for GS5 to split the canopy into three layers: top, middle and bottom. We calculated the sum of the areas of all triangles and then divided this sum by ground area. The cumulative LAI as a function of depth through the canopy was calculated using Equation 2.
Leaf angle distributions were calculated for each canopy and averaged at each canopy depth by using the angle of each 2D triangular face relative to horizontal; where an angle of 0 indicates a more horizontal leaf section and an inclination angle of 90 indicates a more vertical leaf section.
The light extinction coefficient of the canopy was calculated using the 3D structural data and the light distribution obtained from ray tracing. In order to calculate fractional interception, F, within a canopy as a function of depth at time t, we first identified all triangles lying above depth, d (Equation 1). We then calculated their contribution to intercepted light by multiplying PPFD received per unit surface area (ray tracing output) by the area of triangle. The light intercepted was summed for all triangles above the set d, and divided by light intercepted by ground area according to Equation 3.
where L 0 (t) is light received on a horizontal surface with a ground area is light intercepted by a triangle i. The light extinction coefficient derived from ray tracing data, k rt , was calculated by fitting (by least squares) the function according to Burgess et al. (2017): to the set of points {cLAI (d) , F (d, t)} calculated by varying depth from 0 to the height at total cLAI with step Δd = 1 mm (Fig. 3), where a in Equation (4) is a fitted parameter.
The response of photosynthesis to light irradiance, L, was calculated using a nonrectangular hyperbola given by the following equation: The nonrectangular hyperbola is defined by four parameters: the quantum use efficiency, ϕ; the convexity, θ; the maximum photosynthetic capacity; P max and the rate of dark respiration, R d . We assumed that the rate of dark respiration is proportional to the maximum photosynthetic capacity, according to the relationship R d = α P max . Values for P max were determined from leaf gas exchange measurements for the two canopy layers: top and bottom. For GS1-4 (prior to flowering), the LRC data was averaged prior to LRC fitting, as the shorter 7-point curves (see Materials and Methods: Gas Exchange) do not give a good fit. For GS5, all individual curves were fit; the mean ± SEM of P max is presented in Fig. 7. Curve fitting was carried out using the Mathematica command FindFit with a minimum constraint on dark respiration at 0.05 and convexity at 0.7.
As each canopy was divided into two layers, each triangle from the digital plant reconstruction was assigned to a particular layer m according to the triangle centre (i.e. with triangle centre between upper and lower limit of a layer depth). Carbon gain per unit leaf area was calculated as daily carbon assimilation over a whole canopy divided by the total surface area of the canopy according to the following equation.
Carbon gain per unit ground area was calculated as daily carbon assimilation over a whole canopy divided by the area inside the ray tracing boundaries according to Equation 7.

Canopy architecture and the light environment under different N treatments
The canopy reconstructions for each treatment for each of the five growth stages during development are provided as a visual representation in Fig. 1, where GS5 indicates full canopy closure and GS1-4 represent vegetative stages 2 weeks apart starting 18 DAT. Visual differences can be discerned between the lines and between treatments, e.g. all lines show a greater amount of plant material under the high N treatment relative to the low N treatment and this is apparent at all stages. Similarly, differences are seen in plant height between treatments ( Fig. 2A). Generally, IR64 plants were observed to be significantly shorter than the Malaysian varieties, in both high N and low N plants. In both Malaysian varieties, significant differences were observed between high N and low N plants (P < 0.05), where low N plants were at least 25 % shorter than high N plants. However, no significant differences were found between the two treatments for IR64 (P > 0.05). There were both varietal and treatment differences in the number of tillers [see Supporting Information- Fig. S1]. IR64 HN plants produced the highest number of tillers (P < 0.0001) relative to the rest of the varieties and treatments. Previous publications validated 3D reconstruction as a means of measuring canopy leaf area and leaf angle (Pound et al. 2014;Burgess et al. 2015). Fig. 2B shows modelled whole-canopy LAI throughout development. It is clear that high N accumulated a greater total LAI after Day 40, consistent with Fig. 1, indicating that greater soil N availability stimulated greater growth. While LAI values are high, particularly at GS4-5, we note published values of 8-14 at high N (e.g. Fagade and De Datta et al. 1971;Zhong et al. 2002). Fig. 2C shows distribution of the angle of plant material according to height above ground, where a higher inclination angle indicating a more upright posture. The variation between lines and treatments is greater towards the top of the canopy (i.e. 60 cm and above), reflecting differences between the lines and treatments in terms of upright versus curled leaf material in the top portion of the canopy (Fig. 1). There does not seem to be a consistent response to N treatment; however, IR64 low N and MR253 high N showed a less vertical posture. Extinction coefficients are greatly influenced by leaf angle (Murchie and Reynolds 2012). The modelled light extinction coefficient (k rt ) values in Table 1 show variety-dependent responses. In IR64 and MR219, but not MR253, k rt values for high N were consistently lower than reduced N indicating a steeper curve for light extinction in the latter. This could be caused by the higher LAI in the upper layers or by an increased chlorophyll density in the upper layers (below). This is consistent with the leaf angles given in Fig. 2 but only for IR64. However, the lack of consistency in angle distributions of leaf material indicates that angle is not substantially influenced by N treatment and that light attenuation is likely to be more influenced by LAI in the upper layers of the canopy. Fig. 3 shows how the accumulation of plant material along the vertical transect of the canopy (cumulative leaf area index; cLAI) and the corresponding interception of light (fractional interception; F) from modelled data. High N treatment plants accumulated a greater LAI at middle and upper portions of the canopy in GS1-GS3. In GS4 there was a higher accumulation of LAI throughout the canopy in the reduced N treatment but with a lower overall plant height, resulting in a reduced overall LAI. This indicates that reduced N treatment had a delayed progression of canopy development. These differences in leaf accumulation influences the canopy depth at which most light interception occurs, indicated by the steepness of the curve in Fig. 3. By GS5, the overall F was equivalent for all treatments but differences in distribution of F between treatment and variety could be seen in intermediate layers. Fig. 4 show how PPFD is distributed throughout the canopy according to the fraction of total canopy leaf surface area. In this way, it becomes possible to visualize the prevalence of different PPFD values. This is analogous to a frequency histogram and was used in a previous study  to examine the distribution of levels of radiation that are saturating and sub-saturating for photosynthesis, which is also relevant here. For example, an erect canopy would be expected to have a higher leaf area exposed to a higher PPFD values overall. This is especially important for the lower layers of the canopy which, in erect canopies, have increased light penetration which would allow leaves in these lower layers to increase photosynthesis. Here we cannot see consistent patterns in terms of treatment. We note that reduced N IR64 has a larger proportion of leaf area exposed to PPFDs of 100 μmol m -2 s -1 or lower, indicating a high level of self-shading consistent with the higher k rt value at GS5 (Table 1) and the particularly low leaf angle in upper layers (Fig. 2C) suggesting that in IR64, N treatment results in altered leaf angle and altered light distribution. Optimizing light distribution at reduced N could, therefore, be achieved by either altering depthdependent leaf area accumulation or leaf angle.

Nitrogen had limited effects on leaf composition
Chlorophyll content in upper leaves was analysed using a hand-held chlorophyll metre (Fig. 5). This revealed similar patterns in cholorophyll content in both high N and reduced N plants-greenness fell sharply at 44 and 79 days after transplanting in a majority of the plants. Significant differences between high N and reduced N plants were observed in all three varieties (P < 0.05), and while this was more obvious later in development, lower N MR219 plants were consistently lower in greenness when compared with high N plants. Reduced N MR253 plants contained significantly higher percentage leaf N in the top layer of the canopy than reduced N MR219 and IR64 plants (Fig. 6). No varietal differences were observed in high N plants. Within varieties, no differences in percentage leaf nitrogen were seen between treatments. At reduced N IR64 and MR253 had higher chlorophyll contents than MR219 indicating that the former were able to retain chlorophyll at reduced N but there is no indication that this is linked to protein or N content of leaves.

Nitrogen enhances carbon gain per unit ground area but not leaf photosynthetic rate
All lines showed a significant reduction in P max from top to the base of the canopy (P < 0.05: see Fig. 7 for GS5); however, it was not clear whether the differences in attenuation according to N treatment consistently matched the differences in light attenuation (k rt values; Table 1). It is possible to conclude that the drop in P max in IR64 between top and middle canopy layers was substantially higher than the other two varieties, consistent with the less erect leaf stature and greater light attenuation; observed under both N treatments. There were no significant differences between cultivars for P max at middle and bottom layers but in the top layer IR64 had significantly higher values than MR253 (P < 0.05). Based on previous studies, this increased P max is likely due to varietal differences as opposed to differential response to N (Herman et al. 2015). N treatment did not have a significant effect on P max at any layer (P < 0.05). A-Ci analysis of the top layer of leaves reveals a reduction in the maximum carboxylation rate of Rubisco (V cmax ) and electron transport rate ( J) in reduced N MR253 plants compared with high N plants at GS3, but this was not evident in the other varieties [see Supporting Information- Table S1], again indicating no significant effect of N availability on photosynthetic response to altered CO 2 concentrations. Carbon gain per unit leaf area and carbon gain per unit ground area is presented in Fig. 8. There were few consistent differences between cultivars or N treatment when expressed per unit leaf area (Fig. 8A) especially in GS4 and GS5. The higher values for reduced N treatment would seem to arise from the improved light distribution as a result of the lower k. However, when expressed per unit ground area there was a consistently higher carbon gain for the HN plants, with IR64 showing the highest values. At GS4 and GS5 there was little difference. This is partially consistent with the accumulation of LAI (Figs 2 and 3), indicating that despite a slightly lower carbon gain per unit leaf area, the increased biomass compensated for this reduction and improved over carbon gain on a land area basis.

Increased nitrogen increases harvestable biomass of selected lines
Altered N nutrition and corresponding changes in canopy development had opposing effects on biomass production in each of the three   varieties at harvest but not throughout development. Biomass (dry weights) ( Table 2) between varieties and treatments were not statistically different from GS1 to GS4. In GS5, all varieties in the reduced N treatment had lower biomass than the high N treatments (P < 0.05).
Similarly, harvest biomass was observed to be significantly different between treatments, with MR253 showing substantially lower biomass in reduced N plants relative to high N plants (P < 0.0001). However, high N treatment showed significantly (P < 0.05) higher harvest DW and seed DW compared with reduced N, except for in IR64 where the effects were minimal suggesting that under these conditions, biomass accumulation in IR64 is not N-dependent. Within the Malaysian lines, the higher LAI, particularly at GS3, may have been critical in driving the increase in biomass at harvest.

DISCUSSION
The influence of architecture on productivity of crops depends on a number of factors including the structure of both the individual plants and the combined or emergent properties of the whole canopy. Previous studies indicate that relatively small changes in canopy architecture can have substantial effects on light dynamics and canopy carbon gain (Zheng et al. 2008;Burgess et al. 2015Burgess et al. , 2017Rahman et al. 2018). As an essential component, N is critical in determining plant growth and structure, hence the light environment characteristics within a canopy; however, this has not previously been addressed using high resolution 3D reconstruction and ray tracing. Here we used such methods for assessing the relationship between different soil N treatments and whole canopy photosynthetic rate. The architectures of three diverse rice cultivars at five different growth stages were captured. The effect of soil N on the accumulation of leaf area and the distribution of light was strongly dependent on both the position in the canopy and the growth stage measured, with differences between treatments diminishing at the highest LAI values. As roots extend to deeper soil regions, more N may have become available to the reduced N treatment, resulting in the convergence of LAI towards the later stages of growth. The modelled canopy extinction coefficient, k rt , was lower for the high N treatment but not convincingly related to leaf angle, implying that leaf area in the upper regions of the canopy were of greater importance. Lower values of k are thought to be advantageous for productivity under high N because they permit a more efficient light penetration and accumulation of a higher LAI. However, under reduced levels of N, the priorities may be different. Previous work has shown the advantages of maintaining a low k value (Verhoeven et al. 1997;Chen et al. 2003;Burgess et al. 2015). Leaf inclination angle is critical in determining the flux of solar radiation per unit leaf area (Ehleringer and Werk 1986;Ezcurra et al. 1991;Falster and Westoby 2003). Steep leaf inclination angles lead to a decreased light capture when the sun is directly overhead (i.e. during midday hours or during summer) but increases light capture at lower solar angles (i.e. start/end of the day or during seasonal changes in the higher latitude regions). This feature has a number of practical applications including the decrease in susceptibility to photoinhibition (Ryel et al. 1993;Murchie et al. 1999;Valladares and Pugnaire 1999;Werner et al. 2001;Burgess et al. 2015); reduced risk of overheating due to reduction in midday heat loads (King 1997); and minimized water-use relative to carbon gain (Cowan et al. 1982). Nitrogen application can influence both light and nitrogen profiles, modifying k values and indeed the ratio between k for light and for nitrogen, something that has been suggested to be the result of management practice affecting cytokinin synthesis although the reasons for these changes may not always be apparent (Gu et al. 2017a). However, a  recent study showed how leaf angle responds to nutrient deficiencies in rice, mediated by strigolactone (Shindo et al. 2020). As leaf angle influences light distribution and there is a functional link between N and light profiles, there is clearly a need to further understand how N deficiency might interact with light profiles to determine canopy photosynthesis.
One of the recommendations for this work, therefore, is that the posture of plants in mild N deficiency should be more upright to enhance photosynthesis. It is also possible that IR64 may benefit from further genetic alteration to improve posture under all N treatments. At most of the growth stages, IR64 showed a consistently higher k rt in comparison to the Malaysian varieties indicating a less upright canopy or a greater accumulation of leaf area in upper canopy layers. Differences in k can also occur due to the pigment content in the upper layers, where high N induces a higher chlorophyll content (Bojović and Marković 2009;Gu et al. 2017b). This is not necessarily a problem for canopy photosynthetic rate since reduced canopy chlorophyll may enhance light penetration and, as long as it does not affect P max or light harvesting in lower layers, it should actually increase canopy carbon gain (Song et al. 2017;Walker et al. 2018). The small effect on P max here suggests that as long as lowered chlorophyll does not substantially impact Rubisco content then it should not adversely affect yield. However, in the context of the current study, we are assuming that lowered chlorophyll may be associated with lowered photosynthesis capacity even if this was not convincing under our treatment conditions.
A tradeoff between leaf area, N and photosynthesis has been seen previously in field-grown rice and can be viewed as a tendency to prevent 'dilution' of canopy N (Chen et al. 2003). N regulates growth rate, such that N is allocated to a smaller number of leaves, resulting in conserved P max values irrespective of N treatment. This is consistent with results in this study, where minimal effects were seen between P max values under different N treatments. Interestingly, MR219 showed slightly more susceptibility to a lowered N status compared with MR253, which could explain why MR253 is suitable for more marginal growing conditions in Malaysia. The residual N levels in this trial were mildly deficient (http://www.knowledgebank.irri.org/). This may have contributed to the results witnessed here; namely no change in light saturated photosynthesis but an increase in biomass in Malaysian lines. The similar leaf photosynthetic rates between N treatments mean that the differences in biomass and yield may come from canopy level processes, influenced by structure and development. Canopy photosynthesis rate can be thought of as the 'sum' of photosynthesis in all leaves in the canopy at any given time point. The light within the canopy will fluctuate on wide spatio-temporal scales according to factors such as solar movement. Hence it is necessary, when considering changes in canopy architecture to use ray tracing and modelling of photosynthesis. A higher canopy photosynthesis rate is seen in high N treated canopies but largely in the early growth stages (GS1-3) when the differences in LAI between treatments were greatest. It is interesting to note the observed higher carbon gain per unit leaf area in IR64 subject to reduced N, particularly during the mid-growth stages. This could be attributed to a possible growth 'advantage' of having a smaller canopy with less tillers, resulting in leaf tissue being exposed to higher light intensities relative to the same positions under high N. Therefore, we can conclude that the accumulation of leaf area and, therefore, light capture during canopy development is important in enhancing canopy photosynthesis rates. Second, the supposedly improved leaf angle in the reduced N treated plants was not sufficient to enable these plants to achieve the canopy photosynthesis rates seen in the high N plants.
Increased LAI also corresponded to greater height and greater dry weight (DW); however, while the two Malaysian lines exhibit a strong response to N treatment, IR64 is less sensitive. These results are consistent with previous studies on the effect of N application on IR64, which indicated that applications above 90-100 kg N ha -1 (using green manure) did not increase the agronomic efficiency of the system (Morris et al. 1989;Dickman et al. 1996) and on MR219 where increases in the N application rate led to concurrent increases in the grain (Nori et al. 2008). Similar patterns can be seen for seed DW per plant: with IR64 exhibiting similar values under both N treatments but a large increase for both MR219 and MR253. Grain weights were not consistent with modelled canopy photosynthesis rates indicating that there may be other factors such as partitioning during the grain filling period. However, the ranking of DW during GS5 are consistent with ranking of modelled photosynthesis (per unit ground area) during most of the growth stages indicating a general correspondence between modelled canopy photosynthesis and measured biomass up to GS5. The results presented here indicate contrasting N uptake and utilization responses of the three varieties, with MR219 and MR253 capable of utilizing the extra N available in the soil.
Contrasting strategies can be seen in different crops in relation to N availability. In potato (Solanum tuberosum), excess N led to enhanced apical branching and prolonged production of vegetative organs leading to a greater number of leaves per plant Biemond and Vos 1992). Conversely, under N limitation leaf size was reduced (via reduced leaf expansion rates) in order to maintain N concentration per unit leaf area and the photosynthetic capacity of the leaf (Vos and Van der Putten 1998). In contrast, maize (Zea Mays) exhibits a more conservative response to changes in leaf size relative to potato and reduces total leaf area by approximately 30 % (Vos et al. 2005). Furthermore, maintaining higher leaf area comes at the expense of decrease N per unit leaf area and a decrease in photosynthetic capacity. This reflects two opposing strategies to N availability: the maintenance of photosynthetic productivity per unit leaf area at the expense of total leaf area or; the maximization of light interception per unit leaf area at the expense of photosynthetic productivity. It is broadly expected, with some exceptions, that these contrasting strategies represent the dicot versus the Gramineae response (see Vos et al. 2005 and references within). While this study did not use limiting amounts of N availability, results suggest that under excess N conditions, N is used for the production of increased tiller number, a greater leaf area and maintenance of photosynthetic capacity per unit leaf area in rice. Table 2. Plant dry weight measurements throughout development in rice varieties grown under high-(HN) or reduced-(RN) nitrogen levels. Mean± SEM, n = 3. a-d Means in a column without a common superscript letter differ (P < 0.05), as analysed by two-way ANOVA and Tukey's multiple comparisons test.  An empirical model of photosynthesis was employed that calculates carbon gain from ray tracing values, parameterized from measured LRCs. This is integrated over the whole canopy over the course of the day for each growth stage. Fitted P max values used during modelling are given in Supporting Information- Table S2.

CON CLUDIN G R E M A R K S
High-resolution 3D canopy reconstruction revealed novel observations concerning the effect of N treatment on canopy architecture and light distribution in rice. First, leaf photosynthetic capacity was generally less responsive than leaf area to N treatment meaning that light capture and light distribution were more important in determining canopy photosynthesis rates and DWs. The reduction in leaf area accumulation during the mild N deficiency occurred in the mid-canopy region and was associated with an improved canopy light distribution in the reduced N treatment resulting in a higher carbon gain per unit leaf area compared with high N. We show the improved canopy light distribution in reduced N is more likely to be due to depth-dependent leaf area accumulation or of pigment distribution than leaf angle in the case of the Malaysian lines. We show key differences between architecture in the Philippine variety IR64 and the Malaysian cultivars: IR64 had a less upright leaves in upper canopy regions in high N which negatively affected light distribution. Improvement of light distribution would be more beneficial for the high N treatment and may improve yields even further than those seen here. This indicates a potential for increasing yields by improving the light distribution in high N treated plants.

SUPPORTIN G INFOR M ATION
The following additional information is available in the online version of this article- Figure S1. Number of tillers for three rice varieties grown under high-(HN) or reduced-(RN) nitrogen levels over time. Data was fitted using a sigmoidal dose-response (variable slope). Shown are the means (n = 5) and SEM. Table S1. Maximum carboxylation rate of Rubisco (Vcmax), RuBP regeneration rate ( J) and triosphosphate utilization (TPU) for three rice varieties grown under high-(HN) or reduced-(RN) nitrogen levels at 25 °C. Measurements were made on the youngest fully extended leaf at GS2 and 3. Values were calculated using the curve-fitting tool by Sharkey et al. (2007). a-b Means in a column without a common superscript letter differ (P < 0.05), as analysed by two-way ANOVA and Tukey's multiple comparison's test. Table S2. P max values taken from fitted light-response curves for three rice varieties grown under high-(HN) or reduced-(RN) nitrogen levels, used to calculate canopy carbon gain (Fig. 6).