High-Resolution Inflorescence Phenotyping Using a Novel Image-Analysis Pipeline, PANorama

Abstract

Variation in inflorescence development is an important target of selection for numerous crop species, including many members of the Poaceae (grasses). In Asian rice (Oryza sativa), inflorescence (panicle) architecture is correlated with yield and grain-quality traits. However, many rice breeders continue to use composite phenotypes in selection pipelines, because measuring complex, branched panicles requires a significant investment of resources. We developed an open-source phenotyping platform, PANorama, which measures multiple architectural and branching phenotypes from images simultaneously. PANorama automatically extracts skeletons from images, allows users to subdivide axes into individual internodes, and thresholds away structures, such as awns, that normally interfere with accurate panicle phenotyping. PANorama represents an improvement in both efficiency and accuracy over existing panicle imaging platforms, and flexible implementation makes PANorama capable of measuring a range of organs from other plant species. Using high-resolution phenotypes, a mapping population of recombinant inbred lines, and a dense single-nucleotide polymorphism data set, we identify, to our knowledge, the largest number of quantitative trait loci (QTLs) for panicle traits ever reported in a single study. Several areas of the genome show pleiotropic clusters of panicle QTLs, including a region near the rice Green Revolution gene SEMIDWARF1. We also confirm that multiple panicle phenotypes are distinctly different among a small collection of diverse rice varieties. Taken together, these results suggest that clusters of small-effect QTLs may be responsible for varietal or subpopulation-specific panicle traits, representing a significant opportunity for rice breeders selecting for yield performance across different genetic backgrounds.

Figure 1.

Inflorescence architecture in rice. A, Rice panicles are characterized by hierarchical transitions between meristems. Plant breeders often use the composite phenotype PL, even though it combines the length of the rachis axis (RL) with that of the terminal primary branch. Thus, several structural components should be accounted for when using image analysis to measure rice panicles. The panicle notch (red arrowhead) and IM abortion point (red arrowhead in B) denote the beginning and end of IM division, respectively. Multiple IM-to-branch meristem transitions can occur at a single internode point (red asterisk), and the distance between two such points can thus be defined as the NL. Measuring PBL is also important but can be complicated by the presence of awns and secondary branches. The vegetative phenotype EL is calculated as the distance from the flag leaf ligule to the panicle notch and is reflective of how far the panicle emerges from the leaf sheath during heading.

Figure 2.

PANorama pipeline. PANorama generates a skeletonized panicle image based on the exact morphology of the panicle. A, Image capture, skeletonization, processing, and phenotype extraction are controlled by five scripts. B, Panicle images from different stages of the pipeline. Numbers correspond to the steps listed in A. 1, A panicle image captured and saved using GetPictures.py. 2, Execution of CreateSkeletons.py batch generates skeletons for every panicle image. 3, A _skel.jpg version of the original panicle image generated by CreateSkeletons.py. The skeleton is depicted in yellow, superimposed over the panicle axes. Using the DefineMainAxis.py GUI, three developmental checkpoints are selected using a mouse: EP (red circle), IMAP (orange circle), and FMAP (blue circle). 4, The _skel.jpg image after ExtractInfo.py has been executed. Using the three developmental checkpoints selected, ExtractInfo.py labels the skeleton and defines axes: exsertion axis (pink), RL (orange), primary branches (blue), and secondary and tertiary axes (yellow). Branches are collapsed into internodes during this step (red arrow). Bars = 500 pixels.

Figure 3.

Schematic depicting awn removal using exact morphological erosions. A, Raw image of a panicle branch. B, Initial binary image after segmentation. C, Erosion of the image at 0.032 cm using the Euclidean distance transform. The binary components of the image object that remain after erosion are labeled with distinct colors. D, Binary components are superimposed over the original panicle image and reconnected by the darkest path (i.e. branches and rachis axis). E, After path reconnection, the awnless binary image is used to calculate the final skeleton. F, Panicle skeleton (yellow) superimposed over axes. All segmentation and skeletonization steps (B–F) are controlled by the CreateSkeletons.py script. G, Final panicle skeleton after the use of DefineMainAxis.py and ExtractInfo.py.

Figure 4.

Skeleton calculation using the image foresting transform algorithm. Using an RGB color image of a tomato leaf (A), a binary mask is generated from the largest eight-connected component (B). A Voronoi discrete map (C) is calculated using the Euclidean distance transform. This process is visualized as wave fronts of distinct colors, which propagate inward from each contour pixel (C). When pixels meet in the center of the shape, they generate central contours (C zoom, arrows). These contours are used to define the skeleton of an image object (D zoom, arrows). By weighting the importance of pixels with respect to their positions along the length of the skeleton (i.e. along the scales), a skeleton can be thresholded using the exact morphology of the shape (E and F). For rice images, this scale parameter is set at 0.1%, as observed in D.

Figure 5.

PANorama in other species. With minimal adjustments to parameters stored in the iftPANorama.h file, PANorama can be easily adapted to measure structures from other organisms, including total plant architecture in Arabidopsis (A), the male inflorescence of maize (B), and the compound leaf (C) and inflorescence (D) of tomato. Phenotypic values are listed in Supplemental Table S3. Bars = 500 pixels.

Figure 6.

PANorama improved our ability to measure and map for quantitative panicle traits. A, Frequency distributions of phenotypic line means for six traits, calculated using 158 RILs and the two parents of the mapping population. The total number of panicle images was 1, 522. The means for IR64 and Azucena are labeled using black or white dashed lines, respectively. The x axes depict length measurements (cm) for each phenotype. B, Genetic map of the 12 chromosomes in rice, constructed with 30,984 single-nucleotide polymorphisms. PVE indicates total PVE for all QTLs within a trait. C, QTL peaks at the distal end of chromosome 1, shown from approximately 140 cM to the end of the chromosome. The y axis depicts the LOD score calculated for each marker across the region. QTL mapping was performed using Haley-Knott regression and 1,000 permutations to determine a significance threshold for each phenotype; the average threshold for all 6 phenotypes is 3.42 (not shown). LOD profiles show the refined QTL locations after forward selection and backward elimination were used to probe the model space. The peak marker for each trait is depicted using a vertical dashed line.

Figure 6.

PANorama improved our ability to measure and map for quantitative panicle traits. A, Frequency distributions of phenotypic line means for six traits, calculated using 158 RILs and the two parents of the mapping population. The total number of panicle images was 1, 522. The means for IR64 and Azucena are labeled using black or white dashed lines, respectively. The x axes depict length measurements (cm) for each phenotype. B, Genetic map of the 12 chromosomes in rice, constructed with 30,984 single-nucleotide polymorphisms. PVE indicates total PVE for all QTLs within a trait. C, QTL peaks at the distal end of chromosome 1, shown from approximately 140 cM to the end of the chromosome. The y axis depicts the LOD score calculated for each marker across the region. QTL mapping was performed using Haley-Knott regression and 1,000 permutations to determine a significance threshold for each phenotype; the average threshold for all 6 phenotypes is 3.42 (not shown). LOD profiles show the refined QTL locations after forward selection and backward elimination were used to probe the model space. The peak marker for each trait is depicted using a vertical dashed line.

Figure 7.

Quantitative variation in rice panicle traits. Phenotypic distributions are shown for seven distinct genetic lines of rice and one wild relative: Azucena (Azu), CO39, IR64, Jefferson (Jeff), Kasalath (Kasa), Moroberekan (Moro), Nipponbare (Nipp), and O. rufipogon IRGC 105491 (Orufi). Phenotypes include RL (A), exsertion thickness (TE; B), PBL (C), and BpN (D). Distributions were calculated using the methods described by Greenberg et al. (2011). Briefly, Markov chain Monte Carlo line means were calculated using 1,000 permutations and the panicle phenotypes from plants grown in four different pot sizes (2, 3, 4, and 6 inches). The total number of panicle images was 1,139.