Molecular Plant ll

Comment

The power of phenomics: Improving genebank

Genebanks are considered the ‘‘crown jewels’’ of research orga- HTP FOR EFFICIENT AND FASTER
nizations, as they safeguard valuable plant genetic resources and
address the emerging challenges posed by climate-related con-
IMPROVEMENT OF AGRONOMIC TRAITS
straints (Anglin et al., 2018). Currently, 7.4 million accessions HTP methods using sensors, LiDAR (light detection and ranging),
have been conserved in 1750 gene banks worldwide. Of these, or RGB cameras mounted on a ground-based or unmanned
only 10% are being used for crop improvement (Commission on aerial vehicle (UAV) accurately measure plant traits such as plant
Genetic Resources for Food and Agriculture, 2010). The limited height, biomass, yield potential, and lodging resistance in green-
availability of passport data and the challenge of characterizing house and field settings (Ghamkhar et al., 2019; Yang et al.,
trait attributes of accessions at large scale are some of the 2020). HTP can also predict grain yield indirectly using traits
challenges to relate their true value (Tadesse et al., 2019). such as early vigor, height, canopy properties, and biomass.
Obtaining traits such as agronomic or nutritional characteristics Automated HTP platforms are being used to measure traits of a
requires extensive field phenotyping and multiple site testing diverse range of perennial ryegrass genotypes and an in vitro
over several seasons. This will create value with accurate collection of (1617) banana accessions at our organizations,
evaluation of genotype-by-environmental variation in a given AgResearch in New Zealand, and the International Musa
trait. The use of low-cost high-throughput phenotyping (HTP) Germplasm Transit Centre, respectively. The International
methods to characterize the genotypic and phenotypic represen- Musa Germplasm Transit Centre employs a quick phenotyping
tations of genebank genetic resources, otherwise known as gen- platform called ‘‘Bananatainer’’ that can simulate diverse
ebank phenomics, will help in understanding crops’ potential for climatic conditions to cultivate up to 504 plants for their ability
use in breeding programs with desirable traits. to withstand abiotic stress (van den Houwe et al., 2020).

Technologies such as imaging, spectroscopy, and robotics Deep learning methods including convolutional neural networks
enable rapid and non-destructive measurements of observable have been successfully applied in the HTP platform. In a recent
characteristics such as plant height, leaf area, seed size, disease study, researchers used multivariate clustering to identify maize
resistance, and abiotic stress tolerance. This information can help accessions with homogeneous or heterogeneous phenotypes
breeders better understand the genetic diversity and relation- using an extensive image collection of 19 867 maize cobs ob-
ships among accessions, inform conservation strategies, and tained from 3449 images of 2484 accessions from the Peruvian
support the development of predictive models for crop perfor- maize genebank at Universidad Nacional Agraria La Molina.
mance under different environmental conditions. Genebank phe- This successful implementation of deep learning in genebank
nomics can increase the productivity and profitability of small- phenomics demonstrates its high potential for future applications
holder farming systems by capturing information about in plant breeding and agriculture (Kienbaum et al., 2021).
germplasm material that requires less resource (such as water
or fertilizers), thereby contributing to global food security.
HTP OF SEEDS
A good example is the peanut accession PI 203396, which was Genebanks evaluate the physical characteristics of seeds such
acquired from Brazil in 1952 and added to the genebank of US as size, shape, color, vigor, and varietal identification as a part
Department of Agriculture (Griffin, GA, USA). It was not docu- of their quality assurance activities. Automated systems for rapid,
mented to confer resistance to tomato spotted wilt virus but accurate,and cost-effective phenotyping of seeds, such as imag-
only to leaf spot, making it challenging for new peanut breeders ing or using cameras or sensors, will enable genebanks to effi-
to recognize its value (Anglin et al., 2018). However, its ciently measure characters of hundreds of seeds, with more ac-
incorporation into commercial peanut varieties has contributed curacy than human eyes, which can be useful in plant breeding
over $200 million annually to the US economy, demonstrating and genetic research.
the importance of characterizing and documenting poorly
represented genebank accessions for valuable traits in crop Algorithms can be trained to recognize and analyze images, al-
improvement. lowing for rapid and accurate characterization of seed traits.
For example, a portable conveyer-based imaging system
The CGIAR centers IPGRI (Nguyen and Norton, 2020) and provides a rapid and accurate means of phenotyping large
Bioversity International (now the Alliance of Bioversity numbers of seeds. Imaging systems and computer-based seed
International and CIAT) have published phenotypic descriptors
for over 100 crops to ensure consistent and comparable data Published by the Molecular Plant Shanghai Editorial Office in association with
across genebanks. Cell Press, an imprint of Elsevier Inc., on behalf of CSPB and CEMPS, CAS.

Molecular Plant 16, 1099–1101, July 3 2023 ª 2023 The Author. 1099
Molecular Plant Comment

Figure 1. Satellite imagery for biomass measurement in a range of perennial ryegrass germplasm plots.
The images were captured between March and May 2018 (left to right) in the Canterbury region, New Zealand, by AgResearch. Dark green indicates more
biomass, and light green means less biomass. Brown and pink colors indicate bare ground or senesced plant material, respectively.

phenotyping programs have been utilized to characterize seed effective and efficient alternative for developing total carotenoid
morphological traits of 589 soybean accessions and genotypes content phenotyping tools in cassava roots with high predictive
(53 909 seeds) at the National Agrobiodiversity Center ability (de Carvalho et al., 2022).
Jeonju, (Korea) based on the guidelines of International Union
for the Protection of New Varieties of Plants. Technologies such as spectral imaging and sensor technologies
have not yet been fully utilized in plant science, but they hold po-
The International Rice Research Institute (IRRI) maintains the In- tential for investigating plant nutrients, including vitamins and
ternational Rice genebank, which houses over 132 000 rice ac- macronutrients. Plant phenomics can also measure forage and
cessions. IRRI uses advanced phenotyping tools such as Video- crop quality, such as flavor and digestibility, and eventually
meter (https://videometer.com/) and germination Scanalyzer to metabolizable energy. Ground truthing, such as sensory assess-
identify and compare new seed samples and perform seed ments or fruit firmness measurements, may be required before
analysis and viability monitoring (Lee et al., 2020). The use of standardization of data across genebanks.
UAVs equipped with thermal imaging for drought tolerance
screening and field-risk management in IRRI is also a promising DISEASE RESISTANCE
tool for preserving the genetic diversity of seed samples through
the regeneration cycle of genebank accessions. Phenomics is also utilized in screening germplasm for plant dis-
ease resistance. Imaging sensors and machine learning (ML)
PLANT PIGMENTS AND NUTRITION models are used to identify links between resistance to diseases
and other plant traits in order to pick strategies to manage germ-
Phenomics can also be used to analyze plant content and com- plasm collections with a disease-based focus. For example, a
pound traits such as pigment concentration, nutrient content, relevant study in this field, the combination of spectral vegetation
and overall quality. indices in random forest ML models, was used to predict yellow
rust scores in wheat genotypes obtained from the Nordic Gene-
Biofortification is a process that aims to increase the bioavail- bank (Koc et al., 2022).
ability of micronutrients in crops with desirable pigment traits,
such as high carotenoid or chlorophyll content, to benefit the hu- Numerous studies have reported using ML approaches to identify
man population. Biofortification has been achieved in staple different levels of stress such as the aflatoxin level in maize (Yao
crops like wheat, rice, maize, cassava, pearl millet, beans, cas- et al., 2013), leaf rust in wheat (Ashourloo et al., 2014), or powdery
sava, and sweet potato in Asia and Africa. In a study that mildew in cucumber (Lin et al., 2019). Developing labeled, broad-
compared traditional laboratory-based methods with high- spectrum plant disease stress datasets across various plant spe-
throughput image-based phenotyping to measure anthocyanin, cies will allow for the development of disease-resistance libraries
chlorophyll, and carotenoid content in 30 red lettuce genotypes, and prevent data duplication. In taking this approach, Selvaraj
the image-based method offered a better alternative for selecting et al. (2019) have developed a smartphone application called
plants with high pigments to develop biofortified crops (Maciel Tumaini that uses deep learning-based models to classify five
et al., 2019). Another study used red, green, blue camera image major banana pests and diseases.
analysis to predict total carotenoid content in cassava roots.
Digital images of 228 cassava genotypes from the Embrapa
Cassava Germplasm Bank were analyzed using colorimetric
SCALED-UP FIELD PHENOMICS
indices to extract the intensity of yellow color and lightness. Crop phenomics using aerial vehicles or satellites help gene-
The study found that digital image analysis is a cost- banks to screen and record crop growth from above rather
1100 Molecular Plant 16, 1099–1101, July 3 2023 ª 2023 The Author.
Comment Molecular Plant
than on the ground. UAVs or aircraft equipped with high- Real-time, non-destructive and in-field foliage yield and growth rate
resolution cameras and sensors can capture images and data measurement in perennial ryegrass (Lolium perenne L.). Plant
of crops and vegetation from above. Satellite-based phenomics, Methods 15:72. https://doi.org/10.1186/s13007-019-0456-2.
on the other hand, gather data on plant growth and development Kienbaum, L., Correa Abondano, M., Blas, R., and Schmid, K. (2021).
over large areas (Figure 1). Although these methods may sound DeepCob: precise and high-throughput analysis of maize cob
expensive for genebanks, in the very near future, its geometry using deep learning with an application in genebank
advantages to ground-based imaging for screening broad ranges phenomics. Plant Methods 17:91. https://doi.org/10.1186/s13007-
of trait statuses within germplasm collections of the same spe- 021-00787-6.
cies will be pronounced. Koc, A., Odilbekov, F., Alamrani, M., Henriksson, T., and Chawade, A.
(2022). Predicting yellow rust in wheat breeding trials by proximal
CONCLUDING REMARKS AND phenotyping and machine learning. Plant Methods 18:30. https://doi.
org/10.1186/s13007-022-00868-0.
PERSPECTIVES
Lee, J.S., Chebotarov, D., Platten, J.D., McNally, K., and Kohli, A.
Phenomics technologies are crucial for unlocking the potential of (2020). Advanced strategic research to promote the use of rice
genebanks to address global crop production challenges. genetic resources. Agronomy 10:1629. https://doi.org/10.3390/
Phenotypic characterization enables early and efficient identifica- agronomy10111629.
tion of germplasm with specific traits of interest, enhancing the
Lin, K., Gong, L., Huang, Y., Liu, C., and Pan, J. (2019). Deep learning-
value of breeding programs. Collaboration among genebanks based segmentation and quantification of cucumber powdery mildew
through facilitators such as Divseek International can enable using convolutional neural network. Front. Plant Sci. 10:155. https://
the affordability and standardization of phenomics technologies. doi.org/10.3389/fpls.2019.00155.
Scaling up germplasm screening using aerial and satellite tech-
Mascarenhas Maciel, G., de Araújo Gallis, R.B., Barbosa, R.L.,
nologies is the next stage for genebank materials characteriza-
Pereira, L.M., Siquieroli, A.C.S., and Vitória Miranda Peixoto, J.
tion in the 21st century.
(2019). Image phenotyping of inbred red lettuce lines with genetic
diversity regarding carotenoid levels. Int. J. Appl. Earth Obs. Geoinf.
ACKNOWLEDGMENTS 81:154–160. https://doi.org/10.1016/j.jag.2019.05.016.
Thanks to the Crops for Nutrition and Health at the Alliance of Bioversity
International and CIAT for encouraging this writeup. No conflict of interest Nguyen, G.N., and Norton, S.L. (2020). Genebank phenomics: a strategic
is declared. approach to enhance value and utilization of crop germplasm.
Plants 9:817.

Ezhilmathi Angela Joseph Fernando1,*, Selvaraj, M.G., Vergara, A., Ruiz, H., Safari, N., Elayabalan, S.,
Ocimati, W., and Blomme, G. (2019). AI-powered banana diseases
Michael Selvaraj1 and Kioumars Ghamkhar2 and pest detection. Plant Methods 15, 92–11. https://doi.org/10.
1
The Alliance of Bioversity International and International Center for Tropical 1186/s13007-019-0475-z.
Agriculture (CIAT), Km 17 Recta Cali-Palmira, Apartado Aereo 6713, Cali
763537, Colombia Tadesse, W., Sanchez-Garcia, M., Assefa, S.G., Amri, A., Bishaw, Z.,
2
AgResearch, Grasslands Research Centre, Palmerston North, New Zealand Ogbonnaya, F.C., and Baum, M. (2019). Genetic gains in wheat
*Correspondence: Ezhilmathi Angela Joseph Fernando (a.fernando@ breeding and its role in feeding the world. Crop Breed. Genet.
cgiar.org) Genom 1, e190005. https://doi.org/10.20900/cbgg20190005.
https://doi.org/10.1016/j.molp.2023.05.011
Van den houwe, I., Chase, R., Sardos, J., Ruas, M., Kempenaers, E.,
REFERENCES Guignon, V., Massart, S., Carpentier, S., Panis, B., Rouard, M.,
Anglin, N.L., Amri, A., Kehel, Z., and Ellis, D. (2018). A case of need: et al. (2020). Safeguarding and using global banana diversity: a
linking traits to genebank accessions. Biopreserv. Biobanking holistic approach. CABI Agric. Biosci. 1:15. https://doi.org/10.1186/
16:337–349. https://doi.org/10.1089/bio.2018.0033. s43170-020-00015-6.

Ashourloo, D., Mobasheri, M.R., and Huete, A. (2014). Developing two Yang, W., Feng, H., Zhang, X., Zhang, J., Doonan, J.H., Batchelor,
spectral disease indices for detection of wheat leaf rust W.D., Xiong, L., and Yan, J. (2020). Crop phenomics and high-
(Pucciniatriticina). Rem. Sens. 6:4723–4740. throughput phenotyping: past decades, current challenges, and
de Carvalho, R.R.B., Marmolejo Cortes, D.F., Bandeira e Sousa, M., future perspectives. Mol. Plant 13:187–214. https://doi.org/10.1016/j.
de Oliveira, L.A., and de Oliveira, E.J. (2022). Image-based molp.2020.01.008.
phenotyping of cassava roots for diversity studies and carotenoids Yao, H., Hruska, Z., Kincaid, R., Brown, R.L., Bhatnagar, D., and
prediction. PLoS One 17, e0263326. https://doi.org/10.1371/journal. Cleveland, T.E. (2013). Detecting maize inoculated with toxigenic and
pone.0263326. atoxigenic fungal strains with fluorescence hyperspectral imagery.
Ghamkhar, K., Irie, K., Hagedorn, M., Hsiao, J., Fourie, J., Gebbie, S., Biosyst. Eng. 115:125–135. https://doi.org/10.1016/j.biosystemseng.
Hoyos-Villegas, V., George, R., Stewart, A., Inch, C., et al. (2019). 2013.03.006.

Molecular Plant 16, 1099–1101, July 3 2023 ª 2023 The Author. 1101