ANA CAROLINA PICININI PETRONILIO 

 

 

 

 

INNOVATIVE TOOLS TO PREDICT AND IMPROVE SEED PHYSIOLOGICAL 

QUALITY: CASE STUDIES IN TOMATO AND SOYBEAN 

 

 

 

 

 

 

 

 

 

 

 

 

Botucatu 

2024 



  



ANA CAROLINA PICININI PETRONILIO 

INNOVATIVE TOOLS TO PREDICT AND IMPROVE SEED PHYSIOLOGICAL 

QUALITY: CASE STUDIES IN TOMATO AND SOYBEAN 

Thesis presented to the São Paulo State 
University (UNESP), School of Agricultural 
Sciences, Botucatu to obtain the title of 
Doctor of Agronomy – Agriculture. 

Advisor: Edvaldo Aparecido Amaral da 
Silva 

Botucatu 

2024 



 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 





  



 
 

 

 

 

 

 

 

 

 

 

 

 

 

Aos meus amados pais, 

 Sandra e Rosival,  

dedico. 

 

 



 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 



 
 

ACKNOWLEDGMENTS 

To God for being my daily support during this journey. To my parents for all their 

support and for believing in my dreams even though I need to be hundreds or 

thousands of kilometers away to make them come true. To my family, especially my 

uncles and great encouragers, Ronaldo Picinini, Ricardo Picinini, José Naief Tayer, 

and Márcio Jampani. 

During these five years of graduate school, I was lucky enough to meet many incredible 

people, have previously unimaginable experiences, and have dreams come true. 

Therefore, I thank each person who contributed to my personal and professional 

development and the completion of this work. 

To Prof. Dr. Edvaldo Ap. Amaral da Silva, for his guidance, encouragement, and 

teachings. For teaching me the importance of always seeking the practical applicability 

of our research and for introducing me to molecular biology. Besides being a mentor, 

he was also a friend many times. 

To Prof. Dr. João Nakagawa for his teachings and for being a professional inspiration 

who contributed so much to the seed sector. 

To Thiago Batista and Gustavo Fonseca for their friendship and teachings. For 

encouraging me always to strive to be better, and for the work developed in partnership 

that generated the results of this thesis. 

To Dennis Souza for all the friendship and fun hours of work shared in the laboratory. 

More than a friend, Dennis became a brother. 

Isabela Lopes for all her help in conducting the experiments. 

To my friends and colleagues at the Seed Physiology and Molecular Biology 

Laboratory: Samara Perissato, Carolina Cardoso, Larissa Chamma, Renata Arruda, 

Brunna Rithielly, Taíse Rodrigues, Vitor Gonçalves, Yago Triboni, Rômulo Lima, 

Mónica Córdoba, Michelane Lima, Felipe Carvalho and Tamires Freitas, for scientific 

learning, help with experiments and emotional support throughout these years. Having 

you made the exhausting graduation routine lighter and more enjoyable. 

To Ruan Marchi for his friendship, all the laughter in the department corridors, and all 

his teachings. 

To Lennis Afraire, Maria Gabriela Andrade and Aline Oliveira for their friendship. 

To Dr. Clíssia Mastrangelo and Cena/USP for the partnership that culminated in the 

results of this thesis. 



To Dr. Fernando Henning and Dr. Francisco Krzyzanowski for their guidance during 

my undergraduate internship, teaching me a passion for research, and encouraging 

me to pursue a graduate degree. 

To Dr. Monica Schmidt for welcoming me as a visiting student, and for all the teachings 

given. 

To my colleagues at Schmidt's lab: Mariam Marand, Cyrus Mou, Alexis Grace, Huang 

Yicheng, and Greg Clark, for welcoming me. Thank you for all your help and friendship 

throughout these 12 months in Tucson. 

To my Brazilian friends and colleagues whom I met in Arizona, Andriele Muller, Bianca 

Ferreira, Ramon Domingues, Felipe Dorigão, Diego Costa, Sara Paim, Caíque 

Nonato, George Barbosa and Vanessa Mendonça. 

To Prof. Dr. Carlos Alexandre Costa Crusciol and for all the encouragement throughout 

this journey. 

To FCA employees Eliane, Iara, Adelina, and Talita. In particular, to Valéria Giandoni, 

you became more than just a laboratory technician; you became a friend. Thank you 

for all the care and affection. I will never forget the endless help with experiments and 

all the delicious cakes and lunches that I will never forget. 

To the examining board members for their availability and for contributing to this 

important step in my life. 

To the Plant Production Department FCA UNESP Botucatu for the structure and for 

being a pleasant working environment. 

To The University of Arizona for providing the structure where I carried out part of my 

thesis. 

To the Coordination for the Improvement of Higher Education Personnel – Brazil – 

CAPES for granting a scholarship. Processes: DS 88887.670559/2022-00; 

CAPES/PRINT 88887.685215/2022-00; DS 88887.924553/2023-00. 

 

 

 

 

 

 



 
 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

“1Filho meu, se aceitares as minhas palavras e esconderes contigo os meus 

mandamentos, 2para fazeres atento à sabedoria o teu ouvido e para inclinares o 

coração ao entendimento, 3e, se clamares por inteligência, e por entendimento alçares 

a voz, 4se buscares a sabedoria como a prata e como a tesouros escondidos a 

procurares, 5então, entenderás o temor do Senhor e acharás o conhecimento de 

Deus. 6Porque o Senhor dá a sabedoria, e da sua boca vem a inteligência e o 

entendimento." (Provérbios 2:1-6). 

BÍBLIA. Bíblia Sagrada. Traduzida por João Ferreira de 

Almeida. Barueri, SP: Sociedade Bíblica do Brasil, 1999.   



 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 



 
 

ABSTRACT 

The seed quality is made up of physical, sanitary, genetic, and physiological attributes. 

Seed physiological quality refers to factors that allow rapid and uniform germination 

and emergence under broad climatic conditions, which ensures vigorous initial plant 

establishment and produces an increased yield. This research sought to elucidate how 

the physiological quality of seeds can be predicted and improved. To this end, this 

research used soybean and tomato species as models. 1 – the physiological quality of 

soybean seeds produced under water and heat stress during the maturation phase 

was determined, and seed multispectral images were acquired. Machine learning 

models associated with multispectral imaging technology were developed to verify the 

possibility of separating seed groups autonomously. The quality of stressed seeds was 

lower compared to non-stressed ones. The multispectral images produced markers 

that made it possible the segmentation between seed groups. The ML models showed 

high performance in recognizing stressed seeds. This work opens up the possibility of 

using this technology as a results manager, as it provides information about the 

environment of seed production and its consequences on the physiological quality of 

the seeds. 2 – we performed soybean genetic transformations for further study of the 

molecular factors involved in seed chlorophyll degrading/retention. For this, the 

GmABI5 gene was overexpressed and silenced (RNAi) using the glycinin promoter in 

a hygromycin-resistant expression cassette. Somatic embryos were induced from the 

cotyledons of immature seeds and when they reached the globular stage they were 

transformed via bioballistics. Transformed embryos were selected for resistance to 

hygromycin. The presence of the expression cassette in the resistant embryos was 

confirmed via PCR and they were regenerated into seedlings. This work will allow the 

functional study of the ABI5 gene and also the investigation into the possibility of 

reproducing a genotype tolerant to chlorophyll retention in soybean seeds. 3  –  we 

performed gene expression studies in tomato seeds during osmo-priming to 

investigate mechanisms involved in the reduction of longevity. For this, we collected a 

total of seven samples: four samples during the 60 hours of priming, one control group 

(without priming), one sample after post-priming drying, and one after post-priming 

thermal shock and drying. We investigated transcripts associated with stress response. 

The seed vigor and longevity after treatment were also determined. Primed seeds had 

their longevity impaired compared to unprimed ones. Genes from the heat-shock 



protein family were down-regulated during the priming process. This research brings 

new insights into the mechanisms involved in the reduction of longevity of primed 

seeds and allows the use of these transcripts to monitor longevity in primed tomato 

seeds.  

 

Keywords: seed quality; multispectral imaging; environmental stress; chlorophyll 

fluorescence; primed seeds. 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 



 
 

RESUMO 

A qualidade de sementes é composta por atributos físicos, sanitários, genéticos e 

fisiológicos. A qualidade fisiológica de sementes refere-se a fatores que permitem que 

a germinação e a emergência ocorram de forma rápida e uniforme sob amplas 

condições climáticas, o que garante um estabelecimento inicial vigoroso das plantas 

e resulta em aumento de produtividade. Esta pesquisa buscou investigar como a 

qualidade fisiológica das sementes pode ser estimada e aprimorada. Para tanto, esta 

pesquisa teve como modelo as espécies de soja e tomate. 1 – foi determinada a 

qualidade fisiológica de sementes de soja produzidas sob estresse hídrico e térmico 

durante a fase de maturação e adquiridas imagens multiespectrais das sementes. 

Modelos de aprendizado de máquina associados à tecnologia de imagem 

multiespectral foram desenvolvidos para verificar a possibilidade de separar os grupos 

de sementes de forma autônoma. A qualidade das sementes estressadas foi inferior 

em comparação às não estressadas. As imagens multiespectrais produziram 

marcadores que possibilitaram a segmentação dos grupos de sementes. Os modelos 

de aprendizado de máquina apresentaram alto desempenho no reconhecimento de 

sementes estressadas. Este trabalho possibilita a utilização desta tecnologia como 

gestora de resultados, pois fornece informações sobre o ambiente de produção de 

sementes e suas consequências na qualidade fisiológica das sementes. 2 – Foram 

realizadas transformações genéticas em soja para futuramente investigar os fatores 

moleculares envolvidos na degradação/retenção da clorofila nas sementes. Para isso, 

o gene GmABI5 foi superexpresso e silenciado (RNAi) utilizando o promotor glicinina 

em um cassete de expressão resistente à higromicina. Embriões somáticos foram 

induzidos a partir de cotilédones de sementes imaturas e quando os mesmos 

atingiram o estágio globular foram transformados via biobalística. Os embriões 

transformados foram selecionados quanto à resistência à higromicina. A presença do 

cassete de expressão nos embriões resistentes foi confirmada via PCR e eles foram 

regenerados em plântulas. Este trabalho permitirá o estudo funcional do gene ABI5 e 

a investigação da possibilidade de reproduzir um genótipo tolerante à retenção de 

clorofila em sementes de soja. 3 – foram realizados estudos de expressão gênica em 

sementes de tomate durante o condicionamento osmótico (osmo-priming) para 

investigar mecanismos envolvidos na redução da longevidade. Para isso, foram 

coletados um total de sete amostras: quatro amostras durante as 60 horas de priming, 



uma amostra do grupo controle (sem priming), uma amostra após secagem pós-

priming e uma após choque térmico pós-priming e secagem. Foi investigada a 

expressão gênica de moléculas associadas à resposta ao estresse. O vigor e a 

longevidade das sementes após o tratamento também foram determinados. As 

sementes condicionadas tiveram sua longevidade reduzida em comparação às não 

condicionadas. Os genes da família das proteínas de choque térmico foram regulados 

negativamente durante o processo de priming. Esta pesquisa traz novas percepções 

sobre os mecanismos envolvidos na redução da longevidade de sementes 

condicionadas e permite o uso destes genes para monitorar a longevidade em 

sementes de tomate submetidas ao priming. 

 

Palavras-chave: qualidade de sementes; imagem multiespectral; estresse ambiental; 

fluorescência da clorofila; sementes condicionadas. 

 

 

 

 

 

 

 

 

 

 

 

 

  



 
 

SUMMARY 

 

GENERAL INTRODUCTION ........................................................................ 17 

CHAPTER 1 - MAPPING FEATURES TO SEGMENT SOYBEAN 

STRESSED SEEDS USING MACHINE LEARNING MODELS BASED ON 

MULTISPECTRAL IMAGING: AN ACCURATE TOOL TO SOLVE CLASSIC 

PROBLEMS ................................................................................................. 21 

1.1 INTRODUCTION .......................................................................................... 21 

1.2 MATERIAL AND METHODS ........................................................................ 23 

1.3 RESULTS ..................................................................................................... 27 

1.4 DISCUSSION ............................................................................................... 35 

1.5 CONCLUSION .............................................................................................. 39 

REFERENCES ............................................................................................. 40 

CHAPTER 2 - SILENCING AND OVEREXPRESSION OF GMABI5 GENE IN 

SOYBEAN GREENISH SEED STUDIES ..................................................... 44 

2.1 INTRODUCTION .......................................................................................... 44 

2.2 MATERIALS AND METHODS ...................................................................... 46 

2.3 RESULTS AND DISCUSSION ..................................................................... 54 

2.4 CONCLUSION .............................................................................................. 57 

REFERENCES ............................................................................................. 58 

CHAPTER 3 - OSMO-PRIMING IN TOMATO SEEDS DOWN-REGULATES 

GENES ASSOCIATED WITH STRESS RESPONSE AND LEADS TO 

REDUCTION IN LONGEVITY ...................................................................... 61 

3.1 INTRODUCTION .......................................................................................... 61 

3.2 MATERIAL AND METHODS ........................................................................ 63 

3.3 RESULTS ..................................................................................................... 66 

3.4 DISCUSSION ............................................................................................... 69 

REFERENCES ............................................................................................. 73 

FINAL CONSIDERATIONS .......................................................................... 75 

REFERENCES ............................................................................................. 77 

 

 

 

 

 

 

 

 

 



 

 



17 
 

GENERAL INTRODUCTION 

In agriculture, seeds are one of the main inputs (Carvalho; Nakagawa, 2012). A 

new generation of plants begins through seeds, assuring plant life and food security 

worldwide. Much attention is given to producing high-quality seeds because it supports 

crop production. Seed quality consists of genetic, physical, sanitary, and physiological 

attributes. Physiological quality refers to viability, desiccation tolerance, vigor, and 

longevity, which ensure proper seedling establishment (Bewley et al., 2013).  

Seed physiological quality can impact crop yield by affecting the speed and 

uniformity of germination and emergence and plant density (Tekrony et al., 1991). Low-

quality seed lots result in a higher rate of abnormal seedlings, reducing population 

density (considering density recommended limits) and impacting productivity. On the 

other hand, high-quality seeds produce seedlings that emerge earlier and uniformly 

during a suitable sowing time. Seedling emergence during the suitable sowing time 

ensures that environmental conditions will be more favorable for growth than if the 

seedling emerges later, when the climate conditions might be less advantageous. Fast 

and uniform emergence also avoids competition between crop plants and weeds 

(Finch-Savage, 1995). Seed physiology studies have contributed to technology 

development and crop yield increase. High-quality seeds improved crop yield by more 

than 4% in corn production (Graven; Carter, 1991). According to Rajala et al. (2011), 

barley seeds with superior quality improved grain yield.   High-quality seeds improved 

rice yield up to 19% in the Philippines (Diaz et al., 1998). The progressive increase in 

the level of vigor in soybean seeds led to an increase of up to 28 kg.ha-1 in grain 

productivity (Bagateli et al., 2019). In this regard, molecular-level seed physiology 

research, seed priming, and imaging techniques are ways to predict and improve seed 

quality and will be addressed in this research. 

In recent years, imaging analysis has been proposed as an alternative to 

traditional seed tests to assess seed quality. The traditional tests depend mainly on a 

visual inspection by the analyst, which can lead to a subjective interpretation of the 

results. Furthermore, these methods are destructive and mostly time-consuming. 

Therefore, there is a need to advance the use of more assertive tools, which promote 

time savings and eliminate subjectivity, and which additionally allow the integration of 

machine learning and process automation. Several imaging technologies are available 

nowadays, such as X-ray, X-ray microtomography, X-ray fluorescence, near-infrared 



18 
 

spectroscopy, and magnetic resonance imaging that can assess seeds (Bianchini et 

al., 2021; Cotrim et al., 2019; Gomes-Junior et al., 2019) and multispectral imaging 

that can assess both seeds and seedlings (Elmasry et al., 2019; Galletti et al., 2020).  

X-rays were used to detect mechanical damage in several species, such as beans 

(Gomes-Junior et al., 2019; Mondo et al., 2009), and were associated with seed 

physiological quality. Autofluorescence multispectral images associated with artificial 

intelligence tools have been used to recognize soybean maturation stages and predict 

seed quality (Barboza Da Silva et al., 2021; Batista et al., 2022). Multispectral imaging 

and machine learning methods can also be used to provide information regarding 

pigments. Chl fluorescence and anthocyanin in peanut seedlings were reported as 

efficient in segregating seeds with superior quality (Fonseca De Oliveira et al., 2022).   

Seed physiological quality is a multigenic trait and is largely influenced by the 

environment (Tripathi; Khare, 2016). Over the years, in plant breeding programs, much 

attention has been given to improving grain quality traits such as oil and protein quality 

and yield (Rajcan; Hou; Weir, 2005), grain yield, and insect and disease resistance 

(Krzyzanowski, 1998). However, little attention is given to associating the selection of 

the aforementioned traits with seed physiological quality traits (Kalaji; Pietkiewicz, 

2004). There is a need to further research seed physiological quality since it is essential 

to enable the expression of the productive potential of cultivars. Some researchers 

have studied the genetic gains regarding seed physiological quality in breeding 

research. In soybean, Monteiro et al. (2021) showed that the heritability of 

physiological quality traits was moderately high, which confirms the possibility of 

breeding to improve seed physiological quality traits. Mello Filho et al. (2004) 

highlighted the possibility of breeding to improve protein quality while keeping high-

quality seeds and satisfactory yield.  Hatzig et al. (2018) suggest the use of molecular 

markers of mean germination time for selection of germination capacity in oilseed rape 

cultivars.  

Molecular-level seed physiology studies aim to investigate the molecular factors 

involved in the seed physiological quality attributes. This approach has been applied 

in the soybean “greenish-seed problem” research. Soybean greenish seeds are 

mature seeds with chlorophyll (chl) retention. This usually happens in response to 

abiotic stresses during the soybean maturation phase. Chl retention has been a 

significant problem in these last decades in Brazil because it impacts seed and oil 

quality (Pádua et al., 2007; Luccas, 2018).  Besides the environmental circumstances, 



19 
 

genetic factors are also involved in the susceptibility of chl retention in soybean seeds 

(Ajala-Luccas et al., 2023; Pádua et al., 2009; Teixeira et al., 2016). Teixeira et al. 

(2016) suggested that in susceptible genotypes, the impaired expression of genes 

related to photosynthesis (D1, D2. PPH2, NYC1) might be involved in chlorophyll 

retention. Batista (2022) showed that the acquisition of longevity is impacted in 

stressed seeds with higher chl content. In the same study, the expression of genes 

related to chl degradation (D2, NYC1) is impacted in immature stressed seeds. 

Moreover, seeds that suffered stresses during the maturation phase had a higher 

expression of the HSP21 gene, which is a protection mechanism against stresses, 

compared to non-stressed seeds (Batista, 2022). Another approach for molecular-level 

seed physiology studies was presented by Ducatti et al. (2022) who reported 

transcripts associated with vigor in soybean seeds.  

Thus, the above-mentioned transcripts can be used to monitor the quality 

attributes and provide knowledge that can be applied in plant breeding programs 

aiming at marker-assisted selection of superior genetic materials for seed quality. In 

addition, the genes discovered through molecular-level seed physiology studies and 

plant breeding might also be applied in genetic transformation and gene editing, which 

are additional tools that might improve seed quality. Genetic transformation and gene 

editing can be applied in gene functional studies to understand the behavior of a 

candidate gene, and furthermore, it can also provide genetically modified varieties for 

high-quality seed production.  

Seed priming is a technology used to increase the speed and uniformity of 

germination and improve seedling emergence and stress tolerance through controlled 

seed hydration. The germination process starts during the treatment, but radicle 

protrusion is impeded. Thus, seeds are dried after a determined priming time. 

Therefore, germination and seedling emergence are faster when the seeds are sown 

because the metabolism was previously activated during priming treatment. Priming is 

particularly helpful in seeds with slow and uneven germination (Marcos-Filho, 2015). 

Besides the increase in seed physiological quality by priming (Brocklehurst; Dearman, 

1983; Forti et al., 2020; Silveira et al., 2023), studies show crop yield improvement. 

Salicylic acid priming promoted a fruit yield increase of 33% and improved plant 

resilience to water stress in tomatoes and also improved yield and high-temperature 

germination tolerance in carrots (Chakma et al., 2021; Mahmood-Ur-Rehman et al., 

2020). In wheat seeds, osmo-priming increased grain productivity by 1.10 tons per 



20 
 

hectare (Farooq et al., 2020). Hussain and collaborators (2006) obtained increases in 

the productivity of sunflower achenes through NaCl and KNO3 osmo-priming. 

Thus, studies of imaging analysis, molecular-physiology level studies, and seed 

priming, are tools that can predict and improve seed quality (Paparella et al., 2015; 

Fonseca De Oliveira et al., 2022; Ducatti, et al., 2022). Therefore, this research aims 

to study how seed physiological quality can be predicted and improved through 

innovative tools using two economically important crops as models, tomato, and 

soybean. 

 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

 

 

 

 



21 
 

CHAPTER 1 

 

MAPPING FEATURES TO SEGMENT SOYBEAN STRESSED SEEDS USING 

MACHINE LEARNING MODELS BASED ON MULTISPECTRAL IMAGING: AN 

ACCURATE TOOL TO SOLVE CLASSIC PROBLEMS 

 
ABSTRACT 
Extreme environmental conditions have been recurrent during the last few years and 

have impacted crop seed quality, mainly but not limited to, soybeans (Glycine max (L) 

Merrill). To overcome this, seed companies often demand innovative tools to address 

seed quality factors. Machine learning models based on multispectral imaging are a 

novel seed quality analysis approach. Thus, we hypothesize that segmenting stressed 

and non-stressed soybean seeds would be possible with this technology, opening a 

new opportunity for seed quality management and elucidating quality factors. Soybean 

seeds (cultivar BR/MG 46-Conquista) were produced under water deficit and heat 

during maturation (from R5.5 onwards). Multispectral images were acquired from 

stressed and non-stressed seeds, and the reflectance, autofluorescence, physical 

properties, and chlorophyll parameters were extracted from the images. In parallel, we 

determined seed vigor. We designed machine learning models using multispectral 

imaging data based on three algorithms: neural network, support vector machine, and 

random forest. Our results demonstrated that the stressed seeds have spectral 

markers that enable their recognition. Concomitantly, these markers had a direct 

relationship with seed vigor. The machine learning models developed based on neural 

network algorithm showed the highest performance in segmenting stressed seeds 

(90% accuracy). Here, we report a new approach for multispectral imaging with the 

potential to identify soybean seeds of lower vigor as a result of unfavorable 

environmental conditions during seed maturation. 

1.1 INTRODUCTION 

Soybean is a significant source of protein and oil and is utilized in several industrial 

products (Contini et al., 2013). Due to the many applications of this crop, a strong 

market has developed worldwide (USDA, 2023). Many inputs are required to boost the 

soybean chain, and one of these inputs is seeds. Seeds are responsible for 

establishing a new set of plants in the field. High-quality seeds are primordial to 

providing uniform plant development that favors maximum crop productivity 



22 
 

(Krzyzanowski et al., 2018). To ensure high-quality seeds, seed companies strive and 

constantly seek innovations to overcome seed quality challenges.  

One of the potential new approaches to be used in seed quality management is 

multispectral imaging associated with artificial intelligence. Multispectral imaging 

consists of reflectance or autofluorescence images obtained through the emission of 

several light wavelengths (Galletti et al., 2020). This technique can provide optical 

markers regarding seed chemical composition, pigments, and physical traits such as 

texture, area, length, width, color, and coat brightness (Bianchini et al., 2021; França-

Silva et al., 2023). Intact seeds in the samples provide precise data since the image 

pixels are transformed into numbers through an algorithm, and consequently eliminate 

the subjectivity of human interpretation, typical of traditional seed quality tests 

(germination, accelerated aging, tetrazolium, etc.) (Batista et al., 2022; Galletti et al., 

2020). The data generated combined with machine learning (ML) models allow the use 

of this technique in seed processing to evaluate their quality autonomously (Fonseca 

De Oliveira et al., 2022). 

Multispectral imaging has recently been the object of study of several crop seed 

quality management programs. For example, Barboza da Silva et al. (2021b) showed 

that autofluorescence signals from the 365/400 nm wavelength associated with ML 

models could identify artificially aged soybean seeds. Chlorophyll (Chl) fluorescence, 

anthocyanin, 660, 690, and 780 nm light reflectance, and physical attributes associated 

with ML models were used to select high-quality peanut seeds (Fonseca de Oliveira et 

al., 2022). Autofluorescence-spectral imaging related to Chl was used to segment 

soybean seeds in different maturation stages (Batista et al., 2022). This technique was 

also used to identify viability, vigor, and hard seeds from Medicago sativa (Wang et al., 

2023; Zhang et al., 2022). Texture, color, and 490 nm reflectance associated with 

artificial intelligence tools were used to discriminate pathogenic fungi in peanut seeds 

(Sudki et al., 2023).  

The soybean seed market deals with several aspects affecting seed quality. In 

recent decades, reports show that soybean seeds have germination, vigor, and 

longevity impacted due to abiotic stress (Ajala-Luccas et al., 2023; Batista et al., 2022; 

Teixeira et al., 2016). One of the reported stresses that impact seeds is a combination 

of water deficit and heat (Pádua et al., 2009; Teixeira et al., 2016), and the occurrence 

of this stress during the plant maturation phase was outlined as being more critical 

than during the vegetative or embryogenic phases (Cohen et al., 2021). Environmental 



23 
 

stress negatively impacts seed physiological quality (Ajala-Luccas et al., 2023; Zorato 

et al., 2007) and is one of the challenges seed companies must overcome to ensure a 

high-quality product. Therefore, a tool that enables seed companies to know the 

causes of seed quality issues quickly and non-destructively would be helpful for seed 

quality management. While stress affects seed quality through physical and 

physiological changes, it could be possible to map this event beyond what the human 

eye can see. Thus, it would be possible to promptly identify reductions in seed quality 

associated with environmental stress. 

 Thus, we hypothesize that ML models based on multispectral imaging can be 

applied in the segmentation of stressed and non-stressed seeds as a tool for 

companies to keep track of the seed quality factors. To confirm this, we investigated 

multispectral imaging features and ML models that could effectively recognize abiotic 

stressed soybean seeds.  

1.2 MATERIAL AND METHODS 

Plant material  

Soybean seeds from a cultivar susceptible to abiotic stresses (BR/MG 46-

Conquista) (Pádua et al., 2009) were propagated in a greenhouse at the School of 

Agriculture, Department of Crop Sciences, UNESP/Botucatu/São Paulo/Brazil 

(Altitude 810m). Supplementary lighting was provided through LED full spectrum lamps 

(28W), with photosynthetically active radiation emission of 150 μmols m−2 s-1. The 

plants were cultivated in 180 pots with 11 L of soil with texture of 778 g kg-1 of sand. 

The soil acidity and fertilization were adjusted according to the crop needs. 

The mean temperature during crop development was 28 ± 1 °C during the day and 

21 °C ± 2 at night. The humidity of the pots was kept close to the field capacity (12 

kPa), which had been previously determined. The humidity was controlled twice a day 

(10:00 am and 4:00 pm). It was measured with a humidity sensor positioned at a depth 

of 10 cm. If there was a difference between the reading and the field capacity humidity, 

the corresponding water volume was replaced to return the soil to the desired water 

potential (12 kPa). 

 The sample groups were composed of different environmental conditions during 

the seed maturation phase: non-stressful (control) and stressful conditions. From the 

R5.5 stage (Ritchie, Hanway, Thompson, 1982) onwards, 50% of the pots were 

randomly assigned to stressful environmental conditions (stressed group), and the 



24 
 

other 50% were kept in the conditions above and constituted the control group (non-

stressed group). The environmental stress condition constituted a daytime average of 

33 ± 1 ºC and nighttime 25 ± 0.4 ºC. Water supplementation was kept to a minimum 

so as to maintain a water potential of 27 kPa (water deficit) until the R9 stage (full 

maturation). The R5.5 stage approximately coincides with the end of the seed 

maturation phase (end of grain filling and disconnection from the mother plant) and 

with the beginning of the late maturation, an important phase for the seed physiological 

quality acquisition, mainly in terms of desiccation tolerance, vigor, and longevity (Lima 

et al., 2017). 

On the seventh day, after the plants were subjected to stressful conditions, the 

third wholly expanded leaf from the top of the plants was collected to characterize their 

oxidative stress. The oxidative stress was determined with four replicates per treatment 

(stressed and non-stressed), constituting one leaf per replicate. The samples were 

immediately frozen in liquid nitrogen and stored in an ultra-freezer (-80 ºC) until the 

evaluation.  

For hydrogen peroxide quantification (H2O2), 250 mg of leaf tissue was ground 

and homogenized in 3 mL of 0.1% (m/v) trichloroacetic acid (TCA) solution with 20% 

PVPP (m/m). The solution was centrifuged at 14,000 g at 4 ºC for 20 minutes (Hettich, 

Universal 320R, Tuttlingen, Germany). A 0.2 mL aliquot of the supernatant was added 

to the reaction of 0.2 mL of 100 mM potassium phosphate buffer solution (pH 7.5) and 

0.8 mL of potassium iodide (KI) solution 1 M. The solution was incubated for one hour 

on ice in the dark. Then, the absorbance reading was performed on a 

spectrophotometer (Shimadzu, UV-2700, Kyoto, Japan) at 390 nm. The H2O2 

concentration was calculated using the H2O2 standard curve at 1,000 µmol mL-1, and 

the results were expressed in mmol g-1 MF (Alexieva et al., 2001). 

For lipid peroxidation (malondialdehyde – MDA), 250 mg of leaf tissue were ground 

in liquid nitrogen, homogenized in 3 mL of 0.1% (m/v) trichloroacetic acid (TCA) 

solution with 20% PVPP (m/m), and centrifuged at 14,000 g at 4 ºC, for 20 minutes 

(Hettich, Universal 320R, Tuttlingen, Germany). A 0.5 ml aliquot of the supernatant 

was added to 1.5 ml of 0.5% (w/v) 2-thiobarbituric acid (TBA) solution prepared in 20% 

TCA (w/v) and incubated at 90ºC for 20 minutes. The reaction was stopped by putting 

the tube in an ice bath for 10 minutes. Afterward, the samples were centrifuged at 

10,000 g at 4 ºC for 5 minutes. Absorbances were determined using a 

spectrophotometer (Shimadzu, UV-2700, Kyoto, Japan) at 532 and 660 nm. The 



25 
 

malondialdehyde acid (MDA) concentration was calculated using the molar extinction 

coefficient of 155 mM-1 cm-1 and expressed as MDA nmol g-1 MF (Cakmak et al., 

1991).  

The seeds were harvested at the R9 stage (Ritchie et al., 1982), and unformed 

seeds were discarded. The seeds were stored in a cold chamber at 12 ºC/55% RH for 

one week, and then the tests were carried out. 

Seed physiological quality assays  

Before carrying out the seed quality tests, the water content was determined by 

the oven method (ISTA, 2020). The water content of all treatments was balanced to 

0.15 g H2O/g DW-1 by pre-conditioning the seeds on a stainless-steel screen 

suspended inside a plastic box (11.0 x 11.0 x 3.5 cm) containing 40 mL of deionized 

water at 25 ºC for 16 hours.  

Four replicates of 100 seeds were weighed on an analytical scale with a precision 

of 0.001 g to determine the weight of 100 seeds. To evaluate the seedling emergence 

and uniformity index, four replicates of 25 seeds were sown at a depth of 5 cm in 200-

cell trays with a commercial substrate composed of sphagnum peat, expanded perlite, 

expanded vermiculite, and roasted rice hulls. The substrate was moistened with 60% 

of its water-holding capacity. The seedlings were cultivated under an average 

temperature of 25 ºC ± 0.8. Emerged seedlings were counted daily at the same time 

until the number of emerged seedlings stabilized (Krzyzanowski et al., 2020). For 

seedling emergence, data were expressed as a percentage of emerged seedlings. The 

uniformity index was expressed through the sum of the percentage of seedlings that 

emerged on the emergence peak day plus the day before and after the emergence 

peak, according to Ebone et al. (2020).  For the accelerated aging test (AA), four 

replicates of 25 seeds were accommodated on a stainless-steel screen inside a plastic 

box (11.0 x 11.0 x 3.5 cm) containing 40 mL of deionized water. The boxes were sealed 

to avoid evaporation and put in an aging chamber at 41 ºC for 72 h. Afterward, a 

germination test was performed on paper towels moistened with a water quantity 

equivalent to 2.5 times the paper’s weight. The paper rolls were kept at 25 ºC for five 

days in the dark when the normal seedlings were counted.   

 

 



26 
 

Multispectral imaging data acquisition 

Multispectral images of 250 seeds per treatment were acquired. The seeds were 

placed on acetate sheets (5 cm x 8.5 cm) with double-sided tape. On each sheet, 50 

seeds were accommodated in the same position (hilum facing left), and the images 

were captured using a VideometerLab4TM equipment (Videometer A/S, Herlev, 

Denmark), according to Batista et al. (2022). 

Reflectance images were acquired at 19 wavelengths – 365 (UV), 405 (violet), 430 

(indigo), 450 (blue), 470 (blue), 490 (cyan), 515 (green), 540 (green), 570 (yellow), 590 

(amber), 630 (red), 645 (red), 660 (red), 690 (dark red), 780 (dark red), 850, 880, 940, 

and 970 nm (the last four wavelengths are in the near-infrared region) (Fonseca De 

Oliveira et al., 2022). Autofluorescence images were acquired using 28 different 

excitation/emission combinations at wavelengths 365/400, 365/500, 405/500, 430/500, 

450/500, 470/500, 365/600, 405/600, 430/600, 450/600, 470/600, 490/600, 515/600, 

540/600, 570/600, 365/700, 405/700, 430/700, 450/700, 470/700, 490/700, 

515/700, 540/700, 570/700, 590/700, 630/700, 645/ 700, and 660/700 nm (Gomes et 

al., 2022). Before image acquisition, the light setting was adjusted to optimize the light 

intensity at each wavelength and allow all captured images to be compared. 

Representative samples were used for this calibration, which was later applied to all 

other samples. 

After image acquisition, data were extracted using VideometerLabTM software 

(version 3.14.9). We built a segmentation mask based on the first image to separate 

the seeds from the background. Thus, the pixels in the background had values equal 

to zero, while each pixel different from zero was considered part of an object (Barboza 

Da Silva et al., 2021a). In the segmented images, a normalized canonical discriminant 

analysis algorithm (nCDA) was applied. The nCDA algorithm reduces the outliers and 

was used to detect the signals per pixel in each combination of wavelength employed 

and thus obtain numerical resolution values. In addition to the reflectance and 

autofluorescence variables, data were determined for the seed physical 

characteristics: area, length, and width, and for color descriptors: CIELab L*, CIELab 

a*, CIELab b* (Oliveira et al., 2021). The color descriptors are a way to describe color 

features in a objective way, thus, CIELab L* corresponds to brightness features, 

CIELab a* indicates how close an object is to red or green and CIELab b* means how 

close an object is to yellow or blue (Melgosa, 2000).  



27 
 

With the same sample preparation, we obtained multispectral images (2448 x 2448 

pixels) using a SeedReporterTM (PhenoVation B.V., Wageningen, The Netherlands) 

instrument. The Chl a index images were acquired in the 710 and 770 nm reflectance 

(Gitelson et al., 2003) and the Chl a fluorescence was obtained by excitation at 620 

nm wavelength and detection at 730 nm (Galletti et al., 2020). The image acquisition 

and data extraction details were obtained using the SeedReporterTM software version 

5.5.1.  

Seed segmentation using machine learning models 

We designed ML models using the 20 most important variables of the multispectral 

data from each single seed (n=500 seeds). The models were based on neural network 

(NN) (solver: Stochastic Gradient Descent; hidden layer sizes: two layers with 25 

neurons in each; Activation function: Tanh; Learning rate: adaptive; the maximum 

number of interactions: 5000) (Barboza Da Silva et al., 2021b), support vector machine 

(SVM) and random forest (RF) algorithms (Batista et al., 2022), in which 70% of data 

(n = 350) were automatically separated to train the algorithms (K-fold = 5), and 30% (n 

= 150) were used for the external validation test.  

Statistical design  

The physiological (n=8) and multispectral (n=500) data were compared by t-test 

(P<0.05). Additionally, in the multispectral data, we performed (i) Gini coefficient to 

rank variables regarding their importance in seed group segmentation, (ii) principal 

component analysis (PCA) to visualize the spatial seed group segregation (stressed 

and non-stressed) concerning the variables using “ggbiplot” R package. Accuracy, F1, 

precision, recall, and specificity were used to test the performance of the ML models 

generated. All metrics were calculated using a confusion matrix. The Python version 

3.11 environment was used to generate the models. Pearson’s correlation coefficient 

was measured to investigate the relationship between the multispectral markers and 

the seed physiological quality traits using the “corrplot” R package.   

1.3 RESULTS  

Seed quality characterization  

The hydrogen peroxide (H2O2) and lipid peroxidation (MDA) content 

(SUPPLEMENTARY FIGURE 1A, B) increased due to the stressful conditions in 

soybean plants. Moreover, seed weight was significantly affected by environmental 



28 
 

stress. Stressed seeds had less mass than the non-stressed seeds, which was 19.62 

g and 23.63 g, respectively (FIGURE 1A). Therefore, these results confirmed the effect 

of plant stress in our research. In parallel, we performed seed physiological quality 

tests to show seed performance. Stressed seeds had their quality impaired in terms of 

seedling emergence and emergence uniformity. Stressed seeds had low seedling 

emergence performance in relation to non-stressed seeds (FIGURE 1B, C), with a 

substantial difference of 20% in the emergence uniformity, a robust index to measure 

disturbances in the seed-seedling transition directly related to seed vigor. However, 

there was no significant difference in the AA test between stressed and non-stressed 

seeds (FIGURE 1D).  

Supplementary figure 1 – Oxidative stress characterization in soybean plants 
submitted to stressful conditions during maturation. Hydrogen peroxide (H2O2) 
in non-stressed and stressed plants (A). Malondialdehyde acid (MDA) in non-

stressed and stressed plants (B). MDA is a product of lipid peroxidation. 
Different letters indicate a significant difference (P ≤ 0.01) by t-test. Error bars 

show the standard deviation from four samples 

  

 

 

 

 

 

 



29 
 

Figure 1 – Effect of environmental stress during maturation of soybean on seed 
weight and vigor. Seed weight (A), seedling emergence (B), emergence 

uniformity index (C), and accelerated aging (D) of non-stressed and stressed 
soybean seeds. Different letters indicate a significant difference (P ≤ 0.05) by t-

test. Error bars show the standard deviation from four samples 

 

Multispectral parameters can segment stressed seeds 

 We applied the Gini coefficient to the 55 multispectral variables (reflectance, 

autofluorescence, physical properties, Chl a fluorescence, and Chl a index) obtained 

from the stressed and non-stressed seeds to rank the 20 most important. We identified 

that they are, in order: Chl a fluorescence, seed width, Chl a index, seed area, 365 nm, 

540 nm, CIELab a*, 660 nm, 515 nm, 690 nm, CIELab b*, 470/500 nm, 645 nm, 880 



30 
 

nm, 570 nm, seed length, 630 nm, CIELab L*, 780 nm, 365/500 nm. The entire Gini-

coefficient report is shown in FIGURE 2. 

Figure 2 – Gini coefficient for the contribution of the multispectral variables 
in the segmentation of stressed and non-stressed soybean seeds (n=500 

seeds). Chlorophyll a fluorescence and the excitation/emission combination of 
405/500 nm were the variables that respectively contributed the most and the 

least to seed group segmentation 

 
 



31 
 

Using the 20 most important variables based on the Gini coefficient, we applied a 

PCA to explore seed segmentation (FIGURE 3). Component 1 (PC1) and component 

2 (PC2) explained 70.9% and 14.7% of the significant variation, respectively. PCA 

efficiently separated the two groups of seeds, in which stressed seeds were associated 

with Chl a index, Chl a fluorescence, CIELab a*, and 365 nm reflectance. Meanwhile, 

the other 16 most important variables were associated with non-stressed seeds.  

Figure 3 – Biplots of principal component analysis (PCA) for chlorophyll a 
index, chlorophyll a fluorescence, physical properties, reflectance, and 

autofluorescence spectral markers of stressed and non-stressed soybean 
seeds. The vectors indicate the correlation between the seed groups (stressed 

and non-stressed) and the dimensions PC1 and PC2 

 

Figure 4 shows the means of the five most important markers for segmenting 

stressed and non-stressed seeds based on the Gini coefficient. Stressed seeds had 

higher values of Chl a fluorescence (FIGURE 4A), Chl a index (FIGURE 4B), and 

reflectance in the 365 nm wavelength (FIGURE 4E) (P ≤ 0.05), as also shown in the 

images (Figure 5). However, they had smaller seed width (FIGURE 4C) and seed area 

(FIGURE 4D) compared to non-stressed seeds. 

 



32 
 

 

Figure 4 – The five most important multispectral variables to separate non-
stressed seeds from stressed soybean seeds based on the Gini coefficient. 

Different letters indicate a significant difference (P ≤ 0.05) by t-test. Error bars 
show the standard deviation from 250 samples 

 

 



33 
 

Figure 5 – Score pixels of non-stressed and stressed soybean seed images. 
RGB images (red, green and blue channels) and corresponding chlorophyll a 

fluorescence (Chl a fluorescence), chlorophyll a index (Chl a index), and 
reflectance images at 365 nm

 

Seed segmentation based on different machine learning models 

Using NN as an algorithm, we obtained accuracy, F1, precision, recall, and 

specificity ≥ 0.91, while RF and SVM presented ≤ 0.89 in the training of the algorithms. 

In the test, the NN algorithm showed metrics ≥ 0.89; on the other hand, RF and SVM 

showed metrics ≤ 0.82 (FIGURE 6). In addition, the confusion matrix showed that using 

NN algorithm in the multispectral data, the probability of success in the prediction is 

greater than 90%. The probability of success for stressed seeds identification was 

94%, demonstrating a high capacity for prediction through the parameters used to 

identify this class. 

 

 

 

 

 

 



34 
 

Figure 6 – Machine learning models and confusion matrices to discriminate 
stressed and non-stressed soybean seeds using multispectral data based on 

Random Forest (RF) (A), Support Vector Machine (SVM) (B), and Neural 
Network (NN) (C) algorithms. Models performance based on accuracy, F1, 

precision, recall, and specificity. Confusion matrices were created using the 
test dataset (n= 150 seeds) 

 

Correlation of the seed physiological traits with Chl a parameters, physical markers, 

reflectance, and autofluorescence 

The 20 most important multispectral markers were inserted in a correlation matrix 

with the physiological data to check the relationship between the variables (FIGURE 

7). The weight of 100 seeds was strongly correlated with Chl a index (-0.95). There 

was also a strong correlation coefficient between the weight of 100 seeds and physical 

properties, especially seed width (0.92). Furthermore, the weight of 100 seeds and 



35 
 

reflectance obtained the strongest correlation coefficient with 365 nm (-0.92) 

wavelength.  

There was a strong correlation coefficient between the seedling emergence and 

seed area (0.70). The uniformity index was negatively correlated with Chl a 

parameters: Chl a index (-0.88) and Chl a fluorescence (-0.77). The uniformity index 

was also strongly correlated with physical markers, in which the highest correlation 

coefficient was obtained with seed width (0.89). Finally, the uniformity index was mainly 

correlated with the wavelength of 365 nm (-0.83).  

Figure 7 – Pearson’s correlation coefficients between seed physiological 
quality, physical properties, reflectance, and autofluorescence-spectral 

markers (nm) in stressed and non-stressed soybean seeds 

 

1.4 DISCUSSION  

Seed companies require seed quality management tools. Multispectral imaging 

associated with ML algorithms is a prospective innovation in this field. We asked 



36 
 

ourselves if this imaging technology could identify seeds that went through a stressful 

production environment and then if ML algorithms could automatically segregate the 

stressed seeds. For this, stressed soybean seeds (seeds produced under stressful 

conditions of water deficit and heat) and non-stressed soybean seeds – our control 

group - were analyzed for physiological quality, and also using multispectral imaging 

tools. Here, we demonstrate for the first time that by using emerging technologies, it is 

possible to classify the soybean seeds' origin environment and, thus, infer the 

consequences on seed physiological quality.  

In croplands worldwide, sowing seeds with high physiological quality is essential, 

because when establishing in the field, it provides fast and uniform germination and 

emergence. Our data showed a reduction in the seed vigor for stressed seeds, as is 

also reported in the literature (Bagateli et al., 2019; Finch-Savage; Bassel, 2016), 

which causes a disturbance in the seed-seedling transition, reducing emergence 

uniformity (FIGURE 1C). In the fields, uneven seedling emergence produces 

dominated plants in the plant community, which although being able to recover their 

height in the late growth stages due to phenotypic plasticity, are less productive 

because of the reduction in the number of nodes and pods per plant (EBONE et al., 

2020). Reduced seedling emergence rate (FIGURE 1B) causes missing plants in the 

plant stand, directly reducing yield (Krzyzanowski et al., 2018). 

To test and determine seed vigor, seed companies often apply AA tests in their lab 

work routine (França-Neto; Krzyzanowski, 2022). In our study, the AA test (72 h 41 ºC) 

could not detect seed vigor difference between the two groups (stressed and non-

stressed) (FIGURE 1D). Thus, seeds with reduced vigor due to stressful conditions 

during the maturation phase might not be detected by traditional lab tests, such as AA. 

This problem is typical because the reduction in vigor is not as apparent as other 

factors that affect quality with a latent or immediate effect (such as mechanical 

damage). 

We achieved an innovative methodology to diagnose reduction in seed vigor 

caused by stressful environments using Chl a fluorescence, Chl a index, seed width, 

seed area, reflectance variables, etc. since the changes in these properties promoted 

by stress during seed maturation were detected using multispectral imaging (FIGURE 

2, 4, and 5). Stressed seeds presented high Chl a levels, as shown by a fluorescence 

and Chl a index (FIGURE 4A, 4B and 5). Stressful conditions during maturation 

promote Chl retention in soybean seeds (Pádua et al., 2007; Teixeira et al., 2016) that 



37 
 

impact the seed-seedling transition (Ajala-Luccas et al., 2023). Thus, Chl levels in 

stressed seeds were higher, which became a strong marker detected by multispectral 

images. This technique has high sensitivity and enables distinguishing groups based 

on the level of chlorophyll that does not degrade due to stressful conditions.  

In our study, the greenish color (high chlorophyll content) of the seeds was not 

visible to the human eye as shown in the RGB images (Figure 5). High chlorophyll 

levels impact legume seeds' lifespan (Nakajima et al., 2012; Zinsmeister et al., 2023), 

but the reduction in the physiological quality of soybean seeds is generally detected 

only at the time of the effective use of the seeds, i.e., after six months of storage waiting 

to be commercialized. Thus, our method based on multispectral imaging, once 

incorporated data related to chlorophyll fluorescence, can be used as a tool for 

predicting severe problems that seeds may face after sewing (i.e., loss of viability and 

consequent commercial standard loss).  

In seeds, several fluorescent chemical components can be detected by 

multispectral images at short wavelengths (Donaldson, 2020; García-Plazaola et al., 

2015). Fluorescent compounds, such as chlorophyll, can alter the color properties of 

seeds, especially those related to brightness (Fonseca De Oliveira et al., 2022; Oliveira 

et al., 2021; Sudki et al., 2023. parallel, these changes potentially influence the 

reflectance properties of seeds, as shown in our study at 365 nm (FIGURE 4 and 5). 

This also occurred in peanut (Fonseca De Oliveira et al., 2022), tomato and carrot 

(Galletti et al., 2020). Therefore, the environmental stress conditions imposed on 

soybean plants generated seeds with a detectable post-harvest spectral pattern. This 

spectral pattern opens up a technological opportunity for assisted seed lot selection 

based on reflectance, physical properties, and color descriptors. This also have the 

potencial for a future seed testing standardization. 

Regarding the shape markers, width and area were the most important variables 

for segmenting stressed seeds (FIGURE 2). Probably, stressed seeds had smaller 

width, area (FIGURE 4C, D), and weight (FIGURE 1A) because the plant needed to 

use its photoassimilates to deal with the oxidative stress (FIGURE SUPPLEMENTARY 

1), and the accumulation of reserves in the seed was reduced (Sionit and Kramer, 

1977; Staniak et al., 2023). Thus, the shape of the seeds was modified, which is a 

common effect in seeds produced over stressful conditions. Integrating shape 

properties with the chlorophyll fluorescence and reflectance reported here allows us to 



38 
 

generate a robust model to classify seeds that go through stress during their 

development (FIGURE 6). 

Additionally, we demonstrated that our data could be interpreted in an autonomous 

way, especially using NN algorithms (FIGURE 6). This opens the possibility to turn our 

model into a “results manager,” i.e., identifying why there is a reduction in physiological 

quality in seeds when there is no noticeable damage, the so-called "silent enemies", 

as residual chlorophylls that impact the seed-seedling transition as reported by Ajala-

luccas et al. (2023) and observed in our data (FIGURE 1, and 4).  

In another way, our results allow more precise and sensitive diagnoses since the 

technology is highly associated with the uniformity index (FIGURE 7), an effective vigor 

test diagnosis. In this context, Barboza da Silva et al. (2021b) developed markers 

based on autofluorescence-spectral combined with ML to classify soybean seeds 

regarding their vigor. Therefore, our results additionally demonstrate the potential of 

these tools in this matter. 

The technology proposed here can also be used in plant breeding studies. With 

global climate change, the tool can help to identify stress-tolerant materials by 

screening different genotypes submitted to stress. It can also be used in stress memory 

tests. Like phenotypic plasticity, stress memory is a plant survival strategy (Auler et al., 

2021). The so-called plant stress memory is produced depending on the stress type 

and level since plants can carry out epigenetic modifications that will result in changes 

in their metabolism to ensure adaptation to the new environment (Lukić et al., 2020). 

This change can be passed on to the following generations and produce adaptations 

to new environmental scenarios (Liu et al., 2022). Stress memory assessments have 

been used in plant breeding programs to select stress-resistant materials (Villagómez-

Aranda et al., 2022). The great advantage of adapting the tool in memory stress assays 

to seed bulking is that it allows quantification on a single seed. 

Our study brings more knowledge about soybean seed quality evaluation through 

multispectral imaging associated with ML algorithms and presents a new approach to 

data management that enables detecting vigor differences caused by stress 

occurrence during seed maturation in a much more precise way than the traditional 

seed physiological quality tests.  



39 
 

1.5 CONCLUSION 

Machine learning models based on multispectral imaging are highly effective for 

recognizing soybean seeds produced under stressful conditions, especially using 

features based on chlorophyll a fluorescence, seed width, chlorophyll a index, seed 

area, and reflectance at 365 nm. This approach can detect stressed soybean seeds 

with 90% accuracy using neural network algorithm. 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 



40 
 

REFERENCES 

AJALA-LUCCAS, D. et al. The Seed–Seedling Transition in Commercial Soybean 
Cultivars with the Presence of Greenish Seeds in the Sample: A Perspective from 
Classical Genetic Parameters. Agronomy, Basel, Switzerland, Vol. 13, Page 1966, 
v. 13, n. 8, p. 1966, 26 jul. 2023.  

ALEXIEVA, V. et al. The effect of drought and ultraviolet radiation on growth and 
stress markers in pea and wheat. Plant, Cell & Environment, Chichester, United 
Kingdom, 24, n. 12, p. 1337–1344, 1 dez. 2001.  

BAGATELI, J. R. et al. Productive performance of soybean plants originated from 
seed lots with increasing vigor levels. Journal of Seed Science, Brasília, DF, v. 41, 
n. 2, p. 151–159, 2019.  

BARBOZA DA SILVA, C. et al. A novel approach for Jatropha curcas seed health 
analysis based on multispectral and resonance imaging techniques. Industrial 
Crops and Products, Amsterdam, Netherlands, v. 161, p. 113186, 1 mar. 2021a.  

BARBOZA DA SILVA, C. et al. Autofluorescence-spectral imaging as an innovative 
method for rapid, non-destructive and reliable assessing of soybean seed quality. 
Scientific Reports, London, United Kingdom, v. 11, n. 1, 1 dez. 2021b.  

BATISTA, T. B. et al. A reliable method to recognize soybean seed maturation 
stages based on autofluorescence-spectral imaging combined with machine learning 
algorithms. Frontiers in Plant Science, Lausanne, Switzerland, v. 13, 2022.  

BIANCHINI, V. DE J. M. et al. Multispectral and X-ray images for characterization of 
Jatropha curcas L. seed quality. Plant Methods, London, United Kingdom, v. 17, n. 
1, 2021.  

BRASIL, M. DA A. P. E A. Regras para análise de sementes. (Assessoria de 
Comunicação Social, Ed.) Brasília, DF: 2009.  

CAKMAK, I.; CAKMAK, W. J. H.; HORST, W. J. Effect of aluminium on lipid 
peroxidation, superoxide dismutase, catalase, and peroxidase activities in root tips of 
soybean (Glycine max). Physiologia Plantarum, Hoboken, United States, 1991.  

COHEN, I. et al. Meta-analysis of drought and heat stress combination impact on 
crop yield and yield components. Physiologia Plantarum, Hoboken, United States, 
v. 171, n. 1, p. 66–76, 2021.  

CONTINI, E.; TALAMINI, D. J. D.; VIEIRA JUNIOR, P. A. Cenário mundial de 
commodities: frango, soja e milho. - Portal Embrapa, Brasília, DF. Disponível em: 
<https://www.embrapa.br/busca-de-publicacoes/-/publicacao/968636/cenario-
mundial-de-commodities-frango-soja-e-milho>. Acesso em: 21 nov. 2023.  

DONALDSON, L. Autofluorescence in plants. Molecules, Basel, Switzerland, v. 25, 
2020.  



41 
 

EBONE, L. A. et al. Soybean Seed Vigor: Uniformity and Growth as Key Factors to 
Improve Yield. Agronomy, Basel, Switzerland, v. 10, n. 4, p. 545, 2020.  

FINCH-SAVAGE, W. E.; BASSEL, G. W. Seed vigour and crop establishment: 
Extending performance beyond adaptation. Journal of Experimental Botany, 
Oxford, United Kingdom, v. 67, n. 3, p. 567–591, 2016.  

FONSECA DE OLIVEIRA, G. R. et al. An approach using emerging optical 
technologies and artificial intelligence brings new markers to evaluate peanut seed 
quality. Frontiers in Plant Science, Lausanne, Switzerland, v. 13, 2022.  

FRANÇA-NETO, J. B.; KRZYZANOWSKI, F. C. Metodologia do teste de tetrazólio 
em sementes de soja. Londrina, PR: Embrapa Soja, 2022.  

FRANÇA-SILVA, F. et al. Advances in imaging technologies for soybean seed 
analysis. Journal of Seed Science, Brasília, DF, v. 45, p. 1–16, 2023.  

GALLETTI, P. A. et al. Integrating Optical Imaging Tools for Rapid and Non-invasive 
Characterization of Seed Quality: Tomato (Solanum lycopersicum L.) and Carrot 
(Daucus carota L.) as Study Cases. Frontiers in Plant Science, Lausanne, 
Switzerland, v. 11, 21 dez. 2020.  

GARCÍA-PLAZAOLA, J. I. et al. Autofluorescence: Biological functions and technical 
applications. Plant Science, Shannon, Ireland, v. 236, p. 136–145, 2015.  

GITELSON, A. A.; GRITZ, Y.; MERZLYAK, M. N. Relationships between leaf 
chlorophyll content and spectral reflectance and algorithms for non-destructive 
chlorophyll assessment in higher plant leaves. Journal of plant physiology, 
Muenchen, Germany, v. 160, n. 3, p. 271–282, 2003.  

GOMES, W. P. C. et al. Application of multispectral imaging combined with machine 
learning models to discriminate special and traditional green coffee. Computers and 
Electronics in Agriculture, Amsterdam, Netherlands, v. 198, p. 107097, 2022.  

KRZYZANOWSKI, F. C. et al. Vigor de sementes: conceitos e testes. Londrina, 
PR: Abrates, 2020.  

KRZYZANOWSKI, F. C.; FRANÇA-NETO, F. B.; HENNING, A. A. A alta qualidade 
da semente de soja: fator importante para a produção da cultura. CT136. 
EMBRAPA, Londrina, PR, n. 136, p. 1–24, 2018. 

LIMA, J. J. P. et al. Molecular characterization of the acquisition of longevity during 
seed maturation in soybean. PLOS ONE, San Francisco, United States, v. 12, n. 7, 
p. e0180282, 2017.   

LIU, H.; ABLE, A. J.; ABLE, J. A. Priming crops for the future: rewiring stress 
memory. Trends in Plant Science, Cambridge, United States, v. 27, n. 7, p. 699–
716, 1 jul. 2022.  

MELGOSA, M. Testing CIELAB-Based Color-Difference Formulas. Color Research 

and Application, Hoboken, United States, v.25, n.1, 2000. 



42 
 

NAKAJIMA, S. et al. Chlorophyll b Reductase Plays an Essential Role in Maturation 
and Storability of Arabidopsis Seeds. Plant Physiology, Rockville, United States , v. 
160, n. 1, p. 261–273, 3 set. 2012.  

OLIVEIRA, N. M. et al. Hormetic effects of low-dose gamma rays in soybean seeds 
and seedlings: A detection technique using optical sensors. Computers and 
Electronics in Agriculture, Amsterdam, Netherlands, v. 187, p. 106251, 1 ago. 
2021.  

PÁDUA, G. P. et al. Incidence of green soybean seeds as a function of 
environmental stresses during seed maturation. Revista Brasileira de Sementes, 
Brasília, DF, v. 31, n. 3, p. 150–159, 2009.  

PÁDUA, G. P. et al. Tolerance level of green seed in soybean seed lots after storage. 

Revista Brasileira de Sementes, Brasília, DF, v. 29, n. 3, p. 112–120, 2007.  

RITCHIE, S. W.; HANWAY, J. J.; THOMPSON, H. E. How a Soybean Plant 
Develops. Des Moines, United States, p.1-24, 1982.  

SIONIT, N.; KRAMER, P. J. Effect of Water Stress During Different Stages of Growth 
of Soybean. Agronomy Journal, Madison, United States, v. 26, p. 274-278, 1977.  

STANIAK, M.; SZPUNAR-KROK, E.; KOCIRA, A. Responses of soybean to selected 
abiotic stresses—photoperiod, temperature, and water. Agriculture, Basel, 
Switzerland, v. 13, Page 146, v. 13, n. 1, p. 146, 2023.  

SUDKI, J. M. et al. Fungal identification in peanuts seeds through multispectral 
images: Technological advances to enhance sanitary quality. Frontiers in Plant 
Science, Lausanne, Switzerland, v. 14, p. 1112916, 2023.  

TEIXEIRA, R. N. et al. Gene expression profiling of the green seed problem in 
Soybean. BMC Plant Biology, London, United Kingdom, v. 16, n. 1, p. 1–15, 2016.  

USDA. United States Department Of Agriculture. Oilseeds: World Markets and 
Trade. Washington, United States, 2023. Disponível em: 
<https://public.govdelivery.com/accounts/USDAFAS/subscriber/new>.  

VILLAGÓMEZ-ARANDA, A. L. et al. Activating stress memory: eustressors as 
potential tools for plant breeding. Plant Cell Reports, Heidelberg, Germany, v. 41, p. 
1481–1498, 2022.  

WANG, X. et al. Multiple omics datasets reveal significant physical and physiological 
dormancy in alfalfa hard seeds identified by multispectral imaging analysis. The Crop 
Journal, Beijing, China, v. 11, n. 5, p. 1458–1468, 2023.  

ZHANG, S. et al. Non-Destructive Testing of Alfalfa Seed Vigor Based on 
Multispectral Imaging Technology. Sensors, Basel, Switzerland, v. 22, Page 2760, v. 
22, n. 7, p. 2760, 2022.  

ZINSMEISTER, J. et al. ABSCISIC ACID INSENSITIVE 4 coordinates eoplast 
formation to ensure acquisition of seed longevity during maturation in Medicago 



43 
 

truncatula. The Plant Journal, Chichester, United Kingdom, v. 113, n. 5, p. 934–953, 
1 mar. 2023.  

ZORATO, M. DE F. et al. Presença de sementes esverdeadas em soja e seus 
efeitos sobre seu potencial fisiológico. Revista Brasileira de Sementes, Brasília, 
DF, v. 29, n. 1, p. 11–19, 2007.  

 

 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

 



44 
 

CHAPTER 2 

 

SILENCING AND OVEREXPRESSION OF GmABI5 GENE IN SOYBEAN 

GREENISH SEED STUDIES 

ABSTRACT 

Soybeans with chlorophyll retention have been frequent over the years, bringing issues 

for the seed and grain market. Greenish soybean seeds have lower physiological 

quality compared to seeds without chlorophyll retention. Besides the occurrence of 

adverse conditions during the seed maturation phase, genetic factors also play an 

important role as a determining factor for the occurrence of greenish seeds. Thus, this 

work aimed to perform silencing (RNAi) and overexpression of the candidate gene 

GmABI5 for future functional investigations. For the overexpression cassette, a 

fragment of 1359 base pairs was inserted between the regulatory elements of the 

glycinin protein. Two 422 bp fragments were inserted at the ends of an intron to form 

a hairpin and avoid gene expression for the silencing cassette. The transformation was 

carried out via a gene gun. Both cassettes had a hygromycin resistance gene used for 

the transformed embryos' selection phase. The presence of the expression cassette in 

transformed and hygromycin-resistant embryos was confirmed via PCR. The first 

embryos obtained were regenerated into seedlings and continue to be cultivated. Two 

hygromycin-resistant lines were confirmed for the presence of the cassette. The 

remaining embryos will be tested for the presence of the cassette; they are being 

cultivated, and the positive ones will be regenerated into seedlings and will be 

cultivated until homozygous lines are obtained. 

 

2.1 INTRODUCTION 

Soybean is an essential source of protein and oil and is widely used in animal feed, 

biodiesel production, and the pharmaceutical, cosmetic, and food industries 

(APROSOJA, 2023; Contini et al., 2018). Brazil has a prominent position in the world 

ranking of soy-producing countries. In 2021, the country was the production and export 

leader, with approximately 134 million tons and 86 million tons, respectively (FAO, 

2023). Seeds are one of the most critical inputs in agriculture since they are the onset 

of a new field; they carry the genome for the new plant and allow the establishment of 

vigorous plants.  



45 
 

 High quality seed production is critical for the seed technology market. In recent 

years, soybean greenish seeds have been a problem for Brazilian soybean growers 

due to the impairment of seed and oil quality. Soybean greenish seeds are more 

susceptible to degradation, leading to reduced physiological quality in terms of viability, 

vigor, and longevity (Pádua et al., 2007; Zinsmeister et al., 2016). These pigments also 

affect the grain industry since they lead to oxidation and changes in flavor, and demand 

additional bleaching steps (Diosady, 2005), causing financial losses.  

Soybean greenish seeds are mature seeds with chlorophyll (Chl) retention. Chl is 

progressively degraded during maturation, and when maturity is reached, soybean 

seeds become naturally yellow (Lima et al., 2017; Ritchie, Hanway; Thompson, 1982). 

Soybean greenish seeds have their chlorophyll degradation process impaired in 

response to stresses such as high temperatures and drought during the maturation 

phase. Besides, the susceptibility to Chl retention may vary according to the genotype 

(Pádua et al., 2009; Teixeira et al., 2016).  

There are few studies about the molecular mechanisms involved in chlorophyll 

retention/degradation in soybean seeds. Teixeira et al. (2016) reported altered 

expression of the STAY-GREEN 1 and STAY-GREEN 2 (D1, D2), 

PHEOPHORBIDASE 2 (PPH2), and NON-YELLOW COLORING 1 (NYC1_1) genes 

in a susceptible cultivar. M. truncatula abi5 mutant and Mtabi5 Mtabi4 double mutant 

seeds presented more chlorophyll and less seed quality than wild-type ones 

(Zinsmeister et al., 2016, 2023). However, the molecular mechanism associated with 

the retention and degradation of Chl is still an object of study. 

 ABI5 is a transcription factor involved in the control of late seed maturation in 

legumes.  ABI5 plays a role in seed physiological traits such as dormancy, desiccation 

tolerance, and longevity through the regulation of raffinose family oligosaccharides 

(RFO) and late embryogenesis protein (LEA) content (Lynch et al., 2022). ABI5 is also 

involved in expressing photosynthetic genes (PhANGs – photosynthesis-associated 

nuclear genes) and dismantling the photosynthetic apparatus during seed maturation; 

therefore, it might also be involved with seed degreening (Zinsmeister et al., 2016). 

Thus, this study aimed to perform soybean genetic transformations with GmABI5 as a 

candidate gene for further functional investigations and possibly reproduce a genotype 

tolerant to environmental conditions that may cause chlorophyll retention. 



46 
 

2.2 MATERIALS AND METHODS 

Candidate gene 

The first steps of this research was perfomed at the School of Agriculture, 

Department of Crop Sciences, UNESP/Botucatu/São Paulo/Brazil. We performed 

gene expression preliminary studies to verify GmABI5 (Primers: 5’ - 

GTGGAGAAGGTGGTGGAG  3’ - CCAGTTCAACAGTGTATGCC) expression in samples 

of commercial soybean cultivars. The RNA was extracted using the NucleoSpin® RNA 

plant commercial Kit (Macherey-Nagel, Düren, Germany) following the manufacturer’s 

instructions. The purity and quantity of total RNA were checked in a Nanodrop-2000 

spectrophotometer (Thermo Scientific, Wilmington, DE, USA). RNA integrity was 

checked in agarose gel at 1%. cDNA was synthesized by using a High-Capacity cDNA 

Reverse Transcription commercial kit (Applied Biosystems, Victoria, Australia) 

following the manufacturer’s instructions. For a 20 μl reaction, we used 2.0 μl 10× RT 

Buffer, 0.8 μl 25× dNTP, 2.0 μl 10× RT Primer, 1 μl Reverse Transcriptase, 10 μl of 

the extracted RNA and 4.2 μl Nuclease-free water. The reaction was incubated in a 

thermocycler following these steps: 10 min at 25°C, 2 h at 37°C, followed by 5 min at 

85°C and the synthesis ended at a constant temperature of 4°C. 

The primers used were efficient above 1.8 and r2 of approximately 1.0, according 

to the parameters of the LinRegPCR program (Ruijter et al., 2013). Gene expression 

analysis was performed in a thermocycler Eco Real-Time (Illumina, San Diego, USA) 

with SYBR Green qPCR ReadyMix (Sigma-Aldrich, St. Louis, USA), using two 

technical samples for each biological sample (three biological samples for each seed 

sample). For a reaction of 10 μl, 5 μl of SYBR Green, 1 μl of cDNA, and 0.25 μl of each 

primer was used, and the volume was adjusted with Nuclease-free water. The 

amplification consisted of 2 min at 50°C, 2 min at 95°C, then 45 cycles of 10 s at 95°C 

and 1 min at 60°C. At the end of the process, the melting curve was performed 

following these steps: 15 s at 95, 65, and 95°C, respectively. The normalized 

expression (NE) was assessed using the Proteasome gene as the reference gene, 

calculated according to NE = 2ΔCt, where ΔCt is the Cq of the reference gene – Cq 

target gene. 

 In the first gene expression experiment, we studied the expression of the ABI5 

gene in two different maturation stages (R7.2 and R9) in yellow seeds of the cultivar 

MG/BR 46 Conquista which were produced according to Batista (2022). In parallel, the 



47 
 

Chl autofluorescence was determined by using multispectral imaging (Batista, 2022). 

In the second study, we performed gene expression of the ABI5 gene in two 

commercial soybean cultivars, NA5909IPRO and 97Y07. The production environment 

of the commercial cultivars is described by Luccas (2018). The seed samples 

comprised greenish and yellow seeds obtained through visual segregation. Seeds 

were considered greenish when they showed any sign of green pigmentation and were 

considered yellow when they did not show any sign of green pigmentation (Luccas, 

2018).  

Expression cassette 

From this stage forward, the research was carried out at Dr. Schmidt’s lab, College 

of Agriculture and Life Sciences, Bio5 Institute, University of Arizona/Tucson/AZ/EUA.  

We performed genetic transformation in soybean plants with expression alteration: 

increase (overexpression – ABI5OE) and silencing (RNAi – ABI5RNAi). The ABI5 

cDNA nucleotide sequence was acquired in the GenBank database (Clark et al., 2016) 

under the accession number (XM_014763030.3).  

For the overexpression cassette, a 1359 base pairs (bp) fragment containing the 

restriction sites for NotI was commercially synthesized in a plasmid with ampicillin 

(AMP) resistance and was cloned by E. coli transformation. The fragment was excised 

with NotI and placed in the open reading frame of an AMP-resistant vector containing 

the seed-specific storage protein glycinin (GLY) promoter. The plasmid also contained 

a hygromycin (HYG) resistance gene under the expression of the potato ubiquitin 3 

regulatory elements (He et al., 2016) (pGLY/HYG). The resultant plasmid 

pGLYABI5OE/HYG (Figure 1a) was sequenced with a GLY promoter primer (5’ -

CCTCATTCACCTTCCTCTCTTC) to ensure the ABI5 open reading frame was placed 

correctly between the regulatory elements.  

A 422 bp fragment containing NotI, XhoI, SpeI, Xba, and HindIII restriction sites 

was synthesized for the RNAi cassette in a plasmid with AMP resistance (pABI5). The 

cassette construction was according to Schmidt and Herman (2008). An intron was 

placed between two ABI5 arms forming a hairpin secondary structure in which the 

protein translation is impaired, and then the gene expression is silenced. The pABI5 

and intron-containing plasmid were cloned by E. coli transformation. Two 422 bp 

fragments were restricted from the pABI5 vector with two separate restriction enzyme 

reactions: Xho/Spe and Xba/HindIII. The pABI5–Xho/Spe fragment was first inserted 



48 
 

into the new restriction sites at the 5’ end of the intron-containing vector. This plasmid 

was at this moment named pIntron-ABI5-HALF. The second arm pABI5-Xba/HindIII 

was inserted into pIntron-ABI5-HALF, generating the cassette containing a 422-bp 

ABI5 hairpin flanking the modified intron-containing plasmid; this vector was named 

pABI5-RNAi. Afterward, the pABI5RNAi open reading frame was NotI digested to move 

the entire ABI5 hairpin between the glycinin regulatory elements of the pGLY/HYG 

vector as described above (Figure 1b). Correct orientation of the hairpin between the 

regulatory elements was confirmed by sequencing using a glycinin promoter primer as 

previously described. The resulting vector was named pGLYRNAiABI5/HYG. 

Figure 1 – Graphical representation of expression cassettes.  The ABI5OE 
cassette contains the ABI5 cDNA sequence and the GLY promoter regulatory 
elements (a). ABI5RNAi cassette containing sense and antisense 422bp ABI5 
fragments inserted between an intron and GLY promoter regulatory 
elements (b). 

 

 

Tissue Culture and genetic transformation 

Soybean somatic embryos (cultivar Jack) were induced from 5 mm long immature 

cotyledons (Figure 2a, b).  The cotyledons were excised at their narrowest portion, and 

the portion with the embryonic axis was removed from the seed coat. The cotyledons 

were placed on MSD40 media plates with the flat side up (Figure 2c). The MSD40 

media, according to Finer and Nagasawa (1988), had the pH modified according to 

Komatsuda and Ohyama (1988), and sucrose according to Lazzeri, Hildebrand, and 

Collins (1987). The plates were kept in the tissue culture room under 22 °C, low light 

intensity (10 μmol photons m−2 s−1), eight hours photoperiod.  

 

 

 

 

 



49 
 

 

Figure 2 – Soybean somatic embryo induction. Soybean pod containing 5 
mm long immature embryos (a, b). Excised embryos with the flat side up 

before being placed on MSD40 plates (c). Raspberry-appearance embryos 
after 4-6 weeks on MSD40. Magnification range: 30x (d). 

 

After four-six weeks, raspberry-appearance embryos (Figure 2d) were transferred 

to FN-Lite proliferation media (Samoylov; Tucker; Parrott, 1998) and kept on a shaker 

at 125 rpm. The embryos were subcultured weekly for four weeks to new media, when 

they got a cluster appearance and became appropriate for gene gun shooting (Figure 

3a, b). Four days before shooting, compact masses of globular-stage embryos were 

transferred to MSD20 (Wright et al., 1991) plates in a 30mm diameter area in the center 

of the plate (FIGURE 4a,b). 

 

     



50 
 

 
 

 Figure 3 – Embryo clusters after four weeks of growth in FNL media 

 
 

Figure 4 – Embryos on MSD20 ready for shooting (a, b) 

 

The somatic embryos were transformed using a Dupont PDS-1000 Biolistic 

Particle Delivery System with gold particles as a DNA microcarrier, accordingly to 

Dinkins et al. (2003), with some modifications. For 12 shots (0.4 mg of gold and 6.25 

µg of DNA per shot), 10 mg of gold were rinsed in 1mL of 100%EtOH and centrifuged 

for 5 min at 7000rpm. The gold was resuspended in 169 µL of 100% EtOH, and three 

aliquots of 35µL were withdrawn. The aliquots were briefly centrifuged for 10 seconds 



51 
 

at 10000 rpm, the supernatant was removed, and 1 mL of sterile H2O was added. 

Another centrifugation was performed for 5 minutes at 2000 rpm, and the supernatant 

was removed. 25µL of DNA was added to the metal pellet for each aliquot, and the 

microcarriers were resuspended. Subsequently, 220 µL of sterile H2O were added, 

followed by 250 µL of 2.5M CaCl2 and 50µL of 0.1M spermidine stock. The aliquots 

were placed on ice for 2-3 min, vortexed for 10 min, and centrifuged for 5 min at 500-

1000 rpm. The supernatant was removed, the DNA/microcarrier was resuspended in 

600 µL of 100% EtOH, centrifuged for 5 min at 500-1000 rpm, the EtOH was removed, 

and the aliquot was resuspended in 36µL of 100% EtOH by flicking. The aliquots were 

kept on ice for 1h.  

The gene gun was prepared before shooting, and all parts were cleaned in 100% 

EtOH. Twenty minutes after the shooting, the embryos’ plates were opened in the 

laminar flow to let any drop of water dry. The aliquots were sonicated for 10-15 sec, 

and 10 µL of the resuspended DNA/microcarriers were withdrawn and dispensed on 

the surface of a microcarrier disk. The shooting was performed using an 1100 psi 

rupture disk at an 8 cm distance from DNA to target tissue. After shooting, embryos 

were placed in FN-Lite media for one week. In this project, we performed a total of four 

gene gun protocols. The first two were the ABI5OE cassette, and the others were the 

ABI5RNAi cassette. There were 6 MSD20 plates for each protocol, and each plate 

received two shots.  Figure 5 shows gene gun during shooting experiments. 

Figure 5. Dupont PDS-1000 Biolistic Particle Delivery System during soybean 

somatic embrios transformation. 

 



52 
 

Selection of Cis/Intragenic plants 

The embryos were submitted to hygromycin B (HYG) selection one week after 

transformation. Hygromycin B (PhytoTech Labs, Lenexa, Kansas – USA, 50mg/mL) 

was added to FN-Lite media in a final 20mg HYG/L concentration. The FN-Lite media 

+ HYG solution was replaced weekly for 6-8 weeks. The embryos were screened using 

a stereomicroscope to select the ones that were green/alive. Each green embryo was 

considered as a different line (Figure 6) and was grown in FN-Lite media until it became 

a large enough tissue for subsequent analysis. The media was replaced every two 

weeks.  

Figure 6 – HYG-resistant embryos were screened after six weeks of selection. 
Picture of the dissection microscope view. Magnification range: 30x. 

 

Resistant embryos obtained from ABI5OE transformation, after nine weeks of 

growth, were screened by PCR to confirm the presence of the inserted expression 

cassette with HYG gene primers (5’ - CTCACTATTCCTTTGCCCTC 3’- 

CTGACCTATTGCATCTCCCG). Genomic DNA was extracted by the cetyl trimethyl 

ammonium bromide (CTAB) method. 150 ng genomic DNA were used in 25 μL total 

reaction containing 200 nM primers and 3 U Taq polymerase (NEB), and the following 

cycling parameters were used: one initial cycle at 95°C 4 min, then 30 cycles at 95°C 

30 s, 55°C 30 s and 72°C 30s, and then a final extension cycle at 72°C for 7 min. 

GAPDH (glyceraldehyde-3-phosphate dehydrogenase) primers (5’ – 

TAACGAGCACGAGTACAAGCC  3’ - AATTCCCTTCAACTTGCCCTC) were used as 

an absolute control for PCR reactions. Soybean cultivar Jack was used as a negative 

control, and we used a HYG positive sample as a positive control.   

Tissue culture embryos were histodifferentiatiated through SHAM media during 

approximately two weeks, according to Schmidt et al. (2005). When they reached the 

cotyledonary stage, they were submitted to desiccation (Figure 7), in which a 

determined number of embryos were placed in a sealed petri dish for about seven 

days. Afterward, embryos were germinated in MSO media (Figure 8a) (Murashige; 

Skoog, 1962; Gamborg et al., 1968), and when there was a visible shoot and roots 



53 
 

(Figure 8b), they were regenerated into plants (Figure 9). Tissue culture took, in total, 

approximately 9 months, from induction until conversion into plants.  

 

Figure 7 – Soybean somatic cotyledonary stage embryos during desiccation. 

 

Figure 8 – Soybean somatic cotyledonary stage embryos during germination. 
Beginning of germination process (a). Germinated embryos with visible shoot 

and roots (b). 

 

 

 

 

 



54 
 

Figure 9 – Regenerated seedling from somatic embryogenesis. 

 

2.3 RESULTS AND DISCUSSION 

Figure 10a shows an increase in the expression of ABI5 transcripts from immature 

(R7.2) to mature (R9) soybean seeds. In parallel, chlorophylls a and b 

autofluorescence decreased in mature seeds (Figure 10b, c). In the first experiment, 

the BR/MG 46-Conquista was used because it is susceptible to abiotic stresses and 

chlorophyll retention (Pádua et al., 2009), therefore we supposed it would provide on-

target results for ABI5 gene expression.   

In addition, we used two soybean cultivars produced in commercial field 

environments according to Luccas (2018).  There was a decreased expression of ABI5 

transcripts in greenish seeds compared to yellow seeds in both commercial soybean 

cultivars (Figure 11).  Thus, when ABI5 expression is increased, the chlorophyll content 

in the seeds tended to decrease.  

 

 



55 
 

Figure 10 – ABI5 gene expression and Chl fluorescence from MG/BR 46 
Conquista cultivar.  (A) Normalized gene expression from two different 

maturation stages. Each bar represents the mean of three biological 
samples ± standard deviation. 3.83-fold represents the fold change between 

R7.2 and R9 sample. (B) and (C) represent, respectively, Chl a and b 
fluorescence. Each bar represents the mean of four samples.  Different 

letters indicate the difference between samples by t-test 95% of the 
confidence interval. The non-overlapping of the confidence interval 

constitutes the difference between the groups evaluated 

 

 

 

 

 

 

 

 

 

 

 

 

 



56 
 

Figure 11– ABI5 gene expression from two different commercial soybean 
cultivars. (A) Normalized gene expression from greenish and yellow seeds 

from NA5909 cultivar. Each bar represents the mean of three biological 
samples ± standard deviation. 0.86-fold represents the fold change between 
greenish and yellow seeds. (B) Normalized gene expression from greenish 

and yellow seeds from 97Y09 cultivar. Each bar represents the mean of three 
biological samples ± standard deviation. 0.71-fold represents the fold 
change between greenish and yellow seeds. Data are not significantly 

different between samples by t-test 95% of the confidence interval 

 

Therefore, this data indicated that ABI5 is a candidate gene for controlling 

chlorophyll degradation in soybean seeds and may be an essential gene in the 

greenish seed problem and environmental stress tolerance studies. In parallel, 

Zinsmeister et al (2016) suggested that chlorophyll degradation is under the control of 

ABI5, once there was remaining chlorophyll in abi5 mutant mature seeds of Pisum 

sativum and M. truncatula. Therefore, with these results and also based on literature, 

we decided to move forward with GmABI5 transformations.  

After genetic transformation, the presence of the ABI5 overexpression 

cassette was confirmed in the first two HYG-resistant lines (Figure 12). The 

remaining ABI5OE and ABI5RNAi lines were not confirmed until this thesis 

publication because there was not enough tissue for DNA extraction yet. This step 

will be performed and subsequent analysis will be reported in a scientific paper.  

 

 

 

 



57 
 

Figure 12 – Electrophoretic pattern of the PCR product of HYG gene 
detection in GmABI5 overexpression studies. GAPDH - glyceraldehyde-3-
phosphate dehydrogenase primers were used as absolute control for the 

PCR. HYG - Hygromycin gene. Jack – soybean cultivar used as HYG 
negative control. HYG+ - HYG positive control. bp – base pairs. 

 

 

2.4 CONCLUSION 

Two genetically modified GmABI5 overexpression lines have been confirmed as 

positive for the cassette presence, which indicates the gene gun transformation had 

succeeded. The remaining lines will be screened and grown until homozygous plants 

are obtained through self-pollination to characterize the GmABI5 overexpression effect 

in plants.  

 

 

 

 

 

 

 

 



58 
 

REFERENCES 

APROSOJA. Associação Brasileira dos Produtores de Soja. A soja: A origem do 
grão. [S. L.], 2023. Available in: <https://aprosojabrasil.com.br/a-soja/>. Accessed 
on: 27 dez. 2023.  

BATISTA, T. B. MATURAÇÃO EM SEMENTES DE SOJA: ESTÁDIOS 
REPRODUTIVOS, DEGRADAÇÃO DA CLOROFILA E AQUISIÇÃO DA 
QUALIDADE FISIOLÓGICA. 2022. Tese (Doutorado em Agronomia/Agricultura). 
Faculdade de Ciências Agronômicas, UNESP, Botucatu, SP, 2022.  

CLARK, K. et al. GenBank. Nucleic Acids Research, Oxford, United Kingdom v. 44, 
n. 1, p. D67–D72, 2016.  

CONTINI, E. et al. Série desafios do agronegócio brasileiro: complexo soja. 
Caracterização e Desafios Tecnológicos. Embrapa, Brasília, DF, p. 1-35, 2018.  

DINKINS, R. D. et al. Recent Advances in Soybean Transformation. In: Jaiwal, P. K.; 

Singh, R. P. Applied Genetics of Leguminosae Biotechnology, Dordrecht, 

Netherlands: Springer, p.3-21, 2003.  

DIOSADY, L. L. Chlorophyll removal from edible oils. International Journal of 
Applied Science and Engineering, Taichung, Taiwan, v. 3, n. 2, p. 81–88, 2005.  

FAOSTAT. Food and Agriculture Organization of the United Nations. Countries by 
commodities. Rome, Italy, 2023.  Available in : 
<https://www.fao.org/faostat/en/#rankings/countries_by_commodity_exports>. 
Accessed on: 27 dez. 2023.  

FINER, J. J.; NAGASAWA, A. Development of an embryogenic suspension culture of 
soybean [Glycine max (L.) Merrill]. Plant Cell, Tissue and Organ Culture, 
Dordrecht, Netherlands, v. 15, p. 125–136, 1988.  

GAMBORG, O. L.; MILLER, R. A.; OJIMA, K. Nutrient requirements of suspension 

cultures of soybean root cells. Experimental Cell Research, Waltham, United 

States, v. 50, n. 1, p. 151–158, 1968.  

HE, Y. et al. Transgenic soybean production of bioactive human epidermal growth 
factor (EGF). PLOS ONE, San Francisco, United States, v. 11, n. 6, p. e0157034, 
2016.  

KOMATSUDA, T.; OHYAMA, K. Genotypes of high competence for somatic 
embryogenesis and plant regeneration in soybean Glycine max. Theoretical and 
Applied Genetics, Heidelberg, Germany, v. 75, p. 695–700, 1988.  

LAZZERI, P. A.; HILDEBRAND, D. F.; COLLINS, G. B. Soybean somatic 
embryogenesis: Effects of nutritional, physical and chemical factors. Plant Cell, 
Tissue and Organ Culture, Dordrecht, Netherlands, v. 10, n. 3, p. 209–220, 1987.  

LIMA, J. J. P. et al. Molecular characterization of the acquisition of longevity during 
seed maturation in soybean. PLOS ONE, San Francisco, United States, v. 12, n. 7, 
p. e0180282, 2017.  



59 
 

LUCCAS, D. A. Caracterização fisiológica, bioquímica e molecular em sementes 
de soja (Glycine max (L.) Merr.) com retenção de clorofila. 2018. Tese 
(Doutorado em Agronomia/Agricultura). Faculdade de Ciências Agronômicas, 
UNESP, Botucatu, SP, 2018.  

LYNCH, T. et al. ABI5 binding protein2 inhibits ABA responses during germination 
without ABA-INSENSITIVE5 degradation. Plant physiology, Rockville, United 
States, v. 189, n. 2, p. 666–678, 1 jun. 2022. 

MURASHIGE, T.; SKOOG, F. A revised medium for rapid growth and bio assays with 

tobacco tissue cultures. Physiologia Plantarum, Hoboken, United States, v. 15, n. 

3, p. 473–497, 1 jul. 1962.  

PÁDUA, G. P. DE et al. Tolerance level of greenish seed in soybean seed lots after 
storage. Revista Brasileira de Sementes, Brasília, DF , v. 29, n. 3, p. 112–120, 
2007.  

PÁDUA, G. P. DE et al. Response of soybean genotypes to the expression of 
greenish seed under temperature and water stresses. Revista Brasileira de 
Sementes, Brasília, DF, v. 31, n. 3, p. 140–149, 2009. 

RITCHIE, S. W.; HANWAY, J. J.; THOMPSON, H. E. How a Soybean Plant 
Develops. Des Moines, United States, p.1-24, 1982.   

RUIJTER, J. M. et al. Evaluation of qPCR curve analysis methods for reliable 
biomarker discovery: bias, resolution, precision, and implications. Methods, San 
Diego, United States, v. 59, n. 1, p. 32–46, 2013.  

SAMOYLOV, V. M.; TUCKER, D. M.; PARROTT, W. A. Soybean [Glycine max (L.) 
Merrill] embryogenic cultures: The role of sucrose and total nitrogen content on 
proliferation. In Vitro Cellular and Developmental Biology - Plant, Heidelberg, 
Germany, v. 34, n. 1, p. 8–13, 1998.  

SCHMIDT, M. A. et al. Towards normalization of soybean somatic embryo 
maturation. Plant Cell Reports, Heidelberg, Germany, v. 24, n. 7, p. 383–391, 2005.  

SCHMIDT, M. A.; HERMAN, E. M. Suppression of soybean oleosin produces micro-
oil bodies that aggregate into oil body/ER complexes. Molecular Plant, Cambridge, 
United States v. 1, n. 6, p. 910–924, 2008.  

TEIXEIRA, R. N. et al. Gene expression profiling of the greenish seed problem in 
Soybean. BMC Plant Biology, London, United Kingdom, v. 16, n. 1, p. 1–15, 1 fev. 
2016.  

WRIGHT, M. S. et al. A simple method for the recovery of multiple fertile plants from 
individual somatic embryos of soybean [Glycine max (L.) merrill]. In Vitro Cellular & 
Developmental Biology - Plant, Heidelberg, Germany, v. 27, n. 3, p. 153–157, jul. 
1991.  

ZINSMEISTER, J. et al. ABI5 is a regulator of seed maturation and longevity in 
legumes. The Plant Cell, Cary, United States, v. 28, n. 11, p. 2735–2754, 12 dez. 
2016.  



60 
 

ZINSMEISTER, J. et al. ABSCISIC ACID INSENSITIVE 4 coordinates eoplast 
formation to ensure acquisition of seed longevity during maturation in Medicago 
truncatula. The Plant Journal, Chichester, United Kingdom, v. 113, n. 5, p. 934–953, 
2023.  

  

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 



61 
 

CHAPTER 3 

 

OSMO-PRIMING IN TOMATO SEEDS DOWN-REGULATES GENES ASSOCIATED 

WITH STRESS RESPONSE AND LEADS TO REDUCTION IN LONGEVITY1 

 

ABSTRACT 

Tomato seeds subjected to osmo-priming show fast and more uniform germination. 

However, osmo-priming reduces seed longevity, which is a complex seed physiological 

attribute influenced by several mechanisms, including response to stress. Thus, to 

have new insights as to why osmo-primed tomato seeds show a short life span, we 

performed a transcript analysis during their priming. For that, we performed gene 

expression studies of the heat-shock protein family genes that were previously 

reported to be associated with the enhancement of longevity in primed tomato seeds. 

Physiological assays of germination, vigour and longevity tests were used to support 

the data. The results show that the short life span of osmo-primed tomato seeds is 

related to the decrease in the expression of transcripts associated with response to 

stress during the priming treatment. These results are important because they add 

information regarding which seed longevity mechanisms are impacted by the priming 

treatment. In parallel, it will allow the use of these genes as markers to monitor 

longevity in osmo-primed tomato seeds. 

 

3.1 INTRODUCTION 

Seed priming is a broadly used treatment by the seed industry aiming to improve 

the physiological performance of seed lots. During priming, the hydration of seeds is 

controlled allowing phases I and II of the germination process to start, and then the 

treatment is interrupted so that phase III (radicle protrusion) is deterred (Bradford, 

1986). Fast and uniform germination are the main benefits of the priming technique. 

Seeds, such as Solanum lycopersicum, may have slow and uneven seed germination 

due to the mechanical restraint to radicle protrusion imposed by the micropylar 

endosperm (Toorop et al., 2008; Nonogaki et al., 2000; Bewley et al., 2013). The 

weakening of micropylar endosperm cells is a prerequisite for radicle protrusion of 

tomato seeds (Toorop et al., 2008). Thus, priming has been used to increase the speed 

 
1 Chapter written in accordance with the Seed Science Research journal’s rules 



62 
 

of germination in tomato seeds (Batista et al., 2020). Apparently, priming improves 

physiological performance by favouring changes in the embryo and in the micropylar 

endosperm. For instance, Anese et al. (2011) showed that in Solanum lycocarpum 

seeds, growth of the embryo occurs concomitantly with weakening of the micropylar 

endosperm during the priming treatment. 

Although the priming treatment improves physiological performance as mentioned 

earlier, it negatively affects seed longevity as was shown by Liu et al. (1996), Bruggink 

et al. (1999), Buitink et al. (2000) and Batista et al. (2020). Seed longevity is the ability 

of a seed to remain viable during storage in the dry state (Leprince et al., 2017) and is 

acquired over seed maturation. Several elements are involved in this process, 

including the non-reducing sugars, which play a role in membrane protection and form 

the cytoplasm glassy state, late embryogenesis abundant proteins (LEAs), the seed 

coat, which protects against oxygen absorption by the embryo, RNA-binding proteins 

that conserve seed mRNA in the dry state, an array of antioxidant molecules, and heat-

shock proteins (HSPs), which are protective proteins that play a role as chaperones 

under stress conditions, according to the review by Zinsmeister et al. (2020). 

According to Batista et al. (2020), some of the mechanisms associated with seed 

longevity are negatively impacted during the priming treatment. Sano and Seo (2019) 

proposed that the short life span of primed Arabidopsis thaliana seeds is associated 

with the progress of the cell cycle during the priming treatment. On the other hand, 

Wang et al. (2018) showed that primed and stored rice seeds presented events 

associated with the reduction in starch metabolism, the consumption of starch reserves 

from the endosperm, the accumulation of malondialdehyde and the decrease of 

antioxidant activities of the enzymes, which could explain the short longevity observed. 

Recently, Batista et al. (2020) showed that a heat-shock treatment is able to 

enhance longevity in primed tomato seeds and preserve the vigour of the seeds during 

storage. The authors showed that transcripts associated with stress response are up-

regulated after the heat-shock treatment in primed tomato seeds. They proposed that 

these transcripts may have a positive influence on the enhancement of longevity. 

Based on these results and considering the importance of stress response molecules 

to maintain seed longevity in primed tomato seeds, we hypothesize that the loss of 

storability observed in osmo-primed tomato seeds is due to the down-regulation of 

genes associated with stress response that apparently occurs during the priming 

treatment. To confirm this, we tested this hypothesis by performing gene expression 



63 
 

studies of transcripts related to stress response during the osmo-priming treatment in 

tomato seeds. 

3.2 MATERIAL AND METHODS 

Seed production  

Tomato seeds from the LA1509 accession, donated by the Tomato Genetics 

Resource Center (https://tgrc.ucdavis.edu/), were produced as described in Batista et 

al. (2020). Red fruits without the presence of greenish color were harvested. The fruits 

were cut with the aid of a knife and the seeds were extracted by hand. The seeds were 

then treated with a sodium hypochlorite solution at an initial concentration of 9% at a 

ratio of 1:1 of seeds for a period of 30 min to remove the remnants of the fruit. Seeds 

were stored at 12°C and 60% ±2% relative humidity (RH) until the beginning of the 

experiment. The seed lot was mixed and empty seeds were discarded. 

Priming, drying and heat-shock treatment 

The priming treatment was performed according to Batista et al. (2020), which 

consisted of placing the seeds into tubes containing a polyethylene glycol (PEG) 6000 

solution with an osmotic potential of −1.0 MPa at 20°C, for 60 h in the dark. The tubes 

were placed over a mixer to shake the solution. The PEG solutions were renewed three 

times to avoid changes in the osmotic potential. After the priming treatment, seeds 

were washed in running water, placed over paper towels, and kept for 24 h at 20°C 

and 60% ±2% RH for drying. In parallel, another group of seeds were placed over 

paper towels and exposed to 38°C and 32% RH, in an oven with air circulation for 2 h 

following the heat-shock treatment protocol as defined by Batista et al. (2020). After 

that, these seeds were maintained in the drying conditions described above. The 

unprimed seeds were used as a control group. 

Physiological assays  

After treatment, the seeds were stored in hermetic glass pots and placed in a cold 

chamber at 12°C and 60% RH ±2% which resulted in a moisture content of 0.08 ± 0.01 

g H2O/g