**E**ff**ect of Seed Priming with Potassium Nitrate on the Performance of Tomato**

**Muhammad Moaaz Ali 1,**† **, Talha Javed 2,3,**† **, Rosario Paolo Mauro 4,\* , Rubab Shabbir 2,3 , Irfan Afzal <sup>3</sup> and Ahmed Fathy Yousef 1,5**


Received: 12 September 2020; Accepted: 23 October 2020; Published: 25 October 2020

**Abstract:** The seed industry and farmers have challenges, which include the production of poor quality and non-certified tomato seed, which ultimately results in decreased crop production. The issue carefully demands pre-sowing treatments using exogenous chemical plant growth-promoting substances. Therefore, to mitigate the above-stated problem, a series of experiments were conducted to improve the quality of tomato seeds (two cultivars, i.e., "Sundar" and "Ahmar") and to enhance the stand establishment, vigor, physiological, and biochemical attributes under growth chamber and greenhouse conditions by using potassium nitrate (KNO3) as a seed priming agent. Seeds were imbibed in 0.25, 0.50, 0.75, 1.0, and 1.25 KNO<sup>3</sup> (weight/volume) for 24 h and then dried before experiments. The results of growth chamber and greenhouse screening show that experimental units receiving tomato seeds primed with 0.75% KNO<sup>3</sup> in both cultivars performed better as compared to other concentrations and nonprimed control. Significant increase in final emergence (%), mean emergence time, and physiological attributes were observed with 0.75% KNO3. Collectively, the improved performance of tomato due to seed priming with 0.75% KNO<sup>3</sup> was linked with higher activities of total soluble sugars and phenolics under growth chamber and greenhouse screening.

**Keywords:** *Solanum lycopersicum* L.; crop establishment; potassium nitrate; seed quality

## **1. Introduction**

Tomato (*Solanum lycopersicum* L.) is a major vegetable crop on a global scale and one of the principle sources of phytonutrients [1,2], which makes it one of the preferred targets by researchers for metabolic engineering, as it is easily docile to biotechnological modifications [3]. Globally, being a vegetable of major economic importance, the tomato is a source of minerals and vitamins, as well as an anticancer agent [4]. Ripe tomatoes contain (average values per 100 g of edible portion) water (94.1%), energy (23 calories), calcium (1.0 g), magnesium (7.0 mg), vitamin A (1000 IU), ascorbic acid (22 mg), thiamin (0.09 mg), riboflavin (0.03 mg), and niacin (0.8 mg) [5]. In tomato, germination and crop establishment are the most crucial physiological stages that are affected by seed quality and genetics [6]. Rapid and uniform germination and seedling establishment is essential for increasing tomato yield and quality [7], which is of economic importance in agriculture. Therefore, various seed enhancement approaches, such as coating, pelleting, and priming, can be responsible to a major extent for improved quality of seeds. Among these approaches, seed priming with suitable priming agents and concentrations can induce some physiological and biochemical changes in the seed, which result in improved crop performances in terms of enhanced germination potential, seedling vigor, and final yield [6,8].

Seed priming is a process of regulating the imbibition and active metabolism phases of germination before radical emergence followed by drying and maintenance of near to original moisture content [9]. It increases the ability of radical to protrude rapidly, as the initial stages of germination are already fulfilled even under environmental stresses [10]. Seed priming helps the plants to cope with the adverse effects of unfavorable environmental conditions [11,12]. According to Liu [13], priming improves the activities of anti-oxidative metabolites, such as superoxide dismutase and peroxidase, during seed germination. Priming helps the plants to accelerate cell division, transport stored proteins and hasten the speed of seed germination [14]. Seed priming improved germination and seedling vigor in tomato [9] by activation of antioxidants [15], reduced membrane permeability, and maintenance of tissue water contents [6].

Exogenous application of priming agents to the seeds have remarkable role for pre-sowing accomplishment of germination phases [16]. Sliwinska [17] reported that 42% of primed tomato root tip cells were arrested in the G2-phase of mitosis and did not complete cell division. Previous studies revealed the positive role of potassium nitrate (KNO3) as a seed priming agent on seedling establishment and vigor [18]. In addition, considerable increase in germination potential and seedling vigor was observed in tomato seed treated with KNO<sup>3</sup> at the concentration of 50 mmol [19]. Similarly, exogenous KNO<sup>3</sup> treatment on rice seeds concurrently improved multiple aspects of germination and physiology. This implies that KNO<sup>3</sup> might play a signaling role in prompting a wide adaptation of rice seedlings [20]. Therefore, the present study was conducted to evaluate the effects of exogenously applied KNO<sup>3</sup> (0.25%, 0.50%, 0.75%, 1.0%, and 1.25%) as seed priming agent in two different tomato cultivars.

#### **2. Materials and Methods**

#### *2.1. Seed Source*

Six months old seeds of the pure line tomato cultivars "Sundar" and "Ahmar" (oval shaped fruit with regular leaves) were obtained from Ayyub Agricultural Research Institute, Vegetable Research Section, Faisalabad-38000, Punjab, Pakistan. The initial germination and seed moisture content before seed treatment were (86% and 10.5% in "Sundar"; 84% and 11% in "Ahmar"), respectively.

#### *2.2. Seed Priming Treatments*

Tomato seeds were primed/imbibed with 0.25%, 0.50%, 0.75%, 1.0%, and 1.25% (weight/volume) KNO<sup>3</sup> for 24 h at 25 ◦C. Pre-weighed seeds (5 g) were imbibed on two blotter papers in 9-cm diameter Petri dishes with appropriate concentration of KNO<sup>3</sup> solutions, followed by covering of dishes with aluminum foil. For aeration, a hole was provided in the center of each Petri dish. After each treatment, seeds were rinsed thoroughly with distilled water and dried back closer to original moisture level under shaded conditions. Nonprimed tomato seeds were maintained as control for comparison.

#### *2.3. Experimental Site and Conditions*

Growth chamber and greenhouse experiments were conducted at the research station of the University of Agriculture, Faisalabad, Punjab, Pakistan (30.37◦ N, 69.34◦ E) from 29 October 2019 to 1 December 2019. Well pulverized soil was collected from the field of the research station, and each plastic tray 35 cm × 25 cm × 15 cm in size was filled with 6 kg of soil. The textural class of soil was sandy loam having pH (6.8), electric conductivity (0.396 dS m−<sup>1</sup> ), available phosphorus (17.67 ppm), and potassium (353.96 ppm). After leveling the soil surface in each tray, moisture was applied up to

field capacity. In each tray, 30 seeds were sown with equal distance in the soil in both experiments and considered as one replicate. Both experiments were laid out in a completely randomized design with four replications. For growth chamber screening (optimal germination and growth conditions), all the trays were placed in the growth chamber with an optimal temperature of 25 ◦C and a light period of 12 h. The relative humidity during the complete execution of the growth chamber experiment was maintained at 65%. For the greenhouse experiment (suboptimal conditions), all the trays were placed in the greenhouse under natural environmental conditions. The climate data during the complete execution of greenhouse experiment is given in Figure 1.

**Figure 1.** Microclimate conditions inside the greenhouse during the experiment at research station of University of Agriculture, Faisalabad, Punjab, Pakistan.

#### *2.4. Seedling Establishment*

Seedling emergence was recorded daily and recorded when the hypocotyl came above the soil surface. Final emergence, expressed on a percentage basis, was calculated as the ratio among number of emerged seedlings and total number of seeds sown at the end of the experiment [21]. Mean emergence time (days) was recorded as per the equation earlier reported by International Seed Testing Association (ISTA) [22]:

$$\text{Mean Energy Time (MET)} = \frac{\sum \text{Dn}}{\sum \text{n}}.$$

where n is the number of seeds which emerged on day D, and D is the number of days counted from the beginning of emergence.

#### *2.5. Seedling Vigor*

Thirty days after sowing (DAS) plant height was determined on 5 randomly selected seedlings. On the same date, both fresh and dry weight of tomato plants were recorded. For dry weight, plants were dried at 70 ◦C till constant weight in an oven.

#### *2.6. Physiological Variables*

At 30 DAS, i.e., with the plants at the stage of 6 true leaves, measurements of CO<sup>2</sup> index (µmol mol−<sup>1</sup> ), net photosynthetic rate (µmol CO<sup>2</sup> <sup>m</sup>−<sup>2</sup> s −1 ), and transpiration rate (µmol H2O m−<sup>2</sup> s −1 ) were made on a fully expanded leaf from the top of the plant canopy by using an open system LCA-4 (ADC BioScientific Ltd., Hoddesdon, UK) portable infrared gas analyzer. Measurements were made between 6:00 a.m. and 7:00 a.m., with the following specifications: ambient pressure (P) 99.95 kPa, leaf chamber molar gas flow rate (U) 251 µmol s−<sup>1</sup> , molar flow of air per unit leaf area (Us) 221.06 mol m−<sup>2</sup> s −1 , temperature of leaf chamber (Tch) varied from 39 to 44 ◦C, Photosynthetically active radiation (PAR) at leaf surface was maximum up to 918 µmol m−<sup>2</sup> , and leaf chamber molar gas flow rate (U) 251 µmol s−<sup>1</sup> .

#### *2.7. Biochemical Variables*

To determine total phenolics, leaves were ground in liquid nitrogen by using pestle and mortar and a 20 µL sample was mixed with 1.60 mL distilled water, 100 µL Folin-Ciocalteu reagent (2N), and 300 µL sodium carbonate solution in a test tube [23]. After 30 min at 40 ◦C in water bath, test tubes were immediately moved to an ice box and absorbance recorded at 765 nm with a spectrophotometer (UV 4000). The total soluble sugars (TSS) in leaf samples were determined by the anthrone method [24]. Ground leaf sample (25 mg) was mixed with 5 mL of 2.5NHCl in a test tube. Tubes were placed in water bath 100 ◦C for 3 h, followed by cooling of tubes at room temperature. By using distilled water, the volume of tube was made to 100 mL and centrifuged at 4000 rpm for 10 min. After that, 0.5 mL supernatant, 0.5 mL distilled water, and 4 mL anthrone (0.2% v/v anthrone on 95% sulfuric acid) was taken in another tube. The tube was heated again in boiling water bath for 8 min. The tube was cooled rapidly and reading was taken at 630 nm by using spectrophotometer (UV 4000).

#### *2.8. Statistical Analysis*

Collected data were subjected to a two-way analysis of variance (ANOVA) (2 genotypes × 6 KNO<sup>3</sup> level) and Tukey's honest significance difference (HSD) test for means comparison at 5% significance level, using the analytical software package 'Statistix 8.1'.

#### *2.9. Greenhouse Microclimate Conditions During the Trial*

During the experiment, the average mean temperature was 19.2 ◦C, with a sharp decrease from 22 to 14 ◦C (on 27 and 31 DAS, respectively), whereas average minimum and maximum temperatures oscillated between 12–19 ◦C and 17–30 ◦C, respectively (Figure 1). The average relative humidity varied between 49% and 78%, with the lowest value recorded at 22 DAS and highest one at 33 DAS (Figure 1).

#### **3. Results**

#### *3.1. Growth Chamber Screening*

#### 3.1.1. Seedling Establishment

Seedling establishment of tomato includes seedling emergence (%) and the number of days required by seeds to germinate (mean emergence time—MET). The results of the present study indicated that the seed priming with KNO<sup>3</sup> improved the stand establishment of tomato (cv. "Sundar" and "Ahmar") grown in growth chamber. Tomato seeds primed with 0.75% KNO<sup>3</sup> had maximum emergence rate in both cultivars (98% "Sundar"; 99% "Ahmar"), so they were showing better performances when compared to the other treatments. Minimum germination (82% "Sundar"; 84% "Ahmar") was observed in nontreated seeds of tomato, while, in the case of MET, the maximum number of days (4–5) was observed in nontreated seeds of tomato. The seeds treated with 0.75% KNO<sup>3</sup> germinated earlier than all other treatments (Table 1).


**Table 1.** Final emergence (%) and mean emergence time (days) of tomato seedlings as affected by the cultivar and seed priming with potassium nitrate (KNO<sup>3</sup> ), under two different growth conditions.

Values sharing the same letters are non-significantly different (*p* ≤ 0.05).

#### 3.1.2. Seedling Vigor

Statistical analysis of data about seedling vigor revealed that the effect of seed priming treatments was significant in both "Sundar" and "Ahmar". All priming treatments significantly improved the seedling length in both cultivars, whereas the cultivar did not exert any significant effect. Maximum seedling length was achieved in tomato seed primed with 0.75% (8.36 and 8.35 cm in "Sundar" and "Ahmar", respectively), followed by 1% KNO<sup>3</sup> solution (7.5 and 7.8 cm), whereas the lowest values for seedling length in both cultivars was observed in control (5.2 and 5.3 cm). In both cultivars, plants raised from seeds treated with 0.75% KNO<sup>3</sup> showed higher values for seedling fresh weight (35.1 and 37.6 mg in "Sundar" and "Ahmar", respectively) and dry weight (17.9 and 19.1 mg) as compared to other treatments (Table 2).

**Table 2.** Seedling length, shoot fresh weight, and shoot dry weight of tomato as affected by seed priming with KNO<sup>3</sup> under two different growth conditions.


Values sharing the same letters are non-significantly different (*p* ≤ 0.05).

#### 3.1.3. Physiological and Biochemical Attributes

The results shown in Tables 3 and 4 indicate that seed priming treatments significantly improved the physiological and biochemical attributes of both tomato cultivars, while the genotype effect was also found significant. The highest photosynthesis rate, transpiration rate, and CO<sup>2</sup> index were linked to tomato seeds treated with 0.75% KNO3, whereas the lowest values were found in the nonprimed seeds (Table 3). A statistical evaluation of data demonstrated that total soluble sugars and phenolic contents were significantly influenced by seed priming treatments. Though all the seed priming treatments proved successful to improving these biochemical attributes, highest values were observed in plants deriving from seeds primed with 0.75% KNO<sup>3</sup> under growth chamber screening (Table 4).


**Table 3.** Variations in physiological attributes of two tomato cultivars under the influence of seed priming with KNO<sup>3</sup> in two different growth conditions.

Values sharing the same letters are non-significantly different (*p* ≤ 0.05).



Values sharing the same letters are non-significantly different (*p* ≤ 0.05).

#### *3.2. Greenhouse Screening*

#### 3.2.1. Seedling Establishment

Seed priming treatments improved the final emergence of both tomato cultivars under greenhouse conditions. The highest emergence values were recorded in tomato seed primed with 0.75% KNO<sup>3</sup> (93.29% and 96.68% in "Sundar" and "Ahmar", respectively), whereas the lowest ones were found in the nonprimed seeds. Seed priming with 1% KNO<sup>3</sup> proved to improve the final emergence of both cultivars too (88.7% in "Sundar" and 90.1% in "Ahmar") (Table 1). In both cultivars, no variation in final emergence was observed among experimental units receiving tomato seed primed 0.50% and 1% KNO3. However, when compared to the other treatments, the lowest MET value was recorded in tomato seeds primed with 0.75% followed by 1% KNO3. Besides, both nontreated cultivars showed the highest values for MET (Table 1).

#### 3.2.2. Seedling Vigor

Seedling length of both tomato cultivars is presented in Table 2, and data revealed that maximum seedling length in both cultivars was achieved in tomato seed primed with 0.75% (7.0 and 7.1 cm in "Sundar" and "Ahmar", respectively), followed by 1.25% KNO<sup>3</sup> solution (6.1 and 6.2 cm in "Sundar" and "Ahmar", respectively), whereas the lowest ones were recorded in the control (3.8 and 3.8 cm in "Sundar" and "Ahmar", respectively). Seed priming with KNO<sup>3</sup> also proved effective in improving the seedling fresh and dry weight; nonetheless, the effect of different cultivars was not pronounced. Plants in both cultivars raised from seeds treated with 0.75% KNO<sup>3</sup> showed higher values for seedling fresh (30.3 mg in "Sundar" and 32.1 mg in "Ahmar") and dry weight (14.6 and 14.4 mg, respectively) as compared to all other treatments. No significant difference in seedling fresh and dry weight was observed among tomato seed treated with 0.50% and 1.0% KNO<sup>3</sup> in both cultivars (Table 2).

#### 3.2.3. Physiological and Biochemical Variables

Seed priming treatments improved the physiological and biochemical of both tomato cultivars under greenhouse conditions. Higher values for photosynthetic rate, transpiration rate, and CO<sup>2</sup> index were recorded in experimental units receiving tomato seed primed with 0.75% KNO<sup>3</sup> as compared to control (Table 3), while in the case of genotypes, "Ahmar" showed better photosynthetic rate and transpiration rate as compared to "Sundar". Seed priming with 1% KNO<sup>3</sup> improved the physiological attributes of both cultivars, too. No variation in physiological attributes was observed among experimental units receiving tomato seed primed 0.50% and 1% KNO<sup>3</sup> in "Ahmar". In the same way, maximum total soluble sugars were observed in tomato seeds primed with 0.75% followed by 1% KNO<sup>3</sup> (Table 4). The lowest values for phenolic contents were recorded in control in both cultivars (Table 4).

#### **4. Discussion**

#### *4.1. Seedling Establishment*

Tomato seed priming with KNO<sup>3</sup> affected the emergence of seedling and the speed of seed germination. Major events in other literature on priming includes metabolic changes, such as repair of DNA and increases in the biosynthesis of RNA [25], and enhancement in the respiration process of seed [26]. This indicates that the time of seed imbibition is very important for seed priming. For the study of seed priming of tomato with different levels of KNO3, it is important to know about the emergence percentage and mean emergence time. The results of the present study indicate that the performance of both tomato cultivars primed with 0.75% KNO<sup>3</sup> was appreciable in growth chamber, as well as in greenhouse screening, meaning that this effect was still appreciable under suboptimal growth conditions. The pattern of seedling emergence and mean emergence time were almost the same in both cultivars, as well as in both growth environments (growth chamber and greenhouse). The time of water intake by the seed during priming can vary within the cultivars, which can affect the performance of the seed priming agent (KNO3) [27]; similarly, in our study, the difference between the performance of both cultivars were seen.

The data shown in Table 1 indicated that priming of tomato seeds with 0.75% KNO<sup>3</sup> was better than other treatments in terms of final emergence and mean emergence time. Our study is in correspondence with another study that revealed that the emergence percentage of wheat seeds was decreased when they were primed with >1% KNO<sup>3</sup> [28]. This indicates that KNO<sup>3</sup> concentration above a certain threshold may not be appropriate to boost seed germination. Seed priming with 1% KNO<sup>3</sup> was found useful in terms of emergence percentage in sorghum [29] and rice [30]. Besides, soybean seed primed with 1% KNO<sup>3</sup> for 1 day enhanced the emergence percentage as compared to nontreated seeds, both in laboratory and field experiments [18].

## *4.2. Seedling Vigor*

Seedling vigor is the combined result of the emerged seeds under a wide range of biotic and abiotic stresses. Seedling vigor is not a single measurable entity, but it is a sum of many growth parameters, such as seedling length, seedling fresh weight, and seedling dry weight [22]. Maximum vigor was observed when seed priming with 0.75% KNO<sup>3</sup> was done. Our study is in line with another study in which seedling vigor of wheat was improved by priming with KNO<sup>3</sup> [28]. Similar results were found in corn when the priming of seed was done with 1% KNO<sup>3</sup> [31]. Our findings are similar to other studies, in which the shoot length of watermelon and tomato were increased by the seed priming with KNO<sup>3</sup> [32,33]. Seed priming with 0.5% and 1% KNO<sup>3</sup> improved the vegetative growth of watermelon [34] and tomato [35], respectively, under salt stress. Seed priming with KNO<sup>3</sup> can cause a significant increase in seedling vigor of the wheat crop as compared to hydro-priming or dry broadcasting [36].

#### *4.3. Physiological and Biochemical Attributes*

Plant growth is based mainly upon photosynthesis, while its performance is mostly dependent on the opening/closing of stomata, which modulates photosynthetic rate, respiration rate, and CO<sup>2</sup> index [37–39]. The results of the present study revealed that the maximum photosynthesis rate, transpiration rate, and CO<sup>2</sup> index was observed in tomato plants grown by seeds primed with 0.75% KNO3, compared to other priming treatments. Our study is in corroboration with another study in which the increased photosynthetic rate, respiration rate, and CO<sup>2</sup> index of cucumber seedlings as the result of seed priming with KNO<sup>3</sup> were reported. The photosynthesis rate of the seedlings has a positive correlation with the growth of seedling [40]. The results of the present study revealed that the biochemical attributes, e.g., total soluble sugars and phenolic content, of tomato plants were enhanced by seed priming with KNO3. The maximum increase was observed when seeds were treated with 0.75% KNO3, while minimum values were seen in nonprimed seedlings. Previous studies expressed that seed priming with KNO<sup>3</sup> significantly improved the biochemical indices of chicory [41] and rice [20].

#### **5. Conclusions**

The performance of tomato is diminished by the poor quality of seed. Therefore, the present study was conducted to improve the quality of tomato seed by priming with KNO3. The results presented in this paper revealed that tomato seeds of both cultivars primed with 0.75% KNO<sup>3</sup> proved to be successful for improving seedling establishment and vigor, as well as physiological and biochemical attributes, under growth chamber and greenhouse conditions. The present study provides the direction towards further molecular investigation related to the seed priming of tomato.

**Author Contributions:** M.M.A. and T.J. conceptualization, conceived and data analysis and original draft preparation. R.S., I.A. and A.F.Y. helped in data analysis. R.P.M. conceptualization, data curation and editing the manuscript. All authors have read and approved the final manuscript.

**Funding:** This research received no external funding.

**Acknowledgments:** The authors thank the valuable contributions for data collection provided by Ahmed Mukhtar. Helpful suggestions were provided by Hafiz Sohaib Ahmed Saqib for data analysis.

**Conflicts of Interest:** The authors declare no conflict of interest.

## **References**


**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **Zinc Seed Priming Improves Spinach Germination at Low Temperature**

**Muhammad Imran 1,2, Asim Mahmood <sup>3</sup> , Günter Neumann <sup>3</sup> and Birte Boelt 1,\***


**Abstract:** Low temperature during germination hinders germination speed and early seedling development. Zn seed priming is a useful and cost-effective tool to improve germination rate and resistance to low temperature stress during germination and early seedling development. Spinach was tested to improve germination and seedling development with Zn seed priming under low temperature stress conditions. Zn priming increased seed Zn concentration up to 48 times. The multispectral imaging technique with VideometerLab was used as a non-destructive method to differentiate unprimed, water- and Zn-primed spinach seeds successfully. Localization of Zn in the seeds was studied using the 1,5-diphenyl thiocarbazone (DTZ) dying technique. Active translocation of primed Zn in the roots of young seedlings was detected with laser confocal microscopy. Zn priming of spinach seeds at 6 mM Zn showed a significant increase in germination rate and total germination under low temperature at 8 ◦C.

**Keywords:** spinach; Zn priming; multispectral imaging; Zn localization; abiotic stress

**Citation:** Imran, M.; Mahmood, A.; Neumann, G.; Boelt, B. Zinc Seed Priming Improves Spinach Germination at Low Temperature. *Agriculture* **2021**, *11*, 271. https:// doi.org/10.3390/agriculture11030271

Academic Editor: Alan G. Taylor

Received: 20 February 2021 Accepted: 17 March 2021 Published: 22 March 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

#### **1. Introduction**

Spinach (*Spinacia oleracea* L.) is an annual crop, usually sown in early spring. Low soil temperature during early spring is one of the major factors affecting seed germination of various crops. In spinach, seed germination and early seedling establishment are inhibited at low temperature [1]. Imbibition and rehydration of dry seeds is a critical process during germination, and rapid absorption of water can cause severe membrane damage, leading to leakage of electrolytes, sugar, and amino acids [2]. Wuebker et al. [3] and Bochicchio et al. [4] reported embryo membrane damage, directly linked to the speed of imbibition at low seed moisture. Under low temperature conditions, these problems can be even more severe due to limited ability of cell membranes to maintain the integrity that is required during imbibition [5] and may compromise germination performance, leading to low germination rate, low uniformity, and final stand establishment. Spinach seeds germinate best between a range of temperatures between 15 and 24 ◦C. The germination speed and/or rate varies with the change in temperature. Germination speed is very slow at a temperature just above freezing. It may take up to three weeks for germination at 5 ◦C, compared to one week at 20 ◦C.

Seed priming is a pre-sowing seed treatment, in which seeds are soaked in water and dried back to storage moisture contents for later use. According to Harris et al. [6], the 'on-farm seed priming' method has become very popular in developing countries. With the 'on-farm seed priming' method, seeds are soaked in a water or nutrient solution and air dried (not to storage moisture contents) prior to sowing. This can speed up germination, improve the tolerance to various stress conditions, and increase crop yield [7].

Chen and Arora [8] have proposed a hypothetical model demonstrating the cellular physiology of priming-induced stress-tolerance, likely achieved via two strategies. First,

seed priming activates germination-related processes (e.g., respiration, endosperm weakening, and gene transcription and translation, etc.) that facilitate the transition of quiescent dry seeds into the germinating state, which improves germination potential. Secondly, priming imposes abiotic stress on seeds that repress radicle protrusion but stimulate stress responses (e.g., accumulation of Late Embryogenesis Abundant proteins (LEAs), potentially inducing cross-tolerance. The authors suggest that these two strategies constitute a "priming memory" in seeds, which mediates greater stress-tolerance of germinating primed seeds after the exposure to various stress conditions.

Stored reserves are the primary source of mineral nutrients during seed germination and early growth and should be adequate to sustain the seedling until the root system mediates nutrient uptake from the soil. Stored mineral nutrients are vital, particularly when seedlings are exposed to conditions of nutrient limitation [9]. In barley and wheat, seeds with low Zn contents showed delayed germination and poor seedling vigor, which negatively affected plant growth and final grain yield [10–12]. In wheat, seeds with high Zn concentrations produced better stand establishment, and seedlings were able to take up more Zn under Zn-deficient soil conditions as compared to plants established from seeds low in Zn seed reserves [13]. During germination and early seedling development, particularly under stress conditions, micronutrients are essential. Zinc is a co-factor of various enzymes (superoxide dismutase (SOD)) involved in the detoxification of reactive oxygen species, such as O<sup>2</sup> <sup>−</sup> (superoxide radical) and H2O<sup>2</sup> (hydrogen peroxide) [14]. Zn is directly involved in membrane stabilization, biosynthesis of auxins [15] and gibberellins [16] in plant growth regulation, and protein synthesis in general.

In "nutrient seed priming", seeds are soaked in a nutrient solution instead of pure water to improve seed nutrient contents in combination with the priming effect, which improves germination and seedling establishment. Ashraf and Rauf [17] found that priming maize seeds with CaC1<sup>2</sup> improved final germination, rate of germination, and fresh and dry biomass of plumules and radicles, compared to untreated control and water-primed seeds under salt stress. Maize seed priming with 1% ZnSO<sup>4</sup> enhanced plant growth and increased final grain yield and Zn content of harvested seed from plants grown on soil with low Zn availability [18]. It has also been shown in maize [19] and rice [20] that primed Zn is translocated to growing shoots during germination and early seedling development. Furthermore, Imran et al. [21] also showed increased maize grain yield via Zn seed priming under low Zn-available soils combined with low temperature climatic conditions.

Based on the findings of seed priming memory in invoking seed stress tolerance [8] and the role of Zn seed priming in stress tolerance in crop plants, this study investigated the functions of water- and Zn-priming of spinach seeds under low temperature. The multi-spectral imaging technique was used to monitor the Zn priming of spinach seeds, and confocal-laser microscopic analysis was performed to study Zn translocation in young spinach seedlings.

#### **2. Materials and Methods**

#### *2.1. Seed Material and Priming*

Commercially available spinach seed (*Spinacia oleracea* L. *cv* Matador) was obtained from the seed company Vikima Seeds A/S, Denmark. Seeds were primed for 24 h with water and ZnSO4·7H2O, according to Imran, Mahmood, Römheld, and Neumann [21], with some modifications, in which seeds after priming were surface dried at room temperature (20 ◦C) for 24 h before the germination test.

#### *2.2. Optimal Zn Concentration Levels for Seed Priming (Experiment 1)*

To determine the optimal Zn concentration for seed priming, 10 g of spinach seeds were soaked in 100 mL of ZnSO4·7H2O solution. Concentrations of Zn in the priming solutions were e.g., 0 (deionized H2O), 1, 2, 4, 6, 8, and 10 mM Zn solutions. Unprimed seeds were used as control treatment. A germination test of primed seeds was performed using the top of paper method at 12 h light and 12 h dark periods at 15 ◦C. Seeds were germinated in petri dishes with four replicates of 25 seeds per treatment. Seed germination data were recorded at Day 7 and Day 14. The seeds with radicle protrusion > 2 mm were considered germinated.

#### *2.3. Germination Test of Water- and Zn-Primed Seeds at Low Temperature (Experiment 2)*

Based on the results of experiment 1, unprimed control, water-priming, and two levels of Zn concentrations were selected as seed treatments to test seed germination at two different temperatures, 8 ◦C (low temperature) and 15 ◦C (optimal temperature). The germination test was performed as mentioned above in experiment 1.

#### *2.4. Mineral Analysis of Seeds and Young Seedlings*

After the priming treatments, seeds were rinsed with deionized water for 1 min to remove any compounds and nutrients adhering to the seed coat before the analysis of seed mineral nutrients. Furthermore, mineral nutrients were also determined in young spinach seedlings. For this purpose, shoots and roots were separated in all treatments. To measure Zn concentration in seeds and seedlings, after drying at 65 ◦C, ground samples were ashed in a muffle furnace at 500 ◦C for 5 h. After cooling, the samples were extracted twice with 2 mL of 3.4 M HNO<sup>3</sup> (*v*/*v*) and subsequently evaporated to dryness. The ash was dissolved in 2 mL of 4 M HCl, subsequently diluted 10-fold with hot deionized water, and boiled for 2 min. After adding 0.1 mL of Cs/La buffer to 4.9 mL of ash solution, Zn and Mn concentrations were measured by atomic absorption spectrometry (UNICAM 939, Offenbach/Main, Germany).

#### *2.5. Confirmation of Zn Accumulation in Spinach Seeds after Zn Priming*

Staining of seed Zn was performed using 1,5-diphenyl thiocarbazone (DTZ), according to Ozturk et al. [22]. For this purpose, water- and Zn- (6 mM Zn) primed seeds were incubated with 500 mg L−<sup>1</sup> DTZ at room temperature for 30 min. The stained seeds were rinsed with deionized water and images were taken with a high-resolution digital camera. To determine the localization of primed Zn in the different seed tissues, fresh water and Zn-primed (6 mM Zn, rinsed with deionized H2O for 20 s) seeds were immediately fixed in NEG 50TM gel. The fixed seeds were dissected to 20 µm thick slices with a freezing microtome (MICROM HM 550, Microm International GmbH, Walldorf, Germany) and placed on microscope slides. Afterwards, 2 µL of 500 mg L−<sup>1</sup> DTZ solution was applied to the specimen to stain with Zn. After 3 min, a few drops of deionized H2O were applied before placing the coverslip on the thinly sliced sample. Photos were taken with a light microscope (Axiovert 200, Carl Ziess Microscopy GmbH, Göttingen, Germany).

To test the characteristic properties of the red color of DTZ stained spinach seeds (water- and Zn-primed), multi-spectral images of water- and Zn-primed seeds were taken according to Shrestha et al. [23]. Images from each seed sample were acquired using a VideometerLab instrument (Videometer A/S Herlev, Denmark). In this instrument, a topmounted camera acquires multispectral images with the help of 19 light emitting diodes (LEDs) at 19 wavelengths (375, 405, 435, 450, 470, 505, 525, 570, 590, 630, 645, 660, 700, 780, 850, 870, 890, 940, and 970 nm). Prior to image acquisition, the instrument was calibrated with respect to color, geometry, and self-illumination to ensure directly comparable images. After images were obtained, VideometerLab software (version 2.13.83) was used to extract and transform pixel data.

#### *2.6. Translocation of Primed Zn in the Roots of Spinach Seedlings*

The localization of Zn in the roots of 10-day old spinach seedlings was examined by using Zinpyr-1 (C46H36Cl2N6O5) fluorescence dye, according to Sinclair et al. [24]. For this purpose, unprimed, water- and Zn-primed (6 and 10 mM) spinach seeds were germinated in filter paper towels at 15 ◦C. Zinpyr-1 was dissolved in dimethyl sulphoxide (DMSO) to make a 1 mM stock solution and stored at −20 ◦C. For root incubation, a working solution of 20 µM Zinpyr-1 was prepared from the stock solution. From each treatment, 10-day old equally grown spinach seedlings were selected for Zinpyr-1 incubation.

−2

Before immersing into a Zinpyr-1 working solution, seedlings were washed alternatively three times in deionized water and 10 mM ethylene-diamine-tetra-acetic acid (EDTA). Seedlings were incubated in Zinpyr-1 solution for 5 h at room temperature in the dark. Afterwards, roots of the incubated seedlings were rinsed again in deionized water to remove the Zinpyr-1 dye from the root surface, immersed in 75 µM propidium iodide to stain cell walls red, and rinsed again. For negative control, roots of water and Zn-primed seedlings were immersed in a Zn-chelator, N,N,N',N-tetrakis(2-pyridylmethyl) ethylenediamine (TPEN), for 30 min. Samples were mounted in 0.9% saline, and images were taken on an Olympus (Hamburg, Germany) confocal laser-scanning microscope (CLSM), using excitation at 488 nm with a 100 mW Ar ion. chelator, N,N,N',N

#### *2.7. Statistical Analysis*

Data on final germination were analysed by one-way analysis of variance (ANOVA) using SigmaStat 3.5 Software. Significant differences between the means were calculated at *p* < 0.05 and marked with different letters.

Differences between means of Zn and Mn concentrations, reflectance, and speed of germination were compared using the standard error (SE) of four replicates (25 seeds in each replicate).

#### **3. Results**

#### *3.1. Optimal Zn Concentration*

At first count after 7 days (Figure 1a), seed germination performance was increased by all priming treatments, as compared to the unprimed control. Water-primed seeds showed 10% higher germination compared to the untreated control. However, at Day 7, none of the Zn-priming and water priming treatments showed statistical difference in germination, but Zn-priming at 1 mM, 6 mM, and 10 mM had almost 22%, 20%, and 18%higher germination, respectively, as compared to the unprimed control.

**Figure 1.** Germination percentage (%) of spinach seeds at Day 7 (**a**) and Day 14 (**b**) after sowing on top of paper in Petri dishes at 15 ◦C. Seed priming treatments included control (unprimed), water- primed, and Zn-primed at various Zn concentrations (1, 2, 4, 6, 8, and 10 mM Zn) in the priming solution. Bars represent the mean and standard error (SE) of four replicates (25 seeds in each replicate). Significant differences between the means were calculated at *p* < 0.05 and marked with different letters.

> At final count on Day 14 (Figure 1b), compared to the first count, differences in % germination between unprimed, water-primed, and most of the Zn-priming treatments were not significant. Compared to other Zn priming treatments, priming at 6 mM Zn

concentration showed significantly higher germination as compared with the control and water-primed treatments. Based on the results shown in Figure 1b, seed priming treatments at 6 mM and 10 mM Zn concentrations, together with unprimed and water-priming, were selected for testing germination at a low temperature.

#### *3.2. Zinc Status of Seeds and Young Seedlings*

Zinc priming largely increased seed Zn concentration at 6 mM and 10 mM priming treatments compared to unprimed and water-primed treatments. There was approximately an increase of 30 and 48 times in seed Zn concentration, with 6 mM and 10 mM priming solutions, respectively, as shown in Table 1. Concentration of Mn was decreased up to 25% in seeds after Zn priming treatments but concentration of Mn in shoots and roots was not affected by Zn seed priming. Mineral analysis of shoots and roots of young spinach seedlings showed a 4 to 10 time increase in shoot and root Zn concentrations after both Zn-priming treatments compared to unprimed and water primed.

**Table 1.** Zinc and Mn concentrations in spinach seeds, shoots and roots of young seedlings after water, and Zn seed priming. Values represent the mean and standard error (SE) of four replicates.


#### *3.3. Detection of Zn Primed Spinach Seeds with VideoMeter Lab and DTZ Staining*

To confirm the addition of Zn in spinach seeds via Zn priming, the DTZ staining method was used. Images of water-, 6 mM, and 10 mM Zn-primed seeds (Figure 2a–c, respectively) showed no visible differences in all three treatments. Compared to normal images, multispectral images of the same samples taken with VideometerLab (Figure 2d–f), viewed at a wavelength of 700 nm in hot view mode, revealed more uniform and higher absorption of red color.

Higher development of red color in 6- and 10-mM Zn-primed seeds after DTZ staining indicates a higher level of Zn compared to water primed seeds. Differences in the mean spectrum of DTZ stained water- and Zn-primed seeds are shown in Figure 3.

Other spectrum characteristics data (Table 2), such as CIELab, intensity, hue, and saturation, also revealed the development of a darker red color in Zn-primed seeds after DTZ staining, as compared to water priming.

**Table 2.** Characteristic color parameters: CIELab L, intensity, hue, and saturation of DTZ-stained spinach seeds after water-, 6 mM, and 10 mM Zn-priming. Values represent the mean and standard error (SE) of four replicates. Significant differences between the means were calculated between seed priming treatments at *p* < 0.05 and marked with different letters.


– **Figure 2.** Visible spectrum images of 1,5-diphenyl thiocarbazone (DTZ)-stained water (**a**), 6 mM (**b**), and 10 mM (**c**) Zn-primed seeds. Images (**d**–**f**) (water-, 6 mM, and 10 mM Zn-primed seeds, respectively) taken with VideometerLab at wavelength 700 nm in hot view mode.

**Figure 3.** The mean visible spectrum of water-, 6 mM, and 10 mM Zn-primed spinach seeds extracted from the multi-spectral images of seeds at 19 wavelengths (375, 405, 435, 450, 470, 505, 525, 570, 590, 630, 645, 660, 700, 780, 850, 870, 890, 940, and 970 nm). Bars represent the mean and standard error (SE) of four replicates (25 seeds in each replicate).

#### *3.4. Zinc Localization in the Roots of Spinach Seedlings*

Laser confocal microscopy was employed to detect Zn (in vascular tissues) in the roots of 10-day-old spinach seedlings, germinated from water-, 6 mM, and 10 mM Zn-primed seeds, (Figure 4a,b). Higher intensities of Zinpyr-1 fluorescence in 6 mM and 10 mM (Figure 5d,e, respectively) compared to unprimed and water-primed seedlings (Figure 5a,b,

respectively) indicates higher accumulation of Zn in the roots of Zn-primed seedlings compared to water-primed seedlings, irrespective of Zn concentration in Zn-primed seeds.

**Figure 4.** DTZ-staining of Zn in a spinach seed of water-primed (**a**) and Zn-primed (**b**) seedlings with 6 mM ZnSO<sup>4</sup> ·7H2O. Red staining indicates Zn localization, especially in the aleurone layer, endosperm, and pericarp.

N,N,N',N **Figure 5.** Confocal laser-scanning microscope images of spinach roots of (**a**) control (unprimed) seedlings, (**b**) waterprimed seedlings, (**d**,**e**) 6 mM and 10 mM Zn-primed seedlings, respectively, and (**c**,**f**) seedlings treated with 200 µM N,N,N',N-tetrakis (2-pyridylmethyl) ethylenediamine (TPEN) for 30 min and then exposed to 15 µM Zinpyr-1 for 5 h.

#### *3.5. Seed Germination at 15* ◦*C and 8* ◦*C*

Spinach seeds started germination 3 days after sowing. Figure 6a,b represents seed germination at 15 ◦C and 8 ◦C, respectively. At 15 ◦C, there was no statistical difference for germination speed between water- and Zn-primed seeds, but all priming treatments showed a significant increase in germination compared to unprimed seeds. These differences were diminished after Day 6. At the final count (Figure 6a), Day 14, Zn-priming reflected relatively higher total germination but differences were not significant compared to unprimed and water-primed treatments.

N,N,N',N

**Figure 6.** Germination speed of control, water-, 6 mM, and 10 mM Zn-primed spinach seeds germinated at 15 ◦C (**a**) and 8 ◦C (**b**) in Petri dishes. Germination was monitored daily up to 7 days after sowing. Data presented are the means of four replicates (25 seeds in each replicate) with standard errors.

At 8 ◦C, Zn-priming at 6 mM Zn concentration showed a significantly higher germination speed compared to unprimed, water-, and 10 mM Zn-priming treatments (Figure 6b). However, there was no significant difference in germination speed between water- and 10 mM Zn-priming treatments, but both treatments showed significant increases compared to unprimed seeds. At final count (Figure 7b), 6 mM Zn-priming showed a significant increase in total germination (>10%) compared to unprimed and water-primed seeds. There was no significant difference between unprimed, water-, and 10 mM Zn-primed treatments. Final germination recorded on Day 14 showed no significant effect of treatments when seeds were germinated at 15 ◦C (Figure 7a); however, at 8 ◦C, treatment with 6 mM Zn improved germination (Figure 7b).

**Figure 7.** Final germination (%) of spinach seeds after 14 days of sowing, germinated at 15 ◦C (**a**) and 8 ◦C (**b**) on top of paper in Petri dishes. Seed priming treatments include control (unprimed), water- primed, 6 mM, and 10 mM Zn in the priming solution. Bars represent the mean and standard error (SE) of four replicates (25 seeds in each replicate). Significant differences between the means were calculated at *p* < 0.05 and marked with different alphabets.

#### **4. Discussion**

Adequate Zn contents in the seeds are essential for vigorous seedlings and resistance against different abiotic stress factors during germination and the early seedling development stage [25,26]. Zn seed priming has been used efficiently in various crops, e.g.,

maize [21,27], barley [28], rice [20], and soybean [29], to improve seed germination and resistance against various abiotic stress factors, such as drought, low root zone temperature, and nutrient deficiencies. However, Zn seed priming improves germination and seedling development, but determining the optimum concentration of Zn in the priming solution to attain the beneficial effects of Zn priming is very important. In the present study, Zn-priming of spinach seeds at 6 mM and 10 mM Zn concentrations showed the best effects on germination (Figure 1b). However, in the beginning (Figure 1a), Zn-priming at 1 mM Zn concentration showed significant increase in seed germination compared to unprimed seeds but that effect had disappeared at the final germination count. Previously, Ajouri et al. [28] and Prom-u-Thai et al. [20] also reported similar findings in barley and rice, respectively, where Zn seed priming showed beneficial effects on seed germination and early seedling development at certain Zn concentrations. Higher levels of Zn concentration in the priming solution may exert a toxic effect during the germination process and development of young seedlings. The results in the present study indicate the advantageous role of Zn seed priming in improving spinach seed germination, in particular during low temperatures. In the current study, final germination recording was performed on Day 14; however, according to International Seed Testing Associsation (ISTA) guidelines [30], final germination in spinach is scored on Day 21. Standard germination tests are performed using either 15 or 10 ◦C [30]. Since "low temperature" was chosen as 8 ◦C in this experiment, an even slower germination would be expected. This implies that seeds in the current study may not have reached full germination capacity, in particular those tested at low temperature.

Development of red color after staining spinach seeds with DTZ revealed an increase in seed Zn levels after Zn-priming treatment. DTZ is a Zn–chelating agent [31,32], which gives red color after binding with Zn. Additionally, it has been used to determine the localization of Zn in different organisms and crop seeds, such as algae [33], wheat seeds [22], and rice seeds [20]. In the present study, in the thick and relatively dark colored seed coat of spinach seeds, it was difficult to differentiate the intensity of the red color developed after DTZ staining in water- and Zn-primed seeds with ordinary camera images (Figure 2a–c). Recently, Shrestha et al. [23,34] successfully employed multi-spectral image analysis on tomato seeds to classify varietal differences based on the variation of spectral characteristics in seeds from different varieties. The mean spectrum of multi-spectral images of DTZ-stained seeds after water- and Zn-priming treatments (Figure 2d–f) showed a clear difference between water- and Zn-primed seeds, but reflectance at each wavelength exhibited a similar trend. The variation in spectra of water- and Zn-primed seeds after DTZ staining can be attributed to increased Zn in the spinach seeds. Development of red color in the inner parts of Zn-primed seeds (Figure 4b) revealed that primed Zn was not only absorbed in the outer pericarp or seed coat tissues, it was also accumulating in the inner tissues of Zn-primed seeds. Interestingly, Freitas et al. [35] found the highest accumulation of Zn in the embryo of the ZnO coated maize seed. This study employed micro-X-ray fluorescence spectrometry and micro-X-ray absorption near-edge spectroscopy for the mapping of Zn distribution and Zn speciation analysis.

Translocation of primed Zn in the young germinating spinach seedlings was determined by using a histochemical technique of Zn visualization based on the formation of the green-fluorescent complex with Zinpyr-1 (C46H36Cl2N6O5). This technique has been successfully applied in different plant species to detect Zn in different shoot and root tissues, for example Arabidopsis, [24] *Zea mays* [36], *Noccaea caerulescens,* and *Thlaspi arvense* [37]. As shown in Figure 5a,b, higher intensities of green fluorescence in the roots of Zn-primed seeds indicate active transport of primed Zn to the young growing roots. Previously, various authors [20,21,29] have also reported the translocation of primed Zn to the young growing roots during early seedling development, which also supports the seedling growth under Zn-deficient conditions. Zn is well known for its functions in plants under various biotic and abiotic stress conditions [38]. Our results suggest that active

translocation of primed Zn to growing roots can be useful in stress tolerance in spinach grown under various stress conditions.

Zn is important in the physiological functions, during germination and seedling development [25]. It is vital in the processes of protein synthesis and gene expression. For the structural and functional integrity of biological systems, almost 10% of proteins require Zn for their synthesis or functioning [39]. Production of reactive oxygen species (ROS) during seed germination is reported by numerous studies [40–42]. Increased oxidative stress is one of the rapid responses under all kind of stress conditions, including suboptimal or low temperatures. This is associated with increased production of ROS, such as superoxide, hydrogen peroxide, and the hydroxyl radical, involved in membrane damage by lipid peroxidation, protein degradation, enzyme inactivation, and disruption of DNA strands [43]. Micronutrients, such as Zn, are important co-factors of different enzymes involved in the detoxification of ROS, such as superoxide dismutase (SOD) [14,44]. In the present study, compared to unprimed and water-primed seeds, an increase in the germination speed and total germination at 8 ◦C by Zn priming can be attributed to increased Zn in the seeds. Increased localization of Zn in the roots of Zn-primed seedlings (Figure 5d,e) also suggested an active role of primed Zn against the adverse effects of low temperatures during germination, but the current study did not provide any further detail on this potential effect.

Concluding remarks: Previously, beneficial effects of increased seed Zn levels have shown to improve seed germination and early seedling establishment in stress conditions. This study demonstrates that increased Zn level of spinach seeds via Zn priming can enhance seed germination and seedling establishment under low temperature stress conditions. The physiological role of primed Zn in membrane stability, reduced oxidative stress and performance under field conditions, needs to be further elucidated and studied. Furthermore, the potential of combining Zn seed priming with agrochemicals, e.g., fungicides, should be evaluated as a tool to reduce pesticide use. Finally, the seed priming technique is simple, cost effective, and can be performed on farms before sowing, e.g., by small-scale farmers.

**Author Contributions:** Conceptualization, M.I. and B.B.; methodology, M.I.; software, M.I.; validation, M.I.; formal analysis, M.I., G.N. and A.M.; investigation, M.I.; resources, B.B.; data curation, M.I.; writing—original draft preparation, M.I.; writing—review and editing, M.I. and B.B.; visualization, M.I.; supervision, B.B.; project administration, B.B.; funding acquisition, B.B. All authors have read and agreed to the published version of the manuscript.

**Funding:** This material is based on work that was supported by the Innovation Fund Denmark grant number 110-2012-1, SpectraSeed and GUDP (Grønt Udviklings- og Demonstrationsprogram) grant number 34009-12-0528, and the Danish Agricultural Agency under the Ministry of Environment and Food of Denmark.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


## *Review* **The Use of Multispectral Imaging and Single Seed and Bulk Near-Infrared Spectroscopy to Characterize Seed Covering Structures: Methods and Applications in Seed Testing and Research**

**Anders Krogh Mortensen , René Gislum , Johannes Ravn Jørgensen and Birte Boelt \***

Department of Agroecology, Faculty of Technical Sciences, Aarhus University, Forsøgsvej 1, 4200 Slagelse, Denmark; anmo@agro.au.dk (A.K.M.); rg@agro.au.dk (R.G.); jrj@agro.au.dk (J.R.J.) **\*** Correspondence: bb@agro.au.dk

**Abstract:** The objective of seed testing is to provide high-quality seeds in terms of high varietal identity and purity, germination capacity, and seed health. Across the seed industry, it is widely acknowledged that quality assessment needs an upgrade and improvement by inclusion of faster and more cost-effective techniques. Consequently, there is a need to develop and apply new techniques alongside the classical testing methods, to increase efficiency, reduce analysis time, and meet the needs of stakeholders in seed testing. Multispectral imaging (MSI) and near-infrared spectroscopy (NIRS) are both quick and non-destructive methods that attract attention in seed research and in the seed industry. This review addresses the potential benefits and challenges of using MSI and NIRS for seed testing with a comprehensive focus on applications in physical and physiological seed quality as well as seed health.

**Keywords:** fruit morphology; multispectral imaging; near-infrared; pericarp; testa; seed coat; seed testing; image analysis; chemometrics

## **1. Introduction**

Multispectral imaging (MSI) and near-infrared spectroscopy (NIRS) are both quick and non-destructive methods that have received much attention in seed testing and seed research. The fact that it is possible to measure different quality parameters in a nondestructive, quick, and for some methods, automatic way makes it very interesting for seed-testing facilities and the seed industry. Some of the challenges before the methods are fully implemented and integrated are: development and validation of appropriate statistical models to classify future seeds and a better understanding of these models, i.e., why did the seeds belong to the specific group. The latter is probably more interesting from a scientific, research, and development perspective. In some cases, e.g., a commercial setting, a prober model might be sufficient and the deeper understanding of it less important. This review concerns methods and applications in seed testing and research using MSI and single seed and bulk NIRS to characterize the covering structures of seeds used as regeneration material.

#### *1.1. Seed Covering Structure and Chemical Composition*

The microstructure and chemical composition of specific seed coat cell layers give rise to species and variety differences in seed coat structure and function. Most morphological features of the seed coat are relatively insensitive to environmental conditions and therefore very useful for taxonomic identification. Seed coat color is influenced by environmental conditions—i.e., climatic conditions during maturation and hence not appropriate for taxonomic purposes [1].

**Citation:** Mortensen, A.K.; Gislum, R.; Jørgensen, J.R.; Boelt, B. The Use of Multispectral Imaging and Single Seed and Bulk Near-Infrared Spectroscopy to Characterize Seed Covering Structures: Methods and Applications in Seed Testing and Research. *Agriculture* **2021**, *11*, 301. https://doi.org/10.3390/ agriculture11040301

Academic Editor: Alan G. Taylor

Received: 20 February 2021 Accepted: 28 March 2021 Published: 1 April 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

Sugar beet (*Beta vulgaris* subsp. *vulgaris* var. *altissima* Doell.) belongs to the *Amaranthaceae* family, and other important crops in this family are red beet (*Beta vulgaris subsp. vulgaris* var*. Conditiva*) and spinach (*Spinacia oleracea* L.). The dry fruit of sugar beet seed is a single achene with the fruit coat (pericarp) composed of lignified cells. The pericarp consists of an outer layer of parenchyma cells and an inner, denser layer of sclerenchyma cells. The fruit coat is a physical and chemical barrier for germination [2]. The seeds of species in this family are characterized by a thick fruit coat consisting of lignified cells.

The typical fruit of the *Poaceae* family (e.g., cereals and grasses) is a caryopsis, comprised by the embryo, the starchy endosperm, and the outer aleurone endosperm, surrounded in turn by the nucellar layer, the testa (seed coat) and the pericarp. In addition, the caryopsis of barley (*Hordeum vulgare* L.) and oats (*Avena sativa* L.) have an adherent outer coat or husk or hull consisting of the glumellae—lemma and palea—or the glumes, which are not removed, enclosing the caryopsis [3–5]. In contrast to species in the *Amaranthaceae* family, the seeds of species in this family are characterized by a thin fruit coat—the husk or hull.

#### *1.2. Seed Coat Function*

The seed coat is the seed's primary defense against adverse environmental conditions [6]. The seed coat functions as preserving the integrity of the interior parts of seeds, protects against pests and diseases, regulates gaseous exchanges between the embryo and the external environment and in many families the seed coat plays a role in the control of water absorption during imbibition and germination. Species in the *Fabaceae* family (e.g., beans and forage legumes) have an outer layer consisting of a waxy cuticle [1]. This represents a barrier to imbibition, which may be conferred by waxy or phenolic substances in the epidermis of the seed coat. Many legume species can produce seeds with seed coats temporarily impermeable to water—"hard seeds"—which is a mechanism of physical dormancy.

The intact seed coat protects the embryo from cellular rupture and the leakage of intracellular substances during imbibition. Soybean seeds (*Glycine max* (L.) Merr.) with seed coat epidermal cracking have higher leakage and low viability [7,8] and rapid imbibition of soybean seeds increases the leakage of intracellular substance and decreases seedling survival [7]. Leakage of intracellular substances from imbibing seeds are indicators of low seed vigor and viability.

Damage to seeds by microorganisms occurs by the production of exocellular enzymes which degrade the seed coat, and therefore microorganism infection may also lead to an increase in electrolyte leakage [9].

#### **2. Near-Infrared Spectroscopy**

Single seed or bulk seed NIRS is a non-destructive measurement of the seed or seeds in the electromagnetic near-infrared (NIR) spectrum from wavelengths 780 to 2498 nm, equivalent to wavenumbers 12,821 to 4000 cm−<sup>1</sup> , respectively, with a spectral resolution of 0.5–5 nm (Figure 1) Thus, NIRS radiation is invisible to the human eye in contrast to the shorter wavelengths used in most image analysis systems. The NIR spectrum emerges when monochromatic radiation at a frequency which corresponds to the vibration of a particular chemical bond is absorbed by that bond, while the rest of the radiation is either reflected or transmitted without interacting with other bonds [10]. The C-H, N-H, S-H or O-H bonds absorb the radiation energy and hence it is possible to measure water and organic compounds such as protein, carbohydrates, alcohols and/or lipids [11]. The NIR spectrum consists of overtone bands when radiation energy makes the molecule go from the ground stage (*v* = 0) to an excited stage (*v* = 2) defined as the first overtone, or from the ground stage to *v* = 3 defined as the second overtone. Furthermore, the NIR spectrum consists of combination vibrations, which typically form broad and complex wavebands making it difficult to relate the spectra to individual chemical components [12]. This direct link between spectral information and the chemical compounds makes it obvious to

future seeds.

develop a calibration model consisting of single seed NIRS measurements (explainable or X variables) and wet chemical measurement (response or Y variables) of the aforementioned chemical compounds. This model can be used to predict the chemical compounds in other

**Figure 1.** Reflectance of incoming light of a spinach seed lot (N = 70) using MSI (discrete points with error bars) and NIRS (blue continuous line). The MSI reflectance values are the mean and standard deviation of the reflectance of individual seeds at 19 discrete wavebands from a single image. The NIRS reflectance is the mean value of five measurements on the same seed lot. Standard deviation of NIRS reflectance measurements is not shown as it is too small. The color bar below the plot shows the corresponding perceived colors of the human visible spectrum. The ranges above the plot show which chemical compounds contribute to which wavebands [13].

#### *2.1. NIRS Spectra with Good Informative Spectra*

The use of NIRS in seed testing and seed research can be through single seed or bulk seed lot measurements. The single seed measurement requires a sample holder with similar form as the seed to reduce the risk of light scatter (light travelling outside the seed to the detector). Near-infrared light can penetrate the seed; however, the depth of the penetration depends on several factors such as the physical proportions of the seed. The NIR light is then reflected, refracted, transmitted, scattered or absorbed in the seed (Figure 2)

**Figure 2.** The possible interaction of incident light (I<sup>0</sup> ) with seed and subsequent reflected, refracted, transmitted, scattered or absorbed light (I).

The method for bulk seed NIRS measurement depends on the available instrumentation, and the output is a mean spectrum of the seeds.

The choice of single seed or bulk seed lot measurement depends on the aim of the project. The main advantage of single seed NIRS is the possibility to obtain a spectral signature, i.e., fingerprint for individual seeds, while bulk analysis is an average spectrum of the measured seeds. The benefit of bulk seed analysis lies in the reduced operation time and the possibility to characterize seed lots with fewer measurements as each spectrum represents the variation within the seed lot.

The raw NIR spectra contain important information in terms of spectral peaks that relate to chemical information. Shrestha et al. [14] showed the NIR spectra of seeds of seven species and even though the trends (spectral peaks) were similar, it was possible to identify spectral differences between the species using principal component analysis (Figure 3).

**Figure 3.** Raw NIR spectra (**a**) and principal component analysis (PCA) analysis (**b**) of seeds of seven species. The NIRS measurements were performed as bulk seed analysis (five repeated measurements on the same samples).

#### *2.2. Spectral Pre-Processing*

Pre-processing of the NIR spectra is the first step in developing informative classification models. The purpose of pre-processing is to identify and to remove spectral information that interferes with the desired predictions [15]. If the pre-processing fails, there will be confusion between the information which is sought and the noise which is of no interest [16]. Several pre-processing methods are available and some of them are thoroughly described and shown in Rinnan et al. [17]. In practice, it is important to evaluate the effect of different pre-processing methods on the final models. Another possibility is to use the raw spectra in the subsequent principal component analysis (PCA) as shown in Figure 3. The use of raw spectra will in most cases lead to the usage of more principal components for the final model to reduce noise in the spectra.

#### *2.3. NIRS Model Development and Validation*

Models for classification, pattern recognition or clustering developed from NIR spectra for one sample of seeds (either bulk or single seed NIRS) are intended to classify other seeds or seed samples of the same species based on their NIR spectra.

The NIRS data are highly correlated, meaning that data points next to each other are more alike than data points far from each other, and a common method to reduce this dimensionality is through PCA [18]. Subsequently, this reduction in dimensionality is used in different linear and non-linear models as described by [18–20]. The classification models are divided into supervised or non-supervised models where the supervision relates to labelled or non-labelled data. The use of labelled data in supervised classification models will inevitably influence the results and makes proper validation of the models even more important to avoid overfitting. There are a few regression-based classification models, such as partial least squares discriminant analysis [21,22] and extended canonical variates analysis [23].

Validation of models is an essential part of the modelling process to ensure that a model can be used to classify other seeds or seed samples, but also to avoid giving unrealistic (i.e., optimistic) estimates of the ability to classify new samples [24]. Any model should be validated for model performance and prediction ability using either cross-validation or test set-validation. Cross-validation is performed by dividing the full dataset into *G* sample set and using *G-1* sample set as the training set and the remaining segment in the

test set. Each segment is successively excluded and used for testing the model based on the remaining samples from the *G-1* segments. Using this method, all samples are used for both calibrating and validating the model. The performance of the model is evaluated by its predictive error in terms of root mean square error of cross-validation. Test set-validation is normally seen as a stronger validation of the obtained models as samples in the test set are not part of the model development. Test set-validation requires the data to be divided into a calibration and a validation set. The calibration set is used to calibrate the model and this model is subsequently tested on the validation set. The model performance using test set-validation is described by root mean square error of prediction.

#### **3. Multispectral Imaging**

Multispectral imaging of seeds is a non-destructive technique for simultaneously measuring spectral and spatial information of seeds by imaging their surface reflectance at selected wavelengths from 365 to 970 nm (Figure 1). The combined spectral and spatial measurements provide information about the seed surface chemistry [25] and seed morphology (color, shape, and texture). Multispectral images acquired through MSI is a middle ground between RGB (red green blue) color images and hyperspectral images. RGB images use three wide overlapping wavebands to mimic the human visual perception of colors. In contrast to hyperspectral imaging, which measures the reflectance at hundreds of continuous narrow wavebands across a large spectral range, multispectral imaging measures the reflectance at fewer (<50) and wider discrete wavebands (10–50 nm).

The workflow for MSI of seeds generally includes the following six steps (Figure 4): (1) preparation of seed samples, (2) calibration of multispectral imaging system, (3) acquisition of multispectral images of the seeds, (4) segmentation of regions of interest (ROIs, e.g., the seeds, part(s) of the seeds or foreign matter) in the acquired multispectral images, (5) feature extraction from the segmented ROIs and (6) analysis of the extracted features. If the aim is to study changes in the seeds over time, for example, to follow the imbibition process or radicle emergence, steps 1 to 3 may be repeated multiple times before proceeding with steps 4 to 6.

**Figure 4.** Illustration of the typical workflow in multispectral imaging applications with output examples from each of the six steps.

#### *3.1. Sample Preparation*

For MSI, seeds require very little preparation beyond the preparation required for the application or experiment at hand. For example, if the aim was to see if it is possible to identify the presence of particular fungi on seeds, the first step might be to work with sterilized seeds before inoculating them with the fungus/fungi of interest [26–30]. Similarly, it may be necessary to artificially age seeds for different lengths of time, to explore the use

of MSI for predicting whether seeds are viable or dead, or parameters related to vigor (e.g., El Masry et al. [31]). On the other hand, in varietal purity applications, the seeds may be imaged without any further preparation.

Due to the spatial nature of the multispectral images, multiple seeds can be imaged simultaneously. Seeds are often placed in a Petri dish and it is important that there is space around each seed. Seeds located too close to each other may touch or even overlap and cause occlusion leading to poorer segmentation and adding noise to the extracted features. To prevent seeds from moving, when placing them in the Petri dish, they may be fixed with double-sided tape [29,32] or placed on an insert with small recesses inside the Petri dish. When placing the seeds, it is important to consider which side is most relevant for the application and thus should be facing the imaging sensor. In applications where multiple sides are equally relevant, such as detection of processing damage, images from multiple sides can be acquired by imaging each seed multiple times [33–35]. For some studies, it may be necessary to keep track of each individual seed through the imaging process to understand the subsequent 'fate' of each seed.

Placing seeds manually in a Petri dish for imaging can be both cumbersome and timeconsuming. A conveyer belt can be used to automate the imaging process and increase the number of seeds imaged over time in applications where the seeds do not require any special preparation or manual assessment (e.g., variety or foreign matter identification [36]).

#### *3.2. Calibration of Multispectral Imaging System*

The MSI system must be calibrated prior to image acquisition to ensure comparable reflectance measurements across wavebands and images, pixel correspondence between wavebands and to enable spatial measurements in world units [37,38]. This includes both a radiometric calibration and a geometric calibration, which is carried out by imaging calibration targets with known reflectance and geometry [39].

Furthermore, the illumination and exposure times must be set to minimize the number of under- and oversaturated pixels, thereby maximizing the dynamic range and the signalto-noise ratio of the images [27,37].

#### *3.3. Image Acquisition*

After calibration, the MSI system is ready to image the prepared samples. The output of a measurement is a multispectral image or "data cube" consisting of *W* × *H* pixels × *C* channels, where *W* and *H* are the width and height of the image, respectively, and each pixel contains *C* channels corresponding to the discrete multispectral bands. When a pixel position overlaps with a seed, the pixel values represent the chemistry on and below the surface of the seed in the small area covered by the pixel [25].

Although multispectral imaging systems can acquire the images through either point scanning, line scanning or area scanning [40], in the vast majority of the applications the images are acquired through area scanning with a charged coupled device (CCD) imaging sensor and sequentially illuminating the seeds using LEDs with the desired wavebands (Table 1). Ideally, these wavebands should be carefully selected to match the application or research question [34]. However, most MSI applications use the same multi-purpose MSI system (all applications with 19 bands in Table 1), where the wavebands and spectral range are selected by the company. However, changing the spectral range will mean changing imaging sensor technology as the spectral range of the current MSI systems is limited by the quantum efficiency of a standard CCD to approximately 400–1000 nm.


**Table 1.** Summary of selected applications of multispectral imaging for seeds analysis.


**Table 1.** *Cont.*

1. Subdivision of species in brackets are made according to the terminology used in the referenced paper. 2. Adaboost = Adaptive Boosting; ANOVA = Analysis of Variance; CDA = Canonical Discriminant Analysis; CNN = Convolutional Neural Network; GLM = Generalized Linear Model; *k*-NN = *k*-Nearest Neighbors; NN = Neural Network; PCA = Principal Component Analysis; PLS = Partial Least Squares; MLR = Multiple Linear Regression; RF = Random Forest; SDA = Stepwise Discriminant Analysis; SVM = Support Vector Machine. 3. The accuracy is the number of correctly classified samples with respect to the total number of samples. "-" means the reference did not report the accuracy and most likely used another error metric

Selecting a high contrasting background material on which the seeds are placed can make the segmentation step easier; however, the intensity level of the background should approximately match that of the seeds to fully use the dynamic range of the multispectral imaging system.

#### *3.4. Segmentation of Regions-of-Interest*

The multispectral images contain not only ROIs, but also background objects, such as the background material, the Petri dish, a conveyer belt, or other inert matter. In the segmentation step, the ROIs are separated from the background objects and extracted from the image. The ROIs in the multispectral images are often limited to only the seeds, but they may also include other objects such as foreign matter [36]. To ensure that only the correct objects are analyzed, the segmentation method must extract only objects regarded as ROIs. Equally important, the segmentation method must return all pixels related to the ROIs, and only those pixels to reduce noise in the subsequently extracted features.

With a high contrast background material and sufficient space around each seed, the segmentation can often be carried out using a simple threshold in either a single channel [36,40], a sum of the channels [33] or on a score image created through canonical discriminant analysis (CDA) [53] or PCA [36]. Ma et al. [49] used Otsu's algorithm [56] to set the threshold automatically.

Although different methods have been explored, their performance have not been/are seldom quantified (e.g., pixel accuracy or intersection over union) beyond visual inspection as the segmentation step is often seen as an intermediate step towards the final analysis.

#### *3.5. Feature Extraction*

In recent applications, several features quantifying the reflectance and morphology of the seed have been explored. These features form four groups related to their characterization of the seed and their relation to the multispectral image: reflectance, color, shape, and texture (Table 1). They are generally extracted from the entire seed; however, they may also focus on only a specific part of the seed such as the endosperm region [42]. The reflectance and color features relate to the spectral dimension (C) of the multispectral image and express the intensity of either reflectance or color of the seed. The reflectance features either treat the wavebands individually by extracting first-order derivatives from the raw wavebands [52] or combine them with a CDA transformation before extracting either a trimmed mean [50] or ratio of pixels above a given threshold [55]. In contrast, the color features combine wavebands overlapping with the human visible spectrum into a well-defined color space, e.g., CIELAB [47], and extracts first-order features from there. The shape features are related to the spatial dimensions (*W* × *H*) of the multispectral image and are therefore derived from the binary image created during segmentation. They include simple descriptors, such as area, width and length [57], but also more complex descriptors, such as ellipse fitting parameters and resemblance to known simple shapes (i.e., circle, ellipse, and rectangle). The texture features combine the spatial and spectral dimensions by quantifying the spatial variation in intensity across the seed. This spatial variation in intensity can be caused by both small changes in the surface structure (valleys and hills) as well as changes in color in the seed surface pattern. The color, shape, and texture features describe the morphology of the seed and are therefore jointly referred to as morphological features. Characters of morphologic features of different seed structures play an important role in the delimitation and identification of species [58].

The type of extracted features is somewhat application-dependent (Table 1). Applications related to fungal presence all use reflectance features and to some extent color and texture features. Shape features are, however, not used as the shape of the seed is not affected by the presence of fungus until the fungus has grown significantly. On the other hand, applications related to varietal purity almost all use reflectance, color, and shape features, but do not consider texture features. Applications on seed viability and vigor favor reflectance and to a lesser extent color and shape.

For a given application, it is important to extract features which are expected to correlate well with the desired response variable. Features may be derived from existing knowledge, such as previous work in hyperspectral imaging, NIRS or crop descriptors [59]. However, the selection of features should be well argued.

#### *3.6. Multivariate Data Analysis*

The multivariate data analysis of the extracted features often includes a descriptive statistic followed by data modelling. The descriptive statistics compares the mean and variation of the individual features for each class. For the reflectance features, this is often visualized as a mean spectrum for each of the classes [26]. Principle component analysis is also widely used to investigate any trends in the features prior to data modelling.

Several linear and non-linear methods have been used for data modelling in MSI. The most frequently used methods include PCA, CDA, support vector machines (SVM), partial least squares and to a lesser extent neural networks and k-nearest neighbors (Table 1)

Despite a large number of features and correlation between features within feature types (e.g., shape features), dimension reduction [44,57] or feature selection [52] prior to modelling is the exception to the rule. However, several applications evaluate the feature types both individually and combined and show an improvement in accuracy when feature types are combined (Table 1).

In applications related to physiological seed quality and seed health, it may be difficult to ensure an equal number of examples from each class. This leads to an unbalanced dataset, where one or more classes are either over- or underrepresented compared to the remaining classes in the dataset. Unbalanced data in the calibration set can lead to a model with poor generalization on future data, while the model is still reporting misleadingly high values in error metrics such as accuracy. The data imbalance may be handled as a pre-processing step (e.g., resampling) through cost-sensitive learning (assigning different costs to each type of misclassification) or at an algorithm-level [60]. Likewise, error metrics less skewed by an unbalanced dataset should be favored.

#### **4. Applications**

The recent applications of multispectral imaging of seeds can be grouped into three categories according to the aspect of seed quality (Table 1): physical seed quality, physiological seed quality, and seed health.

#### *4.1. Physical Seed Quality*

The physical seed quality applications include (a) varietal identity and purity, (b) presence of other seeds and inert matter and (c) seed coat integrity.

#### 4.1.1. Varietal Identity and Purity

The microstructure and chemical composition of specific seed coat cell layers give rise to species and varieties differences. Most morphological features of the seed coat are relatively insensitive to environmental conditions and therefor very useful for taxonomic identification.

Multispectral imaging has been employed for varietal discrimination and identification in several species such as tomato (*Solanum lycopersicum* L.), rice (*Oryza sativa* L.) and soybean (reviewed in Boelt et al. [55]). Color, shape and spectral features have been used in the classification models (Table 1). Since then, studies in alfalfa and pepper (*Capsicum annuum* L.) have been reported [41,43].

In pepper, three commercial varieties were analyzed for varietal identification [43]. Each variety was represented by at least 450 seeds and seed material was harvested at different locations. Samples were divided into training and test set in the ratio 9:1. The study employs different multivariate data analysis and resulting classification accuracies are in the range of 86–98%. The multispectral imaging system used in this study has 19 bands. Interestingly, a successive projection algorithm identified nine bands, which

provide a classification accuracy almost identical with the outcome with all 19 bands (97%). Still, the authors suggest that a data analysis with lower classification accuracy (93%) may be used as this is easier to operate and has a sufficiently high accuracy for the purpose. This illustrates how feasibility and ease of operation is of importance in the commercial seed industry.

Twelve alfalfa cultivars (*Medicago sativa* L.) with diverse geographic origin were obtained from a genebank [41]. A total number of 200 seeds were split 70:30 in training and testing set, respectively. Different multivariate data analysis was used to classify cultivars (Table 1). When only morphological features were employed, classification accuracy was low (42–44%) but combined with spectral features, accuracy increased to 92–93%. It is noticed that based on spectral reflectance cultivars were classified into three groups correlating with geographic origin. This may be based on common genetic background or seeds may have been produced in different environments. Seed coat color is influenced by environmental conditions—i.e., climatic conditions during maturation and hence not appropriate for taxonomic purposes [1]. However, variation in texture and chemical composition will also be reflected in the spectral features and they are highly relevant for taxonomic discrimination.

Seed accessions in genebanks may not be as uniform as commercial varieties; however, the description of the seed morphology is very important to manage the large accession numbers (for example during the regeneration procedure). Already Hansen et al. [25] demonstrated high classification accuracy among 20 diverse rice varieties (93%) and suggested MSI as an important tool in management of genebank accessions. A recent study used MSI for the assessment of the genetic diversity in a collection of pigmented rice accessions from the Philippines [45]. Geometric seed traits were quantified (area, length, width, roundness, and seed color parameters). The study identified pigmented rice accessions, which represent a valuable genetic resource for the future improvement of commercial rice varieties.

In conclusion, MSI may both be used to distinguishing among commercial varieties in the test of varietal purity and to describe diversity in seed traits during conservation management of plant genetic resources.

#### 4.1.2. Presence of Other Seeds and Inert Matter

Sendin et al. [34] reported the use of MSI for the determination of other crop seeds and plant debris in white maize (*Zea maize* L.). Seeds of crop species were wheat (*Triticum aestivum* L.), sorghum (*Sorghum bicolor* L.), soybean and sunflower (*Helianthus annuus* L.) and all were classified with 100% accuracy. Plant debris was also classified with 100% accuracy and the authors point to the benefit of MSI contra hyperspectral imaging in relation to shorter analysis time and lower cost. Recently Hu et al. [39] published findings on the differentiation of sweet clover (*Melilotus* ssp.) in alfalfa with a classification accuracy of >99% by MSI. Combining morphological features and spectral data in the models increased the accuracy. The survey included six alfalfa varieties and two species of sweet clover: One seed lot of *Melilotus officinalis* and five seed lots of *Melilotus albus*. All seed lots consisted of 200 seeds, divided in training and model testing in the proportion 70:30. Reflection mean intensity showed discrimination both in visible and NIRS wavelength bands.

As indicated in the two above-mentioned studies, very high accuracies may be expected when classifying seeds belonging to different species, and hence this may not attain much consideration in research; however, the determination of other seeds in crop seeds is a very time-consuming task in the seed industry. It appears relevant to develop robust models of crop seeds containing the variability in seed morphology from site to site, year to year for the use in seed testing. There are examples from the food industry in the detection of "foreign matter" which would include some of the same constituents as the inert matter fraction in a seed sample (soil, stones, plant debris) [34,36].

#### 4.1.3. Integrity of Seed Covering Structures

The intact seed coat protects the internal structures of the seed and controls water uptake, but seed coat disrupture may occur due to insect infestation during seed production or storage or mechanical damage during harvest and processing. Seed coat damage negatively affects vigor and viability potential, and the "openings" of the damaged seed coat may be an entrance for pathogenic fungi.

#### Insect Infestation

When insect infestation occurs during seed production, the damaged seed is often discarded during harvest and processing due to a lower seed mass. Insect infestation occurring in the later developmental phase may not be identified and has the potential to develop during storage. Insect infestation has a direct effect on seed quality by consuming the seed reserves but there is also an indirect effect as it allows the establishment of secondary pests and fungi, for the storage pests lay eggs on the seed surface for the larvae to penetrate the seed coat and the larvae may undergo different larvae stages and finally produce a pupa inside the seed. X-ray and MSI have been tested for the identification of grain moth (*Sitotroga cerealella*) in wheat [28]. The study showed the potential of X-ray for the study of internal structures in the seed, whereas MSI showed the potential for identifying eggs on the seed surface.

#### Mechanical Damage

Species containing germination inhibitors in the seed coat (for example sugar beet) undergo different treatments during processing to remove these inhibitors. The inner pericarp layer contains crystals of chemical compounds in the sclerenchyma cells [61], and the crystals dissolve in water during washing. This process alters the outer surface structure of sugar beet seed [2]. MSI can detect changes in surface color and reflectance during maturation in sugar beet seed [62], and the study verified a concomitant increase in the content of phenolic compounds. Removal of the pericarp by polishing is another approach for the removal of inhibitory compounds. The polishing process removes most of the large parenchyma cells of the pericarp and hence alters the surface of the sugar beet seed [2]. The ideal treatment will remove the outer pericarp layer, whereas the inner pericarp layer remains intact. Besides removal of germination inhibitors, polishing also makes seed more uniform for pelleting and improves water uptake.

As with any mechanical operation, excessive processing can cause damage to the seed, and this damage can be extended to the interior parts of the seed and affect physiological quality of the seed (Figure 5) Mechanical injuries decrease the seed longevity, expose the seed to the fungal infection and reduce viability.

Due to their sensitivity to water uptake, damaged seeds may result in heterogenous field performance, and there is also evidence from soybean, sweet corn and maize that the damaged seeds are more likely to produce abnormal seedlings [61].

A study by Salimi [35] displayed the potential of MSI in classification of various damage types, without additional analytical evaluation. The study demonstrated MSI as a tool for the identification of mechanical damage from polishing during processing and hence demonstrates MSI as a tool in seed quality assessment. A classification model based on MSI derived information about surface characteristics and multivariate data analysis enabled discrimination into five damage classes with 82% overall accuracy.

Barley (*Hordeum vulgare* L.) grains without hulls will imbibe water and germinate more rapidly than those with firmly adhering husk [63]. During harvest, the hull acts to protect the embryo during the abrasive threshing process in the harvester [64]. However, the husk may be partially or wholly detached at harvest and during post-harvest handling (Brennan, Shepherd et al. 2017). MSI may be a potential tool for the characterization of de-hulled barley grains.

**Figure 5.** Processing damage in sugar beet seeds. (**A**): RGB images; (**B**): nCDA transformed multispectral images [35]. (**a**). Partially broken pericarp and/or outer testa, (**b**). Completely broken pericarp and outer testa, (**c**). Fractured pericarp and outer testa, partially crushed inner testa with sound embryo, (**d**). partially broken pericarp and/or outer testa, damaged inner testa with intact embryo, (**e1**–**e4**). Different types of severe damages to the embryo or seeds without any embryo like the pericarp or outer testa. Reproduced with permission from ref. [35]. Copyright 2019, MDPI.

#### *4.2. Physiological Seed Quality*

Tannins, phenols, waxes, pigments, germination inhibitors and other substances are found in the seed covering structures of different species, and these may influence the function of the seed coat and subsequently the physiological development of the seed.

#### 4.2.1. Viability

Olesen et al. [50] identified viable castor bean (*Ricinus cummunis* L.) seeds with 92% accuracy and showed good correlation between results from tetrazolium tests and MSI. Three seed lots were included in the study. In castor bean, seed coat color was related to the development and the darker seeds were the most developed. In this study, seeds from four ecotypes were studied. The calibration set consisted of 120 seeds from two ecotypes, and they were divided into three groups in depending on seed coat color (visible inspection). The validation set was two other ecotypes, and the seeds of those were also divided into three groups. After acquisition of MSI images, seeds were germinated for the phenotyping of viability, and a tetrazolium test was performed as the viability reference. A high correlation was found (92%). The supervised nCDA model showed 96% precision accuracy in the classification of viable and dead seeds in the validation set. The study showed high differentiation between viable and non-viable seed in mean intensity reflection in the wavelength interval 375–970 nm with the largest difference in the NIR-regions, which is supported by Shetty et al. [52] in a study predicting germination ability in spinach. This latter reference combined the use of single seed NIRS and MSI.

Liu et al. [53] also found a high prediction accuracy (91–92%) for high-quality watermelon (*Citrullus lanatus* (Thunb.)) seed, in two different varieties using both spectral and morphology features in MSI. From each variety 500 seeds were classified into pure, viable; low vigor; other varieties and dead seeds by means of a grow-out trial. Prediction accuracy concerns two classes: pure, viable, and all other seeds for each variety.

#### 4.2.2. Vigor

Several species in the *Fabaceae* family can produce hard seeds (physical dormancy) which are impermeable or semi-permeable and hence do not absorb water. Physical dormancy is often associated with a layer of wax in the outer layers of the seed coat.

Hu et al. [54] examined seeds of six species within the Fabaceae family with MSI for the detection of hard seeds. For each species, 400 seeds were examined 70:30 in training and testing set, respectively, and following image acquisition seeds were imbibed for germination. Hard seeds were identified as un-imbibed, whereas seed which adsorbed water was classified as "soft" non-dormant seed. For three species (sweet clover, alfalfa and galega (*Galega officinalis* L.)) MSI combined with multivariate data analysis has accuracies in the interval of 88–92% in detecting hard seeds, whereas for the other three species they could not be identified. In all three species studied, hard seeds showed a higher reflectance compared to non-hard seeds. Hu et al. [54] used SVM analysis and found that wavelengths in the NIR-region, i.e., 970 nm (water) and 940 nm (lipids, were of highest importance in the separation of the two groups. However, for each species only one seed lot was represented in the analysis and there were proportionally fewer non-hard seeds which made the two groups unbalanced.

Single seed NIRS spectroscopy and MSI have been employed for the assessment of viability after controlled deterioration or artificially seed ageing in spinach [38] and cowpea (*Vigna unguiculata L.*) [31]. In spinach, two seed lots with viability percentages of 90% and 97% were chosen for the examination by single seed NIRS after artificially ageing of both seed lots [38]. In cowpea, variation in germination performance was generated by artificially ageing in four treatments (ageing intervals 24–96 h) [31]. Olesen et al. [38] used Extended Canonical Variates Analysis (ECVA) assigned differences of scatter corrected absorbance spectra from aged and non-aged seeds to CH2, CH<sup>3</sup> and HC = CH structures, which are some of the functional groups in lipids. Lipids play a major role in both ageing and germination. During accelerated ageing lipid peroxidation leads to deterioration of cell membranes and contributes in that way to reducing seed viability of the seed sample. These biochemical changes may be the reason for a clear grouping between aged and non-aged seeds with misclassification in the range of 4–11% when performing the ECVA. In cowpea, the overall correct classification was in the interval 97–98% between aged and non-aged seeds, whereas the classification was lower in the detection of germinated versus non-germinated seed (79–82%). A recent paper reports a strong relationship between X-ray and MSI and seed physiological potential in *Jatropha curcas L.* seed [51]. Both viability and vigor were studied, and the authors find that reflectance data in the NIR wavelength 940 nm showed 96% accuracy.

Ruptured seed coats allow for the diffusion of leachates, which serve as substrates for pathogen growth, and broken seed coats serve as infection sites for seed pathogens. Common measurements of seed leakage in water are the conductivity of electrolytes and ultraviolet (UV) light absorbance (254 and 280 nm) [65,66]. Leaked solutes may be amino acids, proteins, sugars, and phenolics. Brassica seed has a high content of phenolic compounds. One of these is sinapine, the content of which increases under unfavorable storage conditions. Hill et al. [67] found sinapine leakage a more accurate method for the identification of viable cabbage (*Brassica oleracea* var. *capitata* L.) seeds than the conductivity test. Sinapine was measured by the absorbance at 388 nm. The compound fluoresces when irradiated with UV light and has maximum absorbance values of 326 and 388 nm. Later work [68] showed that seeds with cracked seed coat leaked faster and that seed coat integrity is a major factor regulating sinapine leakage. Sinapine does not leak from viable seeds [67].

Since leaked solutes have been measured using absorbance of light in the UV region, a future perspective of MSI would be analyzing single dry seeds for diffused solutes often associated with cracks in the seed coat.

#### *4.3. Seed Health*

Detection of seeds infected by fungi is traditionally performed by visual inspection of dry seeds, washing tests, incubation methods, embryo count method or seedling symptom tests as well as identification of sporulation [69,70]. These methods require expert knowledge and can be time-consuming. However, the combinations of the features from multispectral images captured by visual light and NIR wavelengths (Figure 1) have proved to be useful in the separation of infected and uninfected seeds (Table 1), but depend on traditional reference methods.

Multispectral imaging for seed health detection has in several studies been based on artificial inoculation of uninfected seeds, with freeze-blotter seed health assay as reference

method. First demonstrated in spinach by detection of *Stemphylium botryosum*, *Cladosporium* spp., *Fusarium* spp., *Verticillium* spp. or *Alternaria alternate* [26], and recently by detection of *Drechslera avenae* and *Helminthosporium avenae* in black oat/oats seeds (*Avena strigosa*) [32], *Fusarium pallidoroseum, Rhizoctonia solani*, and *Aspergillus* sp. in cowpea [29].

The simplest approach is to use a visual score as reference for fungal infection. However, the method depends on an expert to classify the seeds in healthy and infected seeds as well as determine the species of the fungi. Weng et al. [30] used artificial inoculation of uninfected seeds by *Ustilaginoidea virens* in rice with a visual scoring as reference method. The seeds used in this study were divided into healthy, slightly infected, and infected seeds. However, the healthy and the slightly infected seeds were difficult to separate by a PCA. It was suggested that this was due to only minor changes in seed surface features or chemical components of the slightly infected seeds.

DNA-based data may be used as reference in combination with MSI. Boelt et al. [55] used next generation sequencing (NGS) of the ITS (Internal Transcribed Spacer) from total DNA as reference method on naturally infected barley seeds collected from a wide range of environments. NGS is highly sensitive and gives information to species level as well as the fungal composition and quantities. This is particularly useful as several fungi may infect seeds simultaneously. NGS made it possible to separate seeds infected by *Alternaria infectoria, Dothidomycetes* sp., *Fusarium graminearum, F. avenaeum* and *Mycosphaerella tassiana* by multispectral imaging.

Magnetic resonance imaging (MRI) to identify anatomical changes in artificial inoculation of *Jatropha curcas* L. was used in combination with MSI by Barboza da Silva et al. [27]. The proposed MRI and MSI methodology allowed the identification of different damage patterns in the endosperm tissues due to infections by *Lasiodiplodia theobromae, Colletotrichum siamense*, and *Colletotrichum truncatum.*

#### **5. Summary and Perspectives**

Multispectral imaging and single seed or bulk seed NIRS are non-destructive techniques for quality assessment both in research and in seed testing. In contrast, hyperspectral imaging requires more resources for operation and is therefore most relevant in seed testing and seed research. Since recent reviews of multispectral imaging [40,55] there has been a growing evidence of the application of MSI in particular in physical seed quality evaluation and in seed health.

For physical seed quality, focus has been to distinguish genetic purity among varieties (alfalfa and pepper) or between crop species and inert matter (alfalfa versus sweet clover; mustard versus foreign and inert matter). In general, high classification accuracies have been obtained but often the number of samples or sample sizes have been limited or even unbalanced. Future studies ought to include more robust training and validation datasets by including higher and more diverse samples. Exploring seed produced at different sites (years and environmental conditions) would strengthen validation of the models by including variation in seed size and seed coat color and eventually lead to robust global models.

A relevant application for MSI is the characterization of the stored seed samples for the preservation of plant genetic resources. For this application features such as shape, texture, reflectance, and color are highly relevant, but they may be combined with a focus on specific parts of the seed, for example, the morphology of the hilum region, which is a relevant feature in the crop descriptors of legume seeds.

In conclusion, MSI may be used both to distinguish among commercial varieties in the test of varietal purity and to describe diversity in seed traits during conservation of plant genetic resources. For the assessment of physical seed quality, very little sample preparation is required, but a large diversity from each species or variety ought to be included by representing different production sites and climatic environments.

In seed research and seed testing, electrolyte leakage is an established method for vigor evaluation, where seeds are imbibed for a certain period of time, and the imbibition water is analyzed by spectrophotometer. Solutes are measured using the absorbance of light in the UV region. A future perspective of MSI would be analyzing single seeds? for diffused solutes in the UV band. The information acquired on the single seed level may even be combined with other features such as color, physical damage, or cracks in the seed coat. The determination of physiological seed quality will often require more sample preparation depending on the physiological process in question, and sample sizes may be unbalanced, for example, there are far fewer non-viable seeds in a commercial seed lot.

Physiological seed quality is often reflected in the chemistry of the seed and therefore information from the NIR-wavelength regions is often very informative. The region of interest for the chemical information defines which method to apply, where MSI will inspect the seed covering surface and single seed or bulk NIRS will inspect the seed beyond the surface cover. However, none of these methods provides information on internal morphological seed structures.

The use of MSI and single seed and bulk NIRS to characterize seed covering structures is only at the beginning, and there is a future potential for the development of specific applications in seed testing. Cross disciplinary studies between seed research and data science may combine the required insight in seed biology and data analysis to provide relevant seed samples for inspection and optimize feature extraction, data analysis, and model validation.

**Author Contributions:** Conceptualization, A.K.M., R.G., J.R.J. and B.B.; writing—original draft preparation, A.K.M., R.G., J.R.J. and B.B.; writing—review and editing, A.K.M., R.G., J.R.J. and B.B. All authors have read and agreed to the published version of the manuscript.

**Funding:** This material is based upon work that is funded by The Ministry of Higher Education and Science, Denmark.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


## *Article* **Using Multispectral Imaging for Detecting Seed-Borne Fungi in Cowpea**

**Carlos Henrique Queiroz Rego 1,\* , Fabiano França-Silva <sup>1</sup> , Francisco Guilhien Gomes-Junior <sup>1</sup> , Maria Heloisa Duarte de Moraes <sup>2</sup> , André Dantas de Medeiros <sup>3</sup> and Clíssia Barboza da Silva <sup>4</sup>**


Received: 13 July 2020; Accepted: 8 August 2020; Published: 17 August 2020

**Abstract:** Recent advances in multispectral imaging-based technology have provided useful information on seed health in order to optimize the quality control process. In this study, we verified the efficiency of multispectral imaging (MSI) combined with statistical models to assess the cowpea seed health and differentiate seeds carrying different fungal species. Seeds were artificially inoculated with *Fusarium pallidoroseum*, *Rhizoctonia solani* and *Aspergillus* sp. Multispectral images were acquired at 19 wavelengths (365 to 970 nm) from inoculated seeds and freeze-killed 'incubated' seeds. Statistical models based on linear discriminant analysis (LDA) were developed using reflectance, color and texture features of the seed images. Results demonstrated that the LDA-based models were efficient in detecting and identifying different species of fungi in cowpea seeds. The model showed above 92% accuracy before incubation and 99% after incubation, indicating that the MSI technique in combination with statistical models can be a useful tool for evaluating the health status of cowpea seeds. Our findings can be a guide for the development of in-depth studies with more cultivars and fungal species, isolated and in association, for the successful application of MSI in the routine health inspection of cowpea seeds and other important legumes.

**Keywords:** *Vigna unguiculata* (L.) Walp; seed health; spectroscopy

#### **1. Introduction**

Cowpea (*Vigna unguiculata* L. Walp) is a leguminous species which is of nutritional and social importance in underdeveloped regions due to the high protein content in its grains [1,2]. For example, most of the production in Brazil comes from family farming, especially in the North and Northeast regions, but it has currently aroused the interest of farmers in the Midwest region who practice commercial agriculture [3].

Despite its adaptability and rusticity, cowpea seeds are susceptible to several fungal diseases. According to Biemond et al. [4], the contamination of cowpea seeds by *Aspergillus flavus*, *Macrophomina phaseolina*, *Fusarium oxysporum* and *Penicillium* sp. contribute to a marked reduction in germination and seed weight, in addition to acting on accelerating deterioration by producing aflatoxins, thus limiting commercialization of its seeds and consumption of grains.

In agricultural industry, the seed health is mainly monitored by detecting fungi species and their percentages present in the sample, which contributes to make decision regarding the suitability of a lot destined for sowing or marketing [5]. The blotter test is the most well-known and used method for detecting seed-borne fungi. However, it is a time-consuming and subjective test since it depends on visual inspections and requires highly trained specialists [6].

Innovative, accurate and rapid light-based methods have been developed to meet the growing demands of the food and agricultural industries, which can produce a consistent assessment of seed health, overcoming the intrinsic subjectivity of conventional techniques [6]. Unaltered samples can be analyzed with non-destructive and real time visualization of the pathological attributes of seeds, with optimization in the quality control process [7,8]. In addition, these tools produce complementary information related to the energy-matter interaction in the context of seed quality.

The multispectral imaging (MSI) technology is based on the use of spectral bandwidths in the ultraviolet, visible and infrared regions in order to obtain spatial and spectral information from the objects under evaluation [9,10], therefore, it can be a useful tool to distinguish healthy seeds from seeds that are carrying important pathogens [11–13]. For instance, there are reports that the use of MSI showed over 80% separation of healthy spinach seeds from those with different fungi species [14].

Considering that each material has intrinsic spectral characteristics that vary according to chemical or physical attributes, this study had two main objectives. Firstly, to evaluate the efficiency of the MSI technique in the evaluation of cowpea seed health. The second objective was to evaluate whether different fungal species can be discriminated using MSI associated with statistical models, before and after seed incubation.

#### **2. Materials and Methods**

#### *2.1. Seed Samples and Fungi Inoculation*

Cowpea seeds from BRS Tucumanque cultivar were used in this study. Seeds were inoculated with three fungi species isolates (*Fusarium pallidoroseum*, *Rhizoctonia solani* and *Aspergillus* sp.). Each species of fungus was grown in three 9-cm Petri dishes containing potato dextrose agar (PDA) medium and kept in a growth chamber with a temperature adjusted to 20 ± 2 ◦C with a 12-h photoperiod of white fluorescent light for a period of 15 days.

The seeds were disinfected for inoculation in sodium hypochlorite solution (1% concentration for 3 min), washed in distilled water and then dried on paper towels at room temperature for 24 h. After drying, 100 seeds per plate were added in order to be in contact with the fungus colony, kept in a growth chamber under the conditions described above for 24 h. After the contact period, the seeds were removed from the plates and placed in a single layer on paper towels at room temperature for 24 h to dry. Afterwards, seeds were divided into two groups for the image acquisition; the first group was called 'Inoculated seeds' (dry seeds), for which 30 seeds were distributed into three Petri dishes (10 seeds per plate), fixed with double tape facing the bottom, positioned one by one in a single layer and equidistant from each other. The second group was called 'Incubated seeds', and a deep-freezing blotter method was used to kill the seeds. Three subsamples of ten seeds were placed in three Petri dishes containing three filter paper sheets moistened with 3.5 mL of distilled water, kept at 20 ± 2 ◦C for 24 h. After this period, the plates were transferred to a freezer at −20 ◦C for 24 h and, subsequently, incubated at 20 ± 2 ◦C with a photoperiod of 12 h with fluorescent lamps, for 4 days; seeds were positioned equidistantly from each other in a single layer.

#### *2.2. Multispectral Imaging Application*

The Petri dishes were positioned under the sphere of integration of the VideometerLab4® instrument (Videometer A/S, Herlev, Denmark) and, after successive illumination of the samples at 19 contiguous light emitting diodes (LEDs), a monochrome charge-coupled chip (CCD) recorded the reflectance of the seeds and generated 19 images (2192 × 2192 pixels) corresponding to the 19 wavelengths (365, 405, 430, 450, 470, 490, 515, 540, 570, 590, 630, 645, 660, 690, 780, 850, 880, 940, 970 nm) of the electromagnetic spectrum.

Data analysis were performed with VideometerLab4 software version 3.14.9 (Videometer A/S, Herlev, Denmark). The multispectral images were transformed using normalized canonical discriminant analysis (nCDA) to minimize the distance within classes and to maximize the distance among classes. Each seed was identified as a region of interest (ROI), and it was built a mask to segment the seeds from the background, which was based on an nCDA transformation of seeds and Petri dish and a simple threshold. The seeds were collected in a blob database, and 36 variables were extracted from the individual seeds, including tristimulus components of color as hue (angular specification for color perceived as red, yellow, blue or green) and saturation (degree of difference between the color and neutral gray).

MultiColorMean feature extracts the reflectance mean of each seed for the 19 spectral bands (from 365 to 970 nm). To eliminate the influence of outliers at both the high and low ends, a trimmed mean excludes 10% of the lowest and highest values before calculating the mean. RegionMSI\_Mean calculates a trimmed mean of transformed pixel values within the blob (each single seed), and RegionMSIthresh measures the percentage of blob region with transformation value higher than threshold, based on the nCDA model (derived from all the classes).

A gray level run length matrix (GLRLM), was generated to identify and distinguish texture patterns. GraylevelRunStatistics feature captures the coarseness of a texture in specified directions according to algorithm described by Galloway [15] and Albregtsen and Nielsen [16]: (0) = Short Run Emphasis (SRE) measures the distribution of short runs, and higher values indicate fine textures; (1) = Long Run Emphasis (LRE) measures the distribution of long runs, and higher values indicate coarse textures; (2) = Gray Level Non-Uniformity (GLN) measures the similarity of gray level values in the image, and GLN values are lower if gray level values are similar throughout the image; (3)= Run Length Non-Uniformity (RLN) expresses the similarity of run lengths throughout the image, with lower values if the run lengths are the same throughout the image; (4) = Run Percentage (RP) determines the distribution and homogeneity of runs in an image in a particular direction. The texture features described by Chu, et al. [17] were also measured: (5) = Low Grey Level Run Emphasis (LGRE) and (6) = High Grey Level Run Emphasis (HGRE). Short run emphasis measures the short run distribution and it is large for fine textures. Long run emphasis calculates the long run distribution and it is large for coarse structural textures.

The CIE color spaces were measured for the axes of lightness (*L\**) and chromaticities (*a\** and *b\**), where CIELab *L\** represents lightness from black to white, CIELab a\* the color appearance from green to red and CIELab *b\** the color appearance from blue to yellow. The CIELab system is a simplified mathematical approximation to a uniform color space composed of perceived color differences [18]. It was defined by the International Commission on Illumination (CIE), and comprises all perceivable colors of the spectrum, even outside the human vision gamut [19]. An intensity-hue-saturation transformation was applied to map the standardized RGB (sRGB) image into intensity, which is independent of color hue that is the dominant wavelength, and saturation which is the colorfulness or the prominence of the dominant color.

#### *2.3. Unsupervised Analysis*

The data obtained from multispectral images were exported to Excel and subsequently subjected to unsupervised multivariate analysis. Multivariate principal component analysis (PCA) was used in this study as an exploratory technique to identify hidden patterns in the data obtained from the MSI analysis. The data obtained for each seed were normalized, and the eigenvalues and eigenvectors were calculated from the covariance matrices. The results were plotted on two-dimensional graphs using the R 4.0.0 software program [20].

#### *2.4. Supervised Discriminant Analysis*

Two models were developed based on the Linear Discriminant Analysis (LDA) algorithm to classify different fungal species associated with cowpea seeds. The first model was developed based on the MSI

information obtained from the inoculated seeds, while the second model used data from the incubated seeds. The classes used in both models were: Class (1) Control—seeds without fungal infestation; Class (2) *Aspergillus*—Seeds infested with *Aspergillus* sp. fungus; Class (3) *F. pallidoroseum*—Seeds infested with *Fusarium pallidoroseum* fungus; Class (4) *R. solani*—Seeds infested with *Rhizoctonia solani* fungus. The data was partitioned so that 70% was used to train the models and 30% was used for independent validation. In addition, a 10-fold cross-validation was applied. The metrics of general accuracy, Cohen's Kappa coefficient, sensitivity and specificity were used to evaluate the performance of the models. The R 4.0.0 software program (R core Team, 2020) was used to develop the models with the LDA algorithm.

#### **3. Results**

The reflectance patterns in classes of healthy seeds, before incubation, and after incubation with *Fusarium pallidoroseum*, *Rhizoctonia solani* and *Aspergillus* sp. were different in images captured at 780 nm (Figure 1). The nCDA method revealed a slight distinction among classes before incubation (Figure 1a) compared to seeds after incubation (Figure 1b): the intense colonization of the fungi after incubation showed greater separation between healthy and unhealthy seeds and also among the different fungal species.

**Figure 1.** Raw RGB images of cowpea seeds and corresponding transformed images into grayscale and by canonical discriminant analysis (nCDA) captured at 780 nm, with reflectance patters in classes of healthy seeds, *Fusarium pallidoroseum, Rhizoctonia solani* and *Aspergillus* sp. before incubation (**a**), and after incubation (**b**).

Figure 2 shows the mean reflectance spectra at 19 wavelengths in a range from 365 to 970 nm. Before incubation (Figure 2a) all classes showed similar spectral signature with the exception of the '*Aspergillus'* class. However, there was an expressive discrimination among classes after incubation (Figure 2b), especially at wavelengths from 365 to 645 nm, and the '*Aspergillus'* class showed a higher distinction from the other classes across the spectrum. At longer wavelengths, there was a difficulty in distinguishing '*F. pallidoroseum*' from healthy seeds, particular in the NIR region.

**Figure 2.** Spectral signature for classes of healthy seeds, *Fusarium pallidoroseum, Rhizoctonia solani* and *Aspergillus* sp. at 19 wavelengths in a range from 365 to 970 nm before incubation (**a**) and after incubation (**b**).

The reflectance data of the 19 spectrum bands and the color and texture resources were submitted to PCA analysis (Figure 3). Before incubation, components 1 (PC1) and 2 (PC2) were responsible for 73.4% and 11.3% of the total variation, respectively (Figure 3a). The contribution of components after incubation was 77.8% in PC1 and 6.8% in PC2 (Figure 3b). In this context, there was similar behavior of the vectors originating from the spectra reflectance (represented in green), indicating that the 'healthy seeds', '*F. pallidoroseum* 'and '*R. solani*' classes had higher reflectance values compared to '*Aspergillus*'. Meanwhile, the *'Aspergillus'* class showed higher values mainly for CIELab, before and after incubation. The classes of 'healthy seeds', '*F. pallidoroseum* 'and '*R. solani*' showed strong interaction before incubation (Figure 3a), but there was less interaction among them after seed incubation (Figure 3b).

**Figure 3.** Biplots of principal component analysis for multispectral reflectance, color and texture features for classes of healthy seeds, *Fusarium pallidoroseum, Rhizoctonia solani* and *Aspergillus* sp. at 19 wavelengths (365 to 970 nm) before incubation (**a**) and after incubation (**b**). Attributes: Color and texture features – (R2): RegionMSIThresh; (R1): RegionMSI\_Mean; (M19): MultiColorMean1\_(18); (M18): MultiColorMean1\_(17); (M17): MultiColorMean1\_(16); (M16): MultiColorMean1\_(15); (M15): MultiColorMean1\_(14); (M14): MultiColorMean1\_(13); (M13): MultiColorMean1\_(12); (M12): MultiColorMean1\_(11); (M11): MultiColorMean1\_(10); (M10): MultiColorMean1\_(9); (M9): MultiColorMean1\_(8); (M8): MultiColorMean1\_(7); (M7): MultiColorMean1\_(6); (M6): MultiColorMean1\_(5); (M5): MultiColorMean1\_(4); (M4): MultiColorMean1\_(3); (M3): MultiColorMean1\_(2); (M2): MultiColorMean1\_(1); (M1): MultiColorMean1\_(0); (I3): IHSSaturationMean; (I2): IHSIntensityMean;(I1): IHSHueMean; (G7): GraylevelRunStatistics\_(6); (G6): GraylevelRunStatistics\_(5); (G5): GraylevelRunStatistics\_(4); (G4): GraylevelRunStatistics\_(3); (G3): GraylevelRunStatistics\_(2); (G2): GraylevelRunStatistics\_(1); (G1): GraylevelRunStatistics\_(0); (C5): CIELab\_Saturation; (C4): CIELab\_L; (C3): CIELab\_Hue; (C2): CIELab\_B; (C1): CIELab\_A. Reflectance – (B19): Band\_19; (B18): Band\_18; (B17): Band\_17; (B16): Band\_16; (B15): Band\_15; (B14): Band\_14; (B13): Band\_13; (B12): Band\_12; (B11): Band\_11; (B10): Band\_10; (B9): Band\_9; (B8): Band\_8; (B7): Band\_7; (B6): Band\_6; (B5): Band\_5; (B4): Band\_4; (B3): Band\_3; (B2): Band\_2; (B1): Band\_1.

Next, models were developed based on the LDA algorithm using reflectance, color and texture features of the seeds. An overall accuracy of 100% and 92% was observed for the training and test set, respectively, in the first model developed before incubation (Table 1). In testing set for class membership of 'healthy seeds' and '*Aspergillus*', the hit rate was achieved with 100% sensitivity, while other classes showed less individual precision. There was confusion between '*F. pallidoroseum*' and '*R. solani*' (Table 1), since the spectral patters of these classes were very similar (Figure 2).

**Table 1.** Confusion matrices of the LDA model in training and testing set using reflectance, color and texture features of cowpea seeds at 19 wavelengths (365 to 970 nm) for class membership of healthy seed and inoculated seed with *Fusarium pallidoroseum, Rhizoctonia solani* and *Aspergillus* sp.


The second model was created using multispectral data after seed incubation, with an overall accuracy of 100% for both training and testing set (Table 2). The metrics also showed high accuracy of the classification model, with values equal to or greater than 97% in cross-validation, pointing out that the multispectral data can be used to distinguish healthy seeds from seeds carrying different fungal species.


**Table 2.** Confusion matrices of the LDA model in training and testing set using reflectance, color and texture features of cowpea seeds at 19 wavelengths (365 to 970 nm) for class membership of healthy seed and incubated seed with *Fusarium pallidoroseum, Rhizoctonia solani* and *Aspergillus* sp.

The two models developed based on LDA algorithm are shown in Figure 4. The first two discriminatory factors (LD1 and LD2) explained 99.47% of the total variation in the first model, and 94.87% in the second model (Figure 4a,b). Again, the '*Aspergillus'*class was clearly distinguished from the other classes in both statistical models, and there was high interaction between classes of '*F. pallidoroseum*' and *R*. *solani*', before incubation (Figure 4a). Figure 4c,d show the importance of each variable obtained from reflectance, color and texture features for discrimination of the different seed classes. Before incubation, 'CIELabHue' (0.99), 'IHSHueMean' (0.98), CIELab A (0.98), Band 19 (0.98) and Band 18 (0.98) were more effective in discriminating seed classes, and after incubation the 'CIELab B', IHSSaturationMean, MultiColorMean1 [14], GraylevelRunStatistics (2)(3)(4), and Bands 8–17 (Figure 4d), with contribution greater than 0.99. These results emphasize the potential of simple features in discriminating different fungi associated with cowpea seeds.

**Figure 4.** Linear discriminant analysis (LDA) score plot based on reflectance, color and texture features of cowpea seeds at 19 wavelengths (365 to 970 nm) for class membership of inoculated seed (**a**), and incubated seed (**b**) with Healthy seeds, *Fusarium pallidoroseum, Rhizoctonia solani* and *Aspergillus* sp. Importance of variables in the models before inoculation (**c**), and after incubation (**d**). Attributes: Color and texture features – (R2): RegionMSIThresh; (R1): RegionMSI\_Mean; (M19): MultiColorMean1\_(18); (M18): MultiColorMean1\_(17); (M17): MultiColorMean1\_(16); (M16): MultiColorMean1\_(15); (M15): MultiColorMean1\_(14); (M14): MultiColorMean1\_(13); (M13): MultiColorMean1\_(12); (M12): MultiColorMean1\_(11); (M11): MultiColorMean1\_(10); (M10): MultiColorMean1\_(9); (M9): MultiColorMean1\_(8); (M8): MultiColorMean1\_(7); (M7): MultiColorMean1\_(6); (M6): MultiColorMean1\_(5); (M5): MultiColorMean1\_(4); (M4): MultiColorMean1\_(3); (M3): MultiColorMean1\_(2); (M2): MultiColorMean1\_(1); (M1): MultiColorMean1\_(0); (I3): IHSSaturationMean; (I2): IHSIntensityMean; (I1): IHSHueMean; (G7): GraylevelRunStatistics\_(6); (G6): GraylevelRunStatistics\_(5); (G5): GraylevelRunStatistics\_(4); (G4): GraylevelRunStatistics\_(3); (G3): GraylevelRunStatistics\_(2); (G2): GraylevelRunStatistics\_(1); (G1): GraylevelRunStatistics\_(0); (C5): CIELab\_Saturation; (C4): CIELab\_L; (C3): CIELab\_Hue; (C2): CIELab\_B; (C1): CIELab\_A. Reflectance – (B19): Band\_19; (B18): Band\_18; (B17): Band\_17; (B16): Band\_16; (B15): Band\_15; (B14): Band\_14; (B13): Band\_13; (B12): Band\_12; (B11): Band\_11; (B10): Band\_10; (B9): Band\_9; (B8): Band\_8; (B7): Band\_7; (B6): Band\_6; (B5): Band\_5; (B4): Band\_4; (B3): Band\_3; (B2): Band\_2; (B1): Band\_1.

#### **4. Discussion**

The diagnosis of pathogens transmitted by seeds is an important measure in the quality control program, as it avoids the spread of pathogens to exempt areas, economic losses and the unnecessary use of chemicals, thus reducing costs and environmental contamination. Traditional techniques have the characteristic of requiring considerable time for analysis, in addition to subjectivity for interpreting the test. Thus, the use of techniques which minimize this problem is very desirable; in this sense, technological and computational advances enable new methodologies to be used for this purpose.

This study sought to verify the efficiency of MSI in recognizing different fungal species associated with cowpea seeds; it was possible to observe distinctions in the spectral signature between the different seed classes. Variations in the reflectance spectra can be attributed to changes in color, texture and chemical composition of the surface, thereby enabling separation between the classes of infested and non-infested seeds as evidenced by exploratory data analysis. The differentiation between classes before incubation can be attributed due to the change in color caused by the fungi which were adhered to the seed coat; the '*F*. *pallidoroseum*' and '*R. solani*' have a simple mycelia formation in their colonies, which could have resulted in less seed covering by the fungi, whereas there is intense spore production by '*Aspergillus'*, covering the seeds completely. Therefore, it seems reasonable to assume that the conditions before incubation were not favorable for complete fungus development, only resulting in changes in the seed color.

In addition to the coloration, there are the changes caused by the enzymatic and oxidative activity of the fungi on the seeds from the incubation ('Incubated seeds'), which enabled distinguishing the classes more sharply. According to Williams et al. [21], the main source of variation in chemical alteration is an alteration to the starch and protein content which constitute the seed reserves; after incubating the seeds, the fungus starts to consume these compounds, directing them for their growth. The ability of the MSI technique to distinguish fungi-bearing seeds based on physical-chemical changes has already been proven in studies with other species [12–14,21].

Spectral data made it possible to separate the seed classes in the 'Incubated seeds', however this distinction did not always occur in the same region, making the selection of a spectral band common to the class complex. This can be attributed to the overlap and complexity of continuous data, making it difficult to clearly identify the positions of the characteristic bands that represent the different components to be evaluated [22]. In this context, the color and texture parameters were quite expressive in distinguishing the classes, as seen in the determination coefficients (Figure 4). In this context, Boelt et al. [6] point out that the CIELab resource is efficient in distinguishing between fungi species present in barley seeds; this feature is an interesting alternative, as it enables distinguishing color variations which are not perceptible to the human eye [23], eliminating subjectivity from visual inspections. Another color and texture resource, the 'RegionMSImean', has also proved to be efficient in classifying seeds and has already been applied efficiently in several studies. Olesen et al. [10] used this parameter to distinguish *Ricinus cummunis* L. seeds based on their viability with 92% precision. Likewise, Shrestha et al. [24] concluded that the 'RegionMSImean' parameter was efficient in predicting tomato varieties, with a hit rate above 95% and in some cases reaching 100%.

The contribution of color and texture features in distinguishing the seed classes was evidenced by applying the supervised model. The application of supervised methods such as LDA combined with imaging techniques has already shown promise in several studies [9,23–25]; this is because LDA aims to minimize the distance within classes and maximize the distance between classes, thereby enabling good discrimination between classes.

Despite the satisfactory results, it is important to highlight that this is a preliminary study, which can be a guide for future research, covering a greater number of cultivars and species of fungi, isolated or together with the seeds, bringing results that allow a greater application practical and viable in the evaluation of cowpea seeds and other leguminous species of agricultural importance.

#### **5. Conclusions**

The multispectral imaging of cowpea seeds provides the necessary information for quickly distinguishing between different seed classes tested and present accuracy above 92% before incubation and 99% after incubation if associated with a discriminant model; these are promising results, since the amount of data obtained through multispectral imaging is large, and therefore a model capable of selecting the variables which most correlate with a given characteristic, in this case the health status, greatly increases the system's effectiveness, confirming the potential of using technology to assess the seed health.

**Author Contributions:** Conceptualization, C.H.Q.R. and F.G.G.-J.; Funding acquisition, C.B.d.S.; Investigation, C.H.Q.R., F.G.G.-J. and M.H.D.d.M.; Methodology, C.H.Q.R. and M.H.D.d.M.; Resources, C.H.Q.R.; Software, C.H.Q.R., F.F.-S. and A.D.d.M.; Supervision, F.G.G.-J. and M.H.D.d.M.; Validation, A.D.d.M.; Visualization, C.H.Q.R. and C.B.d.S.; Writing—Original draft, C.H.Q.R.; Writing—Review & editing, C.H.Q.R., F.F.-S., F.G.G.-J., M.H.D.d.M., A.D.d.M. and C.B.d.S. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by Fundação de Amparo à Pesquisa do Estado de São Paulo - FAPESP: Grants 2017/15220-7 and 2018/03802-4.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

**Pedro Bello and Kent J. Bradford \***

Seed Biotechnology Center, Department of Plant Sciences, University of California, Davis, CA 95616, USA; pbello@ucdavis.edu

**\*** Correspondence: kjbradford@ucdavis.edu

**Abstract:** *Brassica oleracea* is an important crop species that at early growth stages may exhibit failure of the apical growing point, an abnormality called "blindness". The occurrence of blindness is promoted by exposure to low temperatures during imbibition and germination, but the causes of sensitivity to such conditions are unknown. We combined three analytical seed technology instruments to explore seed physical properties that are highly correlated with quality parameters and might be used directly for grading or sorting seed lots into subpopulations varying in potential susceptibility to blindness. For image analysis, we used the VideometerLab instrument, which can scan 19 wavelengths from ultraviolet to infrared and utilize that information in any combination to potentially identify unique criteria related to seed quality. The iXeed CF Analyzer was utilized to obtain chlorophyll fluorescence values for individual seeds. Chlorophyll contents of many seeds can be used as an indicator of seed maturity, a major contributor to seed quality. Finally, oxygen consumption measurements of individual seeds as obtained with the Q2 instrument are highly correlated with their performance under a wide variety of conditions. Six Brassica seed lots differed in their susceptibility to induction of blindness or loss of viability due to 48 h hydrated incubation at 1.5 ◦C. Analysis of physical and respiratory parameters identified some measurements that were highly correlated with the occurrence of blindness. Higher chlorophyll content, as detected by the CF-Mobile and certain wavelengths in the Videometer, was associated with greater occurrence of blindness or death following the induction treatment, suggesting that more immature seeds may be susceptible to blindness. Further research is required, but methods to detect and sort such seeds based on physical characteristics appear to be feasible.

**Keywords:** *Brassica oleracea*; blindness; multispectral; chlorophyll content; seed respiration; seed vigor

#### **1. Introduction**

*Brassica oleracea* is a morphologically diverse species that has been selected and bred for its leaves (cabbage, collards and kale), stems (kohlrabi), flower shoots (broccoli and cauliflower) and buds (Brussels sprouts). During early seedling growth, plants of all of these crops may lose the apical growing point, an abnormality called "blindness", which usually occurs at low incidence but can cause major losses in the field for growers under some conditions. The occurrence of blind *B. oleracea* plants was described already in the 1940s. It is characterized by termination of leaf primordia initiation and disorganization in the shoot apical meristem (SAM) [1]. The occurrence of blindness is promoted by low temperature combined with low light conditions, and seed production conditions can play a role in the seed lot sensitivity as well [2]. Recent studies have confirmed that both genetics and seed production environment contribute to the occurrence of blindness [1]. Seed treatments that reduce susceptibility to blindness also have been developed [3]. Early identification of affected plants before transplanting them into the field has not been possible, resulting in high economic losses that can be up to 95% in broccoli under some conditions [2].

**Citation:** Bello, P.; Bradford, K.J. Relationships of *Brassica* Seed Physical Characteristics with Germination Performance and Plant Blindness. *Agriculture* **2021**, *11*, 220. http://doi.org/10.3390/ agriculture11030220

Academic Editor: Alan G. Taylor

Received: 26 January 2021 Accepted: 2 March 2021 Published: 8 March 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

Seed production for these crops can be complex due to their indeterminate flowering habit [4]. At a given time during production, these indeterminate species will have immature, mature, over-mature and shattering seeds present simultaneously. Early seed harvest can result in poor seed quality and low germination due to immaturity [5], while delayed harvest may sacrifice up to 50% of seed yield under adverse conditions [6]. In addition to losses due to shattering, a significant fraction of the seed lot is discarded during seed processing due to the removal of smaller, immature seeds. This variability in maturity levels can impact seed lot quality, as immature seeds will lose vigor and viability at a faster rate than mature seeds [5]. These problems can prevent sale of seed lots that do not reach minimum germination levels, with economic losses from discarded lots.

Here we combine three analytical seed technology instruments to explore seed physical and physiological properties that could be highly correlated with quality parameters and potentially used for grading or sorting seed lots to remove lower quality subpopulations. For image analysis, we used the VideometerLab instrument, which can scan 19 wavelengths from ultraviolet to infrared and utilize that information in any combination to measure seed size, detect microbes, classify damage, and potentially identify unique criteria for assessing seed quality [7–11]. The iXeed CF Analyzer was utilized to obtain chlorophyll fluorescence values for individual seeds. Seed chlorophyll contents of many species can be used as an indicator of seed maturity, a major contributor to seed quality [5,12–16]. Finally, oxygen consumption measurements of individual seeds as obtained with the Q2 instrument are highly correlated with their performance under a wide variety of temperature, water potential, hormonal, priming, aging, and other conditions [17–19]. We used these methods to explore the possibility of identifying early indicators of susceptibility to induction of blindness in kohlrabi seeds.

#### **2. Materials and Methods**

#### *2.1. Seed and Plant Materials*

Six kohlrabi (*B. oleraceae* L. var. *gongylodes*) seed lots comprised of three F1 varieties (A, B, C) with two lots of each (1 and 2, 3 and 4, 5 and 6, respectively) exhibiting different susceptibilities for blindness were provided by Bejo Zaden (Warmenhuizen, The Netherlands).

#### *2.2. Blindness Induction*

After initial measurements of physical characteristics of dry seeds, a blindness induction treatment was performed on the seeds preceding respiration measurements. We adapted a published protocol that demonstrated the ability of low temperature treatments to cause shoot apical meristem arrest in *Brassica oleracea* seedlings [1]. Seeds were imbibed in microtiter plate wells in 60 uL of water and incubated at 1.5 ◦C in a foil-covered incubator (Benchmark IS-1010R placed in a 4 ◦C room) for 48 h in darkness and then transferred to respiration tests, maintaining individual seed positioning and identities from previous seed imaging throughout.

#### *2.3. Physical Characteristics Measurements*

Chlorophyll content. The iXeed CF Analyzer (Figure 1; CF-Mobile; SeQso B.V., The Netherlands) was utilized for chlorophyll fluorescence (CF) measurements [5]. A set of 46–48 seeds per lot were placed on blue metal trays with proper-sized pockets, organized in six rows by eight columns, corresponding to the plate layout and capacity for seed respiration measurements. Three measurements were captured on each tray and seed parameters registered by the CF software were recorded and exported to Microsoft Excel (version 16). The average and standard deviation of CF level and CF size per seed were calculated and combined with parameters gathered subsequently. CF information was captured for a total of 174 seeds for each lot. The parameters derived from the CF-Analyzer data are defined in Table 1.

**Figure 1.** Visual workflow for every seed measured in the study. Starting with chlorophyll content and multispectral imaging taken of the dry seeds (left), followed by the blindness induction treatment, additional multispectral imaging, then seed respiration measurements, additional multispectral data collection and finally seeds were transferred to the greenhouse for growth and plant blindness evaluation.

Multispectral imaging. The VideometerLab instrument (Videometer A/S, Herlev, Denmark) was used for multispectral analyses [7,8,20]. The instrument is equipped with a camera inside an integrating sphere along with diodes that emit light at the following 19 wavelengths: 375, 405, 435, 450, 470, 505, 525, 570, 590, 630, 645, 660, 700, 780, 850, 870, 890, 940 and 970 nm. The same 174 seeds initially scanned for CF information were imaged and analyzed in the VideometerLab instrument for each lot. Multispectral pictures were taken and grouped for each 46–48 seeds, respecting the CF measurements positioning. These seeds were placed in coded 6 (rows) by 8 (columns) cells with each seed placed slightly lower than the previous one, aiding the Videometer software sequence numbering. Over two experimental repetitions, multispectral images of the same seed were captured in different stages of the experiment (see Figure 1): (1) dry seed (all 174 seeds per lot); (2) after blindness induction treatment (110 seeds per lot); and (3) after 72 h at 15 ◦C for seed respiration measurements (96 seeds per lot) where these seeds were rapidly transferred to marked microtiter plate lids and Videometer images were acquired to quantify seedling area, respecting the seed positioning from the Q2 equipment. A series of up to eight pictures was taken per Q2 plate to avoid overlapping tissues and data were combined together in the BLOB (Binary Large Objects) collection. Selected parameters derived from the VideometerLab data are defined in Table 1.

#### **Type Parameter Description** Multispectral Area Average or individual projected area (mm<sup>2</sup>) calculated for the seed. CIELab CIELab refers to a color space defined by the International Commission on Illumination (CIE); it expresses color as three numerical values: L for lightness from black (0) to white (100), A from green (−) to red (+) and B from blue (−) to yellow (+). Saturation Average amount of pixels that exceed a maximum value of brightness in the image. Hue Average angular position related to a color space coordinate enclosing all colors. 395–970 nm Average reflectance of specific wavelengths (in nanometers) for individual seeds. The specified wavelength followed by "SD" refers to the average standard deviation or pixel reflectance variation for individual seeds. Tissue Area Area of uncoated or visbible tissue present in individual seeds, quantified by the number of pixels. White Spots Area of white coloration in seed coats as a percentage of the individual seed area. Chlorophyll Fluorescence CF Value Average or individual chlorophyll fluorescence measured for the seed (total seed fluorescence divided by fluorescence size). CF Size Average or individual calculated size (mm<sup>2</sup>) of the CF area in the seed (sum of the pixels that have a CF level above a threshold, and converted to mm<sup>2</sup>). Seed Respiration R75.Time, R50.Time and R25.Time Time in hours for individual seeds to deplete oxygen in vials to 75, 50 and 25%, respectively. R75, R50, R25 POD curves Cumulative population oxygen depletion (POD) time course plotting the percentage of seeds depleting the oxygen level to 75, 50 or 25% of the initial, respectively, at each time. R50 (50) Time in hours to 50% level reached on the R50 POD curve. R75.Final, R50.Final and R25.Final Final percentages of the R75, R50 and R25 POD curves, respectively, of the initial value. Final-O2 Final oxygen concentration in the vials after 72 h of imbibition. Greenhouse Plant Evaluation Plant Blindness Score Plants were ranked as dead, normal or blind. Blind plants included: plants without shoot apical meristem [SAM]; plants with needle shaped (first) leaf; plants with funnel shaped (first) leaf; plants without SAM, but with lateral branches; plants with SAM and lateral shoots that are needle- or funnel-shaped or otherwise malformed; plants with oversized first leaf but lackingSAM;plantswithabnormalbrancharchitecture.

#### **Table 1.**Parameter Definitions.

#### *2.4. Seed Respiration Measurements*

The Q2 instrument (now called Seed Respiration Analyzer; Figure 1; Fytagoras B.V., Leiden, The Netherlands) measures oxygen consumption (respiration) rates of individual seeds repeatedly during imbibition and germination. Individual seeds were placed into 2 mL screw-cap vials containing 1.45 mL of agar (0.4% *w/v*) and 0.2% Plant Preservative Mixture (PPMTM), which were sealed with caps that have a dot of a fluorescent polymer centered on their internal side. The polymer contains a dye that changes its fluorescent properties in response to oxygen concentration [21]. As the seed respires, it depletes the oxygen in the sealed well or vial, which changes the fluorescence intensity of the dye. This change is detected by a light source that shines on the dot and a sensor that measures the fluorescence intensity. A robotic arm sequentially moves the light source/sensor over each well, measuring the oxygen concentration inside the wells. Up to 16 plates of 48 vials can be positioned in the apparatus at a time and automatically measured by the robotic sensor, and the measurements can be repeated frequently to obtain time courses of oxygen consumption activity. Measurements reported here were collected every 30 min. Seeds were transferred individually to 2 mL vials using tweezers after pictures were taken in the CF-Analyzer and the VideometerLab. Sample temperature was controlled to ±0.5 ◦C using Peltier heating/cooling units and fans. In preliminary tests, 48 non-induced seeds (control) per lot were measured at 20 ◦C, followed by plant blindness evaluation. Thereafter, all seeds were measured in separate 48-well plates at 15 ◦C for 50 or 72 h, and placement in the Q2 cells was paired to the Videometer and CF-Mobile codes for each seed when applied. Two experimental repetitions were utilized. The parameters derived from the Q2 data are defined in Table 1.

#### *2.5. Plant Blindness Evaluation*

After the blindness induction treatment and the respiration measurements, seeds were transplanted to marked trays with numbered cells and placed in a greenhouse at 22 ◦C and natural light. The day length during plant growth varied from 13 to 15 h in the initial experiment (carried in April through May 2019) and 13 to 11 h during the experiment repetition (September through October 2020). Individual plants grown for 2.5–5 weeks were evaluated and ranked as dead (no seedling emerged), normal or blind plants (Figure 1). The parameters derived from the plant evaluations are defined in Table 1.

#### *2.6. Data Analyses*

Data analyses for Videometer images were performed using VideometerLab and the Classifier Design Tool (CDT) software version 3.18.11 (Videometer A/S). We separated seeds from the image background using normalized canonical discriminant analysis (nCDA) transformation followed by simple threshold segmentation within the Videometer software. Several multispectral images were used to create the nCDA transformation model with selected areas of images representing the 2 classes: areas of seeds or background. Automatic normalization was performed to maximize the Rayleigh quotient and input data received a preprocessing band normalization (at 645 nm) and output data was centered around the overall mean between the classes and scaled with the two classes showing means at +1 or −1. The actual data were visualized in a scaling between −2 and 2.

A similar approach was used to quantify the visible tissue area after seed respiration measurements, but in this case using the input with multispectral images featuring the seed coat or embryo/seedling tissue area as the two classes to be separated. The input normalization in this case was performed using preprocessing band normalization at 470 nm, while output normalization was similar. The seedling or tissue areas were quantified by numbers of pixels.

Data from the CF-Analyzer, Q2 and plant evaluations were exported or compiled using Microsoft Excel (version 16). The compiled data were then analyzed in R version 4.0.3 using RStudio version 1.2.1335. Type I analysis of variance (ANOVA) was run on models with experiment as random effect. Normality and heteroscedasticity of the data was visually inspected with histograms and diagnostic plots for all parameters reported using the linear regression analysis (lm) models in R. Tukey Honest Significant Differences were then calculated using the TukeyHSD() function in R and the HSD.test() function of the R package *agricolae* [22].

Boxplots were made using ggplot from the ggplot2 R package [23]. Correlation matrices were calculated using the rcorr.test function from the psych R package [24] and plotted with the corrplot R package [25]. Family-wise error rate was accounted for with adjusted *p*-values for multiple comparisons using the Holm method [26]. Multiple factor analysis (MFA) was performed in R with the FactoMineR package [27] and additional tools from the factoextra package [28]. Only quantitative variables without missing values and statistically correlated with blindness (*p* < 0.01) were used for the MFA. All comparisons mentioned were statistically significant at *p* < 0.05 unless otherwise stated.

#### **3. Results**

#### *3.1. Brassica Blindness: Initial Assessment*

All seed lots were first tested for seed respiration and plant growth to identify blindness present in the seed lots and to characterize their vigor prior to the blindness induction treatment. The initial seed respiration test was conducted at 20 ◦C and all seed lots displayed largely homogeneous and rapid oxygen depletion rates (Supplemental Figure S1). At least 80% of seeds in all lots depleted oxygen in the vials to the 50% level within two days after imbibition and at 72 h most seeds were in anaerobic conditions in the vials. These seeds were then transferred to the greenhouse for plant growth evaluation, and no blind plants were identified after 2.5–5 weeks (data not shown).

To further investigate the blindness potential and susceptibility in all lots, we introduced moderate temperature stress during germination by lowering the temperature to 15 ◦C in the Q2 test. As expected, oxygen depletion rates were slower for all lots and larger variation in respiratory patterns within lots was also evident (Supplemental Figure S2). The modest temperature stress did not induce blindness, with only one seed showing some blindness symptoms in Variety B, lot number 3. Additionally, a significant number of seeds in most seed lots (except A-2 and B-4) did not reduce the oxygen within vials to the 50% level following imbibition for 50 h. The results demonstrated that little or no blindness was expressed in the seed lots tested under optimal or moderately low temperatures during imbibition and germination.

#### *3.2. Physical Characteristics Assessment and Brassica Blindness Induction*

We tested whether the CF-Analyzer, the VideometerLab or the Q2 were able to detect differences among lots and the potential presence of blind seeds. Chlorophyll fluorescence measurements were initially captured for all dry seeds (Figure 2). The seed lots within B and C varieties displayed a significant difference between each other (*p* < 0.001) while the lots in Variety A did not. Seed lots 5 (Variety C, 90.7) and 3 (Variety B, 78.4) displayed higher median chlorophyll contents when compared to the other lots (27.9–37.6). Additionally, these two lots had higher CF level variation (Figure 2, CF Level—larger bars/standard deviation on lots 3 and 5) in comparison to all other seed lots, which had more homogeneous low CF levels, although some outliers were present in these lots (Figure 2, black dots on lots 1, 2, 4 and 6). These lots with low median CF levels were not statistically different from each other but were significantly different from lots 3 and 5 (Figure 2, *p* < 0.001). Similar differences among lots were detected for CF size (or area) (Figure 2).

**Figure 2.** CF-Analyzer parameters (see Table 1) across varieties (columns) and seed lots (color-coded). Letters indicate significant differences among all lots for each parameter (rows) as calculated by ANOVA and Tukey HSD.

Multiple seed features were measured using the Videometer (123 parameters in total over all experimental stages, Supplemental Table S1), including seed size, shape, color and multispectral characteristics. Selected features that displayed some relationship with plant blindness, viability, or parameters gathered by the other analytical equipment used here are described in Table 1. Individual data were obtained for 174 seeds (46 seeds in the first repetition and 128 in the second repetition) for each seed lot (Supplemental Table S1). Several captured seed features differed significantly among lots (Figure 3). The calculated seed area was largest for seed lot 2 (Variety A, 3.84) and smallest for seed lot 4 (Variety B, 2.96). Color space (CIELab L, B) and saturation values together with average reflectance at longer wavelengths, such as red (645 nm) and near-infrared (870 nm) showed consistently and significantly higher (*p* < 0.001) values for seed lots 5 (C) and 3 (B) compared to the other lots (Figure 3), as observed for their CF values. A similar result was obtained for the ultraviolet (375 nm) wavelength, but it also included lot 1 (*p* < 0.001) along with the other two high-valued lots. Color and multispectral values for lot 5 also were significantly higher than for other lots (*p* < 0.001) but comparable with lot 1 for reflectance measured at the indigo color (435 nm) and also similar to the two Variety A seed lots (1 and 2) for Hue values. Wavelengths 435 and 645 nm are indications of chlorophyll A and B levels, respectively.

The spectrum reflectance standard deviation within seeds (or pixel variation) was also calculated on an individual seed basis for all wavelengths. This value quantifies the color or spectrum variation of each seed; seeds with a uniform color will display small values while seeds with a diversity of colors or shades will display larger values. Here we show the reflectance standard deviation for the near-infrared (NIR) 970 nm wavelength, which displayed some relationship with plant performance when measured at the dry seed and after blindness induction phases, although the standard deviation for other wavelengths also displayed similar results (Supplemental Figure S3). The calculated standard deviation for the NIR wavelength (970 nm) showed larger median variation (5.75–6.02) for seed lots 3 and 5 with lowest values for lots 6 (4.86) and 2 (4.50) (Figure 3).

The post-blindness-induction (PBI) seed area had a median increase of about 20% for all lots compared to dry seeds, reflecting expansion due to imbibition (Figure 4). The color space parameter CIELab B (blue (−) to yellow (+)) had a median increase of about 50% in most lots with a smaller increase of 32.2% observed in lot 5 (Variety C), which presented the higher (*p* < 0.001) value for CIELab B before induction (Figure 3). Similar relationships were observed for the saturation, hue and 970 nm-SD values (Figure 4), in which the seed lot with the lower initial value had the largest relative increase after blindness induction.

**Figure 3.** Selected Videometer features (Table 1) across varieties (columns) and seed lots (color-coded). Letters indicate significance between all lots for each parameter (rows) as calculated by ANOVA and Tukey HSD.

**Figure 4.** Selected VideometerLab features presented as percentage change in the feature levels post-blindness induction (PBI) relative to initial dry seed analyses for the same seeds. Letters indicate significant differences among all lots for each parameter (rows) as calculated by ANOVA and Tukey HSD.

Q2 measurements at 15 ◦C on the six seed lots after the cold temperature imbibition treatment indicated an overall delay in respiration rates (Figure 5, Supplemental Table S2—bottom section, mean R75 values ranging from 26.2 to 42 h) compared to measurements at a similar temperature on seeds prior to the induction treatment (Supplemental Figure S2, Supplemental Table S2—middle section, mean R75 values ranging from 22.5 to 30.6 h). In addition, a much larger fraction of induced seeds did not consume oxygen after 72 h (Figure 5, Supplemental Table S2—bottom section, e.g., final R75 POD curves values ranging from41 to 98%) compared to non-induced seeds tested earlier for 50 h (Supplemental Figure S2 and Table S2—middle section, e.g., final R75 POD curves ranging at 80–93%). The mean oxygen level at different time points is also a convenient parameter to quantify the respiration profile and variation (Supplemental Table S2, O<sup>2</sup> at 48 and 72 h, mean and standard deviation, respectively). Lack of oxygen consumption usually indicates lack of seed viability [19], suggesting that the blindness induction treatment had killed some seeds, as was reported previously regarding this treatment [1].

**Figure 5.** Oxygen depletion curves for individual seeds of kohlrabi seed lots tested at 15 ◦C for 72 h after blindness induction treatment at 1.5 ◦C for 48 h. Dashed lines represent the median (red) and average (black) oxygen depletion time courses for the entire seed population.

Based on the median oxygen depletion and R75 POD curves (Figure 6), lots 2 (light green) and 4 (light blue) had overall faster median oxygen depletion rates and higher percentages of seeds depleting the oxygen to at least the 75% level. Lot 5 (orange) contained a fraction of seeds respiring even faster than lot 4, but only around 72% of seeds in that lot consumed more than 75% oxygen in the vials, compared to 89 and 100% in lots 2 and 4, respectively. Lots 1 (dark green) and 3 (dark blue) displayed the lowest oxygen consumption (final medians of 65.5 and 84.5% oxygen remaining, respectively) and slower oxygen consumption (2 lowest R75 POD curves), while lot 6 (yellow) performed somewhat better (final median O<sup>2</sup> depletion curves of 53.7% and slightly faster R75 POD curves).

Some Q2 parameters were selected for comparison among lots (Figure 7). Lots 1, 3 and 6 displayed the slowest median times to 75% (36.2 to 42 h) and 50% (47.6 to 58.2 h) remaining oxygen levels. Area under the curve parameters (R75.Area and R50.Area) are highly correlated with the time to required to lower the oxygen to the same levels (R75.Time and R50.Time) but add more detailed information regarding the oxygen consumption profiles and shapes of depletion curves. As expected, distributions of areas under the curve for the 75% oxygen remaining level also showed lots 1, 3 and 6 as slower ones (larger area values ranging from 30.95 to 36.44), but at the 50% oxygen level lot 6 displayed a somewhat faster but significant (*p* < 0.001) oxygen consumption compared to lot 3 (Figure 7, R50.Area). Seed lots 2, 4 and 5 usually showed lower median times and area values compared to the slower lots and could be considered significantly (*p* < 0.001) faster respirators in most cases. It is important to point out that the lower the remaining oxygen level chosen to compute these values, the smaller the fraction of seeds that are used to calculate them. In this case at least 40% of the seeds were used to calculate parameters generated based on the 75% oxygen level (lot 3 with lowest final percentage in the R75-POD curve, Figure 6) but a little over 20% of the seeds were used to calculate values based in the 50% oxygen level (slow lots 1 and 3—R50 POD curves, Supplemental Table S2—bottom section). To overcome this issue for less vigorous or stressed lots, the remaining oxygen levels at a particular

time for all seeds can be used. For example, the final oxygen levels in the vials after 72 h (before seeds were transferred to the greenhouse), clearly showed the wide distributions of respiration rates among seeds in these lots after the blindness induction treatment. Seed lots 2 and 4 had relatively homogeneous low median final oxygen levels (19.4 and 22%, respectively), lot 3 displayed an intermediate variance but at the highest final oxygen level (66.9%), while lots 1, 5 and 6 displayed more intermediate final oxygen levels (41.7, 39.4 and 33.1%, respectively) but with large heterogeneity among seeds (Figure 7, Final-O2).

**Figure 6.** Median oxygen depletion curves (top) and R75 POD curves (bottom) of kohlrabi seed lots tested at 15 ◦C for 72 h after blindness induction treatment at 1.5 ◦C for 48 h.

Seedling or tissue area measurements from the Videometer using nCDA models (Figure 8—top sections) after seed respiration measurements exhibited significant differences among seed lots. Seed lot 3 had the smallest exposed tissue area (median at 1602 pixels) with little to no seedling tissue visible in a large fraction of the seeds after the respiration measurements (Figure 8—see bottom panels for BLOB collection with seeds marked with dark blue circles compared to seed lot 4 of the same variety marked with light blue circles). Seed lot 5 displayed an intermediate median tissue area (2809 pixels), while lots 1, 2, 4 and 6 had higher median tissue areas at around 4200 pixels (Figure 9). The largest seedling size variation was present in seed lot 1, where a fraction of seeds showed little to no embryo tissue while another fraction displayed the largest seedlings in the study. This is consistent with the large variation among seeds in this lot for Q2-derived values (Figure 7).

**Figure 7.** Selected Q2 parameters (see Table 1) of seed lots after blindness induction treatment. Letters indicate significance between all lots for each parameter (rows) as calculated by ANOVA and Tukey HSD.

Finally, seeds were transferred to marked trays and placed in the greenhouse at 22 ◦C for 2.5–5 weeks, when plants were scored as normal, blind or dead (failed to emerge). Lots 2, 4 and 6 had the largest fractions of normal plants (≥80%) while lots 1, 3 and 5 had smaller fractions of normal plants (>62%) (Figure 10). Additionally, lots 3 (19.1%) and 5 (20.9%) had the largest fractions of blind plants, followed by lot 1 (10.9%), lot 4 (7.3%) and lot 6 (2.7%); lot 2 did not display any blind plants after the induction treatment. The percentages of non-viable plants were higher in lots 3 (43.6%), 1 (27.3%) and 5 (26.4%), while lots 4 (12.7%), 6 (10.9%) and 2 (2.7%) exhibited lower percentages of seed death (Figure 10).

**Figure 8.** Examples of seedling or plant tissue area in pixels (**top panel**). Area was calculated using nCDA models and proper threshold value to include only relevant tissue and seedlings pixels. Sample of nCDA transformed image where every pixel is scored and only pixels above a certain threshold are counted (**middle panel**). Seedlings from all treatments were isolated and measured, blob (binary large object) collections with samples of these seeds and seedlings sorted by tissue/seedling area for seed lots 3 (dark blue circles) and 4 (light blue circles) are illustrated here (**bottom panels**).

**Figure 9.** Seedling area (in pixels) of induced seeds after seed respiration measurements and before transfer to the greenhouse for plant growth and evaluation. Letters indicate significance difference among all lots as calculated by ANOVA and Tukey HSD.

**Figure 10.** Plant evaluation for induced seeds and scores for dead, blind and normal seedlings after 2.5–5 weeks of growth in the greenhouse following 48 h at 1.5 ◦C and 72 h at 15 ◦C for Q2 measurements. Plants were scored per the descriptions in Table 1.

All individual seed data for chlorophyll content, multispectral reflectance, seed respiration and greenhouse plant evaluation scores were consolidated in one data file along with the variety, lot and repetition number (Supplemental Table S1). A full correlation matrix (Supplemental Figures S3 and S4) was constructed using all the data, enabling inspection of relationships among a large number of parameters at once to direct further analyses. To summarize the most critical information and avoid duplicating data, some primary parameters were selected and are presented in a smaller correlation matrix (Figure 11). The correlation numbers presented here provide an indication of their potential for use in individual seed sorting using the intersected parameters.

As expected, some parameters collected in the same instruments throughout all stages of the study were highly correlated with each other (Figure 11; Supplemental Figures S3 and S4, darker blue and red clusters). Correlations between the different analytical instruments used were also expected and observed in some cases. The chlorophyll fluorescence parameters from the CF-Analyzer were correlated with each other and also exhibited a strong correlation with multispectral parameters collected in the VideometerLab at different experimental stages (Figure 11, rows 1 and 2). This relationship was anticipated as both instruments are based on spectral imaging, with the CF-Analyzer targeting only chlorophyll content measurements with specific excitation wavelength and fluorescence wavelength filter while our VideometerLab version measures a broad range of wavelengths but lacks fluorescence filters (although addition of these is possible in the instrument).

**Figure 11.** Pearson correlation matrix of selected main parameters (Table 1) from the CF Analyzer (rows and columns 1–2, dashed green), the VideometerLab (rows and columns 3–9 and 11, dashed purple) the Q2 respirometer (row and column 10, dashed orange) and greenhouse tests (rows and columns 12–13) for all induced seeds. Parameters are also organized in the order they were captured in the different stages of the study: dry seeds (rows and columns 1–9, brown bars), induced seeds during germination (rows and columns 10–11, orange bars) and in the greenhouse (rows and columns 12–13, green bars). The Pearson correlation coefficients between the row and column parameters are graphically displayed as circles; their size and color shadings indicate the strength of the correlation coefficients from 0 to 1 for a positive relation (blue color) or 0 to −1 for a negative relation (red color) (only correlation coefficients with significance level for p values below 0.01 are displayed).

The CF level (Figure 11, row 1 and all columns) had a significant relationship with VideometerLab color space parameters collected from dry seeds (CIELab L and B, r = 0.53 and 0.67, respectively, *p* < 0.001), saturation (r = 0.65, *p* < 0.001) and several wavelengths, including the ultraviolet (UV, 375 nm) and the near-infrared (NIR, 875 nm) ranges, with the highest correlation with 780 nm (r = 0.62, *p* < 0.001). This seed maturity indicator was also correlated (positively or negatively) to measurements performed at different stages of the study, such as multispectral imaging after the blindness induction (Supplemental Figure S3), during germination (final oxygen concentration in the Q2 and seedling area after the Q2, r = 0.37/0.45, respectively, *p* < 0.001) and with plant performance in the greenhouse (viability and blindness, r = −0.31/0.29, respectively, *p* < 0.001). Similar relationships were observed for the CF size parameter with weaker but still highly significant (*p* < 0.001) correlations for the dry seed and after induction, germination and greenhouse stages (Figure 11, row 2 and all columns).

The VideometerLab provides numerous parameters to quantify seed characteristics related to size, shape, spectra and others, but it also allows the application of customized features such as the tissue area and white spots/markings that we developed and used in this study (Supplemental Figure S4). In addition to the strong relationship with CF parameters, several of these features collected in dry seeds were correlated with VideometerLab features collected at different stages, Q2 measurements, and plant performance scores in the greenhouse. Some VideometerLab features collected at the dry seed stage were strongly correlated with data collected after blindness induction; these included the color space and saturation parameters from dry seeds and the percentage change in the same parameters after blindness induction (Supplemental Figure S3). The percentage change in the NIR

reflectance variation (PBI-970 nm-SD, Figure 4) was the PBI parameter with the highest (negative) correlation coefficients with Q2 parameters (Final-O2, r = −0.38, *p* < 0.001), seedling area and plant performance (blindness, r = −0.19, *p* < 0.001) (Supplemental Figure S3). These Videometer parameters obtained after the blindness induction were, in most cases, highly correlated with original parameters from dry seeds (e.g., PBI-970 nm-SD with the dry seed 970 nm-SD) and exhibited lower correlations with seed respiration and plant performance. Thus, their relevance for sorting purposes was diminished and they were not included in the correlation matrix.

Videometer color space parameters and saturation values from dry seeds were further associated with the final oxygen measured in the Q2 (r values ranging from 0.23 to 0.27, *p* < 0.001), seedling area after the Q2 (r = −0.31 to −0.34, *p* < 0.001) and viability (r = −0.17 to −0.18, *p* < 0.001) and blindness (r = 0.27 to 0.28, *p* < 0.001) in the greenhouse. Average reflectance of several wavelengths collected in dry seeds also displayed strong correlations with the changes after induction but also with seed respiration, seedling area and plant performance. Some examples of the main wavelengths include UV (375 nm), red (645 nm) and NIR (870 nm) wavelengths that displayed associations with final oxygen level (r = 0.16 to 0.24, *p* < 0.001), seedling area after the Q2 (r = −0.29 to −0.38, *p* < 0.001), viability (r = −0.15 to −0.18, *p* < 0.001) and blindness (r = 0.21 to 0.28, *p* < 0.001). These wavelengths displayed a similar relationships with the quality parameters, but their correlation with CF level was somewhat distinct, with the 375 nm wavelength displaying a lower correlation (r = 0.35, *p* < 0.001) while the red and NIR wavelengths were more closely associated with CF level (r = 0.58 and 0.62, respectively, *p* < 0.001).

Respiration measurements in the Q2 also displayed associations with seedling area and plant performance. The oxygen percentage after 72 h (Final-O2, Figure 11—row and column 10) was highly negatively correlated with seedling tissue area (r = −0.63, *p* < 0.001) and plant viability (−0.42, *p* < 0.001) and positively with blindness (r = 0.28, *p* < 0.001). These results reinforce that seed respiration is a good indicator for germination timing; seeds with a higher oxygen consumption rate also germinated earlier and had more time for seedling growth and development. Furthermore, the seedling tissue area was highly correlated with plant viability (r = 0.55, *p* < 0.001) and negatively with blindness (r = −0.28, *p* < 0.001).

The correlation among these selected traits across all seed lots (Figure 11) is mostly preserved when this dataset is analyzed separately within each seed lot or within each variety (data not shown), although the strength of correlations varied among lots and varieties. The only exception was found in the seed lots from variety A where the relationship between CF level and blindness was not present, likely due to the absence or limited number of seeds displaying the blindness phenotype (Figure 10).

A multiple factor analysis (MFA) with integrated parameters correlated to blindness from the different stages of the study was utilized to verify whether the distinct plant scores (normal, blind or dead) could be discerned (Figure 12; Supplemental Table S3). The MFA reinforced the complexity of distinguishing among these classes using an unguided approach, but it also revealed a clear trend, with blind and dead classes having considerable overlap but being clearly separated from the normal class (Figure 12, left panel). The first dimension (Dim1 accounting for 57.9% of the total variation, x axis Figure 12) distinguished the majority of blind and dead seeds from the cluster of normal seeds. The main parameters that contributed to this separation were the Videometer parameters 645 nm, saturation, 870 nm, 375 nm, CIELab L and A, 970 nm-SD, followed by CF level and size (Figure 12, right panel). The second dimension (Dim2 accounting for 14.3% of the total variation, y axis Figure 12) was most effective in distinguishing between the normal and dead classes. The top parameters in this dimension were the Final-O<sup>2</sup> and the tissue area (antagonistic), which contributed together more than 80% of the total dimension, followed by smaller contributions from the 970 nm-SD, CIELab A, CF level, 645 nm and the saturation (Figure 12, right panel).

**Figure 12.** Multiple Factor Analysis (MFA) of features obtained from imaging, respiration and plant growth observations of kohlrabi seeds after blindness induction treatment. (**Left**): Scatter plot of individual seeds identified as normal, blind or dead separated in two dimensions (Dim1 and Dim2) according to their analytical features. Ellipses define confidence areas (95%) for each plant score, while squares represent their corresponding centers of gravity. Additional supplementary qualitative variables for seed variety (A, B or C) and lots (1 to 6) are shown in black. (**Right**): Vector representation of the influence of different measured factors in relation to their contribution to the two principal dimensions. Only complete observations for all parameters shown were used to generate the MFA.

#### **4. Discussion**

The potential for kohlrabi (and other *Brassica*) seeds to exhibit blindness, particularly after exposure to low temperatures, creates risks for both seed companies and growers. It has been difficult to identify the specific genetic and/or environmental factors during seed development that result in susceptibility to blindness. While conditions that can promote expression of blindness are known [1] and seed pretreatments can ameliorate susceptibility [3], it would be valuable to identify correlated traits that could be used for prescreening for blindness susceptibility to assign problematic lots for seed treatment or for sorting seed lots to remove the seeds that are susceptible to blindness. Thus, we examined both physical (CF-Analyzer, VideometerLab) and physiological (Q2) approaches to screening individual seeds prior to and after inducing blindness in order to identify whether it would be possible to predict which seeds or lots would be more likely to exhibit blindness.

All of the tested lots exhibited good initial performance at 20 ◦C, based on seed respiration time courses (Supplemental Figure S1). Lowering the temperature to 15 ◦C, however, resulted in much larger variances among seeds and discrimination among the seed lots (Supplemental Figure S2), with lots 2 (Variety A) and 4 (Variety B) exhibiting the greatest respiratory capacity at 15 ◦C (most seeds consuming most of the available oxygen). The subtle temperature stress resulted in few or no blind plants, so a blindness induction treatment was required to reveal the desired phenotype for this study. Following the induction treatment, lots 2 (Variety A) and 4 (Variety B) once more exhibited more active respiratory profiles at 15 ◦C (Figure 5), and also lower percentages of dead or blind seeds (Figure 10). In contrast, lots 1 (Variety A) and 3 (Variety B) showed greater impairment in respiratory activity at 15 ◦C and the highest susceptibility to blindness/death due to the induction treatment (Figures 5, 6 and 10). The behaviors of lots 5 and 6 from Variety C were intermediate, as these lots displayed a split respiratory behavior with about half of the seeds consuming most of the oxygen available while the other half consumed little to none (Figures 5 and 6). This result for lot 6 was rather anomalous, as it exhibited relatively poor respiratory capacity at 15 ◦C (Figure 5), but low susceptibility to blindness/death

(Figure 10). As the effect of the induction treatment increases progressively with longer times of exposure [1], it could be that lot 6 would show greater effects after a longer induction treatment. For the purposes of this experiment, the seed lots exhibited a range of susceptibility to death/blindness from the induction treatment, making it possible to test whether physical measurements would be related to this physiological behavior.

Differences among the seed lots at the dry seed stage were evident from parameters determined by the CF-Analyzer and the VideometerLab. Lots 3 (Variety B) and 5 (Variety C) exhibited higher CF values (Figure 2), as well as higher values for CIELab L, CIELab B, Saturation, 645 nm, 870 nm and the variation of the 970 nm reflectance (970 nm-SD) (Figure 3). The 375 nm and 435 nm values also identified lot 1 within the same group (exhibiting high values) as lots 3 and 5 (Figure 3), in agreement with these lots having more dead and blind seeds (Figure 10). Thus, the shorter wavelength parameters, specially 375 nm, suggest a new possibility for sorting, as that wavelength was not as highly correlated with CF Level as the 645 nm and 870 nm measurements (Figure 11). The 375 nm measures may detect another factor that could also be related to seed maturity. As higher values for all these measurements were associated with greater blindness and fewer normal seedlings (Figure 11), more immature seeds, indicated by higher chlorophyll levels, therefore appear to be associated with greater susceptibility to damage by low temperature imbibition.

Measurements performed after the low temperature induction treatment overlapped among seed lots (Figure 4). The saturation value change displayed significant relationships with blindness, but the more relevant relationship was observed with the NIR variation parameter (PBI-970 nm-SD) (Supplemental Figure S3), which was also somewhat more efficient to separate lots (Figure 4). This feature is derived from the variation in the NIR 970 nm reflectance (970 nm-SD) among dry seeds and exhibited a high correlation with that parameter. Both parameters separated out the three lots with higher blindness+death scores within varieties (lots 1, 3 and 5; Figure 3), suggesting that seeds with the larger initial variation could be linked to higher susceptibility to blindness, adding another signal option to aid sorting. The NIR 970 nm wavelength has been used as an indicator of water status in different substances [29,30], and it could be quantifying moisture distribution in seeds in this study, but further research is required to confirm this. Cracking or splitting of the testa precedes radicle emergence from *Brassica* seeds by a few hours [31], and this could be a factor that would add seed surface variation (e.g., exposure of seed tissues and contrast with the seed coat) as well as some moisture differences that could have been measured between active live and damaged or dead tissues.

Seed respiration during germination after the induction treatment also revealed differences among lots. Seed lots 1 (Variety A) and 3 (Variety B) were ranked with the lowest respiratory potential over several parameters (Figure 7, higher values), which agrees with their higher blindness+death scores (Figure 10). However, seed lots from Variety C (lots 5 and 6) usually had overlapping or inverted results when compared to their blindness+death scores (Figures 7 and 10). This issue can be better visualized with the POD curves (R75-POD Curves, Figure 6), as lot 5 has a fraction of faster respirators while lot 6 has a similar fraction of slower respirators; both lots had approximately 30% of seeds that did not consume 25% of the available oxygen (see also Figure 5). The use of POD curves provides a clear view of all seeds tested and avoids the selective calculation of averages and medians that do not account for seeds that did not reach a certain oxygen level. Additionally, a decreasing number of seeds is used to calculate the parameters when lowering the oxygen level threshold used. The final oxygen level or oxygen level at certain times can address this issue and show a realistic performance comparison at that time for all seeds tested. In this study, the final oxygen level (Final-O2) was the Q2 parameter with stronger correlation with seedling area and plant performance in the greenhouse (Figure 11). This close connection between oxygen consumption and seedling area was expected and highlights the critical role of respiration in supporting early stages of plant growth [19]. Seed lots 1, 3 and 5 were ranked

as the lots with higher final oxygen levels in agreement with their blindness+death scores, but due to the larger overall variance present, seed lot 6 was also included in that group.

The seedling areas determined after the respiration measurements was also efficient in identifying lots 3 and 5 as presenting smaller seedling areas, and the large variation present in lot 1 (Figure 9). As expected, this parameter displayed a significant correlation with viability, with larger seedlings at the time of transplanting to greenhouse trays resulting in more viable plants. The relationship with blindness was also significant, although not as strong as with viability (Figure 11). The MFA analysis also shows how tissue area and Final-O<sup>2</sup> clearly separate dead and normal seedlings on opposite vectors (Figure 12).

To summarize the sorting opportunities at the seed lot level, mean values for chlorophyll, CIELab B, saturation, hue, NIR 870 nm and tissue area were able to distinguish lots with higher susceptibility to blindness in varieties B and C. The mean Q2 parameters (R75 and R50, Final-O2) were able to separate the more susceptible lots of varieties A and B. Finally, mean values for CIElab L, 375, 435 and 645 nm, and 970 nm-SD were capable of differentiating these lots within all varieties. Most of these parameters displayed significant positive correlations and could be indicative for blindness susceptibility when high values are observed but also negatively correlated with the presence of normal seedlings. The MFA illustrates these contributions and the direction of separation, increasing with higher presence of blind seedlings and dead seeds while lowering towards higher frequency of normal seedlings (Figure 12, Dim1 both panels). On the contrary, tissue area had the opposite relationship with blindness and normal seedling percentages and a clear antagonistic relation with the Final-O<sup>2</sup> parameter (Figure 12 Dim2, both panels). While some of these parameters can be used to rank and sort all seed lots within these varieties, the importance of most parameters to identify blindness susceptibility seem to be variety-dependent. Sorting opportunities at the individual seed level may require larger sample sizes within lots and varieties to increase the pool of reference seeds displaying the phenotype of interest, expand the information available to properly account for the variation in seed physical characteristics and refine traits important for separation to provide higher confidence and accuracy.

#### **5. Conclusions**

The methodology and approaches used here demonstrate how a set of relevant parameters correlated to a phenotype of interest can be obtained using analytical instruments to assess individual seeds at different stages, starting from the dry seed through to the manifestation of the phenotype. The collection of the parameters from the different analytical instruments and/or stages combined can give valuable insight on how early or late these relevant parameters can be identified and used for seed lot management or upgrading. High-throughput equipment has been developed to physically separate individual seeds based on CF level (www.seqso.com (accessed on 13 January 2021)). Videometer A/S also recently developed a sorter with less speed and capacity, but capable of sorting seeds individually using multiple seed features or combinations of them. Additionally, new instruments and software are becoming available in which artificial intelligence is used to generate powerful algorithms to analyze seed images based on training sets (e.g., Seed-X, Magshimim, Israel). The procedure described here, of making digital images followed by assessing susceptibility to induction of blindness on a seed-by-seed basis, could be used for such training sets by identifying the greater or less susceptible seeds in the original images. At a minimum, the methods utilized here can efficiently identify lots with potential for injury or blindness in response to cold imbibition, which could then be processed further or pretreated to reduce their susceptibility. While further work is required to more fully confirm these approaches, the data provided here also demonstrate the possibility of sorting *Brassica* seed lots to remove individual seeds most susceptible to blindness following exposure to low temperatures.

**Supplementary Materials:** The following are available at https://www.mdpi.com/2077-0472/11/3 /220/s1, Table S1: Seed parameters database, Table S2: Q2 parameters, Table S3: MFA Eigenvalues, Figure S1: Brassica control seed respiration curves at 20 ◦C, Figure S2: Brassica control seed respiration curves at 15 ◦C, Figure S3: Pearson correlation matrix of selected parameters, Figure S4: Full correlation matrix.

**Author Contributions:** Conceptualization, P.B. and K.J.B.; methodology, P.B. and K.J.B.; software, P.B.; validation, P.B. and K.J.B.; formal analysis, P.B.; investigation, P.B.; resources, P.B. and K.J.B.; data curation, P.B. and K.J.B.; writing—original draft preparation, P.B.; writing—review and editing, P.B. and K.J.B.; supervision, P.B. and K.J.B.; project administration, P.B. and K.J.B.; funding acquisition, K.J.B. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by the Western Regional Seed Physiology Research Group.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** The work of the second author is supported by the Lomonosov Moscow State University under grant "Modern Problems of the Fundamental Mathematics and Mechanics".

**Acknowledgments:** The authors thank Corine de Groot for helpful suggestions, technical assistance and providing seed samples. We would like to acknowledge our colleagues Marlen Navarro Boulandier and Vincent Chiu for their initial experimental work and contribution to some of the original data presented here. Thanks to Peter Marks and Aginnovation for providing access to some analytical instruments utilized in these studies. We also thank Allen Van Deynze and Daniel Runcie for critically reviewing the manuscript prior to submission.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


MDPI St. Alban-Anlage 66 4052 Basel Switzerland Tel. +41 61 683 77 34 Fax +41 61 302 89 18 www.mdpi.com

*Agriculture* Editorial Office E-mail: agriculture@mdpi.com www.mdpi.com/journal/agriculture

MDPI St. Alban-Anlage 66 4052 Basel Switzerland

Tel: +41 61 683 77 34 Fax: +41 61 302 89 18

www.mdpi.com ISBN 978-3-0365-1769-8