Assessment of diverse cicer milkvetch (Astragalus cicer L.) germplasm for agro-morphological traits under a stockpiling system

Abstract Cicer milkvetch (Astragalus cicer L.) is a non-bloating perennial forage legume suitable for stockpiled grazing in the fall because of its rapid regrowth and high nutritive value. Genetically diverse germplasm are needed for the development of improved cicer milkvetch cultivars that can provide consistent production across variable climatic conditions. The objective of this research was to assess the diversity and relationship of 27 cicer milkvetch populations to inform the selection of populations for future cultivars that have superior agro-morphological traits during summer and fall growth. A completely randomized field trial was established in 2019 near Clavet, Saskatchewan. In 2020 and 2021, forage dry matter yield (DMY), maximum stem length, leaf number per stem, rhizome spread, and stem density were recorded on 27 populations of cicer milkvetch in late June at a first harvest and mid-October at a stockpile harvest. All five traits were different (p < 0.05) among the populations at both harvests except for leaf number per stem in late June. Principal component analysis identified that the first three principal components described 89% of the variation in agro-morphological traits at the first and stockpile harvests. Of the agro-morphological traits, maximum stem length had the greatest correlation with forage DMY at the first harvest (r = 0.69) and stockpile harvest (r = 0.6). Our research demonstrates that there is a high morphological diversity among cicer milkvetch populations, and plant introductions, PI 362266, PI 576963, PI 440143, and PI 362254 could be used as novel genetic resources for the development of climate-resilient cultivars.


Introduction
Cicer milkvetch (Astragalus cicer L.) is a non-bloating, perennial legume native to Eurasia (Stroh et al. 1973). It is an auto-allo-octoploid (AAAABBBB, 2n = 8x = 64) that behaves largely as a diploid species (Latterell and Townsend 1994). Cicer milkvetch is well suited to a stockpiled forage system, which is a practice of allowing a forage stand to regrow after harvest for the purpose of fall grazing (Hitz and Russell 1998). Using stockpiled forage to extend the grazing season can reduce the cost of winter feed for cow-calf producers in western Canada (Baron et al. 2016). This system can also reduce the use of synthetic fertilizers needed to grow winter feed by depositing manure and urine nutrients directly on to the forage stand (Baron et al. 2016).
Selecting forage species that maintain their yield and nutritional value at an advanced stage of maturity reduces the need for supplemental feed that would decrease the economic return of a stockpiled forage system (Baron et al. 2005;Kulathunga et al. 2016). The benefits of cicer milkvetch for stockpiling include rapid regrowth after defoliation (Foster et al. 2019), a high leaf:stem ratio, leaf retention after a hard frost (Loeppky et al. 1996), and a low neutral detergent fibre content in the fall (Peel and Waldron 2022). Cicer milkvetch is known to be persistent under proper stand management, maintaining the long-term productivity of a forage stand by contributing atmospherically fixed nitrogen to the soil (Townsend et al. 1990;Foster et al. 2019). Compared with alfalfa (Medicago sativa L.), cicer milkvetch has similar yield in areas with greater than 400 mm of annual precipitation (Loeppky et al. 1996;Peprah 2018). It does not cause frothy bloat because of the reticulate venation and thick epidermal layer of its leaves (Lees et al. 1982).
Cicer milkvetch was introduced to Canada from the USSR in 1931 and the first Canadian cultivar Oxley was released in 1971 (Johnston et al. 1971). Subsequently, there have been two cicer milkvetch cultivars released in Canada, which are AC Oxley II and AC Veldt (Acharya 2001(Acharya , 2008. The three Canadian cultivars are genetically related and were not selected for a stockpiled forage system (Johnston et al. 1971;Acharya 2001Acharya , 2008. As an outcrossing species, recurrent mass selection with or without a progeny test has been a common method to develop new cultivars of cicer milkvetch (Townsend 1993a;Acharya 2008). Thus, a high degree of genetic diversity is useful for advancing populations through recurrent plant selection. The introduction of novel germplasm could enhance the genetic diversity of new cultivars and improve the traits that currently restrict the adoption of cicer milkvetch, which include slow seedling establishment due to low seed germination and low seedling vigour (Acharya et al. 2005).
High levels of diversity for phenotypic and agronomic characteristics (agro-morphological traits) have been found in individual populations of cicer milkvetch. This despite there being limited research on the genetic variation and relationship within the germplasm collections of this species. Smoliak and Johnston (1976) studied a population of cv. Oxley and found significant variability for forage dry matter yield (DMY), seed germination, seedling leaf area, and seedling growth. Similarly, Townsend (1985) found highly variable seed germination among polycross progenies of cv. Monarch. In a controlled environment with a decreasing photoperiod to simulate late summer, there were differences in plant height and forage DMY among the polycross progenies of cv. Monarch (Townsend 1993b). Greater regrowth and plant height in the late summer were selected for in the development of cv. Windsor (Townsend 1994). In a grazing study, Townsend (1986) reported that progenies of cv. Monarch had different palatability for grazing sheep. The diversity within germplasm collections of alfalfa (Jenczewski et al. 1998;Li et al. 2009) and sainfoin (Onobychis viciifolia Scop.) (Delgado et al. 2008;Jafari et al. 2014;Bhattarai et al. 2018) has been studied extensively. These studies have shown that plant height (Li et al. 2009;Bhattarai et al. 2018), stem number per plant (Jafari et al. 2014;Bhattarai et al. 2018), and the presence of rhizomes (Jenczewski et al. 1998) can differentiate populations within germplasm collections for these two species.
We hypothesize (1) that there will be differences between populations of cicer milkvetch based on the agromorphological traits and (2) that most of the agromorphological traits are correlated with forage DMY. The objective of this research was to assess the diversity and relationship of 27 cicer milkvetch populations to inform the selection of populations for future cultivars that have superior agro-morphological traits during summer and fall growth.

Plant material
A set of 25 novel cicer milkvetch populations (plant introductions) was obtained from the United States Department of Agriculture (USDA) Western Regional Plant Introduction Station (Pullman, WA, USA). Two Canadian cultivars, AC Oxley II and AC Veldt, were also included as check cultivars for comparison (Table 1). Based on the information available from the USDA Germplasm Resources Information Network (USDA-GRIN), the 27 populations were grouped into six categories: wild, landrace, botanical garden, breeding, cultivar, and other. Wild populations were collected from their natural habitats. The landrace was a regionally adapted popula-tion developed in an agroecosystem. There were a large number (n = 10) of populations sourced from botanical gardens, therefore this was designated as a population type. The breeding population originated from a breeding program but had not been released as the cultivars have been. Populations classified as "other" were of unknown origin or a miscellaneous genebank population according to the USDA-GRIN.

Experimental design
Sixty unscarified seeds per population were imbibed on top of two layers of filter paper (Whatman 597; Whatman PLC, Maidstone, UK) in 90-mm diameter sterilized plastic Petri dishes moistened with 6 mL of distilled water. Each Petri dish was then sealed with a polyethylene bag to prevent desiccation and placed in a dark germination cabinet at a constant temperature of 20 • C according to the Canadian Methods and Procedures for Testing Seed (Canadian Food Inspection Agency 2012). The germinated seeds were transplanted to individual pots and grown for approximately 8 weeks in the College of Agriculture and Bioresources greenhouse at the University of Saskatchewan in Saskatoon, SK, Canada. After this time, the seedlings were transplanted to a field nursery at the University of Saskatchewan Livestock and Forage Center of Excellence (LFCE) located south of Clavet, SK, Canada (51 • 56 N, 106 • 22 W). In the field, the seedlings of the 27 populations were arranged in a completely randomized design with three replications per population. A replication of each population in the nursery contained five plants on 1m centres. The soil texture at the site is described as Elstow Orthic Dark Brown Chernozemic loam soil (SKSIS Working Group 2018) with an E.C. of 1.0 dS m −1 and a pH of 6.61. During the establishment year in 2019, each plant received 2-3 L of water on 6, 10, and 14 June. In the fall of 2019 and 2020, a fertilizer blend of 29-28-0 was broadcast on the nursery at a rate of 173 kg of product ha −1 to provide 50 kg N ha −1 and 44 kg P 2 O 5 ha −1 .
The 2020 and 2021 growing seasons (May-October) can be summarized as being dryer than normal (Table 2). Considering temperature data, 2020 was on average cooler than normal with the average monthly temperatures of June, July, September, and October being below the long-term average. June was the only month of the 2020 growing season that received above average precipitation. The months from July-October 2020 received less than half of the longterm monthly average precipitation. In 2021, four of the six months during the growing season had temperatures above the long-term monthly average. The 2021 growing season received below average precipitation, with June, July, and October receiving less than half of the long-term average precipitation.

Data collection
Data were collected on the cicer milkvetch nursery in 2020 and 2021 in late June for the first harvest, and then the plants were allowed to regrow until the second data collection at the stockpile harvest in mid-October. At both harvests, the data collected by individual plant included forage DMY, maximum stem length, leaf number per stem, stem density, rhizome  spread score, and plant area score. Data collection for the first harvest occurred on 2 July 2020 and 29 June 2021, and for the stockpile harvest on 14 October 2020 and 12 October 2021.
For forage DMY, cicer milkvetch plants were handharvested at 5 cm above the soil surface and fresh weight was recorded by individual plant. Two of the five plants per replication were dried at 60 • C in a forced air oven for 48 h and were then weighed to determine the forage DMY. The average water content of these two plants was used to calculate the forage DMY of the plants in the replication. To measure maximum stem length, the longest stem of each plant was measured from the soil surface to the tip of the leaves. The same stem was then used to determine the leaf number per stem by counting the number of complex pinnate leaves along the stem. This trait was measured on six populations that were selected based on a visual assessment in June 2020 for high (Veldt, Monarch), moderate (PI 576963, PI 362254), and low (PI 362233, and PI 297335) vigour. Stem density was measured by counting the number of stems within a 50 cm × 50 cm quadrat centered on each plant. The rhizome spread score was measured using a 1-4 scale in reference to a 50 cm × 50 cm quadrat (Fig. 1). For the rhizome spread score, 1 = rhizomes emerged within <25% of the quadrat area; 2 = rhizomes emerged from 25% of the quadrat area up to 5 cm of the quadrat border; 3 = rhizomes emerged within 5 cm of the quadrat border; and 4 = rhizomes emerged outside of the quadrat border. The plant area score was based on a 1-4 scale and was only recorded in 2021, therefore an Analysis of variance (ANOVA) was not conducted for this trait (Fig. 2). The plant area score was estimated visually based on the percentage of the quadrat area occupied by plant material with 1= <25%; 2 = 25%-50%; 3 = 50%-75%; and 4 = >75% (Fig. 2).

Statistical analysis
Statistical analysis was conducted using R software (R Core Team 2021) and R Studio version 4.0.3 (R Studio Team 2020). Mixed models were developed using the lme4 package (Bates et al. 2021) considering data from 2020 and 2021 with population as a fixed effect and the random effects were the experimental replication nested within year. An ANOVA was conducted to identify differences among the populations for forage DMY, maximum stem length, leaf number per stem, stem density, and rhizome spread score using the car statistical package (Fox et al. 2019). The rhizome spread score data from the first harvest in 2020 were not used in the analysis because of a collection error. If the ANOVA indicated that there were significant differences (p ≤ 0.05) between populations, means were separated using the R package predictmeans (Dongwen et al. 2021) to derive the least significant difference (LSD) value for each trait at the first and stockpile harvests. Pearson's correlation coefficients were estimated between forage DMY and the agro-morphological traits at the first and stockpile harvests considering data from 2020 and 2021. The relationship among the cicer milkvetch populations was assessed using principal component analysis (PCA) that considered the data of maximum stem length, stem density, rhizome spread, plant area, and forage DMY collected at the first and stockpile harvests. Trait values for the populations at each harvest time were a mean of 2020 and 2021, except in the case of plant area where only 2021 data were available. As many of the traits were on different scales, the data were standardized prior to the PCA. This procedure was conducted using the FactoMineR (Husson et al. 2022) and factoextra (Kassambara and Mundt 2020) packages in R, while the figure to represent the PCA was developed using the ggpubr (Kassambara 2022) package.

Forage DMY
There were significant differences (p = 0.0029) in the forage DMY at the first harvest among the 27 populations of cicer milkvetch (Table 3). The mean forage DMY at first harvest across the 27 populations was 111 g plant −1 , ranging from 39 (PI 206405) to 184 g plant −1 (AC Veldt) (Table 4). Of the 27 populations, the forage DMY of the check cultivars AC Veldt  At the stockpile harvest, forage DMY was also different (p < 0.001) among the cicer milkvetch populations (

Maximum stem length
At both the first and stockpile harvests, maximum stem length differed significantly (p < 0.001) among the 27 populations (Table 3). At the first harvest, the mean stem length across populations was 48 cm with a range from 37 (PI 362247, PI 206405) to 70 cm (Monarch) ( Table 4). At the stock-pile harvest, the mean stem length across populations was 50 cm with a range from 31 (PI 206405) to 70 cm (Monarch) ( Table 5). At the two harvests, the check cultivar AC Veldt had a maximum stem length that was similar to Monarch, and these two populations had a longer stem length than AC Oxley II.

Stem density
At both the first and stockpile harvests, stem density differed significantly (p < 0.001) among the 27 populations (Table 3). At the first harvest, the mean stem density among populations was 28 stems 0.25 m −2 with a range from 16 (PI 206405) to 39 stems 0.25 m −2 (AC Veldt) (

Rhizome spread
The rhizome spread score at the stockpile harvest differed significantly (p < 0.001) among the 27 populations ( Table 3). The mean rhizome spread across populations was 3.4 with a range from 2.6 (PI 362247) to 3.9 (AC Veldt, AC Oxley II, and PI 362246) ( Table 5). The mean rhizome spread of AC Veldt and AC Oxley II was similar to 17 other populations.
Leaf number per stem Leaf number per stem was similar (p = 0.35) among the selected six populations at the first harvest (Table 3). The mean leaf number per stem across populations was 9.3 with a range from 8.6 (PI 297335) to 10.5 (AC Veldt) ( Table 6). At the stockpile harvest in mid-October, the leaf number per stem of AC Veldt was significantly less (p < 0.05) than the other five populations. The mean leaf number per stem among the selected six populations at the stockpile harvest was 13.3 with a range from 11.8 (AC Veldt) to 14.3 for PI 576963 (Table 6).

Relationships among cicer milkvetch populations
The first three principal components (PCs) cumulatively explained 89% of the variation in the five agro-morphological traits. Including the fourth PC would have increased the cumulative variation explained to over 90%, but the eigenvalue was only 0.34 (Data not shown). Principal component one (PC1) had an eigenvalue of 6.36 and explained 64% of the variation. The five agro-morphological traits that were examined at the first and stockpile harvests were significantly correlated (p < 0.05) with PC1 and had positive eigenvectors. Plant area at the stockpile harvest and forage DMY at the stockpile and first harvests had the largest eigenvectors in relation to PC1 of 0.36, 0.35, and 0.34, respectively. When examining the position of individual populations along PC1, populations with relatively high values (>1) included AC Veldt ( (Fig. 3). All cultivars had values >1 along PC1 accompanied by landrace PI 576963, other population PI 362266, botanical garden PI 362246, and wild PI 362254. Principal component two (PC2) had an eigenvalue of 1.34 and explained 13% of the variation among the agromorphological traits. Maximum stem length (both harvest times), plant area at the first harvest, and forage DMY (both harvest times) had negative relationships to the variation in PC2. The remaining five combinations of trait by harvest time had positive relationships to the variation within PC2. Stem density at the stockpile harvest (eigenvalue 0.58), maximum stem length at the first harvest (eigenvalue −0.45), and stem density at the first harvest (eigenvalue 0.42) were significantly correlated with PC2. Principal component three (PC3) had an eigenvalue of 1.18 and explained 12% of the variation among the agro-morphological traits. Rhizome spread (both harvest times) and plant area (both harvest times) had positive relationships with the variation described by PC3. The remaining six combinations of trait by harvest time had negative relationships with the variation in PC3. Rhizome spread (eigenvalue 0.58) and plant area (eigenvalue 0.43), both when recorded at the stockpile harvest were significantly correlated with the variation in PC3.

Discussion
This study was undertaken to understand the diversity and relationship between cicer milkvetch populations for a stockpiled forage system. To our best knowledge, this is the first attempt to characterize a relatively large set of cicer milkvetch populations (n = 27) for various agronomic and morphological traits. First, we found a high degree of variation among populations for the measured traits at both the first and stockpile harvests. Several of the superior populations are unrelated to Canadian cultivars, providing an opportunity for further selection and introgression into future cicer milkvetch cultivars. Second, forage DMY was most highly correlated with maximum stem length at both the first and stockpile harvests. Finally, PCA identified that the first three principal components described 89% of the variation in the agro-morphological traits at the two harvest times.
This study found a high degree of diversity for agromorphological traits among the cicer milkvetch populations except for leaf number per stem at the first harvest. The high interpopulation variability found in our study was also seen in the agronomic and morphological traits of other perennial species, including alfalfa (Jenczewski et al. 1998;Li et al. 2009), sainfoin (Delgado et al. 2008;Jafari et al. 2014;Bhattarai et al. 2018), and crested wheatgrass (Agropyron cristatum (L.) Gaertn.) (Tandoh et al. 2019). The cicer milkvetch populations PI 362266, PI 576963, PI 440143, and PI 362254 had high values for forage DMY, maximum stem length, and stem density at both harvest times. These four populations also had relatively large values along PC1, which had significant and positive correlations with the five agro-morphological traits. They are not related to the check cultivars based on their pedigrees (Johnston et al. 1971;Acharya 2001Acharya , 2008, thus these four populations could be used as novel sources of genetic variability for future varietal development. As the current Canadian varieties are genetically related (Acharya 2001(Acharya , 2008, the introduction of new diverse germplasm can help to maintain the productivity of cicer milkvetch across a wide range of growing regions that are experiencing a warming climate (Morales-Castilla et al. 2020). Differentiation along PC1 was most clear by population type with cultivars having relatively high values and botanical garden populations having moderate values. There was not a clear pat- Fig. 3. The relationship among 27 cicer milkvetch populations in a space plant nursery based on principal component analysis (PCA) calculated from five agro-morphological traits recorded at the first and stockpile harvests of 2020 and 2021 (Clavet, SK, Canada). X-axis values have significant (p < 0.05) positive correlations with the five agro-morphological traits (first and stockpile harvests). Y-axis values have significant (p < 0.05) positive correlations with stem density (first and stockpile harvests) and a negative correlation with maximum stem length (first harvest).
tern of geographical separation among the cicer milkvetch populations based on the PCA. However, for botanical garden populations, an exact geographical origin was unknown. In other perennial forages, Li et al. (2009) found there to be no relationship between the geographic proximity of alfalfa populations and their phenotypic trait values. However, the genetic variation of crested wheatgrass germplasm was found to be associated with geographical locations (Baral et al. 2018). In the future, a detailed genomic study of the 27 populations would confirm the degree of genetic diversity within and among populations. A high degree of variation in agro-morphological traits was also linked to a high degree of genomic variation within populations of sainfoin (Bhattarai et al. 2017(Bhattarai et al. , 2018 and crested wheatgrass (Baral et al. 2018;Tandoh et al. 2019). Ploidy differences among the cicer milkvetch populations were not examined in this study. However, further research could examine the ploidy levels of the 27 populations and if present, whether these ploidy differences inform part of the variation in morphology that was seen in this study.
The cicer milkvetch populations in our study had high levels of intrapopulation morphological variability, which indicated a possibility of further plant selection within populations. This is useful for the continuous improvement of specific traits through recurrent selection of a superior performing population (Vogel and Pedersen 1993). In addition, fur-ther study on the genetic diversity and relationship among the 27 populations based on molecular markers would be useful to understand the genomic basis and to identify markers associated with the agro-morphological traits. There has been little genetic selection in cicer milkvetch compared with other perennial forages, and thus relatively high levels of polymorphism are expected in this outcrossing species (Latterell and Townsend 1994). A combination of selection within and among the cicer milkvetch populations would be valuable to capture high-performing genotypes and genetic variability among populations that could improve the resilience of future cultivars to a variety of climate conditions. The forage yield of cicer milkvetch, especially the stockpile yield, is important for western Canadian cattle producers for extending the grazing season in the fall to reduce yearly feeding costs (Baron et al. 2016). This research showed that the forage DMY of cicer milkvetch had the highest positive correlation with maximum stem length and stem density at both harvest times. The maximum stem length of sainfoin was found to be more highly correlated with forage DMY (Bhattarai et al. 2018) than what was found in our study. Indirect plant selection can be a valuable tool to improve traits with low heritability like forage yield as suggested by de Araújo (2001). While both maximum stem length and stem density showed associations with the forage DMY of cicer milkvetch, maximum stem length could be a better trait for indirect selection because it has a consistent high correlation and is quick to measure. The results of this study as they relate to measuring intrapopulation variability based on their agro-morphological traits and the association between these traits are limited by the single testing location. Increasing the number of plants per population or testing locations is difficult due to the small quantity of seed that is received when making requests for plant introductions and the relatively low germination rates across the populations studied (data not shown). Therefore, examination of this nursery for more than 2 years could increase the confidence in these results with a better understanding of the genotype by environment interaction.
The correlation between forage DMY and the agromorphological traits decreased from the first harvest to the stockpile harvest. For F1 progeny of a cross between Medicago falcata and M. sativa (alfalfa), the phenotypic correlation between forage DMY and plant height was lower for an early September harvest when compared with harvests in early June and early July (Robins et al. 2007). It was found that the genes associated with forage yield appeared to be different than those associated with fall regrowth, which may indicate the effects of dormancy on the early September forage yield of alfalfa. Previous work has shown that there was variability among polycross progeny of cicer milkvetch cv. Monarch for sensitivity to a decreasing photoperiod (Townsend 1993b). An investigation into the fall dormancy of the 27 cicer milkvetch populations of this study could improve the explanation for the differences in their stockpiled forage DMY.
Nutritional value relating to crude protein, fibre, and energy is also key to livestock performance while stockpile grazing (Baron et al. 2004;Kulathunga et al. 2016). While these nutritional characteristics were not measured in this study, cicer milkvetch has shown to have adequate crude protein to meet the needs of dry pregnant cows under both whole season or partial season stockpiling lengths and harvested at various dates from late summer to late fall (National Research Council 2000;Peel and Waldron 2022). However, supplemental feed may still be needed when stockpile grazing under cold temperatures and high wind speeds (Kulathunga et al. 2016).

Conclusion
This study provides a baseline understanding of the variation in agro-morphological traits across a set (n = 27) of cicer milkvetch populations managed under a stockpiled forage system. Considering the first and stockpile harvests, there were significant (p < 0.05) differences between populations of cicer milkvetch for all agro-morphological traits used in this study except for leaf number per stem at the first harvest. For many traits at both harvests, existing cultivars from either Canada or the United States had the highest mean values. The populations PI 362266, PI 576963, PI 440143, and PI 362254 could act as novel sources of variation for future cultivars developed for stockpiled forage systems in western Canada. This is because of their high values for forage DMY at the first and stockpile harvests and they do not share similarities in their pedigrees with the two check cultivars. The value of the agro-morphological trait maximum stem length was further reinforced by its high correlation with forage DMY. PCA demonstrated that the variability described by the agromorphological traits grouped populations with high trait values closely together. Overall, this study confirms the presence of morphological diversity among cicer milkvetch populations that could be selected to develop future cultivars for a stockpiled forage system.