The Th. intermedium accession used was provided by the Heilongjiang Academy of Agricultural Sciences, Harbin, Heilongjiang, China. In 2006, it was crossed as the paternal parent with annual wheat-Th. intermedium partial amphiploids Maicao 8 (919-C/Yuan16-3) and Maicao 9 (2A2/Xiaoyan 68) as the maternal parents. Yuan16-3 and Xiayan 68 are wheat-Th. intermedium and wheat-Th. ponticum partial amphiploids developed by the Shanxi Academy of Agricultural Sciences, Taiyuan, Shanxi, China, and the Northwest Institute of Botany, Chinese Academy of Science, Yangling, Shanxi, China, respectively. Twenty perennial wheatgrass lines were obtained after two rounds of three-year artificial and natural selection based on the recurrent phenotypic evaluation (HNU-C1 and HNU-C2) in the field of Harbin Normal University, Harbin, Heilongjiang (126°57’E, 45°87’N) from 2007 to 2013 (Yan et al., 2020). From 2013 to 2022, another three cycles of selection for perennial wheatgrass lines, HNU-C3, HNU-C4, and HNU-C5, were carried out under the natural cold winter conditions. Accessions of Ps. strigosa, Th. bessarabicum, and D. villosum were used as sources of DNA to prepare probes for the detection of St-, J^b-, and V-genome chromosomes or chromosomal segments in the IWG lines.

From 2007, the newly developed IWG lines were investigated in an experimental farm of Harbin Normal University, Harbin, China. The agricultural climate of Harbin is warm semiarid, with spring drought and semi-humid summers. From October to April of each year during 2019-2022, the temperature of the winter season in Harbin was below freezing point. During the domestication breeding for cold tolerance of perennial wheatgrasses, the average temperature ranged from -16.5°C and -21.26°C for the four consecutive years, with the lowest temperature of -32.9°C in January 2021 ( Figure 1 ).

In autumn of 2019 and 2020, perennial wheatgrass lines were sown in the experimental field plots for HNU-5C. The field management, seeding, and agronomic characterization were carried out as described previously (Yan et al., 2020). Defoliation was performed in August for harvesting both grains and forages each year. Post-harvest regrowth of aboveground parts and underground rhizomes were observed from harvest to the onset of winter in October and November. From November to early April in the following year, the plants in fields were exposed to the natural cold environments to examine their cold-hardiness based on their survivability during winter and regrowth in spring. Plants that were able to regrow for at least three consecutive years in fields were regarded as cold-hardy perennials. Maicao 8, Maicao 9, and the Th. intermedium accession were grown in 1m rows and 0.3 m spacing between rows as the annual and perennial controls, respectively.

During the growing seasons, plant height (cm), tiller number, spike length (cm), seed set percentage, grain number per spike, and 1,000-kernal weight (g) were investigated for the perennial wheatgrass lines. Plant height was determined from the ground level to the top of a spike. Spike length was measured from the base of a rachis to the top of a spike. The numbers of kernels and spikelets per spike were enumerated and spikes were threshed in a bench micro-thresher to determine the 1,000-kernel weight. Common practices of irrigation (one before the overwinter stage immediately after defoliation in November, and the other after grain and forage harvesting in July), fertilizers (nitrogen, phosphorus, and potassium), and pesticides were applied following the local farming system (Yan et al., 2020).

Seeds were germinated at 23.5°C for 24 h on moist filter paper in Petri dishes prior to incubation at 4°C for 48 h and 23.5°C for 27.5 h. Root tips were treated with ice water at 0-4°C for 24 h, fixed in Carnoy’s fixative (anhydrous alcohol: acetic acid = 3:1, v/v) for 24 h, and squashed in 45% acetic acid. Chromosome preparations were observed under a phase contrast microscope (BH-2, Olympus, Tokyo, Japan).

Genomic DNA was isolated from young leaves of Ps. strigosa, Th. bessarabicum, and D. villosum to prepare the St-, J^b-, and V-genome probes using the cetyltrimethylammonium bromide (CTAB) method (Doyle and Doyle, 1987). Five seeds randomly selected from each line were subjected to smGISH analysis. In the first round of GISH, the genomic DNA of Ps. strigosa and Th. bessarabicum were separately labeled with fluorescein-12-dUTP (green) (BioNick Labeling System; Invitrogen, Waltham, MA, USA) and tetranmethyl-rhodamine-5-dUTP (red) (DIG-Nick Translation Mix, Roche Diagnostics, Mannheim, Germany). In the second round of GISH, the genomic DNA of D. villosum and Ps. strigosa were separately labeled with fluorescein-12-dUTP (green) and tetranmethyl-rhodamine-5-dUTP (red). The protocol of smGISH was described previously by Yan et al. (2020) and Cseh et al. (2019). All slides were visualized using a Leica DM5500B epifluorescence microscope (Leica Microsystems, Wetzlar, Germany) equipped with different filters for detecting DAPI [2-(4-Amidinophenyl)-6-indolecarbamidine dihydrochloride] (blue), Alexa Fluor 488 (green), and Alexa Fluor 594 (red). A Leica DM6000B fluorescence microscope (Leica, Mannheim, Germany) was used for observing the hybridization signals, and photographs were captured with a Leica digital camera (Model DFC480) and processed by Photoshop 8.0 (Adobe, San Jose, California, USA).

Values of the agronomic traits investigated were subjected to analysis of variance (ANOVA) using genotype, year, and genotype by year interaction as the main factors. Fisher’s least significant difference (LSD) was used to perform multiple comparisons for differentiating the significant differences among genotypes and plant ages (i.e., the first, second, and third-year plants) based on the means of each parameter examined. Statistical analysis was performed using IBM SPSS Statistics for Windows Version 19 (SPSS Inc., Chicago, IL, USA). Significant differences in the growth years and various traits of different materials were identified by the t-test of multiple samples at P < 0.05.

Four-hundred seeds of each line with the same seed shape were placed into the sample cup. A DA7200 multifunctional near infrared spectrometer (Perten, Switzerland) was used for testing seed protein content. Each line was tested three times.

From 2013 to 2022, three cycles of selection were performed under natural field conditions ( Figure 1 ). Plants without cold tolerance and/or weak growth ability were eliminated by the cold winter in Harbin. Four-hundred cold-hardy perennial plants were obtained during the third cycle of selection (HNU-C3) from 2013 to 2016. Twenty plants were selected because of their high seed setting, cold tolerance, and strong post-harvest regrowth ability. In the fourth cycle of selection (HNU-C4) from 2016 to 2019, 200 perennial plants were selected. Finally, 24 perennial cold-hardy wheatgrass lines, designated 19HSC-Q and 20HSC-Z, were developed in 2019 and 2020. From 2019 to 2022, as HNU-C5, twenty-one 19HSC-Q and three 20HSC-Z lines were subjected to cold-hardiness tests and molecular cytogenetic analysis ( Figure 2 ).

From 2019 to 2022, twenty-one 19HSC-Q perennial lines were evaluated for their cold hardiness, perennial habit, and maturity. All of them survived for more than 3 years, showing perennial growth habit, which resembled their perennial parent Th. intermedium. The coleoptile of these lines was purple. The seedlings stopped growing and were dormant during snow-cover in the natural cold winter, but the leaves and stems remained green during the whole winter. In 2020, all the 19HSC-Q lines resumed normal growth in April, flowered for a week from 22 June, and matured by the end of July. Post-harvest regeneration was observed after cutting the aboveground parts in early August 2020. From September to October, new stems and leaves emerged from crowns or roots, and the regenerated plants grew vigorously. The growth of lines slowed down in November and entered a dormant period after snow-cover in December ( Figure 3A ). From November 2020 to the end of March 2021, some stems and leaves around the crowns remained green in all the regenerated lines, and the new rhizomes appeared in the soil. In April 2021, the leaves and stems were withered. After cutting, new leaves regrew from the original crowns and the lines returned to the vegetative state for a new growth cycle in May 2021. The plants resumed growth again, and the underground rhizomes developed into new plants. The rates of returning green was 100% for all the 19HSC-Q lines. The subsequent performance of the plants was consistent with the first growth cycle. All lines were harvested at the end of July 2021. The 19HSC-Q lines still maintained the post-harvest regeneration. In the third growth cycle from 2021 to 2022, the plants kept a stable rate of returning green and regeneration after cutting ( Figure 3B ), and the dates to flowering and maturity were consistent with those of the previous two growth years. Lines 19HSC-Q5 and 19HSC-Q17 also were cold-tolerant, but the vitality of plants was weak. The ability of regeneration and returning green were decreased also, and the crowns were gradually shrunk. Currently, all the 19HSC-Q lines are experiencing their fourth growth cycle.

From 2020 to 2022, three 20HSC-Z lines in HNU-5C were tested for cold tolerance and maturity in the natural cold winter. These cold-hardy perennial lines had post-harvest regrowth and their growth habits were consistent with other 19HSC-Q lines. At present, they have entered their third growth cycle.

All the 19HSC-Q and 20HSC-Z perennial wheatgrass lines with cold tolerance and post-harvest regeneration (HPR) were seeded in Harbin during 2019 and 2020. Compared to the previous four selection cycles, dates to flowering and maturity of the 19HSC-Q plants were about 40 days earlier. It is more adapted to the cold winter of Harbin.

The agronomic traits of twenty-one 19HSC-Q perennial lines were investigated for three consecutive growing seasons from 2019 to 2022. Except 19HSC-Q5 and 19HSC-Q17, all lines maintained the wheatgrass characteristics in plant height, heading date, and tillering ability. The agronomic performances of these lines varied and the land cover area per plant was inconsistent after three growth cycles.

The average plant height of 19HSC-Q lines over 3 years ranged from 139.9 cm to 147.6 cm ( Figure 4A ). The average spike length ranged from 22.35 cm to 26.92 cm ( Figures 3D , 4B ). The average number of spikelets ranged from 19.37 to 21.11 with four to five pairs of florets per spikelet, and the highest number of 24 spikelets was observed in 19HSC-Q14 in 2022 ( Figure 4C ; Table 1 ). The average spike length and spikelet number per spike of 19HSC-Q5, 19HSC-Q7, 19HSC-Q9, 19HSC-Q12, 19HSC-Q15, and 19HSC-Q17 were lower than those of the control Th. intermedium. Although 19HSC-Q5 and 19HSC-Q17 have grown for three years, 19HSC-Q5 was sterile and 19HSC-Q17 produced only a few seeds in 2022.

The tiller number of 19HSC-Q lines increased over the years. For example, line 19HSC-Q2 produced 57 tillers in the first growth cycle, while all the other lines had 14 to 38 tillers. In the third growth year, the tiller number of 19HSC-Q2 was 147. Lines 19HSC-Q14 (335) and 19HSC-Q20 produced 335 and 467 tillers, respectively. The tiller numbers of 19HSC-Q14 and 19HSC-Q20 in the first life year were only 33 and 31. In the third growth cycle, there were nine lines with more than 200 tillers ( Figure 4D ). In 2022, the lines with the largest number of tillers was 19HSC-Q20, which also had the largest land cover area per plant (0.615 m²), followed by 19HSC-Q14 (0.335 m²) and 19HSC-Q8 (0.272 m²). The other lines ranged from 0.080 m² to 0.615 m² for the land cover area per plant.

The agronomic characteristics of three 20HSC-Z lines were measured from 2020 to 2022 ( Table 1 ). The average plant height ranged from 55 cm to 138 cm in the two growth cycles. The spike length and the number of spikelets per spike ranged from 18.37 cm to 27.97 cm and 19 to 24, respectively. The number of florets per spike were four to six pairs. In 2022, the plant height and spike length of 20HSC-Z1 were 155 cm and 27.97 cm, respectively ( Figure 3D ). The highest number of spikelets was 24 for 20HSC-Z2. In 2022, the largest tiller number of the three 20HSC-Z lines was 173 for 20HSC-Z3. The largest land cover area per plant was also observed in 20HSC-Z3 (0.151 m²), followed by 20HSC-Z2 (0.119 m²) and 20HSC-Z1 (0.058 m²).

During 2019-2022, the number of grains per spike and 1,000-kernal weight varied significantly among the 21 19HSC-Q lines, and their grain protein content was all greater than 15%. In the first life year, the number of grains per spike of 15 lines exceeded 60, but 14 of them decreased each year. Seven lines increased in the number of grains per spike during the three growth cycles. In the third growth cycle, lines 19HSC-Q2, 19HSC-Q11, 19HSC-Q14, 19HSC-Q18, 19HSC-Q20, and 19HSC-Q21 had more than 80 grains per spike in comparison with 54.6 of Th. intermedium ( Figure 4E ; Table 1 ). The grain numbers per spike of the top three lines 19HSC-Q14, 19HSC-Q20, and 19HSC-Q18 in 2022 were 115, 102, and 97, respectively ( Figures 3C , 4E ; Table 1 ). The average seed set for these lines was around 50%, and the 1,000-kernal weight in 2022 was more than 6 g, which was higher than that of IWG (5.128 g). Lines 19HSC-Q14, 19HSC-Q18, and 19HSC-Q20, with the highest grain numbers per spike, also had a 1,000-kernal weight of 7.67, 7.52, and 7.37 g, respectively ( Figure 4F ; Table 1 ). In 2022, the average grain length and width of the six lines with greater number of grains per spike were 4.13-5.54 mm and 1.22-1.47 mm, which were higher than those of the IWG (3.42 mm and 0.92 mm, respectively). The largest average length (5.54 mm) and width (1.47 mm) of grains were observed in line 19HSC-Q14 ( Figures 3E, F ; Table 1 ). The protein content of 10 lines was more than 17%, and 19HSC-Q14, 19HSC-Q18, and 19HSC-Q20, with the best grains per spike and 1,000-kernal weight, had protein contents of 17.36%, 19.14%, and 19.89%, respectively ( Table 1 ).

In 2022, the numbers of grains per spike of the three 20HSC-Z lines were all greater than 80, and the seed setting rates were over 50%. The highest number of grains per spike was 116 for 20HSC-Z1. 20HSC-Z3 had the highest setting rate of 55.69% ( Table 1 ). The 1,000-kernal weight of 20HSC-Z1, 20HSC-Z2, and 20HSC-Z3 were 7.146, 6.523 g, and 5.916 g, respectively. The grain length of these lines ranged from 3.728 mm to 6.314 mm, and the average width ranged from 1.146 mm to 1.365 mm ( Figure 3E ; Table 1 ). The protein contents of these three lines were more than 17% protein, with the highest of 20.01% for and 20HSC-Z1.

In summary, we obtained six lines, 19HSC-Q2, 19HSC-Q11, 19HSC-Q14, 19HSC-Q18, 19HSC-Q20, and 19HSC-Q21, with the most outstanding agronomic traits and yield components in HNU-5C. Lines 19HSC-Q14, 19HSC-Q18, and 19HSC-Q20 had the best agronomic traits over the other perennial lines ( Figure 5 ).

The smGISH analysis for twenty one 19HSC-Q lines with the same combination of probes found that the chromosome compositions of most lines, for example, line 19HSC-Q2 and 19HSC-21, were the same as IWG, i.e., 2n=6x=42, StStJ^rJ^rJ^vsJ^vs. However, four of six lines with prominently agronomic traits, 19HSC-Q11, 19HSC-Q14, 19HSC-Q18, 19HSC-Q20, and three 20HSC-Z lines had the translocation and introgression between subgenome chromosomes.

The pattern of smGISH analysis for line 19HSC-Q11 showed that an IJ^vs chromosome had the introgression of St-subgenome chromosomes across the centromeric regions ( Figures 7A1–C1 ). Line 19HSC-Q14 showed a chromosome with a bright-green signal of the St genome in the short arm and stained dark-red by the J^b-genome probe in the long arm. There was no significant green signal of the V-genome probe on the whole arm of the chromosome. It was a whole arm translocation between the short arm of a St chromosome and the long arm of a J^r chromosome ( Figures 7A2–C2 ). The smGISH analysis of 19HSC-Q18 detected that a St-subgenome chromosome had a large light-green hybridization signal produced by the V-genome probe at the proximal end of the long arm. It was the terminal translocation of a large segment with the J^vs-St subgenome chromosome ( Figures 7A3–C3 ). The smGISH analysis of 19HSC-Q20 identified that a half of the J^r chromosome arm was translocated into the end of an IIJ^vs chromosome ( Figures 7A4–C4 ). In line 20HSC-Z1, a translocation of a half chromosome arm between the J^r chromosome and the IIJ^vs chromosome was observed ( Figures 7A5–C5 ). The introgressions of J^vs-J^r chromosome concentrated on the terminal parts of both long and short arms, as well as the centromeric regions, were found in line 20HSC-Z ( Figures 7C6, C7 ). In addition, a small segment of J^r chromosome in line 19HSC-Q20 was inserted into a St chromosome at the proximal end of the short arm ( Figures 7A4, B4, C8 ). This proximal end small translocation of J^r-St chromosome was also observed at the long arm in lines 20HSC-Z1 ( Figures 7A5, B5, C9 ), 20HSC-Z2 ( Figures 7A6, B6, C10 ), and 20HSC-Z3 ( Figure 7C11 ).

The results of the smGISH analysis for lines 19HSC-Q and 20HSC-Z detected seven types of chromosome variations based on the size and locus of the translocated fragments. The translocation of large fragments included four types, St-J^r, J^vs-St, J^r-IIJ^vs, and J^vs-J^r, and the small fragment translocations and introgressions contained three types, J^r-St, St-IJ^vs, and J^vs-J^r.