Comprehensive Evaluation and Main Identification Indexes of Herbicide Resistance of High-Quality Foxtail Millet (Setaria italica L.)

Abstract

Foxtail millet (Setaria italica L.) is an important crop grown worldwide as a food and fodder crop owing to its potential nutritional and feed values. High-efficiency herbicide varieties is crucial to achieving efficient weeding and ensuring successful foxtail millet production. Herbicides affect several morphological and physiological indicators of foxtail millet. In this study we aimed to evaluate the damage caused by herbicides, improve their effectiveness, and select indicators that accurately reflect herbicide resistance in foxtail millet. Jingu 21, which has the largest planting area in Shanxi province and even the whole of China, is selected as the experimental material to be sown in the field in 2022. A total of 31 herbicides were applied, and 21 traits, including morphological, physiological, and yield-component traits, were measured to assess millet resistance. Principal component analysis was employed to transform these 21 conventional traits into seven independent and comprehensive indexes. These indexes were screened using regression analysis, resulting in the selection of the following indicators: the surface area of the second leaf from the top, peroxidase activity, catalase activity, malondialdehyde content, chlorophyll (a + b), grain weight per ear, and yield. Through membership function and cluster analyses, the resistance of Jingu 21 to 31 herbicides was divided into five categories: extremely weakly resistant, weakly resistant, moderately resistant, strongly resistant, and extremely strongly resistant. Jingu 21 exhibited extremely strong resistance to lactofen, butachlor, and anilofos. After an investigation into the effectiveness of herbicides, it was found that eight herbicides had good effects.

1. Introduction

Foxtail millet (Setaria italica L.) is a gramineous plant from the genus Setaria, is diploid, and is a self-pollinated crop [1,2]. Millet is the main crop grown in northern China and is characterized by drought and resistance and strong adaptability [3]. It is an environmentally friendly crop that plays an indispensable role in food diversity and planting structure adjustment [4]. Millet has both conventional and hybrid varieties. Conventional millet varieties are sensitive to herbicides. Jingu 21 is a typical representative of high-quality conventional varieties that are extremely sensitive to herbicides. It is a well-known cereal variety selected and bred by the Economic Crop Research Institute of the Shanxi Academy of Agricultural Sciences. It has a recognized national-level millet quality, with a golden and shiny color, and a nutritional composition that exceeds the traditional Shanxi tribute millet, Qinzhou Huang, and the first-class quality millet standard [5]. The planting area of millet in Shanxi Province is about 200,000 hectares, of which nearly 100,000 hectares are planted with Jingu 21, but Jingu 21 does not have the herbicide resistance gene. Therefore, we have been exploring the for several years, but we have never investigated several types of herbicides at the same time and the resistance level of Jingu 21 to various herbicides. We have not yet proven or found a rapid identification method.

Weeds in millet fields severely affect millet yield and quality, and have become an obstacle to the industrialization of millet production [6,7]. Currently, chemical weeding technology is labor-saving, time-saving, and highly efficient and is one of the most effective methods for weed prevention and control in modern agricultural production [8,9]. However, the grain seedlings of Jingu 21 are weak, seedling growth is slow, and this cultivar is extremely sensitive to herbicides. Prometryn, sethoxydim, bromoxynil octanoate, and monosulfuron are registered herbicides in millet fields. Single-weed control types and suitable millet varieties are the main factors restricting effective weed control and successful millet production [10,11,12,13]. Therefore, weeding in millet fields consumes considerable human and material resources for manual weeding, and the willingness of farmers to plant is not sufficiently strong, which limits the development of the millet industry [9,14]. There are various specific or broad-spectrum herbicides for major crops, such as rice (Oryza sativa L.) and wheat (Triticum aestivum L.); however, there is a lack of specific herbicides for grain crops, such as millet fields. Therefore, selecting new herbicide varieties for use in millet fields is crucial to the development of technological systems in the millet industry.

Recently, traits related to millet resistance have been reported. Various phenotypic characteristics exist in millet, including the yield components, plant height, number of earlets, number of grains per ear, and 1000-grain weight [15,16,17,18]. Furthermore, at the physiological and biochemical levels, the chlorophyll content, abscisic acid accumulation, and antioxidant enzyme activity can be used as reference indexes for evaluating millet resistance [19,20,21,22]. The aforementioned studies used a single morphological or physiological index as the identification index for crop herbicide resistance, making it difficult to objectively reflect on the role of herbicides [18,23]; thus, it is important to screen effective and simple identification indicators. Principal component analysis (PCA), the membership function method, and regression analysis have been used to study the effects of glyphosate on Brassica plants, the effects of pyrazine + flufenoxazine on soil microorganisms, and herbicide resistance caused by herbicide use in wheat fields in China and other countries [24,25,26]. Nonetheless, there is no unified standard for accurate and reliable herbicide resistance identification indicators and evaluation methods, and the published research on crop resistance identification methods is limited. The PCA membership function method and regression analysis have been used to analyze the effects of imazethapyr and drought on wheat [27,28], the salt tolerance of oat varieties [29], and lentil and sweet potato resources [30,31]. This method can accurately determine the weight of each index and transform the original data into several new, relatively independent, and comprehensive indexes. They can then be weighed using a membership function to obtain a comprehensive resistance evaluation objectively and comprehensively evaluate the resistance of grains to each herbicide [32]. The membership function method can provide a more comprehensive evaluation to determine multiple indicators, thereby avoiding the limitations and inaccuracies of a single evaluation index. Using a comprehensive evaluation, the differences between herbicides are comparable, and the resistance levels of crops to each herbicide can be divided via cluster analysis. Through correlation analysis and stepwise regression analysis, a mathematical model of herbicide resistance can be established to screen several representative indicators.

In this study, we cultivated Jingu 21, sprayed it with 31 herbicides in a field experiment, and measured 21 resistance traits. The differences in resistance to the 31 herbicides were comprehensively evaluated based on agronomic traits, physiological indicators, yield composition, and other indicators. Principal component analysis, cluster analysis, stepwise regression analysis, and other methods were used. Our aim was to establish a key index system for identifying the resistance of foxtail millet to herbicides and to provide a theoretical basis for a comprehensive evaluation of the safety of herbicide use on foxtail millet.

2. Materials and Methods

2.1. Experimental Design

This study was conducted at the millet demonstration base of Shanxi Agricultural University (38°59′ N, 112°90′ E) in Yujiazhuang Village, Dingxiang County. The study area has a temperate continental climate and an altitude of 760 m. The 0–30 cm soil layer has a pH of 8.31 and an available phosphorus content of 19.53 mg/kg, a potassium content of 177.33 mg/kg, and a total nitrogen content of 0.96 g/kg. Before the experiment, the field was planted with summer maize in rotation, and all maize straw was crushed and returned to the field after harvesting. The monthly average maximum and minimum temperature and monthly precipitation at the test site in 2022 are shown in

Table 1. The monthly average maximum and minimum temperature and monthly precipitation at the test site in 2022.

Month	Mean Temperature (°C)		Total Precipitation (mm)
	Max.	Min.
May	26	9	45.4
June	32	16	52.8
July	30	17	125.7
August	28	17	262.1
September	26	9	49.2
October	19	2	3.3

.

On 18 May 2022, the millet field was rotary tilled 10–15 cm, and the soil was prepared to flatten the land. After that, the millet was sown on 20 May. When the millet grows at the 4-leaf stage, the field millet is interplanted and fixed to have 330,000 millet plants per hectare, and herbicides are sprayed on the leaves when the millet grows to the 5-leaf stage and the weeds grow to the 3–5-leaf stage. The herbicides were applied using a hand-operated knapsack sprayer (300 kPa pressure) (China Delixi Holding Group Co., Ltd.; Yueqing, China) with a flat fan nozzle. The experiment was conducted in a randomized block design with three replicates, including no weeds, weed control treatments, and 31 herbicides (

Table 2. The chemical results, molecular targets, and dosage of herbicides were tested.

Treatment	Common Name	Type of Chemistry	Molecular Target	Application Rate (g a.i ha−1)
T1	MCPA-Na	Phenoxy carboxylic acids	Auxin mimics	960.96
T2	Picloram	Pyridine-carboxylates	Auxin mimics	1260.00
T3	Chlorotoluron	Ureas	Photosystem II D1 protein	2400.00
T4	Bromoxynil octanoate	Nitrile	Photosystem II D1 protein	562.50
T5	Prometryne	Triazines	Photosystem II D1 protein	504.00
T6	Bentazone	Benzothiadiazinone	Photosystem II D1 protein	1568.00
T7	Diflufenican	Phenyl ethers	Phytoene desaturase	113.16
T8	Cloransulam-methyl	Triazolopyrimidine-type	Acetolactate synthase	42.00
T9	Florasulam	Triazolopyrimidine-type	Acetolactate synthase	6.00
T10	Flumetsulam	Triazolopyrimidine-type	Acetolactate synthase	50.00
T11	Monosulfuron	Sulfonylureas	Acetolactate synthetase	17.78
T12	Pyribenzoxim	Pyrimidinyl benzoates	Acetolactate synthase	50.00
T13	Fomesafen	Diphenyl ethers	Protoporphyrinogen oxidase	162.50
T14	Fluoroglycofen	Diphenyl ethers	Protoporphyrinogen oxidase	93.33
T15	Lactofen	Diphenyl ethers	Protoporphyrinogen oxidase	144.00
T16	Saflufenacil	Miazines	Protoporphyrinogen oxidase	78.75
T17	Mesotrione	Triketones	Hydroxyphenyl pyruvate dioxygenase	260.00
T18	Butachlor	α-Chloroacetamides	Very long-chain fatty acid synthesis	1704.00
T19	Anilofos	α-Thioacetamides	Very long-chain fatty acid synthesis	420.00
T20	Halosulfuron	Sulfonylureas	Acetolactate synthetase	60.00
T21	Thifensulfuron methyl	Sulfonylureas	Acetolactate synthetase	24.75
T22	Gratil	Sulfonylureas	Acetolactate synthetase	40.00
T23	Bensulfuron methyl	Sulfonylureas	Acetolactate synthetase	40.00
T24	24-D iso-octyl ester	Phenoxy carboxylic acids	Auxin mimics	600.00
T25	Isoproturon	Ureas	Photosystem II D1 protein	800.00
T26	Flumioxazin	N-Phenylphthalimides	Protoporphyrinogen oxidase	61.20
T27	Clomazone	Isoxazolones	Carotenoid synthesis	950.40
T28	Pendimethalin	Dinitroanilines	Microtubule assembly	115.20
T29	Trifluralin	Dinitroanilines	Microtubule assembly	80.00
T30	Fluthiacet	α-Oxyacetamides	Very long-chain fatty acid synthesis	147.60
T31	Napropamide	Amides	Cell-division inhibitors	600.00

). Each test plot had an area of 4 m × 5 m (20 m²). The agronomic traits and physiological parameters of the millet seedlings were recorded 20 d after treatment, and the yield was determined at the maturity stage. Details of the specific experimental design are shown in

Table 3. Details of the experimental design.

18 May 2022 The depth of rotary soil preparation is 10–15 cm		23 June 2022 The millet grows to the 5-leaf stage and the weeds grow to the 3–5-leaf stage	13 July 2022 20 days after spraying	6 October 2022 Harvest
20 May 2022 Millet sowing	During the 4-leaf period, the seedlings were set to 330,000 plants per hectare.	31 herbicides were sprayed. such as prometryn (504 g a.i ha−1) Monosulfuron(17.78 g a.i ha−1)	Effectiveness of herbicides, plant height, calculate the leaf area, the aboveground fresh weight, the aboveground dry weight, SOD activity, POD activity, CAT activity, MDA content, soluble protein content, chlorophyll content.	Ear length, ear diameter, ear size, ear weight, grain weight per ear, 1000-grain weight, yield.

.

2.2. Morphological Parameters

Plant height, along with the length and width of the second leaf from the top, in uniformly growing foxtail millet in each treatment, was measured using a ruler 20 days after treatment. The fresh and dry weights of the aboveground parts were measured using a precision balance, and stem diameter was measured using a Vernier caliper [14]. To determine the height of the plant (X₁), the length (in centimeters) from the base to the tip of the longest leaf was measured after the plant was straightened. To calculate the leaf area (X₂), the length and width of the two leaves were measured and calculated using the following formula:

$$\mathrm{l}\mathrm{e}\mathrm{a}\mathrm{f}\mathrm{a}\mathrm{r}\mathrm{e}\mathrm{a}=\mathrm{l}\mathrm{e}\mathrm{a}\mathrm{f}\mathrm{l}\mathrm{e}\mathrm{n}\mathrm{g}\mathrm{t}\mathrm{h}\times \mathrm{l}\mathrm{e}\mathrm{a}\mathrm{f}\mathrm{w}\mathrm{i}\mathrm{d}\mathrm{t}\mathrm{h}\times 0.75$$

To determine the aboveground fresh weight (X₃), the aboveground parts of the millet seedlings were intercepted and the fresh weight (g) was determined. To determine the aboveground dry weight (X₄), the aboveground part was placed in an oven, deactivated at 105 °C for 15 min, dried to a constant weight at 75 °C, and weighed. To calculate stem diameter (X₅), we measured the internode diameter between the first and second internodes.

2.3. Physiological Parameters

Fresh foxtail millet leaves were weighed out to a quantity of 0.1 g after the veins were removed, then were cut into pieces and placed into a grinding vessel, and 2 mL of pH 7.8 phosphate buffer was added. The mixture was ground into homogenate in an ice bath and centrifuged at 13,201× g at 4 ° C for 15 min using a Sigma 3-18KS benchtop refrigerated centrifuge (Sigma Laborzentrifugen GmbH, Osterode am Harz, Germany). The supernatant was then removed and stored at 4 °C until further analyses. Superoxide dismutase (SOD) activity (X₆) was measured using the photochemical reduction method with nitrogen blue tetrazolium [14]. The control and enzyme treatments were exposed to light at 25 °C for approximately 20 min, while the blank treatment was placed in the dark as a control. The absorbance was measured at 560 nm using a Sunrise light absorption microplate reader (Tiangen Biochemical Technology Co., Ltd., Beijing, China). Peroxidase (POD) activity (X₇) was determined using the guaiacol method [14]. Briefly, 200 μL of enzyme solution was added to 3 mL of reaction solution, and the initial and 3 min absorbance values at 470 nm were measured using a UV 2400 UV–visible spectrophotometer (Sunny Hengping Instrument, LLC, Shanghai, China), using phosphate buffer pH 6.0 as a reference. Catalase (CAT) activity (X₈) was determined using the hydrogen peroxide method [14]. In a quartz cuvette, 100 μL of the enzyme solution was added to 1.5 mL of phosphate buffer pH 7.8, 1 mL of distilled water, and 0.3 mL of 0.1 M peroxide. The absorbance was measured at 240 nm every 30 s for 3 min using a UV 2400 UV–visible spectrophotometer (Sunny Hengping Instrument, LLC., Shanghai, China). Malondialdehyde (MDA) content (X₉) was determined using the thiobarbituric acid method [14]. Leaf samples (0.4 g) were weighed and ground in 5 mL 0.1% trichloroacetic acid solution. Five milliliters of 0.5% thiobarbituric acid was added, and the mixture was shaken. The test tube was placed in boiling water for 15 min, cooled to room temperature, transferred to a 10 mL centrifuge tube, and centrifuged for 15 min. A Sunrise light absorption microplate reader (Tiangen Biochemical Technology Co., Ltd., Beijing, China) was used to determine the absorbance values at 532 and 600 nm. The soluble protein content (X₁₀) was determined using Coomassie Brilliant Blue G-250 staining [33]. Fifty microliters of enzyme solution was added to 3 mL of Coomassie Brilliant Blue G-250 solution, and the solution was shaken several times in the reverse direction. Absorbance at 595 nm was determined using a Sunrise light absorption microplate reader (Tiangen Biochemical Technology Co., Ltd., Beijing, China). Chlorophyll a (X₁₁), chlorophyll b (X₁₂), total chlorophyll (X₁₃), and carotenoid (X₁₄) contents were determined according to the method described by Yuan et al. [14]. Twenty days after treatment, three millet seedlings with uniform growth were randomly selected from each pot, and two leaves were collected from each seedling. After the removal of veins, 0.1 g of fresh sample was weighed, cut into pieces, and placed into 10 mL test tubes, and 10 mL of 96% anhydrous ethanol was added to the extract for 24 h until the leaves turned white. The test tube was shaken every 6–8 h during the extraction period to thoroughly mix the photosynthetic pigments with ethanol. Chlorophyll a, chlorophyll b, and carotenoid concentrations were measured using a UV 2400 UV–visible spectrophotometer (Sunny Heng Ping Instrument, LLC., Shanghai, China). Absorbance was measured at 470, 649, and 665 nm [11].

C_a = 13.95 × A665 − 6.88 × A649

(1)

C_b = 24.96 × A649 − 7.32 × A665

(2)

C_car = 1000 × A470 − 2.05 × C_a − 114.8 × C_b/245

(3)

$$\mathrm{Pigment} \mathrm{content} (\mathrm{mg} {\mathrm{g}}^{-1} {\mathrm{F}}_{\mathrm{W}})=\mathrm{C} \times {\mathrm{V}}_{\mathrm{T}} \times \frac{n}{\mathrm{F}\mathrm{w}}\times 1000$$

(4)

where C is the pigment concentration (mg L⁻¹), F_W is the fresh weight (g), V_T is the total volume of the extraction (mL), and n is the dilution ratio.

2.4. Yield-Associated Traits

After harvesting, the following traits were measured using a ruler, Vernier caliper, and an analytical balance (Mettler-Toledo, LLC., Shanghai, China): the ear length (X₁₅), ear diameter (X₁₆), ear number (X₁₇), ear weight (X₁₈), grain weight per ear (X₁₉), and 1000-grain weight (X₂₀). The entire plot was harvested, threshed, and dried. The millets in each plot were accurately weighed and converted to 667 m² yield (X₂₁).

2.5. Resistance Analysis

The calculation of the resistance coefficient, membership function, weight calculation, and comprehensive evaluation D value is widely used in the study of drought resistance and salt tolerance [34,35,36,37]. The specific formulas are as follows:

DC = X_HR/X_CK

(5)

μ(X_i) = (X_i − X_min)/(X_max − X_min)

(6)

$${\mathrm{W}}_{\mathrm{i}}={\mathrm{P}}_{\mathrm{i}}/\sum _{\mathrm{i}=1}^{\mathrm{i}}{\mathrm{P}}_{\mathrm{i}}$$

(7)

$$\mathrm{D}=\sum _{\mathrm{i}=1}^{\mathrm{i}}\left[\mathrm{U}\right({\mathrm{X}}_{\mathrm{i}}\left){\mathrm{W}}_{\mathrm{i}}\right]\mathrm{i}=\mathrm{1,2},3,\dots ,\mathrm{n}$$

(8)

where DC denotes the resistance coefficient [38]; X_HR and X_CK are the trait values evaluated under herbicide and clear-water conditions, respectively; μ is the resistance membership function value based on traits; X_i is the i-th comprehensive index; X_max and X_min are the maximum and minimum values of the i-th comprehensive index, respectively; W_i is the weight of the i-th comprehensive index; and P_i is the variance contribution rate of the i-th comprehensive index.

The membership function value (μ) of each comprehensive herbicide index was calculated using Equation (7). The principal component weight was calculated using the contribution rate of each comprehensive index, and the comprehensive evaluation value (D value) of the resistance was calculated using the index weight and membership function value. The D value was the comprehensive evaluation value of each resistance determined via comprehensive evaluation of each index under the action of each herbicide.

2.6. Effectiveness of Herbicides

Twenty days after treatment, 0.5 m × 0.5 m (0.25 m²) was taken from each plot by a five-point sampling method to investigate the number of broad-leaved and gramineous weeds in each treatment plot, plant height, and weight, and later converted into the weed number control effect, plant height inhibition rate, and fresh weight control effect.

2.7. Data Analysis

The data were collected through experiments, and the herbicide resistance coefficient was calculated using Formula (5) after data processing. Through correlation analysis, principal component analysis, and membership function analysis, the comprehensive evaluation D value of each herbicide for Jingu 21 was calculated. The resistance difference of Jingu 21 to different herbicides was classified using cluster analysis. Stepwise regression analysis was used to find out the main indicators affecting the comprehensive evaluation D value. Microsoft Excel 2021 (Microsoft, Redmond, WA, USA) was used to summarize and analyze the data. Each trait was measured in triplicate. Analysis of variance, PCA, and stepwise regression analysis were performed using the IBM SPSS Statistics 19.0 (IBM Corp., Armonk, NY, USA) software. Images were obtained using the R 4.1.0 (R Development Core Team, Vienna, Austria).

3. Results

3.1. Herbicide Resistance Coefficient and Simple Correlation Analysis for Each Trait of Jingu 21

The herbicide resistance coefficients for the relevant traits were calculated based on the findings for each trait in 31 herbicide-treated and control groups. The descriptive statistics are presented in

Table 4. Descriptive statistics of herbicide resistance coefficients for each individual indicator of Jingu 21.

Trait	Maximum	Minimal	Average	Standard Deviation	Coefficient of Variation (%)
X1	1.12	0.46	0.94	0.14	14.31
X2	1.28	0.57	1.09	0.18	16.12
X3	1.14	0.18	0.82	0.22	26.91
X4	1.36	0.27	0.97	0.22	22.74
X5	1.25	0.60	0.94	0.17	18.59
X6	3.51	0.98	2.06	0.63	30.51
X7	3.86	1.05	1.96	0.59	30.27
X8	2.13	0.80	1.41	0.29	20.57
X9	1.16	0.78	0.96	0.09	9.38
X10	1.12	0.59	0.83	0.16	19.57
X11	1.74	0.34	0.96	0.36	37.60
X12	1.06	0.70	0.87	0.09	10.46
X13	1.24	0.59	0.90	0.17	19.23
X14	1.19	0.72	0.94	0.13	14.27
X15	1.14	0.60	0.85	0.14	16.27
X16	1.36	0.88	1.04	0.12	11.88
X17	1.29	0.75	1.02	0.11	10.4
X18	1.45	0.54	0.90	0.21	22.92
X19	1.81	0.42	1.19	0.33	28.01
X20	1.26	1.00	1.13	0.04	3.92
X21	1.59	0.47	1.17	0.25	21.05

. Among the herbicide resistance coefficients, the coefficients of variation for SOD activity, POD activity, and chlorophyll a were 30.51%, 30.27%, and 37.60%, respectively. The coefficients of variation for the aboveground fresh weight and grain weight per ear were 26.91% and 28.01%, respectively. The coefficients of variation for the aboveground dry weight, CAT activity, ear weight, and yield were 22.74%, 20.57%, 22.92%, and 21.05%, respectively. These results indicated that herbicides had certain effects on these traits. The herbicide resistance coefficients showed that the values of relevant agronomic traits decreased to different degrees under the effect of herbicides compared with those of the control; however, the values of physiological traits increased to different degrees. These results indicated that Jingu 21 grains could resist herbicide damage by regulating their relevant metabolic processes. The magnitude of variation in each index varied among the Jingu 21 samples, and it was difficult to determine the herbicide resistance of Jingu 21 based on the herbicide resistance coefficient of each index. Correlation analysis showed that the plant height was positively correlated with the leaf area, shoot dry weight, shoot fresh weight, stem diameter, and chlorophyll a, total chlorophyll, and carotenoid contents. However, the MDA content was negatively correlated with the yield. Therefore, each individual indicator was significantly or very significantly correlated with at least one other individual indicator, leading to overlaps in the information they provided. This indicated that herbicide resistance in Jingu 21 is a complex and comprehensive trait (Figure 1). The use of any single index of an herbicide resistance coefficient to evaluate the herbicide resistance of millet is one-sided and unstable, and it is more reliable to use multiple indicators for a comprehensive evaluation.

3.2. PCA Comprehensive Evaluation and Screening of Herbicide Resistance Traits in Foxtail Millet

PCA was performed on the herbicide resistance coefficients of 21 conventional traits, and seven principal components were extracted (

Table 5. Coefficient of principal component scores and contribution of each trait.

Principal Factor	CI1	CI2	CI3	CI4	CI5	CI6	CI7
X1	0.803 *	−0.317	0.043	0.200	−0.125	0.296	0.186
X2	0.803 *	−0.382	−0.038	0.235	0.042	0.214	0.047
X3	0.833 *	−0.443	−0.163	0.064	−0.156	0.047	0.009
X4	0.775 *	−0.455	−0.229	0.132	−0.048	0.004	−0.010
X5	0.588 *	−0.400	−0.319	−0.023	0.294	−0.236	−0.184
X6	0.027	−0.383	0.303	−0.376	0.488 *	−0.169	0.361
X7	0.002	0.328	0.184	−0.151	0.526	0.254	0.527 *
X8	0.194	−0.078	0.459	0.211	0.631 *	−0.080	−0.052
X9	−0.263	−0.350	−0.402	−0.521 *	−0.014	0.107	0.265
X10	0.031	0.554 *	−0.192	0.002	−0.547	−0.123	0.426
X11	0.805 *	0.501	0.013	−0.207	−0.005	−0.007	−0.134
X12	0.752 *	0.467	−0.170	−0.247	0.095	−0.084	−0.041
X13	0.803 *	0.523	−0.055	−0.23	0.018	−0.020	−0.044
X14	0.790 *	0.502	−0.006	−0.262	0.112	−0.055	−0.007
X15	0.236	−0.262	0.736 *	−0.112	−0.319	−0.035	0.068
X16	0.074	0.050	−0.587 *	0.547	0.087	−0.189	0.410
X17	−0.136	0.344	−0.390	0.611 *	0.229	0.172	0.039
X18	−0.231	0.250	−0.024	−0.127	0.155	0.762 *	−0.205
X19	0.232	0.000	0.686 *	0.056	−0.320	0.001	0.223
X20	−0.080	0.317	0.299	0.276	0.186	−0.493 *	−0.114
X21	0.239	0.21	0.591 *	0.563	−0.034	0.203	0.025
Eigenvalue	5.770	2.890	2.698	1.920	1.689	1.260	1.033
CR (%)	27.47	13.76	12.85	9.14	8.05	6.00	4.92
CCR (%)	27.47	41.24	54.09	63.23	71.27	77.2	82.19
Factor weights	0.334	0.167	0.156	0.111	0.098	0.073	0.060

* Indicates the maximum absolute value of an indicator for each factor; CI, comprehensive index; CR, contribution ratio; CCR, cumulative contribution ratio.

). The variance contribution rates of the composite index (CI) for CI1–CI7 were 27.47%, 13.76%, 12.85%, 9.14%, 8.05%, 6.00%, and 4.92%, respectively, with a cumulative contribution of 82.19% and cumulative variance contribution of ≥80%. This cumulative variance contribution was considered highly representative of the information [28,29,30,31]. The original 21 single traits were converted into seven new independent comprehensive indices, covering most of the information. The eigenvalue of CI1 was 5.770, and the contribution rate was 27.47%. The first principal component was primarily determined by the plant height, leaf area, aboveground fresh weight, aboveground dry weight, stem thickness, chlorophyll a, chlorophyll b, chlorophyll (a + b), and carotenoids, suggesting that these indicators could reflect the resistance of Jingu 21, among which the aboveground fresh weight indicator had the largest absolute value. The eigenvalue of CI2 was 2.890, and the contribution rate was 13.76%. The second principal component was determined primarily according to the soluble protein content. The eigenvalue of CI3 was 2.698, and the contribution rate was 12.85%. The third principal component was primarily determined by the ear length, ear thickness, ear grain weight, and yield, among which the absolute value of the ear length was the largest. The eigenvalue of CI4 was 1.920, and the contribution rate was 9.14%. The fourth principal component was determined primarily according to the MDA content and ear number, and the absolute value of the ear number was the largest. The eigenvalue of CI5 was 1.689, and the contribution rate was 8.05%. The fifth principal component was primarily determined via SOD and CAT activity, and the absolute value of the CAT activity index was the largest. The eigenvalue of CI6 was 1.260, and the contribution rate was 6.00%. The sixth principal component was primarily determined according to the ear weight and 1000-grain weight, and the absolute value of the ear weight index was the highest. The eigenvalue of CI7 was 1.033, and the contribution rate was 4.92%. The seventh principal component was primarily determined via the POD activity. In this analysis, the first principal component was broadly categorized as the plant morphology and photosynthetic pigmentation factor, the second, fourth, fifth, and seventh principal components were categorized as the physiological regulation factor, and the third and sixth principal components were categorized as the yield composition factor.

3.3. Comprehensive Evaluation and Cluster Analysis of Jingu 21 Herbicide Resistance

Herbicide resistance coefficients were calculated for each indicator (Table 4), and eigenvectors were calculated for each principal component (Table 5). Under the action of herbicides, the higher the value of the same composite indicator, the more resistant the millet, and vice versa. The resistance of Jingu 21 was evaluated based on the values for each herbicide (CI1–CI7), and the weights assigned to these seven principal components influenced the importance attributed to each component in assessing millet resistance. Therefore, Equations (6) and (8) were used to calculate the D values of the indicators of Jingu 21 resistance under the action of herbicides, using the affiliation function based on the integrated indicators (

Table 6. Overall evaluation of the D values for Jingu 21.

Processing	μ1	μ2	μ3	μ4	μ5	μ6	μ7	D	Sort by
T1	0.776	0.555	0.365	0.569	0.943	0.417	0.411	0.620	20
T2	0.246	0.483	0.267	0.000	0.484	0.000	0.422	0.277	1
T3	0.575	0.554	0.777	0.707	0.582	0.586	0.358	0.606	18
T4	0.809	0.500	0.509	0.382	0.851	0.642	0.434	0.632	22
T5	0.799	0.321	0.937	0.635	0.579	0.325	0.299	0.636	24
T6	0.662	0.196	1.000	0.809	0.527	0.351	0.416	0.603	16
T7	0.949	0.428	0.864	0.733	0.793	0.139	0.125	0.701	28
T8	0.840	0.752	0.556	0.544	0.743	0.373	0.205	0.666	27
T9	0.771	0.804	0.626	0.474	0.384	0.444	0.374	0.635	23
T10	0.709	0.466	0.489	0.303	0.917	0.413	0.347	0.566	13
T11	0.738	0.420	0.881	0.739	0.565	0.542	0.445	0.658	26
T12	0.000	0.960	0.622	0.657	0.972	0.634	0.073	0.477	6
T13	0.380	0.791	0.854	0.532	0.000	0.273	0.434	0.498	8
T14	0.502	0.503	0.728	0.802	0.499	0.860	0.397	0.590	15
T15	1.000	0.991	0.318	0.745	0.601	0.501	0.254	0.743	29
T16	0.421	0.824	0.924	0.801	0.534	0.213	0.471	0.608	19
T17	0.216	0.519	0.372	0.854	0.623	0.423	0.554	0.437	3
T18	0.760	1.000	0.272	0.799	1.000	0.479	1.000	0.746	30
T19	0.959	0.780	0.633	0.580	0.666	0.851	0.455	0.769	31
T20	0.843	0.382	0.237	0.566	0.416	0.190	0.000	0.500	9
T21	0.587	0.000	0.920	0.632	0.890	0.345	0.779	0.569	14
T22	0.621	0.325	0.136	1.000	0.815	0.121	0.211	0.496	7
T23	0.911	0.396	0.700	0.221	0.881	0.292	0.677	0.653	25
T24	0.594	0.186	0.582	0.491	0.896	0.834	0.081	0.529	11
T25	0.492	0.229	0.349	0.475	0.566	1.000	0.246	0.454	5
T26	0.505	0.215	0.445	0.433	0.316	0.617	0.626	0.436	2
T27	0.856	0.558	0.562	0.707	0.186	0.520	0.385	0.625	21
T28	0.702	0.173	0.524	0.780	0.652	0.383	0.207	0.536	12
T29	0.753	0.207	0.382	0.588	0.250	0.736	0.535	0.522	10
T30	0.881	0.705	0.448	0.528	0.004	0.509	0.401	0.603	17
T31	0.616	0.154	0.000	0.987	0.414	0.349	0.582	0.442	4

). The larger the value, the stronger the resistance. The D value was used as a criterion to determine and rank the strength of herbicide resistance in Jingu 21. T₁₉ had the highest D value and the strongest resistance, whereas T₂ had the lowest D value and weakest resistance.

A correlation analysis between the DC and D values of the 21 indicators showed that the plant height, second leaf area, POD, CAT, chlorophyll a, chlorophyll b, chlorophyll (a + b), carotenoids, grain weight per ear, and yield were positively correlated with the D value. However, the MDA content was negatively correlated with the D value (

Table 7. Correlation analysis between the DC and D values for each indicator.

Characteristic	Correlation Coefficient between DC and D Values	Sort by
X1	0.551 **	7
X2	0.513 **	8
X3	0.344	12
X4	0.303	13
X5	0.147	16
X6	0.024	20
X7	0.408 *	10
X8	0.465 **	9
X9	−0.561 **	6
X10	0.080	18
X11	0.739 **	2
X12	0.632 **	5
X13	0.730 **	3
X14	0.745 **	1
X15	0.238	14
X16	0.016	21
X17	0.123	17
X18	0.035	19
X19	0.387 *	11
X20	0.203	15
X21	0.637 **	4

** and * indicate significant differences at the 1% and 5% levels, respectively.

). Based on the correlation analysis, an optimal regression equation was constructed using the D value and traits, and the evaluation index for millet herbicide resistance was screened. The optimal regression equation constructed via stepwise regression analysis was D’ = 0.036 + 0.112 X₂ + 0.038 X₇ + 0.077 X₈ – 0.258 X₉ + 0.349 X₁₃ + 0.038 X₁₉ + 0.115 X₂₁, where X₂, X₇, X₈, X₉, X₁₃, X₁₉, and X₂₁ represent the inverted leaf area, POD activity, CAT activity, MDA content, chl (a + b) content, grain weight, and yield, respectively. The correlation coefficient (r) and the coefficient of determination (R²) were 0.986 and 0.971, respectively.

Based on the D values of the different herbicides after treatment, the herbicides were clustered using the Euclidean distance and the unweighted average method for their resistance (Figure 2). At an Euclidean distance of 0.0660, the 31 herbicides were classified into five categories: T₂ as category I (extremely weakly resistant); T₁₇, T₂₆, T₃₁, and T₂₅ as category II (weakly resistant); T₁₂, T₁₃, T₂₀, T₂₂, T₂₄, T₂₉, and T₂₈ as category III (moderately resistant); T₁₈, T₁₅, and T₁₉ as category IV (extremely strongly resistant); and T₁, T₂₇, T₄, T₅, T₉, T₃, T₁₆, T₆, T₃₀, T₁₄, T₁₀, T₂₁, T₇, T₈, T₁₁, and T₂₃ as category V (strongly resistant).

3.4. Analysis of Resistance Index in Stepwise Regression Analysis

As shown in Figure 3, the leaf area of T₂ and T₁₂ decreased significantly, by 37.20% and 42.96%, respectively, compared with that of clear-water conditions (CK). Please note that p < 0.05. The leaf areas of T₈, T₉, T₁₃, T₁₆, and T₁₇ decreased by 4.71 3.38%, 17.23%, 7.5 0%, and 13.19%, respectively, compared with that of CK, but there was no significant difference compared to CK (Figure 3A). The POD activities of T₂₀, T₂₂, and T₃₁ were not significantly different from those of CK. The POD activities of T_4, T₈, T₁₀, T₁₂, T₁₆, T₁₈, T₁₉, T₂₁, and T₂₃ were significantly increased, by 139.69%, 150.97%, 155.59%, 141.79%, 122.64%, 285.75%, 217.57%, 179.02%, and 135.50%, respectively, compared to CK (p < 0.05) (Figure 3B). The CAT activities of T₇, T₂₃, and T₂₄ were significantly increased, by 53.15%, 48.39%, and 43.27% (p < 0.05), respectively, compared to those of CK. The CAT activities of T₁₃, T₂₆, and T₂₉ were reduced by 20.38%, 1.08%, and 1.8%, respectively, compared with those of CK, and there was no significant difference compared to CK (Figure 3C). The MDA contents of T₂, T₂₃, T₂₅, and T₂₆ were significantly increased, by 9%, 12.98%, and 13.77% (p < 0.05), respectively, compared to those of CK. The MDA contents of T₅, T₇, and T₁₅ were significantly decreased, by 15.96%, 17.68%, and 21.77% (p < 0.05), respectively, compared to CK (Figure 3D). The chlorophyll (a + b) content of T₁₇ and T₂₁ decreased significantly, by 40.82% and 37.76%, respectively, compared to CK (p < 0.05). The chlorophyll (a + b) of T₁, T₄, T₁₅, T₂₃, and T₃₀ increased by 3.92%, 4.85%, 19.67%, 10.09%, and 13.27%, respectively, compared to CK, and there was no significant difference compared with CK (Figure 3E). Compared to CK, the grain weight per ear of T₁₁, T₁₃, T₂₁, T₂₃, T₂₈, and no weeds increased significantly, by 40.57%, 42.94%, 44.78%, 37.12%, 34.95%, and 36.75% (p < 0.05), respectively. Compared with CK, the grain weight per ear of T₂, T₄, T₁₀, T₁₂, and T₁₇ decreased by 28.56%, 6.35%, 33.57%, 10.86%, and 24.26%, respectively, but the difference was not significant (Figure 3F). Compared with CK, the yields of T₁, T₁₁, T₁₄, T₂₄, T₂₇, and no weeds were significantly increased, by 13.1%, 37.61%, 36.52%, 1.36%, 37.07%, and 26.26%, respectively (p < 0.05). Compared with CK, the yields of T₂, T₁₀, T₂₀, T₂₃, T₂₅, T₂₆, and T₃₁ were significantly decreased, by 53.42%, 2.74%, 17.81%, 6.85%, 21.92%, and 10.96%, respectively (p < 0.05) (Figure 3G).

3.5. Effectiveness of Herbicides

From Figure 4, it can be seen that the plant control effect, plant height inhibition rate, and fresh weight control effect of T₁, T₂, T₃, T₁₇, and T₂₆ on broadleaf weeds are all 100%. The fresh weights of broadleaf weeds in T₁₁, T₁₄, T₁₆, and T₂₄ were 77.77%, 86.15%, 53.77%, and 96.46%, respectively, which increased by 71.61%, 74.37%, 58.94% and 77.11% compared with T₁₉. The plant control effect, plant height inhibition rate, and fresh weight control effect of T₁₇ and T₂₄ on gramineous weeds were all 100%. The fresh weight control effect of T₃ and T₅ on gramineous weeds was 94.92% and 94.73%, respectively, which was increased by 86.3% and 86.28% compared with T₁₉. The fresh weight control effects of T₈ and T₉ on gramineous weeds were 6.8% and 8%, which had decreased by 47.69% and 38.46% compared with T₁₉.

4. Discussion

The current study was based on agronomic traits, physiological indicators, yield components, and other indicators, through PCA, cluster analysis, stepwise regression analysis, and other analytical methods to clarify the different modes of action of herbicide resistance of the high-quality conventional millet Jingu 21. In general, compared to the control treatment, Jingu 21 showed significant single-trait changes with different herbicides. All the genotypes showed significant single-trait changes under stress conditions. The effects of herbicides on cereals are multifaceted, affecting almost every aspect of growth and development. All traits in this study showed varying degrees of variation with high magnitudes and coefficients of variation. Crop resistance to herbicides is a comprehensive trait [32]. Therefore, it is important to select more comprehensive traits and appropriate methods for evaluating herbicide resistance. There was a certain degree of correlation among many indicators, resulting in overlapping responses in the sources of crop resistance traits. However, it was difficult to identify this using a single indicator. The use of multiple indicators reflected the strength and characteristics of crop resistance at different stages and various aspects. An increasing number of screening studies have recently used a combination of PCA, affiliation function, and cluster analysis to evaluate results more reasonably and reliably, avoiding the bias and instability of a single trait [39,40]. PCA can reduce multiple variables to several potential factors and missing data to effectively group herbicide resistance. The resistance membership function value is a multivariate index that integrates the resistance coefficients of different traits and effectively reflects the overall performance of plants under herbicidal action. Furthermore, the use of PCA in conjunction with membership functions and cluster analysis makes the assessment of stress resistance in crops reliable and practical. However, herbicide resistance has rarely been identified in cereals. Therefore, in this study, the herbicide resistance coefficients of each index of Jingu 21 were used as indicators to measure the strength of herbicide resistance of Jingu 21, and the 21 relevant indicators were transformed into 7 integrated indicators using PCA, which comprehensively reflected the information of the original indicators. The D values of the different herbicides for Jingu 21 were obtained based on seven integrated indicators to compare the differences in the resistance of Jingu 21 to various herbicides. The 31 herbicides were classified into five categories based on their D values, with each category representing a different level of resistance. Through the correlation analysis between the D value and each resistance index, the resistance traits with p < 0.05 were selected for the stepwise regression analysis of 11 resistance indexes, and the model for evaluating the resistance of Jingu 21 to herbicides was established, which further improved the accuracy of the regression analysis model and the scientific reliability and evaluation of the identified traits.

Liu et al. pointed out that the leaf morphological index and MDA content can be used as important reference indexes in the evaluation of the drought tolerance of Alsophila spinulosa in China [41]. Magwanga et al. used the POD activity, CAT activity, and chlorophyll content as important reference indexes in the evaluation of the drought and salt tolerance of cotton [42]. Sun et al. used yield as an important index to evaluate the drought tolerance of cotton [43]. It is consistent with the identification indicators selected in the herbicide resistance evaluation in this study. Leaf development, size, and area are important indicators of millet growth, stress resistance, and yield formation [44]. Both POD and CAT are key enzymes in the enzymatic defense system of plants under stress conditions and are responsible for catalyzing the removal of H₂O₂ and improving the stress resistance of plants [45]. The degree of membrane lipid peroxidation in plants depends on the MDA content, which is proportional to it [16]. The change in chlorophyll (a + b) content shows that foxtail millet is very sensitive to herbicides and directly affects photosynthetic efficiency and yield [46]. The grain weight per ear and yield are closely related to each other and are macroscopic data obtained under the combined action of growth and physiological indicators [44]. These data are of great significance for the safe use of herbicides and the study of grain production and food security. Therefore, the seven indicators screened are important for effective and simple identification to evaluate the strength of resistance. There are representative physiological indexes and yield components that provide strong guidance for the rapid evaluation of the herbicide resistance of grain after herbicide application.

The 31 herbicides were classified into five categories based on their D values, and each category represented a different level of resistance. Picloram led to a decrease in leaf area and chlorophyll (a + b), an increase in POD and MDA, a decrease in ear weight and yield, and a comprehensive effect in multiple indexes; therefore, it was as defined as extremely weakly resistant. Mesotrione caused a decrease in leaf area and chlorophyll (a + b), whereas POD and CAT increased the resistance to stress, eventually leading to a decrease in panicle weight. Shi et al. also showed that 300 g ha⁻¹ mesotrione was less safe for Jingu 21 [47]. Therefore, mesotrione was defined as weakly resistant. Pyribenzoxim led to decreased leaf area and chlorophyll (a + b), while POD increased and resistance increased, which eventually led to a decrease in grain weight per spike. Zhao et al. showed that 6.00 L ha⁻¹ of trifluralin and 9.00 L ha⁻¹ of pendimethalin had no significant effect on the seedling rate or agronomic traits of Zhangzagu 10, which was consistent with the results of this study [48]. Therefore, pyribenzoxim, trifluralin, and pendimethalin were defined as moderately resistant. Prometryne and Saflufenacil led to an increase in leaf area, an increase in CAT, an increase in resistance, and an increase in grain weight per spike and yield. Therefore, it is classified as moderately resistant. Zhang et al. and Ma et al. showed that 300 g ha⁻¹ of bensulfuron-methyl was relatively safe when applied to Jingu 21 [20], which was demonstrated in this study. Therefore, prometryne, saflufenacil, and bensulfuron methyl were defined as strongly resistant. Anilofos led to an increase in leaf area, chlorophyll (a + b), and photosynthesis in millet. Simultaneously, a sharp increase in POD increased its resistance, which eventually led to an increase in grain weight. Therefore, anilofos was defined as extremely strongly resistant. Due to the action of different herbicides, the representative indexes of agronomic traits, physiological indexes, and yield components of Jingu 21 showed different degrees of change, among which the extremely weakly resistant index changed significantly, and the extremely strongly resistant index changed slightly, and then quickly returned to normal growth. Therefore, according to the cluster analysis, three types of herbicides with extreme resistance and sixteen types of herbicides with strong resistance can be studied in Jingu 21.

Among the 19 herbicides with strong and extremely strong resistance to Jingu 21, MCPA-Na, chlorotoluron, prometryne, monosulfuron, fluoroglycofen, saflufenacil, thifensulfuron methyl, and bensulfuron methy showed a comprehensive weeding effect of more than 80%. Therefore, the next stage of research can be carried out. For the 11 herbicides with poor weeding effect, the dose can be increased to study the effectiveness of weeding and to evaluate their resistance.

5. Conclusions

In this study, 31 herbicides that have been introduced locally and worldwide were comprehensively identified to cause resistance in millet, and each trait was affected to a different degree. Based on the results of the multi-index and multi-method resistance evaluation, cluster analysis revealed that foxtail millet exhibited strong resistance to three herbicides, lactofen, butachlor, and anilofos, with anilofos demonstrating the highest resistance. Stepwise regression analysis further revealed that the leaf area, POD activity, CAT activity, MDA content, chlorophyll (a + b), grain weight, and yield could be used as indicators for the rapid and accurate evaluation of resistance in cereal crops. These findings provide a solid foundation for the future application of herbicides and the evaluation of resistance in cereal crops. Among the herbicides with strong resistance and extremely strong resistance, eight herbicides were found to be suitable for the next study due to the effectiveness of their herbicides. These results divided the resistance of Jingu 21 to 31 herbicides and screened out the identification indexes that are of great significance for the evaluation of herbicides, determined the degree of phytotoxicity of each herbicide to Jingu 21 and its effectiveness, and selected superior herbicides for the next field trials to improve the use of these herbicides in the field. The opportunity provides a solid foundation for the application and resistance evaluation of cereal crop herbicides in the future.