Method for precise breeding of excellent individuals of pinus densata based on multivariate genomic selection model and application thereof

Abstract

The application discloses a method for accurately breeding excellent individuals of Castanopsis hystrix based on a multivariate genomic selection model and application thereof. The application constructs a reference population by using asexually propagated clones of the Castanopsis hystrix, constructs a candidate population by using half-sib family offspring, focuses on growth traits and wood property traits, adopts a GOF genomic relationship matrix, differentiates the number of SNP markers according to trait genetic characteristics, matches an optimal auxiliary trait combination to construct a multivariate GBLUP model, combines 5-fold cross validation to complete model optimization, and then adopts an equal-weight selection index method and a Breeding multiple trait evaluation method to jointly screen excellent single plants. The application greatly improves the prediction accuracy of multiple traits of the Castanopsis hystrix, shortens breeding generations, improves breeding annual genetic gain, and can cooperatively and accurately breed multiple-target excellent single plants at an early stage of a seedling period. The application is operable, cost-controllable, suitable for scale breeding of fine broad-leaved tree species of the Castanopsis hystrix and the Fagaceae, parent selection and matching and high-generation breeding, and has a wide application prospect.

Description

Technical Field

[0001] This invention belongs to the field of forest tree molecular genetics breeding and genome selection technology, specifically involving a method for precise selection of superior individuals of the Red Cone tree based on a multivariate genome selection model and its application. Background Technology

[0002] Red cone ( Castanopsis hystrix The Chinese pine (Castanopsis fargesii) is a precious broad-leaved timber species endemic to South my country, belonging to the Fagaceae family. It is characterized by rapid growth, excellent material quality, and strong ecological adaptability, and is widely used in furniture, construction, ecological restoration, and other fields, with extremely high economic and ecological value. Currently, the mainstream breeding method for Chinese pine is traditional phenotypic selection, which has three major bottlenecks: (1) The breeding cycle is long. The wood properties of Chinese pine need to be expressed stably when the tree is over 16 years old. The generation interval of traditional breeding is as long as 20 years, resulting in extremely low breeding efficiency; (2) It is difficult to improve multiple traits in a coordinated manner. There are complex synergistic and antagonistic relationships between the growth traits, wood properties, and wood chemical properties of Chinese pine. Traditional single-trait selection is prone to neglecting one aspect and missing another; (3) The breeding effect of traits with low heritability is poor. The heritability of traits such as microfibril angle, cellulose, and hemicellulose is extremely low, and the accuracy of phenotypic selection is insufficient.

[0003] Genomic selection (GS) is a core technology for efficient forest tree breeding. It can predict the genomic breeding value of an individual by using whole-genome SNP markers, enabling early selection in the seedling stage and significantly shortening the breeding cycle. Existing genomic selection studies on *Castanopsis fargesii* mostly use univariate GBLUP (Genomic Best Linear Unbiased Prediction) models, which only predict single traits and cannot utilize genetic covariance information between traits. This results in poor prediction accuracy for traits with low heritability and fails to configure marker density and optimize auxiliary trait combinations for different traits, making it difficult to meet the actual breeding needs of simultaneous improvement of growth and multiple timber properties in *Castanopsis fargesii*.

[0004] Multivariate GBLUP models can integrate genetic information related to multiple traits, enabling cross-trait information borrowing and significantly improving prediction accuracy. They are particularly suitable for multi-target trait improvement. However, there is currently no systematic application scheme for this model in Castanopsis fargesii and other precious broad-leaved tree species in the Fagaceae family. Therefore, developing a multivariate genomic selection breeding method that is adapted to the genetic characteristics of Castanopsis fargesii and has optimized parameters is key to breaking through the breeding bottleneck of Castanopsis fargesii and promoting the development of molecular breeding of precious broad-leaved tree species. Summary of the Invention

[0005] The primary objective of this invention is to overcome the shortcomings of existing technologies, such as long breeding cycles, difficulty in synergistic improvement of multiple traits, and low prediction accuracy of univariate genomic selection. This invention provides a method for precisely selecting superior individuals of the *Pinus thunbergii* species based on a multivariate genomic selection model. This method achieves precise selection of multiple traits in the seedling stage of *Pinus thunbergii* through population construction, genetic parameter analysis, multi-dimensional parameter optimization of the model, and joint screening, thereby shortening the breeding cycle and increasing genetic gain.

[0006] The present invention also provides supporting application scenarios for the above method.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] A method for precisely selecting superior individuals of the Red Cone plant based on a multivariate genomic selection model includes the following steps: population construction → phenotypic determination + leaf sampling → DNA extraction and genotyping → SNP quality control and imputation → genetic parameter analysis → multi-parameter optimization of the multivariate GBLUP model (matrix + auxiliary traits + marker density) → cross-validation model → GEBV calculation of candidate populations → joint screening of superior individual plants using two methods → breeding application; the preferred method includes the following specific steps:

[0009] (1) Construction of experimental populations: A reference population and a candidate population of Conophytum purpureus were constructed respectively; the reference population was a clone of Conophytum purpureus and the candidate population was a half-sib family;

[0010] (2) Phenotypic determination and sample collection: Growth traits and wood properties were measured for the reference population to obtain phenotypic data. Among them, growth traits were tree height (H) and diameter at breast height (DBH), and wood properties were wood density (WD), fiber length (FL), fiber length-to-width ratio (LWR), microfibril angle (MFA), lignin content (LIG), and cellulose content (CE). Several sample plants were randomly selected from each Conophytum purpureus clone, and the average value was taken as the phenotypic data of the clone. Healthy young shoots and leaves from the current year of the reference population and candidate population were collected, preserved, and used for DNA extraction.

[0011] (3) Genotyping and SNP screening: Genomic DNA was extracted from leaves and subjected to high-throughput sequencing, sequence quality control, alignment, variation detection and filtering to obtain high-quality SNP markers; the filtering criteria were: site sequencing depth ≥4, population deletion rate ≤20%, minor allele frequency ≥0.01, Hardy-Weinberg equilibrium P ≥0.0001; genotyping was completed using Beagle 5.0;

[0012] (4) Genetic parameter analysis: Based on the GOF genome relation matrix, the restricted maximum likelihood method (REML) was used to estimate the broad heritability, genetic correlation, environmental correlation and phenotypic correlation of the eight traits, and to clarify the genetic hierarchy of traits and the synergistic / antagonistic relationship between traits;

[0013] (5) Construction and parameter optimization of multivariate GBLUP model:

[0014] A. Determine the standard matrix for genomic relationships: Compare the predictive capabilities of four matrices—GOF, GD, G05, and GMF—to determine the standard matrix;

[0015] B. Optimize auxiliary trait combinations: Construct bivariate and multivariate GBLUP models with a single trait as the target trait and the remaining traits as auxiliary traits. Select trait combinations with an absolute value of genetic correlation coefficient > 0.6 as the optimal auxiliary combinations, and distinguish between three types of trait patterns: saturated response mode, cumulative gain mode, and weak response mode, to determine the practical optimal combination that balances prediction accuracy and measurement cost.

[0016] C. Differentiated configuration of SNP label quantity: Based on step B, label gradients are divided according to the trait prediction saturation features, and the saturation label quantity is determined by combining 5-fold cross-validation;

[0017] D. Model Optimization: Based on B and C above, a hierarchical 5-fold cross-validation strategy is adopted to test and evaluate the model's predictive ability and accuracy, and to complete the overall optimization of the multivariate GBLUP model.

[0018] (6) Calculation of candidate population genomic breeding value (GEBV)

[0019] Using the optimized multivariate GBLUP model, the genomic estimated breeding value (GEBV) for all target traits in the candidate population was calculated.

[0020] (7) Joint screening of superior single plants

[0021] A. Equal-weighted selection index method: The GEBV of the 8 core traits is standardized, an equal-weighted selection index is constructed, and the top 5% of individuals are selected as preliminary superior single plants.

[0022] B. Brekin Multi-Track Evaluation Method: The GEBV of negative traits MFA and LIG is converted, the comprehensive evaluation value is calculated, and the top 5% of individuals are selected as preliminary excellent single plants.

[0023] C. Cross-validation: Individuals selected by both methods are considered to be excellent single plants of the Red Cone variety.

[0024] The reference population mentioned in step (1) is the Red Cone clone, which refers to the superior Red Cone varieties selected through asexual reproduction methods such as grafting; preferably, it consists of 226 16-year-old Red Cone clones.

[0025] The candidate group mentioned in step (1) is a half-sib family, which refers to a group of offspring that share only the same father or the same mother; preferably, it is 479 offspring individuals derived from 23 half-sib families.

[0026] The method for determining the material properties described in step (2) is as follows: a wood core penetrating the pith is drilled at a height of 1.2 m on the trunk using a growth cone with an inner diameter of 5.15 mm, and various material property indicators are tested.

[0027] The preferred number of plants mentioned in step (2) is 3 to 5 plants.

[0028] The number of blades in step (2) is preferably 5 to 10.

[0029] The preferred storage conditions described in step (2) are liquid nitrogen freezing followed by storage at -80°C.

[0030] The genomic DNA selection process described in step (3) also includes testing the following indicators: concentration, integrity, and purity.

[0031] The high-throughput sequencing described in step (3) uses the Illumina HiSeq PE150 platform.

[0032] The sequence quality control described in step (3) is preferably implemented using Fastp, and the standard is: sequences with the removal of sequencing adapter sequences, sequences with an unknown base ratio of >10% at one end, and sequences with an low-quality base ratio (Q≤5) of >50% at one end.

[0033] The alignment described in step (3) is preferably performed using BWA-MEM to align the red cone reference genome and using SAMtools to perform PCR repeats.

[0034] The mutation detection described in step (3) is preferably performed using mpileup to detect SNP mutations.

[0035] The heritability classification criteria described in step (4) are as follows: H and DBH are highly heritable, with values ​​of 0.479 and 0.517, respectively; WD, FL, and LWR are moderately heritable, with values ​​of 0.233, 0.349, and 0.328, respectively; LIG and CE are low heritable, with values ​​of 0.208 and 0.036, respectively; and MFA is extremely low heritable, with a value of 1.570 × 10⁻⁶. -6 .

[0036] The basic formula of the GBLUP model described in step (5) is as follows: ;in For phenotypic matrix, , To design the matrix, For fixed effects, Due to genetic effects, This is due to the residual effect.

[0037] The predictive ability described in step (5)A is examined by setting up a breeding scenario.

[0038] The preferred breeding scenarios are internal growth, growth + physical properties, growth + fiber, growth + chemical properties, and pure properties.

[0039] The standard matrix mentioned in step (5)A is preferably a GOF matrix.

[0040] The absolute value of the genetic correlation coefficient mentioned in step (5)B is preferably determined by the restricted maximum likelihood method (REML).

[0041] The saturated response mode mentioned in step (5)B refers to the prediction plateau that can be reached with the help of a single trait, such as H, DBH, LIG, CE;

[0042] The cumulative gain mode mentioned in step (5)B refers to the need to superimpose multiple auxiliary traits to improve predictive ability, such as FL and LWR;

[0043] The weak response pattern mentioned in step (5)B refers to auxiliary traits that have no significant effect on improving predictive ability, such as WD and MFA.

[0044] The gradient selection in step (5)C is as follows: growth traits H and DBH are labeled with 79K SNP, fiber morphology traits FL and LWR are labeled with 16K SNP, wood chemical traits LIG and CE are labeled with 3K SNP; WD and MFA use the corresponding functional class label density.

[0045] The number of tests described in step (5)D is preferably 100 to 125.

[0046] The formula for the equal-weighted selection index mentioned in step (7)A is:

[0047] SI i =0.125 (Z Hi +Z DBHi +Z WDi +Z FLi +Z LWRi +Z CEi -Z MFAi -Z LIGi );

[0048] Among them, SI i For individuals, an index value, Z, is selected. Hi Z DBHi Z WDi Z FLi Z LWRi Z CEi Z MFAi Z LIGi The standardized GEBV has an absolute weight of 0.125 for each trait.

[0049] The formula for the Brekin multi-trait assessment method described in step (7)B is:

[0050] ;

[0051] in: For individuals i Overall evaluation value j Characteristics j , n For the quantity of traits, For individuals i Properties j GEBV value, Characteristics j The maximum value of GEBV.

[0052] The above methods are applied in the breeding of superior varieties of Conophytum rubrum, the cultivation of asexual lines, the selection and mating of hybrid parents, and the high-generation breeding of Conophytum rubrum.

[0053] Application of the above methods in genome selection breeding of broadleaf tree species

[0054] The broad-leaved tree species are preferably at least one of the following: species from the genera *Castanopsis*, *Lithocarpus*, and *Castanopsis* in the family Fagaceae.

[0055] The present invention has the following advantages and effects compared with the prior art:

[0056] (1) Breeding efficiency is greatly improved: This invention shortens the generation interval of red cone breeding from the traditional 20 years to 3 years. The annual genetic gain of tree height and diameter at breast height is 5.3 and 4.5 times that of traditional phenotypic selection, and 10.6 and 2.8 times that of univariate genomic selection, which greatly solves the problem of long breeding cycle in traditional breeding.

[0057] (2) The prediction accuracy is significantly improved: By adopting the multivariate GBLUP model and making full use of the genetic covariance between traits, the prediction accuracy of tree height reaches 0.55 and diameter at breast height reaches 0.44, which is far better than the univariate model (H=0.15, DBH=0.05); for some low heritability traits, the prediction effect is greatly improved.

[0058] (3) Scientific and economical parameter configuration: SNP marker density is configured according to the differences in trait genetic characteristics, eliminating the need for full-scale marker sequencing and significantly reducing the cost of genotyping; the combination of auxiliary traits is optimized to reduce the workload of phenotyping and balance accuracy with industrialization costs.

[0059] (4) High reliability of screening results: Two classic multi-trait evaluation methods were used for cross-validation. The four excellent single plants selected had stable comprehensive traits and no obvious shortcomings. At the same time, they were verified by previous single-trait models, so the breeding risk was extremely low.

[0060] (5) High versatility: The entire technical system is not only applicable to Castanopsis chinensis, but can also be extended to other precious broad-leaved tree species in the Fagaceae family such as Castanopsis, Lithocarpus, and Castanopsis, filling the technical gap in multivariate genome selection for Fagaceae tree species. Attached Figure Description

[0061] Figure 1 This is a comparison chart of the predictive capabilities of various traits of the Red Cone based on a stratified label quantity strategy. Detailed Implementation

[0062] The present invention will be further described in detail below with reference to specific embodiments. These embodiments are only used to explain the present invention and are not intended to limit the scope of protection of the present invention.

[0063] Example 1: Collaborative Breeding of Superior Individuals of the Red Cone Based on a Multivariate GBLUP Model

[0064] (1) Experimental group formation

[0065] Reference population: 226 16-year-old clones from the Red Cone gene bank of Longyandong Forest Farm in Guangdong Province, managed by South China Agricultural University, were selected. The germplasm covers 17 sampling points including Pubei and Bobai in Guangxi Zhuang Autonomous Region, Xinyi and Luhe in Guangdong Province, Jinshan in Fujian Province, Ledong in Hainan Province, and Jinghong in Yunnan Province. The genetic diversity is rich, and the clones were used as the model training population.

[0066] Candidate population: Healthy seeds from the reference population were collected in 2020, disinfected and sown. After the seedlings were cultivated to a height of 75-80cm, they were planted in Yangjiang State-owned Forest Farm, Foshan Yunyong Forest Farm and Yunfu Forest Farm respectively. Using a randomized block design, 23 half-sib families and 479 offspring were finally selected as candidate populations.

[0067] (2) Phenotypic determination and leaf sampling

[0068] Phenotypic determination: Phenotypic determination of the reference population was completed in 2018. Three samples were randomly selected from each clone in the reference population, and the tree height (H) and diameter at breast height (DBH) were measured. The average value was taken as the phenotypic value of the clone.

[0069] Wood property determination: A growth cone with an inner diameter of 5.15 mm was used to drill through the pith core at a depth of 1.2 m in the trunk, and six wood property indicators were determined: basic density (WD), fiber length (FL), fiber length-to-width ratio (LWR), microfiber angle (MFA), lignin content (LIG), and cellulose content (CE).

[0070] Leaf sampling: Leaves from the current year's shoots free of diseases and pests were collected from all individual plants. 5 to 10 leaves were collected from each sample. Three biological replicates were performed. The leaves were wrapped in aluminum foil, flash-frozen in liquid nitrogen, and stored at -80°C for genomic DNA extraction.

[0071] (3) DNA extraction and genotyping

[0072] DNA extraction and quality control: Leaf DNA was extracted using the Tiangen Plant Genomic DNA Kit. Concentration was detected by Qubit, integrity was detected by 1% agarose gel (150V, 20min), purity was detected by Nanodrop, and qualified DNA samples were screened.

[0073] High-throughput sequencing: Samples were fragmented to 300–350 bp, libraries were constructed, and high-throughput sequencing was performed using the Illumina HiSeq PE150 platform.

[0074] Data quality control and SNP screening: Fastp was used to filter adapters and low-quality sequences. The sequence filtering criteria were: removing sequencing adapter sequences, sequences with more than 10% unknown bases at one end, and sequences with more than 50% low-quality bases (Q≤5) at one end; BWA-MEM was used to align with the Red Cone reference genome; SAMtools was used to remove PCR duplicates; mpileup was used to detect SNP variations; VCFTools was used to screen SNPs according to the following criteria: sequencing depth ≥4, population deletion rate ≤20%, minor allele frequency ≥0.01, Hardy-Weinberg equilibrium P ≥0.0001; Beagle 5.0 was used to fill in missing genotypes, resulting in 790,877 high-quality SNP markers.

[0075] (4) Estimation of genetic parameters

[0076] Based on four genome relation matrices (GOF, GD, G05, GMF), primarily the GOF matrix, the REML method was used to estimate the broad-sense heritability, genetic associations, environmental associations, and phenotypic associations of eight traits; it was clarified that:

[0077] ①Heritability tier: H (0.517), DBH (0.479) (high); WD (0.233), FL (0.349), LWR (0.328) (medium); LIG (0.208), CE (0.036) (low); MFA (1.570×10⁻⁶) (high). -6 (Extremely low);

[0078] ② Trait association: H and DBH (rg=0.665) and FL and LWR (rg=0.799) showed extremely significant positive correlations (rg>0.66), and LIG and CE showed extremely significant negative correlations (rg=-0.906), providing a basis for auxiliary trait combinations.

[0079] (5) Parameter optimization of multivariate GBLUP model

[0080] Basic formula of the model: ;in For phenotypic matrix, , To design the matrix, For fixed effects, Due to genetic effects, This is due to the residual effect.

[0081] ① Optimized Genomic Relationship Matrix: Five breeding scenarios were set up (internal growth, growth + physical properties, growth + fiber, growth + chemical properties, and pure properties). The predictive capabilities of four matrices were compared. The GOF matrix was determined as the standard matrix due to its transparent construction, strong stability, and high versatility. Among them, internal growth traits include H and DBH, physical properties include WD, fiber morphology traits include FL, LWR, and MFA, and chemical properties include LIG and CE.

[0082] ② Optimization of auxiliary trait combinations: Bivariate and multivariate GBLUP models are constructed with a single trait as the target trait and the remaining traits as auxiliary traits. Trait combinations with an absolute value of genetic correlation coefficient > 0.6 are selected as the optimal auxiliary combinations. Three types of trait patterns are distinguished: saturated response, cumulative gain, and weak response. Practical optimal combinations that balance prediction accuracy and measurement cost are determined.

[0083] ③ Differentiated configuration of SNP marker density: Based on step ②, 10 marker gradients, namely 1K, 3K, 8K, 16K, 40K, and 79K, are further set up, and the saturation marker amount is determined by combining 5-fold cross-validation: 79K SNPs are used for growth traits (H, DBH), 16K SNPs are used for fiber morphology traits (FL, FW, LWR), and 3K SNPs are used for wood chemical traits (LIG, CE).

[0084] ④ Model Optimization: Based on steps ② and ③, repeat the test 100–125 times, calculate the predictive power and accuracy, and the results are as follows: Figure 1 As shown, the final optimization of the model was completed, namely the multivariate GBLUP optimized model, which covers the optimal auxiliary trait pattern and the saturation number of hierarchical SNP markers. Based on the above optimized multivariate model, the prediction accuracy of tree height (H) reached 0.55, diameter at breast height (DBH) reached 0.44, FL reached 0.57, FW reached 0.56, LWR reached 0.82, LIG ​​reached 0.86, and CE reached 0.96.

[0085] (6) Calculation of the estimated breeding value (GEBV) of the candidate population genome

[0086] Based on the multivariate GBLUP optimization model constructed from the reference population, the genotyping data of the candidate population and the reference population were combined, the model was rerun, and the GEBV of 8 core breeding traits of 479 individuals in the candidate population was calculated in batches.

[0087] (7) Multi-objective joint screening of superior single plants

[0088] ① Equal-weighted selection index method: Eight core traits, including H, DBH, WD, FL, LWR, MFA, LIG, and CE, were selected. GEBV was standardized, and an equal-weighted selection index was constructed (each trait weight was 0.125). The top 5% (24 plants) of preliminarily superior single plants were selected in descending order of the index, and the core families were F4 and F6. This method allows for complementary trait advantages and is suitable for comprehensive breeding of superior varieties.

[0089] The formula for calculating the weighted selection index is as follows:

[0090] SI i =0.125 (Z Hi +Z DBHi +Z WDi +Z FLi +Z LWRi +Z CEi -Z MFAi -Z LIGi );

[0091] Among them, SI i For individuals, an index value, Z, is selected. Hi Z DBHi Z WDi Z FLi Z LWRi Z CEi Z MFAi Z LIGi The standardized GEBV has an absolute weight of 0.125 for each trait.

[0092] ②Brekin multi-trait evaluation method: The GEBV of negative traits MFA and LIG is negative to achieve directional unification. The comprehensive evaluation value is calculated, and the top 5% (24 plants) of preliminary excellent single plants are selected in descending order. The core families are F24 and F10. This method avoids the extreme advantage of a single trait and is suitable for breeding of trait-balanced varieties.

[0093] The formula for calculating the Brekin multi-trait assessment is as follows:

[0094] ;

[0095] in: For individuals i Overall evaluation value j Characteristics j , n For the quantity of traits, For individuals i Properties j GEBV value, Characteristics j The maximum value of GEBV.

[0096] ③ Cross-validation: Individuals selected by both methods are the final comprehensive excellent single plants of the Red Cone in this invention (numbers: 4016, 4042, 4044, 4236).

[0097] Comparative Example 1: Collaborative Breeding of Superior Individuals of the Red Cone Based on a Univariate GBLUP Model and Multi-Objective Selection

[0098] (1) Same as Example 1.

[0099] (2) Same as Example 1.

[0100] (3) Same as Example 1.

[0101] (4) Same as Example 1.

[0102] (5) Parameter optimization of univariate GBLUP model

[0103] The univariate GBLUP model predicts only a single trait. Based on the optimized univariate model described above, the prediction accuracy is 0.15 for H, 0.05 for DBH, 0.05 for FL, 0.06 for FW, 0.03 for LWR, 0.16 for LIG, and 0.15 for CE.

[0104] Matrix selection: Comparing 4 types of matrices in 5 scenarios, the GOF matrix has stable predictive ability and is determined to be the standard matrix.

[0105] Label density configuration: H and DBH use 2K SNPs, FL, LWR, LIG ​​and CE use 3K SNPs.

[0106] 5-fold cross-validation: Each labeled gradient is repeatedly sampled and cross-validated to determine the final prediction accuracy of the model.

[0107] (6) Calculation of GEBV for candidate population

[0108] The univariate GBLUP optimization model, constructed based on the reference population, combines the genotyping data of the candidate population and the reference population, re-runs the model, and then calculates the GEBV of 8 core breeding traits for 479 individuals in batches.

[0109] (7) Screening of superior individual plants

[0110] Fifteen superior progeny individuals were selected from the candidate population of Conophytum purpureus using the Brekin multi-trait comprehensive evaluation method for each trait of the GEBV. Compared with the reference population, their genetic gain ranged from 0.319% to 2.671%.

[0111] Example 2: Efficiency Comparison of Multivariate GS and Univariate GS

[0112] Using H and DBH as representative core growth traits, the expected annual genetic gain of four breeding strategies—traditional phenotypic direct selection, traditional indirect selection, single-trait genome direct selection, and multi-trait genome direct selection—was analyzed. ) to conduct quantitative comparisons.

[0113] Expected annual genetic gain ( The calculation formula is as follows:

[0114] ;

[0115] In the formula, To select intensity (this study uniformly uses a 10% selection rate, corresponding to...) ); To ensure accuracy, the correlation coefficient between the selection indicators used and the individual true breeding value (TBV) is used. The genetic standard deviation of the target trait ( ); The generation interval (in years) is the number of years required to complete a full breeding cycle.

[0116] ① Traditional phenotypic selection

[0117] Strategy description: Direct selection based on measured phenotypes during the physiological maturation period of the target trait.

[0118] Select accuracy ( ): , That is, the square root of the generalized heritability based on pedigree.

[0119] Genetic standard deviation ( ): ,in The phenotypic standard deviation of the target trait.

[0120] generational gap ( Based on the stable age of the properties of Conifer argentis, the following is set: Year.

[0121] ② Univariate genomic selection

[0122] Strategy Description: During the seedling stage, the target trait's GEBV is predicted using a univariate GBLUP model based solely on its genome-wide marker information, and direct selection is performed accordingly.

[0123] Select accuracy ( ): To verify the genomic breeding value of individuals in the population ( ) and its true phenotypic value ( y The Pearson correlation coefficient between them It is the square root of the generalized heritability.

[0124] Genetic standard deviation ( ): ,in The phenotypic standard deviation of the target trait.

[0125] generational gap ( ): Year 1: Field planting of breeding materials, determination and sampling of target phenotypic traits; Year 2: SNP genotyping, genomic selection model training and estimation of whole-population genomic breeding values; Year 3: Selection of superior individuals, parental configuration and controlled pollination, to complete the construction of a new generation of breeding, significantly improving the efficiency of forest tree breeding.

[0126] ③ Multivariate genomic selection

[0127] Strategy Description: During the seedling stage, the multivariate GBLUP model is used to simultaneously integrate information on the target trait and the auxiliary traits for optimization, jointly predict the GEBV of the target trait, and make direct selection based on this prediction.

[0128] Select accuracy ( The prediction accuracy of the multivariate GBLUP model optimized by this invention is calculated using the same method as the univariate GS model.

[0129] Genetic standard deviation ( ): ,in The phenotypic standard deviation of the target trait.

[0130] generational gap ( ): Same as the single-variable GS setting, Year.

[0131] ④ Efficiency Comparison

[0132] Based on traditional phenotypic direct selection, the relative efficiency ratio (ER) of each strategy is calculated:

[0133] .

[0134] Based on the above calculations, the following results can be obtained:

[0135] Table 1. Comparison of expected annual genetic gains of growth traits in Conophytum purpureus under different breeding strategies.

[0136]

[0137] Note: Genetic parameters of traditional selection strategies ( , Estimate based on matrix A, where H is DBH Genetic related The accuracy of traditional indirect selection is based on... Calculation. The generalized heritability of the GS strategy is estimated based on the GOF matrix, and H's ( ), DBH Generation interval: 20 years for traditional strategy, 3 years for GS strategy. Selection strength is uniformly set to... .

[0138] As shown in the table above, based on the multivariate GBGLUP, the generation interval of Conophytum purpureus breeding is shortened from the traditional 20 years to 3 years. The annual genetic gain of tree height and diameter at breast height is 5.3 and 4.5 times that of traditional phenotypic selection, and 10.6 and 2.8 times that of univariate genomic selection, which greatly solves the problem of long breeding cycles in traditional breeding.

[0139] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. A method for precisely selecting superior individuals of the Red Cone genotype based on a multivariate genomic selection model, characterized in that... The specific steps include the following: (1) Construction of experimental populations: A reference population and a candidate population of Conophytum purpureus were constructed respectively; the reference population was a clone of Conophytum purpureus and the candidate population was a half-sib family; (2) Phenotypic determination and sample collection: Growth traits and wood properties were measured for the reference population to obtain phenotypic data. Among them, growth traits were tree height (H) and diameter at breast height (DBH), and wood properties were wood density (WD), fiber length (FL), fiber length-to-width ratio (LWR), microfibril angle (MFA), lignin content (LIG), and cellulose content (CE). Several sample plants were randomly selected from each Conophytum purpureus clone, and the average value was taken as the phenotypic data of the clone. Healthy young shoots and leaves from the current year of the reference population and candidate population were collected, preserved, and used for DNA extraction. (3) Genotyping and SNP screening: Genomic DNA was extracted from leaves and subjected to high-throughput sequencing, sequence quality control, alignment, variation detection and filtering to obtain high-quality SNP markers; the filtering criteria were: site sequencing depth ≥4, population deletion rate ≤20%, minor allele frequency ≥0.01, Hardy-Weinberg equilibrium P ≥0.0001; genotyping was completed using Beagle 5.0; (4) Genetic parameter analysis: Based on the GOF genome relation matrix, the restricted maximum likelihood method was used to estimate the broad heritability, genetic correlation, environmental correlation and phenotypic correlation of the eight traits, and to clarify the genetic hierarchy of traits and the synergistic / antagonistic relationship between traits; (5) Construction and parameter optimization of multivariate GBLUP model: A. Determine the standard matrix for genomic relationships: Compare the predictive capabilities of four matrices—GOF, GD, G05, and GMF—to determine the standard matrix; B. Optimize auxiliary trait combinations: Construct bivariate and multivariate GBLUP models with a single trait as the target trait and the remaining traits as auxiliary traits. Select trait combinations with an absolute value of genetic correlation coefficient > 0.6 as the optimal auxiliary combinations, and distinguish between three types of trait patterns: saturated response mode, cumulative gain mode, and weak response mode, to determine the practical optimal combination that balances prediction accuracy and measurement cost. C. Differentiated configuration of SNP label quantity: Based on step B, label gradients are divided according to the trait prediction saturation features, and the saturation label quantity is determined by combining 5-fold cross-validation; D. Model Optimization: Based on B and C above, a hierarchical 5-fold cross-validation strategy is adopted to test and evaluate the model's predictive ability and accuracy, and to complete the overall optimization of the multivariate GBLUP model. (6) Calculation of candidate population genomic breeding value Using the optimized multivariate GBLUP model, the genomic estimated breeding value (GEBV) for all target traits in the candidate population was calculated. (7) Joint screening of superior single plants A. Equal-weighted selection index method: The GEBV of the 8 core traits is standardized, an equal-weighted selection index is constructed, and the top 5% of individuals are selected as preliminary superior single plants. B. Brekin Multi-Track Evaluation Method: The GEBV of negative traits MFA and LIG is converted, the comprehensive evaluation value is calculated, and the top 5% of individuals are selected as preliminary excellent single plants. C. Cross-validation: Individuals selected by both methods are considered to be excellent single plants of the Red Cone variety.

2. The method for precisely selecting superior individuals of the Red Cone genotype based on a multivariate genomic selection model according to claim 1, characterized in that: The reference population mentioned in step (1) consists of 226 16-year-old clonal red cone plants; The candidate population mentioned in step (1) consists of 479 offspring individuals derived from 23 half-sib families.

3. The method for precisely selecting superior individuals of the Red Cone genotype based on a multivariate genomic selection model according to claim 1, characterized in that: The method for determining the material properties described in step (2) is as follows: a wood core penetrating the pith is drilled at a height of 1.2 m on the trunk using a growth cone with an inner diameter of 5.15 mm, and various material property indicators are tested. The number of plants mentioned in step (2) is 3 to 5; The number of leaves mentioned in step (2) is 5 to 10; The preservation conditions described in step (2) are: frozen in liquid nitrogen and then stored at -80°C; The genomic DNA described in step (3) is also tested for the following indicators: concentration, integrity, and purity; The high-throughput sequencing described in step (3) uses the Illumina HiSeq PE150 platform.

4. The method for precisely selecting superior individuals of the Red Cone genotype based on a multivariate genomic selection model according to claim 1, characterized in that: The sequence quality control mentioned in step (3) is implemented using Fastp, and the standard is: removing sequencing adapter sequences, sequences with an unknown base ratio of >10% at one end, and sequences with a low-quality base ratio of Q≤5 at one end of >50%; The alignment described in step (3) involves using BWA-MEM to align the red cone reference genome and using SAMtools to remove PCR repeats; The mutation detection described in step (3) is to use mpileup to detect SNP mutations.

5. The method for precisely selecting superior individuals of the Red Cone genotype based on a multivariate genomic selection model according to claim 1, characterized in that: The heritability classification criteria described in step (4) are as follows: H and DBH are highly heritable, with values ​​of 0.479 and 0.517, respectively; WD, FL, and LWR are moderately heritable, with values ​​of 0.233, 0.349, and 0.328, respectively; LIG and CE are low heritable, with values ​​of 0.208 and 0.036, respectively; and MFA is extremely low heritable, with a value of 1.570 × 10⁻⁶. -6 .

6. The method for precisely selecting superior individuals of the Red Cone genotype based on a multivariate genomic selection model according to claim 1, characterized in that: The basic formula of the GBLUP model described in step (5) is as follows: ;in For phenotypic matrix, , To design the matrix, For fixed effects, Due to genetic effects, This is due to residual effects; The standard matrix mentioned in step (5)A is the GOF matrix; The absolute value of the genetic correlation coefficient mentioned in step (5)B is determined using the restricted maximum likelihood method; The saturated response mode mentioned in step (5)B refers to the prediction plateau that can be reached with the help of a single trait; The cumulative gain mode mentioned in step (5)B refers to the need to superimpose multiple auxiliary traits to improve predictive ability; The weak response pattern mentioned in step (5)B refers to the auxiliary trait having no significant effect on improving predictive ability; The gradients described in step (5) C are as follows: growth traits H and DBH are labeled with 79K SNPs, fiber morphology traits FL and LWR are labeled with 16K SNPs, and wood chemical traits LIG and CE are labeled with 3K SNPs; WD and MFA use the corresponding functional class labeling density. The number of tests described in step (5)D is 100 to 125.

7. The method for precisely selecting superior individuals of the Red Cone genotype based on a multivariate genomic selection model according to claim 1, characterized in that: The formula for the equal-weighted selection index mentioned in step (7)A is: SI i =0.125 (From Hi +Z DBHi +Z WDi +Z FLi +Z LWRi +Z CEi -Z MFAi -Z LIGi ); Among them, SI i For individuals, an index value, Z, is selected. Hi Z DBHi Z WDi Z FLi Z LWRi Z CEi Z MFAi Z LIGi The standardized GEBV has an absolute weight of 0.125 for each trait. The formula for the Brekin multi-trait assessment method described in step (7)B is: ； in: For individuals i Overall evaluation value j Characteristics j , n For the quantity of traits, For individuals i Properties j GEBV value, Characteristics j The maximum value of GEBV.

8. The application of the method described in claims 1 to 7 in the breeding of superior varieties of Conophytum rubrum, the cultivation of asexual lines, the selection and mating of hybrid parents, and the high-generation breeding of Conophytum rubrum.

9. The application of the method described in claims 1 to 7 in genomic selection breeding of broadleaf tree species.

10. The application according to claim 9, characterized in that: The broad-leaved tree species mentioned are at least one of the species of the genus Castanopsis, the genus Lithocarpus, and the genus Castanopsis in the family Fagaceae.

Images