Protocols for In Vivo Doubled Haploid (DH) Technology in Maize Breeding: From Haploid Inducer Development to Haploid Genome Doubling

Fine mapping of major QTL qshgd1 for spontaneous haploid genome doubling in maize (Zea mays L.)

KEY MESSAGE

A large-effect QTL was fine mapped, which revealed 79 gene models, with 10 promising candidate genes, along with a novel inversion. In commercial maize breeding, doubled haploid (DH) technology is arguably the most efficient resource for rapidly developing novel, completely homozygous lines. However, the DH strategy, using in vivo haploid induction, currently requires the use of mutagenic agents which can be not only hazardous, but laborious. This study focuses on an alternative approach to develop DH lines-spontaneous haploid genome duplication (SHGD) via naturally restored haploid male fertility (HMF). Inbred lines A427 and Wf9, the former with high HMF and the latter with low HMF, were selected to fine-map a large-effect QTL associated with SHGD-qshgd1. SHGD alleles were derived from A427, with novel haploid recombinant groups having varying levels of the A427 chromosomal region recovered. The chromosomal region of interest is composed of 45 megabases (Mb) of genetic information on chromosome 5. Significant differences between haploid recombinant groups for HMF were identified, signaling the possibility of mapping the QTL more closely. Due to suppression of recombination from the proximity of the centromere, and a newly discovered inversion region, the associated QTL was only confined to a 25 Mb region, within which only a single recombinant was observed among ca. 9,000 BC1 individuals. Nevertheless, 79 gene models were identified within this 25 Mb region. Additionally, 10 promising candidate genes, based on RNA-seq data, are described for future evaluation, while the narrowed down genome region is accessible for straightforward introgression into elite germplasm by BC methods.

Development of doubled haploid inducer lines facilitates selection of superior haploid inducers in maize

Haploid inducers are key components of doubled haploid (DH) technology in maize. Robust agronomic performance and better haploid induction ability of inducers are persistently sought through genetic improvement. We herein developed C1-I inducers enabling large-scale in vivo haploid induction of inducers and discovered superior inducers from the DH progenies. The haploid induction rate (HIR) of C1-I inducers ranged between 5.8% and 12.0%. Overall, the success rate of DH production was 13% on average across the 23 different inducer crosses. The anthesis-silking interval and days to flowering of inducer F1s are significantly correlated with the success rate of DH production (r = -0.48 and 0.47, respectively). Transgressive segregants in DH inducers (DHIs) were found for the traits (days to flowering, HIR, plant height, and total primary branch length). Moreover, the best HIR in DHIs exceeded 23%. Parental genome contributions to DHI progenies ranged between 0.40 and 0.55, respectively, in 25 and 75 percentage quantiles, and the mean and median were 0.48. The allele frequency of the four traits from inducer parents to DHI progenies did not correspond with the phenotypic difference between superior and inferior individuals in the DH populations by genome-wide Fst analysis. This study demonstrated that the recombinant DHIs can be accessed on a large scale and used as materials to facilitate the genetic improvement of maternal haploid inducers by in vivo DH technology.

Self-supervised maize kernel classification and segmentation for embryo identification

Introduction

Computer vision and deep learning (DL) techniques have succeeded in a wide range of diverse fields. Recently, these techniques have been successfully deployed in plant science applications to address food security, productivity, and environmental sustainability problems for a growing global population. However, training these DL models often necessitates the large-scale manual annotation of data which frequently becomes a tedious and time-and-resource- intensive process. Recent advances in self-supervised learning (SSL) methods have proven instrumental in overcoming these obstacles, using purely unlabeled datasets to pre-train DL models.

Methods

Here, we implement the popular self-supervised contrastive learning methods of NNCLR Nearest neighbor Contrastive Learning of visual Representations) and SimCLR (Simple framework for Contrastive Learning of visual Representations) for the classification of spatial orientation and segmentation of embryos of maize kernels. Maize kernels are imaged using a commercial high-throughput imaging system. This image data is often used in multiple downstream applications across both production and breeding applications, for instance, sorting for oil content based on segmenting and quantifying the scutellum's size and for classifying haploid and diploid kernels.

Results and discussion

We show that in both classification and segmentation problems, SSL techniques outperform their purely supervised transfer learning-based counterparts and are significantly more annotation efficient. Additionally, we show that a single SSL pre-trained model can be efficiently finetuned for both classification and segmentation, indicating good transferability across multiple downstream applications. Segmentation models with SSL-pretrained backbones produce DICE similarity coefficients of 0.81, higher than the 0.78 and 0.73 of those with ImageNet-pretrained and randomly initialized backbones, respectively. We observe that finetuning classification and segmentation models on as little as 1% annotation produces competitive results. These results show SSL provides a meaningful step forward in data efficiency with agricultural deep learning and computer vision.