Our analysis of ripening-related gene expression in Cnr showed striking similarities to WT in the number and functions of genes changing between stages. Moreover, 69.5% of ripening related DEGs in Cnr were shared with WT . These results further support the hypothesis that Cnr is not exclusively a ripening mutant. Instead, Cnr fruit undergoes gene expression changes consistent with WT “ripening.” However, the ripening related changes in gene expression that occur in Cnr are not enough to compensate for the large defects accumulated in the fruit during growth and maturation. In a recent report, a knockout mutation to the gene body of CNR yielded little visible effects on fruit development and ripening , which suggests that the Cnr mutant phenotype may result from more than just a reduced expression of the CNR gene as previously reported . It has also been demonstrated that Cnr fruit have genome-wide methylation changes that inhibit ripening-related gene expression . The developmental defects observed in Cnr are likely caused by these methylation changes, directly or indirectly caused by the Cnr mutation . Thus, to better understand the Cnr mutation, more physiological data at earlier stages of development needs to be analyzed and complemented with more in-depth functional analysis of gene expression alterations at the corresponding stages. In addition, further molecular and genetic studies need to be performed and compared against complete CNR knockout mutants. Our data support that the mutants never produce a burst in ethylene production, even at the OR stage where more ripening phenotypes are observed . The orange-red pigmentation in nor OR fruit and the similarities of rin OR fruit in texture and taste-related attributes to WT RR fruit occur independently of an ethylene burst. These observations evidence that other regulatory mechanisms exist to initiate ripening events outside of ethylene .

Unlike previous reports, our data consistently showed that Cnr presented increased ethylene levels at the MG stage compared to WT . Interestingly, Cnr fruit produced more of the ethylene precursor ACC than WT at the RR stage. Also, rin made equivalent levels to WT fruit. Ethylene biosynthesis is divided into two programs: System 1 produces basal levels of the hormone during development, black plastic plant pots and System 2 generates the climacteric rise in ethylene during ripening . Each of these systems is catalyzed by a different set of ethylene biosynthetic enzymes . It is clear that all mutants show defects to System 2 of ethylene biosynthesis, but they also appear to have alterations specific to System 1. For example, we observed that SlACO3, a System 1- specific ACC oxidase, was higher expressed in Cnr fruit than WT .The role of ABA in climacteric ripening is not as well explored but has been reported to be complementary to ethylene . Previous reports in WT fruit have shown that ABA increases until the breaker stage, just before the ethylene burst . ABA has also been shown to induce ethylene production and linked to the NOR transcription factor . We found that nor and rin fruit did not show decreases in ABA concentration during ripening like WT did . For nor, the constant levels of ABA between MG and RR stages are another example of how fruit ripening events are delayed or inhibited. RIN and ABA have been demonstrated to have an inverse relationship where RIN expression is repressed with the induction of ABA . The significant increase of ABA accumulation in rin during ripening suggests that ABA biosynthesis and metabolism are misregulated in this mutant. rin fruit appear to present a delayed peak in ABA levels compared to WT fruit. Our results support the indirect interaction between the TFs and ABA during ripening. More developmental stages, genetic manipulations, and exogenous hormone treatments are needed to investigate further the trends of ABA accumulation seen in the ripening mutants.

The interactions between the CNR, NOR, and RIN in ripening have been debated in the literature . The TF RIN directly interacts with NOR and CNR, binding to their respective promoters, and therefore has been proposed to be the most upstream TF among the three regulators . Here we provided evidence that the three TFs display at least indirect effects on each other. We have argued that the Cnr mutant shows a wide breadth of defects across fruit development before ripening begins, and thus, we propose the Cnr mutation is acting before NOR or RIN. This further supports the hypothesis made in Wang et al. that Cnr acts epistatically to nor and rin. The gene expression patterns of CNR, NOR, and RIN across ripening stages were decreased or delayed in each of the single ripening mutants. The most substantial variation in gene expression was the downregulation of NOR and RIN expression across all stages in the Cnr mutant . We present for the first time double ripening mutants, homozygous for both loci, that can be used to see the combined effects of each mutation on fruit development and quality traits. We successfully generated the double mutants by establishing reliable and high throughput genotyping protocols for each mutation and evaluating segregation of the mutant phenotypes in field trials across multiple growing seasons. We obtained double mutants from both reciprocal crosses but saw no fruit phenotypic differences between them, suggesting that the ripening mutations are not influenced by maternal or paternal effects . Because the nor and rin mutants look so similar, it was hard to visually determine the individual effects of each mutation on the appearance of rin/nor fruit. However, when specific fruit traits were measured, we could detect additive or intermediate fruit phenotypes in this double mutant, supporting the proposed relationship in Wang et al. . Thus, nor and rin appear to influence similar fruit traits and act in coordination.

The Cnr mutation had a significant effect on the Cnr/nor and Cnr/rin mutants resulting in fruit with similar appearance and ethylene production to the Cnr fruit . When analyzing the gene expression profiles of the Cnr/nor fruit, we also observed multiple similarities to the Cnr parent, but also several deviations . Surprisingly, Cnr/nor was also reminiscent of nor, as it displayed few ripening-related gene expression changes, suggesting the inhibition or delay of specific ripening events in nor carried over to the double mutant. Here, we proposed that the Cnr mutation causes defects throughout fruit development while the nor mutation causes defects predominantly in ripening. However, the Cnr/nor double mutant showed additional phenotypic and transcriptional defects before ripening than both mutant parents . These observations indicate that in combination with Cnr, nor may contribute to alterations in early fruit development and the inhibition of ripening progression.Fruit breeders actively selected several morphological and quality phenotypes during the domestication of the garden strawberry , an allo-octoploid of hybrid origin. F. × ananassa was created in the early 1700s by interspecific hybridization between ecotypes of wild octoploid species , multiple subsequent introgressions of genetic diversity from F. virginiana and F. chiloensis subspecies in subsequent generations, and arti-ficial selection for horticulturally important traits among interspecific hybrid descendants. Domestication and breeding have altered the fruit morphology, development, and metabolome of the garden strawberry, distancing modern cultivars from their wild progenitors. Approximately 300 years of breeding in the admixed hybrid population has led to the emergence of high yielding cultivars with large, firm, visually appealing, long shelf life fruit that can withstand the rigors of harvest, handling, storage, and long-distance shipping. Fruit shape is an essential trait of agricultural products, particularly those of specialty crops, owing to perceived and realized relationships with the quality and value of the products. Image-based fruit phenotyping has the potential to increase scope, throughput, and accuracy in quantitative genetic studies by reducing the effects of user bias, enabling the analysis of larger sample sizes, and more accurate partitioning of genetic variance from environments, management, and other non-genetic sources of variation. Many fruit phenotyping approaches rely on the human eye to sort fruit into discrete, descriptive categories for planar shapes. Categories are either nominal, existing in name only, or ordinal, referring to a position in an ordered series or on a gradient. Classification into categories is often labor-intensive and prone to human bias, black plastic garden pots which can increase with task complexity and time requirements. Alternative scoring approaches rely on morphometrics and machine learning to automate classification; e.g., sorting fruit into shape categories in both tomato and strawberry. Unsupervised machine learning methods , unlike supervised methods, are useful for pattern detection and clustering, while supervised machine learning methods are useful for prediction and classification.

Unsupervised clustering enables the calculation of several measures of model performance and overfitting to balance compression and accuracy. However, the categories derived from these techniques are without order, resulting in the need for a suitable transformation to an ordinal scale more appropriate for quantitative genetic analyses. In this context, ordinal categories give the interpretation of relationship with, or distance from, other shape categories in a series. To enable this interpretation, we developed a method for asserting the progression through fruit shape categories derived from unsupervised machine learning methods. The Principal Progression of k Clusters allowed us to nonarbitrarily determine the appropriate shape gradient for statistical analyses using empirical data. The advantages of PPKC, relative to a manually determined ordinal scale, are that it does not require arbitrary, a priori decisions and is unsupervised, which obviates additional operator bias. Here, we describe approaches for translating digital images of strawberries into computationally defined phenotypic variables for identifying and classifying fruit shapes. Fruit shape and anatomy are complex, multi-dimensional, and, potentially, abstract phenotypes that are often not completely or intuitively described by planar descriptors and individual qualitative or quantitative variables. Beyond the qualitative definitions used in plant systematics, references to fruit shape encompass a wide variety of mathematical parameters and geometric indices that establish quantitative measurements of plant organs . Much like human faces or grain yield, fruit shape and anatomy are products of the underlying genetic and non-genetic determinants of phenotypic variability in a population. Quantitative phenotypic measurements have allowed researchers to uncover some of the genetic basis of fruit shape in tomato, pepper, pear, melon, potato, and strawberry. However, the major genetic determinants of fruit shape remain unclear, or understudied, in octoploid strawberry, in part because researchers have not yet translated fruit shape attributes into holistic, quantitative variables, which may empower the identification of underlying genes or quantitative trait loci through genome-wide association studies and other quantitative genetic approaches. Quantitative features often rely on linear metrics of distance and are generally modified into compound descriptors that remove the effects of size. However, compound linear descriptors often have limited resolution compared to more comprehensive, multivariate descriptors. Elliptical Fourier analysis quantifies fruit shape from a closed outline by converting a closed contour into a weighted sum of harmonic functions . Generalized Procrustes analysis quantifies the distance between sets of biologically homologous, or mathematically similar, landmarks on the surface of an object. Fruit shape can also be described using linear combinations of pixel intensities from digital images extrapolating from analyses generally used to quantify color patterns and facial recognition. Similar pixel-based descriptors have recently been referred to as ”latent space phenotypes” and arise from unsupervised analyses that allow a computer to produce novel, independently distributed features directly from images. Here, we generate a dictionary of 68 quantitative features, including linear-, outline-, landmark-, and pixel-based descriptors to investigate the quality of different features in preparation for quantitative genetic analyses. The ultimate goal of our study was to develop heritable phenotypic variables for describing fruit shape, which could then be used to identify the genetic factors underlying phenotypic differences in fruit shape. The phenotyping and analytic workflow for this study are summarized in Figs 1 and 2. We first describe and demonstrate the application of PPKC, which transforms categories discovered from unsupervised machine learning methods to a more convenient and analytically tractable ordinal scale. We then explore the relationship between machine acquired categories and 68 quantitative features extracted from digital images. Next, we apply random forest regression to select critical sets of quantitative features for classification and use supervised machine learning methods, including support vector regression and linear discriminant analysis , to determine the accuracy of shape classification.