Daucus accessions (N=1381) are maintained through the U.S. National Plant Germplasm System (NPGS) at the North Central Regional Plant Introduction Station (NCRPIS) in Ames, IA, with information on the accessions (also known as genotypes or plant introductions) in the Germplasm Resources Information Network (GRIN) database of the NPGS (GRIN-Global, 2023). Each carrot accession is a genetically unique, heterogeneous, heterozygous population. Accessions were selected from the GRIN system for our diversity panel if passport information suggested the presence of domestication traits (N=695). These cultivated carrots represent global carrot germplasm, collected over multiple plant exploration trips between 1947 and 2015 from 60 countries, with over 80% of accessions originating from the Eurasian supercontinent: 53% from Asia, 34% from Europe and the Caucasus, and 13% (in descending order) collected from the Americas, Africa, Australia, and New Zealand. This collection includes 148 total accessions from the primary center of diversity in central Asia (modern-day Afghanistan and surrounding countries) and secondary center of diversity in western Asia (modern-day Turkey) (Vavilov, 1951; Banga, 1957). This collection includes landraces and heirloom cultivars with annual, biennial, or mixed flowering habits. Although biennial flowering habit is a known domestication trait in carrot, reliable accession flowering habit data was not available prior to this study. GRIN-Global maintains flowering habit data for each accession (‘lifecycle’), but this data is not reliable due to being recorded in many different environments and on variable numbers of plants. As such, this data was not a criterion for identifying domesticated germplasm in this evaluation. This study is the first and largest (N = 695 accessions) multi-year field evaluation of flowering habit in a diverse carrot germplasm that includes landraces. The carrot accessions in this study are maintained by the United States Department of Agriculture National Plant Germplasm System (USDA-NPGS). All or parts of this global USDA germplasm collection have previously been evaluated in studies on canopy vigor (Loarca et al., 2024b), core collection curation (Corak et al., 2019), demographic history of carrot domestication and breeding (Coe et al., 2023), genetic structure, phyologeny, and carotenoid presence (Ellison et al., 2018), taproot shape (Brainard et al., 2021), plant growth traits (Acosta-Motos et al., 2021), antioxidant capacity (Pérez et al., 2023), resistance to the necrophytic fungal pathogen Alternaria dauci (Tas, 2016), and several studies on seed germination under abiotic stress (Bolton et al., 2019; Bolton and Simon 2019; Simon, 2019; Simon et al., 2021).

We used RStudio Version 2023.6.1.524 (Posit team, 2023) and R Version 4.3.1 (R Core Team, 2023) to perform all statistical analyses. Rosner’s Test in the EnvStats identified multiple simultaneous potential outliers for each trait in each year (Millard, 2013).Various utility packages were crucial to our analysis, such as ggthemes (Arnold, 2021), beepr (Bååth, 2018), flextable (Gohel and Skintzos, 2023), and the tidyverse suite of packages (Wickham et al., 2019).

F-tests of significance were performed to identify significant sources of variation for each trait in each year using fixed effects models in a two-way Analysis of Variance (ANOVA) with Type III sums of squares with the car package (Fox et al., 2012/). For each year, a fixed effects model was structured to calculate the proportion of variance in each trait (percentage of flowering plants in plot at 60 DAS or 100 DAS) attributable to genotype and block: T_ik = u + g_i + b_k + e_ik , where T_ik = phenotype measured on the trait of interest, u = intercept, g_i = genotype, b_k = block, and e_ik = error with e_ik ~ i.i.d. N(0, σ²).

The multi-year fixed effects model includes accessions with trait data across all years and accounts for variation across years: T i j k = u + g i + y j + ( g y ) i j +   b k ( j ) + e i j k , where T= phenotype of the trait of interest, g_i = genotype, y_j = year, (gy)_ij = genotype*year interaction, b_k(j) = block within year, and e_ijk = error with each component assumed to follow a respective, independent normal distribution [e_ijk~ i.i.d. N(0, σ²)]. Due to unbalanced data from abnormal weather events (destructive hail), we ran two multi-year analyses: one that included the 2017 flowering data and one that excluded the 2017 flowering data.

Variance components (V) for each trait were estimated using within-year (single-year) and across-years (multi-year) random effects models with the lme4 package (Bates et al., 2015). These used the same model as above with all effects random. Broad-sense heritability (H2) for each trait, within years (single-year model) and across years (multi-year model), was estimated from variance components, including genotypic variance (Vg) and phenotypic variance (Vp). Single-year broad-sense heritability (for each year 2016-2018) was calculated for each trait:

Multi-year broad-sense heritability was estimated for each trait:

As of 2023, GRIN-Global describes three categories for carrot flowering habit (or ‘life form’ in the GRIN system): annual, biennial, or a mixture (of annual and biennial plants), however, it was not clear what criteria or thresholds were used to categorize accessions in to one of the three flowering categories. In absence of this information, we created a theoretical construct (‘2023 GRIN-Global Flowering Habit’ in Figure 2 , right panels) with these three traditional flowering habit categories: annuals had 100% flowering at 100 DAS, biennials had 0% flowering at 100 DAS, and mixtures had between 1% and 99% flowering at 100 DAS.

Starting with these assumptions, we made minor threshold adjustments for the CarrotOmics Flowering Habit Ontology presented in this paper. Given that annual flowering habit is intolerable in temperate regions of carrot root production, we maintained the 0% flowering at 100 DAS threshold for the biennial category. We characterized annual-flowering plants as plots where the vast majority (85% - 100%) of plants were flowering at 100 DAS – despite this slight allowance for non-flowering plants, these are not considered mixed populations. Mixtures had between 1% and 85% of plants flowering at 100 DAS ( Figure 2 , right panels).

Using this three-category ontology as a baseline, we then subdivide it into different types of annuals and mixtures. We added an additional time-point (60 DAS) for evaluating flowering habit to capture precocity and uniformity of flowering among annuals ( Figure 2 , left panels). We designated annual accessions with 85% - 100% at 60 DAS as uniformly-early annuals. Similarly, we designated uniformly-late annuals as those with uniform-flowering by the end of the season, with no or low flowering at 60 DAS (≤15%) and high amounts of flowering (85% - 100%) at 100 DAS. The remaining annual accessions flower non-uniformly and sporadically between 60 DAS and 100 DAS – these accessions may contain various proportions of uniformly-early and uniformly-late plants, all of which ultimately flower (85% - 100%) by end of season (100 DAS). As described above, mixtures have both biennial and annual plants (1% - 85% flowering at 60 DAS and 100 DAS). Recognizing the need for biennial germplasm for breeding and root production in temperate climates, we partitioned some low-flowering mixtures into a predominantly biennial subcategory, which we characterize as having between 1% and 15% flowering at 60 DAS and 100 DAS ( Figure 2 , left panels). This flowering trait ontology ( Table 1 and Figure 3 ) is stored in the CarrotOmics database (www.carrotomics.org/) (Rolling et al., 2022).

Grouped by flowering habit, we evaluated correlations between vegetative growth traits, including seed viability, seed weight, stand count, and canopy height. Pearson correlations were calculated and smoothed trend lines were visualized in a correlation matrix using GGally, along with boxplots and scatterplots.

The GRIN-Global three-category model characterized biennials (29.5%; n=197) and annuals (9.1%; n=61), with mixtures (61.4%; n=410) representing the vast majority of the collection ( Table 2A ). By 100 DAS, 70.5% of the collection expressed some proportion of flowering. These mixtures are not well-characterized except that accessions in this category contain both annuals and biennials in various proportions.

Summary statistics are presented using our proposed new ontology to characterize the collection at both the 60 DAS and 100 DAS data ( Table 2B ). Flowering at 60 DAS was the earliest time point at which we observed signs of flowering ( Figure 1 ) in 4%-10% of plots in two of our three studies. Flowering thresholds and subcategories did not change the number of biennials identified (n=197), but did increase the number of annuals (10% of accessions; n=61) with fewer than 2% being uniformly-early annuals (n=5) and uniformly-late annuals (n=6). In every year of our evaluation, uniformly-early annuals are a distinct group from uniformly-late annuals at 60 DAS, and both are distinct from the non-uniform flowering annuals (n=50) ( Figure 4 ). These remaining annuals flowered non-uniformly with 15% - 85% flowering at 60 DAS and 100 DAS.

While flowering is typically measured at 100 DAS, the addition of the 60 DAS time-point enabled differentiation between uniformly-early annuals and uniformly-late annuals, a distinction that would otherwise be lost by 100 DAS, as indicated in every year of our study ( Figure 5 ). The predominantly biennial subcategory accounts for 24% of accessions (n=160) in the germplasm collection. Flowering percentage among low-flowering mixtures is indiscernible from predominantly biennial populations at 60 DAS, but distinct at 100 DAS. After partitioning the predominantly biennial population, there are 250 mixed-flowering accessions, which have between 15% and 85% annual plants at 100 DAS.

ANOVA results indicate that genotype was a highly significant factor influencing flowering percentage at 60 DAS (Table 3A), with broad-sense heritability (H²) estimates also consistently very high in each year ( Table 2A : 0.87< H²< 0.93). The block effect was not significant in 2016 or 2018 significant at the p<0.05 level and in 2017. Similarly, F-tests of significance at 100 DAS flowering had a highly significant genotype effect and high broad-sense heritability ( Table 3B : 0.81< H²< 0.84). Genotype and genotype x year interaction are highly significant factors across all three years in the multi-year ANOVA ( Table 4A ). Multi-year broad-sense heritability (H²) is moderate at 60 DAS (0.57< H²< 0.64) and high at 100 DAS. (0.88< H²< 0.89). Excluding 2017 data did not substantially change heritability estimates or which factors were considered significant ( Table 4B ). P-values for flowering-percentage ANOVA results are available in Supplementary Tables 1 - 4 .

Using the strictest definitions for annual (100% flowering) and biennial (0% flowering), we explored vegetative trait correlations among the three traditional flowering habit categories ( Figure 6A ). Correlations among shoot-growth traits vary by flowering habit category. Correlation between seed viability and emergence was high for biennials (r = 0.68) and mixtures (r = 0.66) and very low and not significant for annuals (r = 0.26). Seed viability and seed weight had low negative correlation in biennials and mixtures, and was uncorrelated in annuals. Correlation between seed viability and late-season canopy height is moderate and positive for annuals (r = 0.56) and very low for biennials (r = 0.17) and mixtures (r = 0.16). Across all three flowering habits, emergence has low correlation with canopy height (80 DAS) and no correlation with seed weight. There also appeared to be a difference in mean and variation between biennial and annual accessions for canopy height. In the simple three-category model, mixtures tend to behave more similarly to biennials than to annuals, sharing very similar correlations among trait pairs, with one exception: the correlation between seed weight and canopy height was slightly higher for mixtures and annuals (r=0.41) than for biennials (r=0.25). There also appears to be a difference in mean and variation between biennial and annual accessions for canopy height.

We evaluated relationships among our trait ontology’s categories and subcategories ( Figure 6B ).

Correlations among shoot-growth traits vary by flowering habit category. As with Figure 6A , correlation between seed viability and emergence is high for biennials (r = 0.68), predominantly biennials (r = 0.67) and mixtures (r = 0.66) and low to non-existent for annual populations (r = 0.0 for late annuals, =0.35 for early annuals and 0.39 for non-uniform annuals). Correlation between emergence and late-season canopy height is very high (r = 0.84) for uniformly-early annuals, as is the relationship between seed viability and seed weight (r = 0.94), and low for all other flowering categories. Data presented are on estimated marginal means across years. Uniformly-early annuals have the highest mean seed weight, emergence, and canopy height. We observed that 37% of accessions in this collection expressed some proportion of flowering by 60 DAS and that this increased to 70.5% by 100 DAS, suggesting that the 60 DAS measurement is insufficient to predict whether plants will flower, as only half of accessions that will flower by 100 DAS are flowering at 60 DAS ( Figure 7 ). Many of the accessions that are flowering at 100 DAS may be late-flowering and could be of use to breeders, but would likely have been discarded using the prior classification system.

A fundamental purpose of crop ontology is to create a descriptive and consistent vocabulary for crop traits, facilitating communication across collaborators and comparison of data between trial years, locations, and breeding programs (Shrestha et al., 2012; Walls et al., 2012). We have provided improved descriptions for flowering habit characteristics in carrot, including standard methodologies and time-frames for trait evaluation. Previous studies in carrot have used “early”/”late” as modifiers to flowering habit in the first growing season and second growing season, respectively. The term “annual” has conventionally been used synonymously with “early-flowering”, while the term “biennial” has been used synonymously with “late-flowering” (Alessandro & Galmarini, 2007; Alessandro et al., 2013; Wohlfeiler et al., 2019; Simon & Grzebelus, 2020; Wohlfeiler et al., 2021). However, confusion arises when early-flowering has also been used to describe wild carrot plants that flower earlier in the first growth season than cultivated annuals, which tend to flower later in the same season (Simon & Grzebelus, 2020). Furthermore, recent research suggests that a gradient of vernalization requirements exists within annuals and biennials (Wohlfeiler et al., 2021) – we propose a standardized vocabulary to refer to this germplasm.

While flowering has been typically measured at 100 DAS, the 60 DAS measurement enabled differentiation between annuals that flower uniformly early in the season (uniformly-early annuals) from annuals that flower uniformly at the end of the season (uniformly-late flowering annuals) ( Figure 5 ). As posited by Wohlfeiler et al. (2019), these accessions could represent a range of vernalization requirements for flowering, with later flowering annuals needing more cold exposure and uniformly-early annuals needing far less, if any, cold exposure. Mixtures, which did not have a clear threshold defined previously, are entries with both biennial and annual plants in intermediate quantities, which we defined as between 1% and 85% flowering. We created a useful subcategory of mixture – predominantly biennial (n=160) – that constitutes a sizable 24% of accessions in this germplasm collection. Predominantly biennial accessions contain fewer than 15% flowering plants at the end of the season. Given the simple inheritance of biennial habit in carrot (Wohlfeiler et al., 2019), it is practical to select biennial plants and eliminate annual plants from a predominantly biennial population with other favorable traits for breeding. The establishment of the predominantly biennial subcategory nearly doubles the availability of germplasm with commercial potential, from 197 biennials to 357 biennials and predominantly biennials, accounting for 54% of the germplasm collection. This subcollection is a useful source of genetic diversity for breeders.

In this crop ontology, we propose categories that more fully describe the flowering habit of cultivated carrot, and we encourage carrot researchers to utilize and expand upon the descriptive terminology we provide in Table 1 and Figure 3 . In this ontology, “uniformly-early flowering” describes both precocity and uniformity of annual flowering, characterized by a high proportion of plants flowering at 60 DAS. “Late” describes only uniformity of end-of-season flowering, characterized by a low proportion of plants flowering at 60 DAS and a high proportion of plants flowering on harvest day or end-of-season. Following this logic, researchers can extend their evaluation of biennial carrot germplasm into the second season of growth. Collecting data on precocity of biennial flowering could identify uniformly-early biennials, which is important for carrot seed producers. This study provides a framework and an opportunity to study carrots in the second season of growth, and lays the groundwork for performing seed-to-seed phenotyping over the crop’s lifecycle.

Logically, the threshold of 85% for annual flowering means that these groups could also include mixtures. It would be more correct biologically to say that annuals are plants with 100% flowering at end of season, and any entries less than 100% flowering, and greater than 0% flowering, is a mixture. However, given that annuals are undesirable in temperate carrot production systems, we found little practical use in defining a “predominantly annual” population. It is more likely that non-flowering plants in an annual population at the end of the season are late-flowering annuals rather than biennials. Furthermore, breeding programs typically avoid intercrossing temperate (biennial) and subtropical (annual) carrots, and rarely use wild germplasm, given that substantial new challenges outstrip the benefits of such a cross (Simon & Grzebelus, 2020). The accessions identified, as well as the methodology provided for the identification of uniformly-late annuals, could prove useful for carrot breeders in subtropical climates or semi-arid climates, where uniformly-late annuals are essential for successful cultivar development; carrots adapted to subtropical regions often flower with reduced exposure to cold and tend to flower prolifically in temperate regions. In subtropical climates, carrots are managed as late-flowering annuals, with some plants used to produce the root crop and the remainder generate the seed stock in the same season. Biennial plants are unsuited to subtropical and semi-arid market, as they do not contribute to the seed crop.

The ability to distinguish between uniformly-early annuals and uniformly-late annuals is present at 60 DAS and lost by 100 DAS ( Figures 2 , 4 , 5 ), illustrating that the additional subcategories are distinct from one another in all three years evaluated. This demonstrates the need to evaluate flowering at two timepoints to capture the phenological variation in this population. The 60 DAS measurement alone is too early to identify the true biennials, but crucial when combined with the 100 DAS measurement for distinguishing uniformly-early annuals from uniformly-late annuals. The 100 DAS measurement alone is too late to differentiate the uniformly-early annuals from uniformly-late annuals, but critical to differentiate uniformly-late annuals from biennials, as well as identifying predominantly biennial populations. Our 100 DAS flowering evaluation allows us to further distinguish between accessions with uniformly-early annual habit (0.75% of accessions), uniformly-late annuals (0.9% of accessions), and the remaining annuals with non-uniform flowering (7.5% of accessions).

Identification of predominantly biennial accessions provides improved characterization of mixed population accessions and nearly doubles the availability of germplasm with commercial potential (n=357) for temperate areas of root production, compared with strictly biennial germplasm (n=197). Custom core collections could be curated from this data for carrot breeders and researchers interested in new sources of genetic diversity for specific traits, climatic conditions, production uses, or market types (Brown, 1989) by filtering on data such as plant morphology, ecogeographical origin, molecular marker data, and genetic relatedness (Berger et al., 2013; Byrne et al., 2018; Corak et al., 2019; Corak, 2021). In the biennial and predominantly biennial custom core collection, a minicore can be curated from our shoot-growth phenotypic data for traits such as plant height, canopy coverage, emergence, or available data for any trait of interest. Cores that maintain diversity while also maximizing desirable traits have great utility to breeders. The upper threshold for emergence is just as high in the biennial and predominantly biennial groups, which demonstrates the advantage of evaluating this custom core collection more closely for agronomically important traits. Agromorphological data has been leveraged to create core collections in sweet potato (Huamán et al., 1999), potato (Huamán et al., 2000), groundnut (Upadhyaya et al., 2003), pigeonpea (Reddy et al., 2005), maize (Malosetti & Abadie, 2001; Li et al., 2005; Risliawati et al., 2023), safflower (Dwivedi et al., 2005), yam (Girma et al., 2018), walnut (Mahmoodi et al., 2019), pomegranate (Razi et al., 2021), lentil (Tripathi et al., 2022), and Indian mustard (Nanjundan et al., 2022). Corak compared methods for creating custom core collections in a subset of 433 accessions from our study’s carrot diversity panel, and found that custom methods combined with representative methods built cores balanced for genetic representation and enriched for desirable phenotypes, though it is important to note that carrot has low population structure (Corak et al., 2019; Corak, 2021). Similarly, the mixtures and annuals we identified in this collection can be used as sources of genetic diversity by carrot breeders and researchers targeting carrots for subtropical/semi-arid climates.

There is a potential bias in every cultivated germplasm collection, as traits that are considered useful are relative to culture, production system, environment, technological access, local economy, and myriad unmeasurable factors. One bias inherent in the USDA cultivated carrot germplasm collection is preference for biennial germplasm, as that is what is grown in the U.S. Other gene banks may have higher diversity and larger samples for annual habits. We expect to see varying proportions of flowering in carrot germplasm, with the least in biennial cultivated carrot bred for cool temperate climates, and increasing amount of flowering in annual cultivated carrot bred for subtropical/semi-arid climates and wild carrot, indicating that the proportion of flowering in a population is relative to the germplasm under evaluation. Previous studies on U.S. commercial carrot cultivars, which have been selected for bolting resistance, had insignificant amounts of bolting, such that they were noted but not analyzed (Luby et al., 2016). In wild germplasm, most were bolting in early planting but all were non-bolting in later plantings (Solberg and Yndgaard, 2015), which could be due to cooler spring with sufficiently low temperatures to induce vernalization, and a warm, temperate summer resulting in no bolting. These varying results likely reflect the genotypic base of the carrot germplasm and the environment under study. Similarly, biennial genotypes in warm climates will never flower, while warm-acclimated annuals in heat-stressed environments may flower readily and prolifically (Simon et al., 2019). This study is limited to three years in one temperate environment. With this said, this environment is a commercially relevant region, with Wisconsin ranking in the top 3 U.S. states for carrot root crop production. Studies are underway to characterize this USDA germplasm collection in multiple other commercially relevant temperate carrot production environments. Accessions characterized in Wisconsin may not be stable in other growing regions. Climate warming could stimulate early-flowering or increased total flowering in germplasm we have already characterized.

Motivations for this flowering ontology were twofold: to attempt to understand the essential nature of carrot flowering phenology and to promote utilization of diverse germplasm by way of its characterization for agronomically critical traits. The former goal was satisfied by combining 60 DAS and end-of-season flowering data and identifying subtle but significant distinctions among annuals that open questions into carrot’s life history and domestication. The ontology provides trait definitions and methods for measuring flowering habit in diverse germplasm. Users of this ontology can set their own threshold based on what they see as tolerable for their own program. Future evaluations can be improved by overseeding accessions with low germination or low emergence to achieve the sample size required for accurate characterization. To better elucidate the relationship between trait stability in other temperate and economically relevant carrot production regions, genomic data and multi-environmental data will be integrated with this study’s phenotype data in future carrot diversity panel studies on QTL x E. Multi-environmental GWAS studies in temperate, subtropical, and semi-arid climates will facilitate molecular characterization of flowering habit in other Daucus germplasm collections.

We were also motivated by utility, which is the primary concern of breeders evaluating diverse germplasm. Identification of predominantly biennial germplasm fulfilled this goal, expanding the availability of genetic backgrounds that can be leveraged in temperate breeding programs. This evaluation has improved access to useful plant introductions in mixtures by identifying predominantly biennial accessions, doubling the size of the commercially promising accession gene pool for temperate carrot production regions. Given the relatively simple inheritance of biennial flowering habit, breeders for temperate root production can select biennial individuals out of predominantly biennial germplasm or other mixed flowering populations. However, selections should be evaluated for other agronomic traits and validated in multi-year, multi-environment trials in target locations. Data from canopy studies can be used in combination with flowering studies to identify accessions with high emergence and vigorous shoot growth for temperate climates. Similarly, the mixtures and annuals we identified can be leveraged as sources of genetic diversity by carrot breeders and researchers targeting carrots for subtropical and semi-arid climates. As such, flowering habit characterization has increased access to genetic resources with baseline commercial potential and provided useful data and methods to global users of carrot germplasm.

While within-accession diversity can be a challenge for ex situ conservation systems (Solberg and Yndgaard, 2015), we propose leveraging it as an opportunity to perform selection for desirable ecotypes, enabling identification of accessions with flowering traits required by local markets. We have provided a roadmap for evaluating and characterizing flowering habit in vegetable crops with mixed lifeforms, and this methodology is immediately useful to breeders and users of carrot PGR. Evaluating flowering habit as a gradient, rather than a binary trait, expands availability of commercially viable germplasm, further lowering the barrier to utilization of carrot PGR.