Whole-vine resources modify within-vine relationships between growth parameters and metabolites in Vitis vinifera L. cv. Cabernet-Sauvignon

Fruit ripening on a perennial tree or vine is typically asynchronous and rarely investigated. In commerce, this is problematic as it impairs sampling to meet commercial maturity and quality standards at harvest. For grapevines, understanding within-grapevine variability in berry maturity will benefit precision management and harvest planning. It is accepted that variation in maturity is approximately equally allocated at the within-and between-vine scale. However, the mechanistic and ecological factors that cause functional variance are poorly understood and rarely documented. This study aimed to identify structural and spatial within-vine attributes associated with berry composition at harvest related to available resources from the whole grapevine. Vegetative and reproductive biomass attributes within-and between-vines were analysed for cv. Cabernet-Sauvignon trained as a cordon, spur-pruned system using a Smart-Dyson trellis in a ‘New World’ industrial-scale vineyard. The vines were located in northern Margaret River, Western Australia. Variability in soluble solids, anthocyanins and phenolics were analysed for clusters and corresponding canopy characteristics. Berry size was the best predictor of variance in maturity among sugar, anthocyanins and phenolics composition within individual vines. Smaller stems or fruit had greater variation in maturity than larger stems or fruit. However, the relationship between whole plant biomass and fruit maturity interacted at different scales; the first report for grapevines. Including different vine-biomass scales helped explain the heteroscedasticity observed when the individual vine effect was excluded from the analysis. These findings suggest that high vigour grapevines could benefit from differential management, regardless of harvest index. Furthermore, these findings may help explain the diversity of response to treatment effects such as cluster or leaf thinning reported previously for perennial fruit.


INTRODUCTION
Fruit ripening in grapevine is asynchronous within an individual vine and varies widely between vines within a management unit.Variation within the grapevine is commonly as large as that between grapevines (Carbonneau et al., 1991;Lefort et al., 1979;Pagay and Cheng, 2010;Rankine et al., 1962;Wisdom, 2018).Varieties with a propensity for high leaf biomass, such as Cabernet-Sauvignon, are more likely to exhibit internal variation in maturity (Wisdom et al., 2020).However, functional relationships between vegetative structures and fruit are rarely considered within-or between-vines.The allocation or partitioning of resources will likely differ depending on stored metabolites, yet this phenomenon is under-researched in grapevines (Zufferey et al., 2012).Modelling the ripening processes within a grapevine is challenging because many analogous activities co-occur (Dai et al., 2011).Thus, understanding ripening patterns within a grapevine should improve our understanding of grapevine physiology and ultimately grapevine management.
Partitioning of resources among organs within a grapevine reflects the biophysics of the underlying processes: synthesis and utilisation, and processes driving the flow from source to sink.Grapevine is likely similar to willow (Salix babilonica) and the vast majority of woody trees that rely on mass flow in phloem vessels that are connected symplastically from source mesophyll to sink in vegetative organs and which possess open minor veins in source leaves (Gamalei, 1989;Rennie and Turgeon, 2009;Turgeon and Medville, 1998).The ripening grape berry differs however, changing from symplastic unloading to apoplastic at véraison (Zhang et al., 2006).While this differentiation of process directs solutes preferentially to the berry, scale and spatial influences are likely to impact partitioning of solutes between organs within an individual vine, together with temporal influences likely to impact variation in the onset of véraison within and between clusters.
Assimilate accumulation is determined by three confounding characteristics of the vine (de Cortázar-Atauri et al., 2009).First, the capacity to provide assimilates, which -as with annual plants-depends on leaf size and number (Winkler, 1958), but is supplemented by the translocation from reserves especially during the period from bud burst to anthesis (Clingeleffer and Krake, 1992;Coombe, 1987;Hardie and Considine, 1976).Second, the efficiency or rate of metabolite production in leaves: orientation, light intensity, exposure, temperature and stomatal aperture (García de Cortázar-Atauri et al., 2005;Poni et al., 1994;Reshef et al., 2019).Third, competition between permanent storage organs and growth influences the assimilate pathway (Dai et al., 2010;Zhu et al., 2019).Specifically, sugar accumulation depends on available carbohydrates and the duration of fruit ripening, whereas anthocyanins (and therefore likely phenolics) are affected by climatic conditions and canopy structural effects that influence the light environment and drive metabolic processes (Dai et al., 2010;Rienth et al., 2021;Shahood et al., 2020).
Distal shoot dominance may be responsible for ripening trends related to distance from the crown in cane-pruned systems (Belvini et al., 1978;Carbonneau et al., 1991), explained by morphology management and its association with acrotonic growth.However, the direction of the trend is not always consistent; thus, inferences made from studies on cane-pruned vines should not be extrapolated to spur-pruned vines (Belvini et al., 1978;Carbonneau et al., 1991;Lefort et al., 1979;Wisdom et al., 2020).Spur-pruned systems exhibit reduced positional trends and greater synchronicity of shoot ontogeny than Guyot or head (cane)-pruned vines.However, there remains significant unexplained internal variation in maturity (Wisdom et al., 2020).
Much attention has been given to the individual vine concerning exposed leaf area and carbon allocation to fruit (Bobeica et al., 2015;Chorti et al., 2010;Kliewer and Antcliff, 1970;Šuklje et al., 2013), but little to individual shoot effects.It is asserted that leaves adjacent to the grape cluster are the primary source of assimilates (Hale and Weaver, 1962), particularly early in the growing season (Poni et al., 2004).The effect on cluster size indicates that individual shoot characteristics are a likely determinant of metabolite allocation to the attached cluster(s) (Caspari et al., 1998;Eltom et al., 2013;Joubert et al., 2016).Interesting observations have been made on fruiting cuttings; however, in the field, there is limited information on the comparative contributions of the components used to model ripening in the whole vine at the within-vine level (Morales et al., 2016).
Surprisingly few studies have compared the impact of in situ, intact, differential vine biomass from similarly managed vines on vine and fruit behaviour; that is, unadulterated in-field vines with differing vegetative biomass (colloquially vigour) due to external factors such as soil type.This may be due to smaller management units that have minimised whole-vine variability through experience.However, greater variability in biomass may be more common in industrial vineyards with larger plantings and management units that suit larger vineyard operations and mechanical equipment.In some regions of the United States of America, New Zealand and Australia, whole-vine vegetative biomass can considerably affect reproductive biomass and photosynthetic behaviour, resulting in variable sugar, acidity, colour, phenolics, yield and ripeness (Cortell et al., 2007;King et al., 2014;Song et al., 2014).Such measures have subsequently informed the development of canopy ideals or 'ideotypes' (Donald, 1962;Kliewer and Dokoozlian, 2005;Smart and Robinson, 1991).
In this study, we used statistical models of increasing complexity to assess fruit composition on individual stems to provide insight into biomass and architectural parameters and how they interact to influence berry composition at harvest.We use the insight gained to evaluate contemporary indices of vine management: harvest index and leaf area models.

Study site
Samples for this study were collected from Jindong Estates vineyard in the Jindong subregion (33°45'S, 115°14'E) of the Greater Margaret River Wine Region, Western Australia.The climatology is Mediterranean (Gladstones, 1999).Soils were either medium brown sandy loam to loam over sandy clay loam and brown clay loam over grey or yellow-brown medium clay (no inclination or elevation).The irrigated V. vinifera L. cv.Cabernet-Sauvignon vines (WA clone, own rootstock) were planted in 1998 in north-south rows.Vines were spur pruned to two nodes and unilaterally cordon-trained, trellised to a Smart-Dyson trellis, where approximately 30-40 % shoots originating from the same cordon are left to hang (Smart and Robinson, 1991).Data were collected in the 2003-2004 vintage.Vines were spaced approximately 2 m apart in rows approximately 3 m apart.Standard vineyard operations of trimming and thinning were excluded from the experimental sites.Whole-vine canopy statistics are summarised in Table 1.

Leaf area analysis
Leaf area per stem was estimated using a linear equation relating leaf area to stem diameter using pipe theory as the basis (Vertessy et al., 1995).Leaf area was determined for 64 stems across three vines using a Li-Cor LI-3100C ® Leaf Area Meter (Nebraska, USA).Stem diameter was calculated using digital Vernier callipers on the same stems at the middle of the second clear node using the equation for an ellipse.A linear regression was calculated between actual leaf area and stem diameter as y = 14.68x + 172.80 (R² = 0.63).The leaf area for all remaining stems with known stem diameters measured in the field at the same time point was estimated in this way (Note: the equation is site-and variety-specific).This methodology allows for the development of a model for non-destructive measurement of stem leaf area.For clarity of nomenclature, 'stem' refers to the vegetative anatomy of the grapevine (season growth) and 'shoot' refers to whole-of-season growth, including vegetative and reproductive anatomy.

Harvest
Thirty clusters were picked by hand from each of nine vines (2004).If a vine yielded more than 30 clusters, then 30 clusters were selected at random from evenly spaced positions along the cordon, taking care to avoid operator bias in choosing cluster position and exposure.If there were less than 30 clusters, all clusters were collected, and appropriate statistical measures used (such as Games-Howell pairwise comparisons).The length of the vine's cordon was measured from the beginning to the end of the permanently trellised cordon (± 0.01 m).
The remaining clusters on each vine were counted, weighed, and used to calculate total vine fruit biomass.Sample collection was as close to harvest as possible and determined by the operating viticultural company to meet harvesting schedule requirements.All clusters were hand-harvested into labelled zip-lock plastic bags and cooled in batches as practical.The samples were kept in a cool room overnight and transported to the laboratory the following day for freezing (-20 °C).Individual stem weights were recorded (± 0.01 g).

Fruit analysis
Clusters were thawed and gently warmed over warm water to release any precipitated tartrates (Fowles, 1992).Each cluster was weighed (± 0.001 kg) then crushed in the bag, so the juice maintained contact with skin, stalks, and seeds (debris) immediately prior to analysis.Clusters were hand-pressed to extract juice.The debris was allowed to settle, and the supernatant used for analysis.The sugar concentration of the clear juice sample was measured as degrees Brix (°Bx), referred to as °Brix, grams of sucrose equivalent per 100 g of solution, with a digital refractometer (Atago PR32 ® Japan).Anthocyanins and phenolics were determined spectrophotometrically (Iland et al., 2000).A subsample of 20 berries per cluster was used to estimate the average berry mass of each cluster.

Data analysis
Data analysis was carried out using Microsoft Excel ® , Addinsoft XLSTAT ® and the R statistical environment (Addinsoft, 2016; R Core Team, 2016).Damaged samples were excluded from analysis.The Dixon test for outliers at 5 % significance was used to exclude outliers later in the maturity analyses (sugar, acid, colour and phenolics).

Univariate analysis
The maturity data per whole vine used for all subsequent analyses are summarised in Figure 1.The box plots highlight the spread of the individual cluster maturity data around the mean and the spread between individual vines.The whole-vine maturity trend by leaf area is generally negative, but values increase at first and then decrease as leaf area increased (respective R 2 and p-values for °Brix (0.27, 0.148), anthocyanins (mg g -1 berry) (0.47, 0.042) and phenolics (absolute units g -1 berry) (0.45, 0.049).The range of values within an individual vine is approximately equivalent to between vines with leaf areas less than ~10.00 m 2 .

Bivariate analysis
The interrelationships for all sampled shoots for all vines are displayed as a matrix of bi-plots, correlations and fits (linear) of the relationships between the maturity variables and stem and fruit descriptors (Figure 2).The diagonal cells display the variable name code and a histogram; those below the diagonal, a bivariate scatter plot with a linear fit; and those above the diagonal present the Pearson coefficients of determination (R 2 ) and their statistical significance.
Berry count (BeCl) was predictive of cluster mass (ClM), and stem mass (StM) was predictive of both (BeCl and ClM).Similarly, anthocyanins (Antho) were predictive of phenolics (Pheno) and vice versa.Sugar accumulation (°Brix) correlated with phenolic and anthocyanin concentrations, but the distributions were unique, though near normal; anthocyanins had the greatest variance and °Brix the least.Tests showed multi-collinearity between ClM and BeCl and maturity indicators with variance inflation factors of 39.3 and 33.7, respectively.From this point on, the number of berries per cluster was excluded from further model building.
°Brix displayed no clear relationship with any of the measured parameters.While generally skewed to the left, anthocyanins and phenolics tended to be negatively related to the correlate: cluster, berry or stem mass, for example.The reproductive parameters of berry count, berry mass and cluster mass generally increased with stem leaf area.
The vine descriptors were skewed to the left: stem mass, cluster mass and °Brix were statistically correlated with shoot position but accounted for a low fraction of variance (< 5 %).Stem mass and leaf area were skewed strongly to the left, and while the two were correlated statistically, the correlation only accounted for about 10 % of the variation, i.e. the two were largely independent despite appearing correlated.The remaining variables and relationships displayed a high degree of heteroscedasticity, indicating either no or a complex relationship (Figure 2).The Breusch-Pagan test for heteroscedasticity was used to verify these observations by determining whether the error variance from the regression depended on the independent variable values (Table 2).Unequal variability was evident, particularly for position and reproductive predictor variables in °Brix, all predictor variables in anthocyanins, and all phenolics variables except for position.The model could not be improved by transformation or weighted least-squares estimation.

Multiple regression
Bivariate analysis is useful for data exploration and to describe the relationships between discrete independent variables, yet it has limitations when multiple parameters interact.Thus, a multivariate approach was adopted using partial least-squares (PLS) regression to rank the contribution of individual  Pos, distance from permanent cordon origin (cm); BeCl, number of berries per cluster; ClM, mass of whole cluster (g); BeM, average individual berry mass per cluster (g); StM, mass of stem (g); StLA, leaf area per stem (cm²); StLApClM, stem leaf area per corresponding mass of cluster (cm 2 g -1 ); Sug, sugar concentration as °Brix; Antho, anthocyanins (mg g -1 berry); Pheno, phenolics (absolute units g -1 berry).TABLE 2. Breusch-Pagan (BP) test for heteroscedasticity where H0 = homoscedasticity using Chi-square for three grape maturity indicators, sugar concentration as °Brix, anthocyanins (mg g -1 berry) and phenolics (absolute units g -1 berry), from individual vegetative and reproductive plant growth components within a grapevine.The table relates to data in Figure 2. Sugar as total soluble solids (°Brix); Colour as anthocyanins (mg g -1 fruit); Flavonoids as phenolics (absolute units g -1 berry); Pos, position along cordon measured as distance from head of vine (permanent cordon-trained) (cm); ClM, mass of whole cluster (g); BeM, average individual berry mass per cluster (g); StM, mass of stem (g); StLA, leaf area per stem (cm²); NS, not significant.
grapevine shoot components to explaining variability in the individual grape cluster maturity (Figure 3).Additionally, PLS was used to assist in parameter number reduction and to minimise the collinear influence of the reproductive and vegetative components of the model (identified in Figure 2).
Q 2 cumulative model strength (highest value) was used to select the number of components (Components 1 to 4 = 0.155, 0.178, 0.161, 0.157), which is similar to R 2 as an estimate of the strength of the least-squares fit but derived from the test set; it also ranges from 0 to 1 but is relatively independent of the number of fitted variables.The lines in the correlation map exhibit the direction and strength of the relationship between the predictors and the response variables in three-dimensional space (only two presented, Figure 3A).
Clear relationships between predictor and indicator variables emerged from this approach.Strong clustering is apparent with the three compositional attributes of maturity, which were negatively related to the biomass variables: BeM and StLA (PLS t1, Figure 3A).The strength of the contribution made by each predictor variable is shown by the variable importance of the projection (VIP) value (Figure 3B).Berry mass was the strongest predictor for each maturity indicator.The model is untrustworthy if the confidence interval falls below reference line; for example, for stem leaf area (StLA), the ratio of stem leaf area to cluster mass (StLApClM), and shoot distance from the crown or start of the crodon (Pos) (Figure 3B).

Analysis of covariance
It is apparent from the values presented in Figure 1 that the observed maturity values partly depended on the whole-vine leaf area.Hence, the individual vine was added to the model, and individual vine growth predictors were explored as covariates with the maturity indicator variables through analysis of covariance (Table 3).All ANOVAs were verified with the Welch statistic as a robust test of equality of means.
No effects were noted.
The effect of the whole vine as a covariate was greatest in the model of each maturity parameter, indicated by a higher whole-vine Fisher's F value than the shoot growth parameter (Table 3).The vine effects were important and additive for cluster mass and stem mass, as indicated by the significant effect of vine on the model but with no significant interaction (Table 3).However, statistically significant interactions were observed when considering individual berry mass and stem leaf area.In all instances, no interaction was expected if the differences were solely due to a temporal factor; for example, if the harvest was staged to ensure all fruit was harvested at a similar maturity.To simplify interpretation, and to place the analysis in a framework more relevant to populations of vines and vineyard management, vines were then grouped according to two popular indices describing canopy characteristics.
The trends in the maturity ranking within each index contrast strongly (Table 4), and pairwise comparisons indicate significant differences between each class within each index across all maturity indicators.For the vine leaf area model, the trend in mean from Low to High was strongly negative (-25 % for °Brix, -60 % for anthocyanin content and -45 % for phenolics).Whereas for the vine harvest index model, the trends were non-linear, with the minima at the 'ideal' level.The High harvest index class had the highest predicted values for °Brix, anthocyanins and phenolic composition.The vine leaf area index was the best predictor of maturity of the two indices, as indicated by the greater Fisher's F values, and was selected for further analysis.TABLE 4. Winegrape canopy indices and their impact on berry composition as °Brix, anthocyanins (mg g -1 berry) and phenolics (absolute units g -1 berry) at maturity.
TABLE 5. Analysis of covariance (ANCOVA) for grape maturity indicators including sugar concentration-°Brix, anthocyanins (mg g -1 berry) and phenolics (absolute units g -1 berry)-against individual stem size variables-berry mass (average individual berry mass per cluster (g)) and leaf area (shoot leaf area (cm 2 ))-within grapevines by, and including interactions with, vine leaf area (whole-vine leaf biomass (cm 2 )).The table relates to data in Figure 4.   5 for the analysis of covariance.
Analysis of covariance was undertaken using the berry and stem size predictors, with whole-vine leaf area across all maturity indicators (Table 5).Berry size and stem leaf area size were selected because they showed interactions between the parameters and the individual, whole-vine covariate and a strong goodness of fit (Table 3).All models were statistically significant, with vine biomass the strongest predictor except for berry phenolic content, where berry mass was the strongest predictor (Table 5).The models are shown in Figure 4.
The trend in slope for the relationships between °Brix and both individual berry mass and leaf area per stem depended on the leaf area class of the vine: from strongly negative for low leaf area vines to nil or even positive for high leaf area vines depending on the composition component assessed (°Brix, anthocyanins or phenolics) (Figure 4).Note that high leaf area class vines had lower maturity levels than the low leaf area class vines.
While the slopes depicted in Figure 4 are consistent with the information provided in Table 4, an important degree of interaction is not apparent in the previous analysis.Generally, the value of the dependent variable (°Brix, anthocyanin or phenolics) decreased as the predictor stem leaf area or individual berry mass increased.However, while the slope decreased with whole-vine leaf biomass, strong interactions are associated with average individual berry mass, especially for °Brix and anthocyanin content.Comparatively low vine leaf area significantly influenced the relationship between berry mass anthocyanins and phenolics.The inverse occurred for the relationship between stem mass and maturity indicators.Adding other within-vine plant growth predictors decreased the fit of the analysis of covariance model (data not presented: °Brix 5.5 %, anthocyanins 1.3 % and phenolics 1.0 %).Thus, there was no added value in adopting multiple independent variables to the model beyond berry size.

All models summary
Table 6 summarises the strength of goodness of fit for the key models evaluated.It allows for easy comparison and demonstrates that the models with whole-vine covariates best explain the variability in the dependent variables (°Brix, anthocyanins and phenolics).In summary, the whole vine does matter; however, something to do with vine physiology or its location affects the level of maturity attained that is not captured by the two most popular grapevine canopy characteristics (leaf area and leaf area to fruit weight ratio).When considering within-vine variation, whole-vine leaf area and individual shoot berry mass together describe the most variability in maturity.

DISCUSSION
Smaller grapevines produced fruit with higher intensity of composition than larger grapevines.Smaller berries exaggerated this trend.While such observations are not new at the whole-vine scale, the novelty of this study is evidence of these trends at the within-vine scale.The solute concentration of berries at maturity typically varies inversely with their size (Barbagallo et al., 2011;Considine, 2004;Dreier et al., 1998;Gray, 2002;Roby et al., 2004).Conversely, this relationship did not hold for vines with much greater vegetative biomass, where larger berries had higher sugar contents, or the maturity progression of sugar with anthocyanin and phenolic development becomes uncoupled.These relationships represent the interplay of the biophysical factors driving source -sink relationships in grapevines.In the context of a viticulture environment where water, sunlight and temperature do not limit grapevine growth, our modelling informs the management of large grapevine canopies such as those found in industrial-scale viticulture.
We demonstrated that leaf area as an estimate of whole-vine vegetative biomass best represent the characteristics of reproductive organs, as opposed to harvest index, harvest mass, or internal spatial location (c.f.Bravdo et al., 1984;Carbonneau et al., 1991;Chapman et al., 2004;Kliewer and Dokoozlian, 2005;Lefort et al., 1979).The observed trends from low to high leaf biomass diverge from the 'ideals' presented in the literature but are consistent with the concept that high vegetative biomass delays or impedes maturity.The harvest index trend, which appears to intuitively offer a standard way of matching traditional to industrial viticultural systems, failed to show the anticipated trends: the berry composition values were lowest for the 'ideal' values suggested by Smart and Robinson (1991).Clearly, matching theory to field observation remains a challenge.
Bivariate analysis of berry composition with shoot growth parameters did not help dissect the importance of these maturity determinants.The single most important factor contributing to the scatter of values in the bivariate analysis was the individual vine, particularly the leaf area of the individual vine.This analysis helped elucidate the observed heteroscedasticity from the bivariate analysis with shoot growth parameters.This also showed that the mass of the reproductive organ, the berry, was the single most important predictor of its composition, though leaf area of individual stems within a vine also exerted a significant influence.Together, these parameters largely determined the observed within-vine variance of fruit composition: °Brix, colour (anthocyanin content) and phenolics concentration.There were complex relationships between °Brix and anthocyanins and phenolics (secondary metabolites), representing the quality potential of berries for wine production (Downey et al., 2006;Ough and Singleton, 1968).The solute concentration of berries at maturity typically varies inversely with their size (Barbagallo et al., 2011;Considine, 2004;Dreier et al., 1998;Gray, 2002;Roby et al., 2004).However, the data presented here showed that size alone did not account for the values found.A novel finding from this study was that the sugar concentration of high biomass vines tended to increase as berry mass increased.Temporal influences seem unlikely because the slopes vary in direction rather than intercept (i.e. and interactive effect Figure 4).It has been proposed that high vegetative biomass vines allocate resources preferentially to shoot and root growth (Hale and Weaver, 1962); however, larger fruit may be more competitive in certain circumstances, as reported for peach (Marini and Sowers, 1994) and grapes (Ollat and Gaudillere, 1998).This response is likely to relate to the biology and physics of the phloem acting as the conduit, as well as the energetics of the apoplastic unloading in the pericarp.Surface area-volume trends have been thought to influence the mass-concentration relationship (Greer and Rogiers, 2009) but recent research suggests that such a relationship may only become apparent once loading has ended (Savoi et al., 2021) and therefore internal spatial factors may underpin the general relationship.This leaves the relationship observed in high biomass vine enigmatic and open to further study.
Canopy geometry factors such as shading likely interacted with our measures.While this was not assessed here, its importance is well understood for anthocyanins.Anthocyanins and phenolics are strongly influenced by vegetative biomass and microclimate, and then post véraison more strongly by climatic conditions post-véraison, especially light intensity and temperature (Azuma et al., 2012;Bergqvist et al., 2001;Guidoni et al., 2008).It is unclear whether the impact of vegetative biomass on this biosynthetic pathway occurs through changes in metabolite availability and/or changes in microclimate.Shading influences biosynthesis directly through the phytochrome system (Matus et al., 2009;Sun et al., 2017) and indirectly through an effect on berry temperature (Costa et al., 2020;Tarara et al., 2008).Temperature also affects other maturity parameters, such as acid and sugar accumulation (Brandt et al., 2019;Rienth et al., 2021).A recent study suggested that limited carbon supply promotes carbohydrate accumulation in berries at the expense of secondary metabolites (Bobeica et al., 2015).Our results support the concept that high vegetative biomass conditions can delay the accumulation of anthocyanins and phenolic polymerisation.
The influence of berry size on the relationship to maturity parameters in this study was highly significant, which suggests a competitive aspect to these processes (see also Gray and Coombe, 2009;Trought et al., 2017).
A limitation to interpreting our observations is that we harvested on a single date, so the fruit was not necessarily all mature.However, leaf area manipulation studies have similar limitations.Thus, the matter is more complicated than the leaf area/fruit weight model assumes and perhaps accounts for the relatively poor performance of the harvest index model in this study.It should be noted that we partitioned variation across space rather than time.Season, management and clonal variabilities could influence vegetative vine biomass, but we present comparative data across one year because spatial variability does not change with these contributing factors.
Lateral growth increases with increasing growth rate, particularly in highly vegetative vines (Keller et al., 2015).
As growth slows post-anthesis, reducing vegetative competition in general, increased growth of sylleptic lateral shoots may provide an alternative competitive sink for vines with high vegetative biomass such as those in this study (Filippetti et al., 2013;Santesteban et al., 2017;Song et al., 2014).Three factors -early shoot growth vigour, degree of partitioning to storage organs and lateral shoot growth -may help us understand the disparity in the reported relationships between berry size and maturity (Casassa et al., 2016;Ferrer et al., 2014).

CONCLUSION
This study identified a previously unrecorded phenomenon in grapevines where shoot-to-shoot variation in berry mass and its relationship with berry composition changes with whole-vine biomass.Thus, biomass management at the whole-vine scale may affect metabolite partitioning and thus fruit composition within vines.Present management indices that have been developed to predict ideal fruit composition at harvest, direct agronomic practice (such as cluster thinning) to low yield per hectare.In locations with the potential for high vine biomass, such as those with non-limiting conditions (abundant water, warmth and light) found commonly in new-world industrial-style agriculture, such practices may exacerbate the high vegetative state, with excess carbohydrates redirected to leaf growth with negative consequences for quality.
The functional relationship between vegetative and reproductive organs in grapevines must be examined in the context of other factors influencing the dynamics of berry development, such as light penetration and water availability.
Our study represents a 'natural' experiment, with the grapevine population exhibiting a range of vegetative biomass in the field rather than experimentally manipulated to achieve this result.Future studies could be based on contemporary understanding of the nature of source -sink relationships in the grapevine.How sugars are not loaded into phloem vessels but diffuse from mesophyll cytosol to phloem vessel, undergo mass flow and then follow a diffusion pathway to sink cytosol is critical to understanding source -sink relationships in the grapevine (Turgeon and Medville, 1998;van Bel, 2003), as is the switch to apoplastic unloading in the pericarp of the berry at véraison (Zhang et al., 2006).
In contrast to expectations, large berries can be competitive sinks for carbohydrates when canopies are large.Consequently, management practices that encourage larger berries (e.g.increased light penetration before anthesis) may be beneficial when resources and growing conditions are not limiting.Ubiquitous generalisations, such as cluster thinning to achieve 'vine balance', may be problematic if the whole viticultural system and contributing factors to fruit development are not considered.This experiment demonstrated how high leaf area and large average berry size per cluster can delay maturity.For grapevines with low or ideal vegetative biomass grown in warm conditions, management practices that encourage smaller berries rather than fewer clusters are desired (Clingeleffer et al., 2002).

FIGURE 1 .
FIGURE 1. Box plots of grape maturity data °Brix (A), anthocyanins (mg g -1 berry) (B) and phenolics (absolute units g -1 berry) (C).One box per vine.Red crosses (+) are mean maturity values, black central horizontal bars are medians, and black lower and upper limits of the box are the first and third quartiles, respectively.Pairwise comparison post hoc test (Games-Howell) results by maturity parameter with a confidence interval of 95 % are indicated by different letters denoting significant difference.Mean vegetative biomass (leaf area) is presented adjacent to the x axis, and vines are ordered accordingly.

FIGURE 3 .
FIGURE 3. Correlation map by partial least-squares regression visualising the first two components of the correlations between Xs and the components, and Ys and the components for grape maturity parameters from individual stem vegetative and reproductive plant growth components within a grapevine (A).Variable importance in the projection (VIP) bar graph with a 95 % confidence interval with reference line (black dashed) (B).Maturity indicators (dependent variable Ys) in red dashed line and open circle: °Brix; Antho, anthocyanins (mg g -1 berry); Pheno, phenolics (absolute units g -1 berry).Plant growth maturity predictors (independent variables Xs) in black line and closed circle: BeM, average individual berry mass per cluster (g); ClM, mass of whole cluster (g); StLA, leaf area per stem (cm²); StM, mass of stem (g); StLApClM, stem leaf area per corresponding mass of cluster (cm 2 g -1 ).Goodness of fit (R²) A = 0.13, B = 0.20, C = 0.36.

TABLE 3 .
Analysis of covariance (ANCOVA) for grape maturity indicators: sugar concentration as °Brix; anthocyanins (mg g -1 berry); phenolics (absolute units g -1 berry), against individual plant growth parameters within grapevines by, and including interactions with, whole individual vines as a factor.Vines significantly differed (P > F < 0.001) for all maturity indicators.Vine factor alone accounted for 58, 53 and 50 % of the variability in °Brix, anthocyanins and phenolics, respectively.Equation fitted was y = a + bx + cz + dxz, where 'x' is the vine, 'z' is cluster mass, berry mass, stem mass or leaf area, respectively, and 'y' is the predicted maturity variable.
NS = not significant.

TABLE 6 .
Summary of the linear regression model goodness of fit values (Pearson's R²) for best models of individual vegetative and reproductive plant growth components with maturity indicators: °Brix, anthocyanins (mg g -1 berry) and phenolics (absolute units g -1 berry).