The impact of non-structural carbohydrates (NSC) concentration on yield in Prunus dulcis, Pistacia vera, and Juglans regia

Successful yield in orchards is the culmination of a series of events that start with plants entering dormancy with adequate energy reserves (non-structural carbohydrates; NSC). These NSC are responsible for the maintenance of activities during dormancy and extending onto the period of activeness. Using multi-year yield information and monthly NSC content in twigs, we show that high levels of carbohydrate in Prunus dulcis, Pistachio vera, and Juglans regia during the winter months are indeed associated with high yield, while high levels of the NSC in late summer often correlate with low yield. An evaluation of monthly NSC level importance on yield revealed that for P. dulcis high levels in February were a good predictor of yield and that low levels throughout summer were associated with high yield. In P. vera, high levels of NSC in December were best predictors of yield. J. regia exhibited peculiar patterns; while high pre-budbreak reserves were associated with high yields they only played a minor role in explaining crop, the most important months for predicting yields were June and July. Results suggest that NSC levels can serve as good predictors of orchard yield potential and should be monitored to inform orchard management.

to how trees physiologically prepare for this quintessential period of quiescence 6,7,9 . The amount of reserves needed to maintain dormancy and a healthy growth resumption (bloom/leafing) can be variable and not easily predicted. The build-up of pre-dormancy reserves may be subject to changing abiotic and biotic conditions, growth, and reproductive activity during the active season 3 especially in alternate bearing species like P. vera. Additionally, not only is the NSC reservoir contingent upon the active season but the length and conditions of the dormant period itself can also vary from year-to-year, further affecting the amount of reserves readily available to sustain phenological transitions. The unpredictable nature of the local climate combined with the selection for yield maximization most likely enforce the need to store more NSC reserves in domesticated plants than what is required for the average dormancy period in most undomesticated perennials 10,11 . While the accumulation of reserves is often seen as a byproduct of an excess of carbohydrates, mounting evidence suggests that it may actually be a sink that actively competes with growth and reproduction rather than merely a passive process 12,13 . To their undomesticated counterparts, this 'excess' might provide a competitive advantage in which long-term survival is promoted over current vegetative growth and reproductive capacity (yield). However, in domesticated fruit and nut species this may instead shift to promoting short-term gains in reproductive capacity in lieu of long-term NSC reserve formation. This, in turn, potentially makes selected varieties potentially more susceptible to the negative impacts of unexpected changes in dormancy conditions and reduces their resilience to additional stresses. Finally, as a healthy and synchronous bloom is a prerequisite for pollination and fruit set, any changes to NSC content and its forms as affected by weather, biological stress or management can result in significant yield variation. Furthermore, since NSC levels and their form can affect a range of physiological activities, it is important to ask if and when NSC content has the greatest impact on tree productivity and if it is always better to assure high NSC content to generate high yields. Therefore, to answer these questions, we used multi-year observations of NSC content in twigs of P. dulcis, P. vera, and J. regia and combined them with reported yields for over 300 orchards located across the Central Valley, CA, USA.

Materials and methods
Using a Citizen Science approach, growers across the entire Central Valley of California sent samples of currentseason twigs of Prunus dulcis (Mill. D.A Webb), Pistacia vera L. and Juglans regia L. (Fig. 1). The study complies with local and national guidelines. The carbohydrate data set used in this study spans from September 2016 to August 2019 with yield data for the 2017-2019 period. Out of over 590 orchards participating in the NSC study, we selected the orchards from which growers shared yield information for at least one year during the 2017-2019 period. This resulted in 132 P. dulcis, 122 P. vera, and 84 J. regia orchards used in the presented analysis. We encouraged growers to collect samples once a month, however frequency and participation level varied over time and therefore the data sets varied from month to month.
Specific details of sample collection and handling were described previously 1 . Briefly, a unified protocol for sample collection required that one current season twig from three trees per orchard be cut at the base where the current season's wood met last year's wood. The bark from the lower 10 cm of the twig was removed using a razor blade. Both the bark and the wood of the three twigs were put in a paper envelope and mailed to the laboratory for NSC analysis. Buds were excluded from the samples. The integrity of the NSC content over shipping time was tested to assure the quality of the results 1 . Upon arrival, samples were put in the dryer for 48 h at 75 °C. The bark and wood were chopped into small < 1 mm pieces separately and ~ 100 mg of each was ground into a fine powder (~ 1 µm) using a ball grinder (MiniBeadbeater-96, Glen Mills Inc., NJ). To analyze for soluble sugar and starch content, we used the previously described protocol 14 with modifications to use smaller sample sizes 15 . Specifically, 25 mg of powder per sample was placed in 1.5 mL tubes. Tubes were then treated with 1 mL of sodium acetate buffer (0.2 M, pH 5.5), vortexed, and incubated in a 70 °C water bath for 15 min and centrifuged (10 min at 21,000g). 50 µL of supernatant was extracted and diluted in ultra-pure (UP) water (1:20, v:v) and vortexed. Soluble sugar content was quantified from diluted supernatant tubes using an anthrone/sulfuric acid colorizing reagent (0.1% (m:v) in 98% sulfuric acid) and reading absorbance at 620 nm in a spectrophotometer.
The remaining centrifuged tubes containing the pellet and buffer were used for starch quantification. To extract the starch, the tubes were boiled at 100 °C for 10 min to allow starch gelatinization, and let sit for 20 min at room temperature (22 °C). Once cooled, 100 µL of amyglucosidase (7 units per mL, Sigma-Aldrich) and 100 µL amylase (0.7 units per mL, Sigma-Aldrich) were added to the tubes and incubated for 4 h at 37 °C in a rotating incubator. Samples were then centrifuged (10 min at 21,000g). Tubes were then centrifuged and 50 µL of supernatant was extracted and diluted in 1 mL of ultra-pure (UP) water (1:20, v:v) and vortexed. Total soluble sugar content was analyzed using the same method described above. Starch content was determined by subtracting the original from the post-digestion soluble sugar content. All samples were plated onto 96-well plates. To account for any procedural variability (chemicals, timing, pipetting, temperature, etc.), each plate contained a glucose standard curve (4 wells per plate) and wood/bark standard tissue samples with known soluble sugar and starch contents (4 wells of each per plate) that were concurrently undergoing all steps in the same chemical analysis. The wood/bark standard tissue is a sample from a homogenous mix of several thousand ground samples leftover from a 2016 preliminary part of the study 1 .
To calculate the coefficient of correlation (r) between yield and carbohydrate content for each of the 12 months preceding harvest (September till August) we used the 'cor' function using the Pearson method (R-core). A linear model Yield = ß 0 + ß 1 × Concentration NSC type (lm, R-core) was used to estimate the slope parameter (ß 1; yield change in kg ha −1 in response to an increase of NSC concentration by 1 mg g −1 of tissue). To determine the most important months, for each carbohydrate type, in predicting the observed yields we used the Random Forest Regressor in PyCaret (Python; PyCaret.org. PyCaret, April 2020, URL https:// pycar et. org/ about. PyCaret version 2.3). The native function 'feature importances' , in which 'months' were assigned as features, was used to indicate which months were the most important predictor of yield.

Results
From post-harvest (September) till harvest (August), NSC concentrations in twigs not only show seasonal variation 1 but are also characterized by high variation within each month of the year in all three species (P. dulcis, P. vera, and J. regia; to access raw NSC data visit http:// zlab-carb-obser vatory. herok uapp. com/). Due to the seasonal variation in NSC content, calculation of the coefficient of correlation between NSC content and yield was performed separately for each month. In general, the analysis revealed that the coefficients of correlation between NSC concentrations and yield were positive and significant (at p-value < 0.1) during mid-winter (January and February) in all three species (Table 1; Figs. 2, 3, 4). Moreover, P. dulcis (almond) was characterized by the presence of multiple periods of significant negative correlations between NSC content and yield; during the active period (April to July), highly significant negative correlations before harvest (August), and following harvest in September. Soluble sugar concentration in the bark was only weakly correlated with yield, while total NSC concentration and NSC concentration in wood was significantly correlated in 5 or 6 months during the year preceding harvest (Fig. 2). P. vera (pistachio) was characterized by significant positive correlations between concentrations of NSC and yield over the period from post-harvest till the end of dormancy (September till March). Specifically, starch content, in wood and bark, was the main driver behind these positive correlations.
During the active period spanning from April till July, NSC contents were not significantly correlated with the current-year yield. A shift to negative correlations occurred in August (before harvest) when total NSC, NSC in bark, and starch in bark showed negative correlations with yield. Interestingly, starch content in wood and bark as well as the total content of NSC was most correlated with yield across all months, while soluble sugar concentration in wood remained uniformly non-correlated to yield through the entire season (Fig. 3). Out of the three analyzed species, NSC concentrations in J. regia (walnut) were the least correlated with yield. Only during the late dormancy period (February-March), the content of NSC and their forms were positively correlated with yield. Unlike in both other species, no significant negative correlations between NSC and yield were observed in J. regia (Fig. 4). Each type of carbohydrate (NSC total, NSC in wood, NSC in bark, starch in wood, starch in bark, soluble sugars in wood, and soluble sugars in bark) was independently analyzed and the relative importance of each month's content as the predictor of yield was determined using the Random Forest Regressor algorithm (PyCaret's Regression Module with the split between training and test group of 0.7 and 0.3 respectively; Figs. 5, 6, 7). The positive or the negative impact of NSC concentrations on yield was assigned using the sign of coefficient of correlation for that month (Table 1). In P. dulcis, NSC content in February was the most important positive feature contributing to yield, while NSC concentration in August was the most important negative feature in predicting yield (Fig. 5). In P. vera, NSC concentrations in December was the most important positive feature in predicting yield, while NSC concentrations in August was the most important negative indicators predicting high yield (Fig. 6). In J. regia, NSC concentration in July was the most important positive indicator for yield followed by May concentrations, while the concentration in June was the most important negative feature in predicting yield (Fig. 7).

Discussion
The two main goals of the presented work were (1) to test for the presence of correlations between NSC concentration in twigs of nut trees and orchards' yields and (2) to determine the months at which carbohydrate concentration were the best predictors of realized yield. In general, several correlations were significant (Table 1), the coefficient of correlation (r) ranged from maximum positive correlations of 0.42, 0.63, and 0.52 and negative correlations of − 0.44, − 0.36, and − 0.36 for P. dulcis, P. vera, and J. regia, respectively. Typically, winter (dormancy) NSC concentration was positively correlated with yield, while the summer (active period) NSC concentration was negatively correlated with yield. Thus, more is not always better. In fact, the apparent exhaustion of NSC just prior to harvest can be linked to high yield, in which case, less is better. The presence of such correlations underlines the importance of monitoring NSC reserves for enhancing yield success.
The inverse relationship between yield and summer NSC suggests that yield comes at the expense of NSC reserve formation. Nevertheless, the positive correlation in the fall and winter requires that NSC reserves be replenished during the short postharvest period, prior to senescence, to assure adequate reserves for bloom. If NSC reserves are not replenished, a lower yield may be expected and may help explain the presence of alternate bearing at either the whole-tree or orchard level in the case of P. vera 16 or at the twig level seen in P. dulcis 17 . Interestingly, despite the general trends mentioned above, there were large differences in the magnitude and temporal patterns of the positive and negative impacts of high NSC levels on yield among the studied species. In P. vera, NSC levels were almost always positively correlated with yield from September through June (i.e. through the post-harvest, dormancy period, bloom, and vegetative part of the season). Out of all the months however, December NSC concentration was the most important positive predictor of yield, this was true in all of its studied forms and locations (NSC total, NSC in wood, NSC in bark, starch in wood, starch in bark, soluble sugars in wood and soluble sugars in bark). NSC concentration was only negatively correlated with yield in the short period preceding fruit maturation (August). The low levels of August NSC associated with high yields, suggest that reserve exhaustion during this period was correlated with a high accumulation of nut biomass. This pattern may reflect sink dominance of fruit over reserve formation. However, if high yield results in the depletion of NSC concentrations to the extent that they cannot be replenished prior to senescence (high levels of NSC are required in December to assure high yield), then this may lead to a reduction in the following year's yield and ultimately explain the alternate bearing habit seen in P. vera 16 . If true, breeding objectives aiming to reduce alternate bearing in P. vera 18 may benefit from selecting varieties that show a strong NSC recovery pattern in the fall.
Prunus dulcis presents a slightly more complicated picture of carbohydrates' impact on yield. NSC total, starch, and sugar concentrations in wood were positively correlated with yield in late fall and during dormancy www.nature.com/scientificreports/ (November through February) with February reserves being the most important positive feature associated with high yield. This would suggest that high NSC levels just prior to and during bloom are the most important prerequisite for achieving higher yields. Hence, a high NSC content during dormancy, achieved either by preservation and/or by the influx of sugars from more distal sources during bloom, to provide sufficient energy and structural material for flowers is the key element to yield success. This finding may also provide an interesting opportunity for P. dulcis breeding efforts, wherein selection could be informed by high NSC levels in February 19 . The sudden change in direction, from a positive relationship in February to a negative in March, most likely reflects the strong dependence on local twig reserves for sustaining a healthy bloom and promoting vegetative bud pushing. The ensuing negative correlation, characterized by a steep decline in NSC concentration beginning in March, continuing through the summer 1,3 suggests that during the most active period, the reproductive NSC sink takes precedence over reserve formation. While reproductive prioritization outweighs reserve formation in both P. dulcis and P. vera, the persistent decline observed in P. dulcis comes in stark contrast to P. vera, where only a narrow time frame, the nut filling period, was negatively correlated with yield. This prolonged period of low NSC content and its associated negative correlation with yield in P. dulcis may be offset by the fact that while P. dulcis has the earliest harvest amongst the three species, its senescence occurs at the same time as in P. vera and J. regia. Thus, in effect allowing more time for the recovery of NSC and potentially avoiding a pronounced alternate bearing habit 17 .
Juglans regia presents the most ambiguous pattern of carbohydrate impact on yield. Much like P. dulcis and P. vera, we also found the strongest positive correlations between NSC content and yield in the months just prior to budbreak. However, in contrast to the two-former species, J. regia is a wind-pollinated species with female flowers developing on new vegetative extension growth and thus relies on the concurrent development of both the vegetative and flowering structures. As a result, this magnifies the burden of bearing enough NSC reserves to initiate both their growths following dormancy. In walnut, this burden exceeds the storage-supply capacity within the twigs and it must therefore import NSC from distant sources to attain sufficient energy. Therefore, in preparation for bloom, J. regia may strongly depend on the redistribution of NSC, via the xylem, from the stem to  15,20 . Furthermore, given that flowering and vegetative growth occur simultaneously, photosynthetic independence promptly follows thereby quickly becoming the main energy source for supporting both their growths. Hence, the dependence on a distal energetic supply for growth initiation and then on current photosynthate for growth sustenance may explain the lack of significant correlations between twig NSC storage and yield. As a consequence, the antagonistic relationship between storage and reproductive sinks that is more apparent in P. vera and P. dulcis is diminished, making J. regia potentially less sensitive to twig reserve carbohydrate content. Interestingly, when analyzing monthly importance there were consecutive shifts; from a high NSC content in June, as the most important predictor of low crop, to a high NSC content in July, as the most important predictor of high yield. These months, in particular, coincide with the transitional phases that occur between growth and storage accumulation. In J. regia specifically, June marks the fastest growth rates and the lowest carbohydrates content while July is the moment at which growth slows but also the point of maximum reserve accumulation rates 1 . Therefore, in the context of phenology and yield, it is thought-provoking that the crux between the interplay of these months is captured as most important for predicting final yields. Such increasing importance on mid-summer carbohydrate concentrations further supports the notion that in J. regia, high crop is not dependent on the competition between storage and yield but rather on an overall high photosynthetic productivity especially at the end of summer. The low dependence of yield on autumnal NSC reserves can be expected from the fact that multi-year observations of NSC content on walnut tree twigs were relatively unaffected by seasonality compared to the two other studied species 1 . In addition, it is important to note that an impact of NSC reserves on yield is not always detected, for example in Olea europea L. (olive tree) no such impact has been reported 21 . However, such a relationship may be very difficult to detect in small, short term experimental studies due to high temporal and year-to-year variation in NSC content 1,3,9 . In all cases, only 1 or 2 months shared a high importance for crop prediction and such distribution of importance may suggest the practical implications in using carbohydrate analyses for orchard management and decision making. A simple NSC concentration analysis in twigs at specific months for example, December for P. vera or February for P. dulcis, can help project yield and provide information for assessing irrigation and fertilization needs. As non-structural carbohydrate content represents a buffer between photosynthetic capacity and needs (base respiration, growth, yield, defense, and dormancy reserve formation), knowledge on their dynamics may provide physiological insights to better understand the physiological status of trees. Sudden and unexpected changes to NSC concentrations may reflect orchard health issues. The introduction of NSC analysis to breeding may open new avenues in the search for high-yielding varieties. We can also expect that adding NSC content analysis to yield prediction models which consider environmental elements (temperature, rainfall) as physiological attributes and encompass a range of abiotic and biotic stressors (tree water status, pathogen infestations, fertilization, etc.) will improve their performance.