The Pennsylvania State University

The Graduate School

College of Agricultural Sciences

Department of Plant Science

EVALUATING THE IMPACTS OF VITICULTURAL AND ENVIRONMENTAL FACTORS ON

ROTUNDONE IN NOIRET GRAPES

A Thesis in

Horticulture

by

Andrew D. Harner, Jr.

Ó 2019 Andrew D. Harner, Jr.

Submitted in Partial Fulfillment

of the Requirements

for the Degree of

Master of Science,

May 2019

ii

The thesis of Andrew Harner was reviewed and approved* by the following:

Michela Centinari

Assistant Professor of Viticulture

Thesis Advisor

Ryan J. Elias

Associate Professor of Food Science

Richard Marini

Professor of Horticulture

Justine Vanden Heuvel

Associate Professor

Horticulture Section, School of Integrative Plant Science

Cornell University

Erin L. Connolly

Professor

Head of the Department of Plant Science

*Signatures are on file in the Graduate School.

iii

Abstract

The grape-derived sesquiterpenoid rotundone is responsible for the ‘black pepper’ aroma

of several wine grape varieties, including the interspecific hybrid Noiret. Numerous studies have

evaluated the effects of major climatic variables, including both temperature and solar radiation,

on the accumulation of rotundone in wine grapes. However, only a few studies have assessed the

effects of common viticultural management practices, and no studies have assessed both climatic

and viticultural influence together in a single study to fully investigate which variables have the

strongest influence on rotundone concentrations. Over the 2016 and 2017 seasons, we evaluated

the influence of 21 different viticultural, meso- and microclimatic variables on the concentrations

of rotundone in Noiret wine grapes at 7 Pennsylvania and New York vineyards with distinct

environmental conditions. Vineyard-scale post-veraison temperatures and solar radiation had

robust and positive correlations with rotundone concentrations measured at harvest. At the level

of the fruiting zone, rotundone concentrations correlated negatively with post-veraison

temperatures below 15 °C, above 30 °C, and pre-veraison fruit sun exposure. Rotundone

concentrations were also strongly correlated to several grapevine tissue nutrients, including

calcium, potassium, and magnesium. A four-variable model was constructed using multiple

linear regression analysis of the vineyard-scale data. This model can be used by growers to

identify vineyards with potential for producing ‘peppery’ wines, as well as assist growers with

manipulating rotundone concentrations via implementation of canopy management practices to

achieve wines with the desired levels of rotundone and related ‘pepperiness.’

iv

TABLE OF CONTENTS

LIST OF FIGURES……………………………………………………………………...v

LIST OF TABLES………………………………………………………………………vi

ACKNOWLEDGEMENTS……………………………………………………………vii

Chapter 1: Climatic and agronomic influence on major grape-derived aroma compounds..........................................................................................................................1

1.1 Grape-derived chemical compounds drive wine aroma...........................................1 1.2 Climatic influence on specific wine grape aroma-active compounds......................4 1.3 Agronomic influence on wine grape aroma compounds..........................................7 1.4 Biological, chemical, and sensorial characteristics of rotundone..........................11 1.5 Climatic and agronomic influence on rotundone concentrations..........................15

Chapter 2: Weather conditions during fruit ripening, vine nutrient status, and vine size collectively influence rotundone concentrations in cool-climate Noiret wine grapes................................................................................................................................20 2.1 Introduction................................................................................................................20 2.2 Methods and Materials..............................................................................................23 2.3 Results.........................................................................................................................37 2.4 Discussion...................................................................................................................54 2.5 Conclusion..................................................................................................................69

Chapter 3: Conclusion.....................................................................................................71 References.........................................................................................................................74 Appendices........................................................................................................................83

Appendix A: Fruit maturity and production metrics........................................83

Appendix B: Nutrient concentrations and water status....................................84

Appendix C: Supplemental details on mesoclimate model selection...............85

Appendix D: Supplemental details on microclimate model selection..............90

v

LIST OF FIGURES

Figure 1: Map of vineyard study sites............................................................................24

Figure 2: Noiret grape rotundone concentrations for 2016 and 2017.........................44

Figure 3: Partial validation plot for the selected model...............................................50

vi

LIST OF TABLES

Table 1: Vineyard characteristics for all study sites.....................................................25

Table 2: All measurements taken during the study period..........................................36

Table 3: Mesoclimatic data for the study period..........................................................38

Table 4: Microclimatic degree-hour data for the post-veraison period......................40

Table 5: Microclimatic solar exposure data for the study period...............................41

Table 6: Correlations between rotundone and measured variables...........................46

Table 7: Examples of SAS RSQUARE option output..................................................48

Table 8: Regression equation for the chosen mesoclimatic data model......................48

Table 9: Regression equation for the chosen microclimatic data model....................53

vii

ACKNOWLEDGMENTS

The success of this project was directly due to the support of all those involved, as this

thesis would not have come to fruition otherwise. I would like to thank my advisor, Michela

Centinari, who offered me the opportunity to pursue viticultural research and has been a

consistent source of support and guidance throughout the entirety of this project, from its

inception and all the way through and past the writing process. Many thanks is extended to my

committee: to Rich Marini, who was instrumental in teaching me multivariate regression and the

necessary statistics for this project; to Ryan Elias, who assisted with all chemical analyses and

rotundone extractions; and to Justine Vanden Heuvel, who helped coordinate and manage all of

the experimental sites in the Finger Lakes. Thanks to Don Smith, and the many members of the

Centinari and Elias labs, past and present, who have provided critical advice, support, and

assistance with data collection and various analyses.

This project would not have been possible without the collaboration of Bryan Hed and

Steve Lerch, who I am indebted to for their aid with data collection and management of

experimental sites. Much appreciation is also extended to Tracey Siebert, Markus Herderich, and

Sheridan Barter of the Australian Wine Research Institute, as we would not have been able to

identify, extract, or analyze rotundone without their thorough help and collaboration. The

support of the AWRI was essential to the success of this project.

I would like to deeply thank my family for their enduring support throughout my time as

a graduate student, and for always being encouraging of my academic pursuits from the very

beginning. And, lastly, my sincere gratitude and love are extended to Clara Miller, who has

provided endless support throughout this project and with my research overall, and who

continues to be a strong source of encouragement in all that I do.

1

Chapter 1: Climatic and agronomic influence on major grape-derived aroma compounds

1.1 Grape-derived chemical compounds drive wine aroma

Perceived wine aroma and flavor is chemically complex, and both are dependent on the

presence of specific aroma-active chemical compounds and their interactions within the wine

matrix. The majority of compounds responsible for wine aroma are grape-derived secondary

metabolites (González-Barreiro et al., 2013). To date, the most studied classes of secondary

metabolites in wine grapes and those most critical for wine aroma are the pyrazines, mono- and

sesquiterpenes, C13-norisoprenoids, volatile thiols, and phenylpropanoids (Robinson et al., 2014).

During the fermentation process some of these compounds are extracted directly from grape

tissue—namely the skins or pulp—in a free, volatile form, while others are glycosylated and

structurally bound to a sugar moiety, and thus rendered inactive unless enzymatically or

chemically hydrolyzed (Robinson et al., 2014). It is this combination of both free and bound

aroma compounds that contribute to and influence perceived wine aroma.

Current knowledge about aroma-active chemicals in wine focused on investigations of

specific varietal aromas, with much research focusing on monoterpene and sesquiterpene

compounds. Monoterpenes, one example of the variety of compounds that comprise the terpene

class, drive the aromas of so-called aromatic white varieties, including Riesling, Müller-Thurgau,

Gewürztraminer, and many of the Muscats (Mateo and Jiménez, 2000). For example, the

monoterpenes linalool and citronellol can impart floral and citrus notes to many varietal wines,

an example being Gewürztraminer (Robinson et al., 2014; Waterhouse et al., 2016). Recent

research has also focused on another monoterpene, 1,8-cineole, or eucalyptol, as it contributes

aromas of ‘eucalypt,’ ‘mint,’ and ‘camphor’ to wines. Concentrations of this monoterpene were

heavily influenced by the inclusion of material other than grapes (MOG) into the must of Shiraz

2

wines, namely leaves of Eucalyptus species, in addition to grapevine leaves and stem tissue from

vines that were near Eucalyptus trees (Capone et al., 2012; Black et al., 2015). A variety of

terpenes, such as the monoterpene cis-rose oxide and the sesquiterpene rotundone, have been

labelled “aroma impact compounds” (Guth, 1997; Wood et al., 2008) due to their low perception

thresholds and ability to impart strong, specific aromas to specific wine varieties. Cis-rose oxide

has been associated with important ‘lychee’ aromas of Gewürztraminer wines (Ong and Acree,

1999), while rotundone imparts potent ‘black pepper’ aromas to Shiraz, Duras, Noiret, and many

other red- and white-fruited varieties (see, for example, Herderich et al., 2015; Homich et al.,

2017).

Another group of compounds within the terpenoids, C13-isoprenoids are abundant in

aromatic varieties and important contributors to the aromas of Semillon, Sauvignon blanc,

Merlot, Shiraz, and Cabernet Sauvignon (Robinson et al., 2014). In terms of wine aroma

perception, the most important C13-norisoprenoids studied have been b-damascenone, 1,1,6-

trimethyl-1,2-dihydronapthalene (TDN), and b-ionone (Waterhouse et al., 2016). b-damascenone

imparts aromas of ‘cooked apple’ and ‘floral’ notes to wines of many varieties, while TDN

contributes a ‘kerosene’ or ‘petrol’ note, which is considered a fault mainly in aged Riesling

wines (Waterhouse et al., 2016). Various carotenoids present in grapes are chemical precursors

of C13-norisoprenoids, and as such research has focused on the manipulation of fruit sunlight

exposure to modulate C13-norisoprenoid concentrations in fruit and wines, with most studies

focusing on TDN given its importance to Riesling wines (Kwasniewski et al., 2010; Waterhouse

et al., 2016).

Pyrazines, another major grape-derived class of aroma-active compounds, have also been

well studied (Robinson et al., 2014). This is namely due to the role that 3-isobutyl-2-

3

methoxypyraize (IBMP) plays in wine aroma and flavor, as IBMP has been associated with

‘green bell pepper’ and related herbaceous notes in Cabernet Franc, Sauvignon blanc, and

Cabernet Sauvignon (Ryona et al., 2008; Robinson et al., 2014). Although low concentrations of

IBMP might impart varietal character and vegetative aromas (Allen et al., 1991), high

concentrations are associated with excessively herbaceous notes that are typically undesirable

(Robinson et al., 2014). Previous work indicated that IBMP concentrations in key wine grapes

varieties like Cabernet Franc and Merlot might be manipulated through management practices

that alter the fruit zone microclimate (Scheiner et al., 2010). Although IBMP is the most

explored methoxypyrazine to date, 3-isopropyl-2-methoxypyrazine (IPMP) is also important in

wine as its presence can be an indication of ladybug taint, caused by the extraction of IPMP from

cluster-borne Multicolored Asian Ladybeetles (Harmonia axyridis) (Robinson et al., 2014). This

compound contributes to aromas of ‘bell pepper,’ ‘asparagus,’ and ‘peanut’ in white wines, and

when the proportion of extracted beetles increases, a subsequent decrease in fruit and floral

intensity of wines has been observed (Pickering et al., 2004; Robinson et al., 2014).

A few other classes of compounds contribute important positive or negative aromas to

wines, including many sulfur-containing volatile compounds. The sulfur-containing aroma

compounds that contribute the most to specific varietal aromas are polyfunctional thiols, named

due to their chemical structure (Waterhouse et al., 2016). The polyfunctional thiols are especially

relevant due to their low perception threshold as impact compounds (Waterhouse et al., 2016)

and their ability to impart specific, fruity notes. Specific and major examples include 4-

mercapto-4-methylpentan-2-one (4MMP), 3-mercaptohexan-1-ol (3MH), and 3-mercaptohexyl

acetate (3MHA), and these compounds have been associated with ‘guava,’ ‘passionfruit,’ and

‘grapefruit’ aromas (Robinson et al., 2014; Waterhouse et al., 2016). These compounds are

4

especially prevalent in Sauvignon Blanc wines, and have also been identified within a variety of

red and white wines of various styles, including Riesling, Semillon, Petit Manseng, Gros

Manseng, and Grenache wines (Waterhouse et al., 2016).

Although a variety of other compounds exist in grapes and wine that contribute to wine

aromatics, the aforementioned chemical groups represent the most widely studied within the

scope of manipulating their concentrations in the fruit and or wine. Given that many of these

compounds are either extracted directly from the grapes or serve as precursors for other aroma-

active compounds, it has been reasoned that specific viticultural practices, such as canopy leaf

removal, can affect their concentrations (González-Barreiro et al., 2013). Moreover, research into

the seasonal accumulation patterns of major aroma compounds is also important in that it could

assist with choosing the right timing of the application of specific viticultural practices (Scheiner

et al., 2010; Zhang et al., 2016). Much of this research has focused exclusively on Vitis vinifera

varieties, however, and further investigation into the aromatic compounds present in Vitis

interspecific hybrid grapes increasingly planted in the Northeastern and Midwest U.S. is

warranted. This would further expand the knowledge of viticultural influence on aroma

compounds, and allow for the application of such knowledge to a wider range of winegrowing

regions.

1.2 Climatic influence on specific wine grape aroma-active compounds

Climate encompasses all the environmental conditions related to temperature, solar

radiation, precipitation, and humidity within a given location, which synergistically influence

both the primary and secondary physiological processes of grapevines (Iland et al., 2011).

Differences between climatic factors among the winegrowing regions of the world drive

regional-driven varietal aroma trends (Robinson et al., 2014). Many studies indicated that the

5

mesoclimate, and more particularly temperature and heat accumulation, can influence the

concentrations of important aroma-active compounds in grapes and wines. Sauvignon blanc

grapes and wines produced in cool-climate regions, for example in New Zealand, tend to have

higher concentrations of IBMP and other methoxypyrazines than grapes and wines produced in

warmer climates (Lacey et al., 1991). Recent research has demonstrated the importance of intra-

season weather variation as well, as pre-veraison temperatures are more influential than total

seasonal heat accumulation in regard to IBMP concentrations in Cabernet Franc grapes and

wines (Scheiner et al., 2012). Rotundone accumulation is also sensitive to regional climate, as

grapes produced from cool climates tend to have higher concentrations of rotundone than those

produced in warmer climates. Furthermore, regional climate strongly influenced rotundone in

Grüner Veltliner wines produced from different Austrian winegrowing areas (Herderich et al.,

2015; Nauer et al., 2018).

The effects of solar radiation on grape-derived aroma compounds is well documented

with specific classes of compounds, though it is important to note that most studies have

evaluated these effects at the fruiting zone scale, and not assessed broad mesoclimatic

differences in radiation across vineyards and winegrowing regions. Most research has evaluated

different levels of fruit solar exposure, either via leaf removal (Lee et al., 2007; Kwasniewski et

al., 2010; Scheiner et al., 2010; Feng et al., 2014; Homich et al., 2017) or artificial shading

(Bureau et al., 2000; Skinkis et al., 2010; Zhang et al., 2015a). Important trends have emerged

from these and other studies; for example, increased fruit sun exposure favorably increased the

concentrations of a variety of C13-norisoprenoids and monoterpenes in various varieties, while

methoxypyrazine concentrations can be inhibited by increased fruit exposure (Robinson et al.,

2014).

6

Less work has evaluated the effects of sun exposure on concentrations of sesquiterpenes,

polyfunctional thiols, and other sulfur-including volatile compounds in grapes and wines, though

the sesquiterpene rotundone might reach higher concentrations in Shiraz grapes from shadier

clusters and vines (Zhang et al., 2015a). Most research has addressed the effects of solar

radiation intensity, though radiation quality may be more important when considering C13-

norisoprenoids, given that they are products of carotenoid degradation (Robinson et al., 2014).

Further investigation into radiation intensity and quality, in addition to the effects of radiation

when considered independent from the effects of radiation-induced warming, is necessary to

better understand the relationships between sunlight and important classes of grape-derived

aroma compounds.

The total amount and patterns of seasonal precipitation are strictly correlated to grapevine

water status which can directly or indirectly after various aromatic compounds (Robinson et al.,

2014; Black et al., 2015). The relationship between concentrations of IBMP in Cabernet

Sauvignon wines and water deficit is not clearly defined. It has been hypothesized that water

deficit might increase various ‘fruity’ compounds, including esters, lowering the perceived

‘vegetal’ and ‘bell pepper’ aromas in the wine (Chapman et al., 2005). However, increasing

levels of water deficit might also negatively affect ester concentrations (Talaverano et al., 2017).

Decreased vine water status also had a variable effect on monoterpene and C13-isoprenoid

concentration: whereas water deficit increased concentrations of b-damascenone in Merlot

wines, no effect on b-ionone was observed (Ou et al., 2010). Similarly, concentrations of the

monoterpenes nerol, geraniol, nerolidol, and citronellol were higher in water stressed vines,

while concentrations of linalool were unchanged (Ou et al., 2010). The sesquiterpene rotundone

is sensitive to vine water status, as Duras grapes grown under higher water deficits had higher

7

concentrations this aroma compound (Geffroy et al., 2014). Despite these associations, it is still

unclear whether these relationships are mainly due to direct influence of water deficit on the

biosynthetic processes responsible for the production of aroma compounds, or rather to indirect

effects on vine canopy size and fruiting zone exposure (Robinson et al., 2014).

Climatic parameters interact with other environmental factors, including soil

characteristics, and cultural practices to definite “terroir,” or more specifically, the wine sensory

attributes of a specific variety within a given geographic context (Van Leeuwen and Seguin,

2006). Similar attempts to classify the distinctiveness of an individual winegrowing region has

led to the development of the concept of ‘regional typicity.’ Riesling wines displayed region-

specific aromatic and flavor profiles across both Germany and Canada (Fischer et al., 1999;

Douglas et al., 2001). Similarly, Malbec wines from both Mendoza and California exhibited

region-specific sensory and chemical traits (King et al., 2014). Previous work has also attempted

to define intra-region typicity of Riesling wines from various locations within the Finger Lakes

of central New York (Nelson, 2011). Identifying regional identities for a region or country’s

commercially important wine grape varieties has economic and marketing implications, but also

emphasizes the underlying role that climate has on wine grape and wine aromatic and flavor,

coupled with variety-specific genotypic traits. More efforts to understand and define regionality

where it exists may shed further light on the various climatic, geological, and environmental

factors that influence and drive wine grape aromatics and flavor.

1.3 Agronomic influence on wine grape aroma compounds

Wine grape growers implement a variety of viticultural practices within a given season to

produce quality grapes and wines in accordance with consumer preferences (Wolf, 2008). These

practices might require crop load (yield to vegetative growth ratio) adjustment or manipulation

8

of the fruiting zone to improve microclimate and reduce pest and pathogen pressure. Whereas

some practices can directly manipulate vine vegetative tissue (e.g., leaf removal, shoot thinning,

canopy hedging, and dormant pruning), others directly affect the vine’s reproductive tissue, the

fruit (e.g., cluster thinning and shoot thinning). The former practices can directly adjust

grapevine canopy density, vine size, and crop load via the removal of vegetative biomass and the

thinning of canopy shoots and leaf layers (Wolf, 2008). Cluster thinning can be used to adjust

yield and crop load through direct removal of fruit. This has the effect of reducing vine yield and

crop load and has the possibility of inducing a compensatory effect such that individual clusters

and berries may reach higher sugar concentrations and greater mass (Dami et al., 2006).

Canopy management practices can alter the accumulation and concentrations of grape-

derived aroma compounds and precursor molecules, giving growers a chance to tailor berry and

wine aroma characteristics to match those that are preferred by consumers. Of these practices,

fruiting zone leaf removal, a management strategy used in many winegrowing regions, has been

the most widely studied and reviewed. Leaf removal is likely the most useful practice available

to growers to manage concentrations of both favorable and unfavorable aroma compounds in the

field. Removing leaves in the fruiting zone enhances sunlight and air penetration and thus

reduces disease pressure and might enhance fruit ripening (Smart and Robinson, 1991; Hed et

al., 2014). By manipulating the microclimate, it is possible to also induce changes on berry

secondary metabolites and many grape-derived aroma and flavor compounds (Alem et al., 2018).

Numerous studies have focused on the intensity of leaf removal and how varying levels

of leaf cover and sun exposure affects grape aroma compounds and important precursors

(Reynolds et al., 1996; Lee et al., 2007; Skinkis et al., 2010; Geffroy et al., 2014; Feng et al.,

2014; Homich et al., 2017), while others have explored the timing of leaf removal (Kwasniewski

9

et al., 2010; Scheiner et al., 2010; Zhuang et al., 2014; Komm and Moyer, 2015; Verzera et al.,

2015; Homich et al., 2017; Hickey et al., 2018). Both types of research will be briefly reviewed

here, in addition to other canopy management practices.

Overall, increased fruit sun exposure due to fruiting zone leaf removal favorably

enhanced the concentration of both free, aroma-active and glycosidically-bound monoterpenes in

several varieties, such as Riesling, Gewürztraminer, Golden Muscat, and Traminette (Reynolds

et al., 1991; Macauley and Morris, 1993; Reynolds et al., 1996; Skinkis et al., 2010).

Furthermore, leaf removal was more effective than shoot hedging in increasing free monoterpene

concentrations (Reynolds et al., 1996). Since leaf removal enhances fruit sun exposure,

overexposure of clusters to sunlight under particularly warmer growing conditions may,

however, reduce final free monoterpene concentrations; this suggests that leaf removal as a

management method for increasing desirable aroma compounds may be better suited to cool

climates (Skinkis et al., 2010). Because of the negative relationship between fruit sun exposure,

temperature, and IBMP concentration (Ryona et al., 2008), fruiting zone leaf removal has also

been used to favorably reduce IBMP concentrations in Cabernet franc and Merlot grapes at

harvest in cool climates (Scheiner et al., 2010). Leaf removal also increased the concentrations of

various C13-norisoprenoids, and namely b-damascenone to the greatest extent, in Nero d’Avola

wines (Verzera et al., 2016). Limited research has focused on the effects of leaf removal on

sesquiterpenes and results are still inconclusive.

The effects of leaf removal on aroma compounds may considerably change depending on

phenological stage of the berries at the time of application. Timing of leaf removal application

can be important for IBMP concentrations in Cabernet franc grapes; early season leaf removal,

applied 10 to 40 days after full bloom, had the most pronounced effect when compared to later

10

treatment timings (Scheiner et al., 2010). Leaf removal timing is also important for major C13-

norisoprenoids, as removing leaves at pre-veraison (~33 days post-fruit-set) was more effective

at increasing TDN concentrations in Riesling juices and wines as compared to earlier (fruit-set

stage) or later (post-veraison) leaf removal applications (Kwasniewski et al., 2010). Given that

excessive concentrations of TDN are typically considered undesirable, tradeoffs may exist

between implementing leaf removal to enhance berry maturation, reduce disease pressure, and to

adjust the concentrations of important aroma compounds to desired levels. Taken altogether, it is

necessary to further study the timing of leaf removal as it relates to major aroma chemical classes

and specific impact compounds, so that wine grape producers can better implement leaf removal

strategies to achieve production goals related to grape quality.

Although less studied than fruiting zone removal, cluster thinning could also modify the

concentration of aroma-active compounds. For example, cluster thinning increased total bound

monoterpenes in Shiraz grapes, but without altering other specific aroma compounds of interest,

such as C13-norisoprenoids (Bureau et al., 2000). Similarly, a study conducted on Chardonnay

Musqué confirmed positive effects of cluster thinning in increasing both free and bound

monoterpene concentrations in grapes (Roberts et al., 2007). Conversely, cluster thinning did not

impact rotundone accumulation in Duras wines, suggesting that responses to cluster removal

might not be uniform across all terpenoids (Geffroy et al., 2014). It is important to note that it is

not the reduction in yield per se which influences development of aroma-active compounds, but

rather the manipulation of the vine crop load. Crop load adjustment via varying degrees of

dormant pruning severity, for example, indicated a strong negative relationship between the

number of viable buds kept (i.e., both the potential vine size and yield) and IBMP in Cabernet

Sauvignon wine grapes (Chapman et al., 2004).

11

The effects of other canopy management practices, such as hedging and shoot thinning,

on the development of aroma compounds have been little explored. A reduction of vegetative

growth and fruit shading obtained through hedging positively influenced free monoterpene

concentrations in grapes at harvest (Reynolds et al., 1996). Similarly, mild shoot thinning

increased berry concentrations of both free and bound monoterpenes, when compared to no

thinning (Reynolds et al., 1994), likely due to an increase of fruit sunlight exposure. While

fruiting zone leaf removal mainly affects canopy microclimate, other management practices such

as shoot thinning influence multiple vine parameters simultaneously (i.e., fruit shading, crop

load, and vegetative growth). Therefore, it is challenging to experimentally decouple the effects

of crop load versus microclimate manipulation on aroma development when applying these

practices.

1.4 Biological, chemical, and sensorial characteristics of rotundone

Rotundone (C15H22O), a sesquiterpenoid ketone of the guaiene family responsible for the

aroma of ‘black pepper’ in wines, has been of the utmost interest to both scientists and the wine

industry since its isolation from Shiraz wine in 2008 (Wood et al., 2008; Mattivi et al., 2011).

Since then it has been identified in several Vitis vinifera varieties, including both red-fruited

(e.g., Shiraz, Mourvèdre, Duras, Gamay, Schioppettino, Pinot Noir, and others) and white-fruited

varieties (e.g., Grüner Veltliner) (Wood et al., 2008; Mattivi et al., 2011; Geffroy et al., 2014;

Logan et al., 2015; Geffroy et al., 2016a). Recently it was extracted from the red-fruited

interspecific Vitis hybrids Noiret and Koshu, in addition to the white-fruited hybrid variety

Muscat Bailey A (Goto-Yamamoto et al., 2015; Takase et al., 2015; Homich et al., 2017). The

presence of rotundone in a variety of both grapes and wines of Vitis vinifera and interspecific

12

Vitis varieties suggests that it may be present in other wine grape varieties and species of

economic importance, in addition to a variety of other sesquiterpene-producing plant species.

Prior to its extraction from grapes and wines, rotundone was originally extracted from

tubers of the purple nutsedge plant (Cyperus rotundus L.) in 1967, from which it derives its name

(Kapadia et al., 1967). It has also been extracted from myriad other non-Vitis plants, including

another Cyperus species, a variety of culinary herbs, such as thyme (Thymus vulgaris L.), basil

(Ocimum basilicum L.), and rosemary (Rosmarinus officinalis L.), and from a variety of plant

species used for incense production, including various Aquilaria spp. (Lam.) and Boswellia

sacra (Fleuck) (Kapadia et al., 1967; Ishihara et al., 1991; Ishihara et al., 1993; Pandey et al.,

2002; Wood et al., 2008; Naef et al., 2011; Niebler et al., 2016). Moreover, rotundone has

recently been identified within and extracted from the peel and juice of grapefruit (Citrus x

paradisi Macfad.), in addition to apples (Malus domestica Bork. nom. illeg. cv. Sun-Fuji) and

commercial mango (Mangifera indica L. cv. Alphonso) puree (Nakanishi et al., 2017a;

Nakanishi et al., 2017b). Lastly, rotundone has been extracted directly from oak wood used for

aging of alcoholic spirits, and is thereby present within many oak-aged spirits in addition to un-

aged tequila, suggesting that it is also present within the blue agave (Agave tequilana F.A.C.

Weber) plant (Genthner, 2014).

Despites its extraction from myriad plant species and plant-based products, analysis of

the physiological processes involved in rotundone synthesis and accumulation has almost

entirely focused on grape-derived rotundone. Rotundone is primarily, if not exclusively,

produced within the berry exocarp during the accumulation period, though non-fruit tissue (i.e.,

rachises, shoots, and leaves) remain sources with high concentrations of potentially extractable

rotundone (Caputi et al., 2011; Takase et al., 2016b; Zhang et al., 2016).

13

In several grape varieties, such as Shiraz, Duras, and Vespolina, accumulation pattern for

rotundone begins immediately following veraison, and increases slowly for about 3 to 4 weeks

after veraison before accumulating rapidly at the end of the ripening period, at about 6 to 8

weeks following veraison (Caputi et al., 2011; Herderich et al., 2012; Geffroy et al., 2014;

Logan, 2015; Zhang et al., 2016). Indeed, in Duras, peak rotundone concentrations are reached

around 44 days following veraison and concentrations thereafter decrease, suggesting that there

is a possible point at which rotundone ceases accumulation or is instead degraded, depending on

ripening period length and environmental conditions (Geffroy et al., 2014).

To understand mechanisms of rotundone synthesis, recent work focused on

transcriptional analysis of the specific phytochrome enzyme responsible for rotundone

production (Takase et al., 2016b) and two specific polymorphisms of a terpene synthase gene

that are related to the production of the precursor, a-guaiene (Drew et al., 2016). Transcription

patterns of the phytochrome responsible for rotundone formation in Shiraz and Merlot berries,

VvSTO2, support the reported trend of post-veraison accumulation and late-season degradation,

as VvSTO2 transcription and rotundone concentrations reached a peak at about 14 weeks in

Shiraz before decreasing afterwards (Takase et al., 2016b). Additionally, VvSTO2 transcription

was higher in a high-rotundone variety when compared to a low-rotundone variety (i.e., Shiraz

versus Merlot), suggesting a possible genetic influence on rotundone synthesis, aside from

regulation via a-guaiene availability (Takase et al., 2016a; Takase et al., 2016b). Aerial

oxidation of a-guaiene to rotundone – via (2R) and (2S)-hydroperoxyguaiene and (2R)- and (2S)-

rotundol intermediaries – can occur, but it is most likely that rotundone is enzymatically oxidized

within the berry exocarp (Huang et al., 2014; Huang et al., 2015; Takase et al., 2016b). Berry-

derived concentrations of a-guaiene exhibit a temporal accumulation pattern in the berry exocarp

14

similar to rotundone (Takase et al., 2016b); it also accumulates to higher concentrations in cooler

conditions (Takase et al., 2016a). Recent elucidation of VvGuaS, an allele of the sesquiterpene

synthase-encoding gene VvTPS24, as the enzymatic source of a-guaiene in Shiraz also highlights

the possibility of a genetic predisposition for high a-guaiene synthesis, and subsequently high

rotundone accumulation, though further research is necessary (Drew et al., 2016)

As an aroma-active sesquiterpene, rotundone is potent at low concentrations. Reported

sensory thresholds for rotundone perception are approximately 16 ng/L in red wine, while in

water it is 8 ng/L, with about 20% of panelists being anosmic (Wood et al., 2008). It is a very

hydrophobic compound with a mostly apolar structure (Log Kow = 4.98); this most likely hinders

its extraction from grape exocarps in the must during red wine fermentation, and leads to only

about 5.0-7.0% of berry rotundone concentrations at harvest being present in the finished wine

(Caputi et al., 2011). It has been suggested that increased ethanol concentrations in wines might

help rotundone extraction (Caputi et al., 2011; Geffroy et al., 2016a), but efforts to increase

Duras wine rotundone concentrations via use of macerating enzymes, thermovinification, and

extended fermentation time all failed to significantly increase rotundone concentrations when

compared to a standard vinification protocol (Geffroy et al., 2017). The inclusion of non-grape

materials into the must, however, can increase rotundone concentrations in wines, as inclusion of

leaves and stem tissue yielded rotundone concentrations in Shiraz wines 6 times greater than that

of the control (Capone et al., 2012).

Rotundone concentrations in Shiraz, Gamay, and Noiret red wines are strongly and

positively associated with perceived pepperiness (Wood et al., 2008; Geffroy et al., 2016a;

Homich et al., 2017). In addition to the noted specific anosmia of rotundone concentrations – up

to concentrations of 4000 ng/L (Wood et al., 2008) – consumer acceptance of peppery notes (i.e.,

15

high rotundone concentrations) in red wines is mixed. Geffroy et al. (2016a) found that those

who preferred Gamay wines with stronger ‘peppery’ notes were professionals and managers that

were more likely to spend more money for a bottle of wine. According to Geffroy et al. (2018),

the panelists’ responses to rotundone was variable and split along 3 main groups: whereas young

panelists with little wine knowledge and tasting experience preferred an unspiked control to

Duras wines with moderate to high concentrations of rotundone, older panelists with strong wine

knowledge and higher tasting experience preferred a moderate amount of rotundone (<46 ng/L)

and rejected wines with concentrations above 30g ng/L. A third cluster of panelists preferred

wines with rotundone concentrations exceeding 94 ng/L, and this cluster was comprised mainly

of managers who both frequently consume wine and appreciate ‘peppery’ notes in wine (Geffroy

et al., 2018). Although these studies are few and relegated to analysis of only Gamay and Duras

wines, they provide critical insights into consumer acceptance and perception of rotundone and

related ‘peppery’ notes in wines, and how variable these response are based on consumer

characteristics.

1.5 Climatic and agronomic influence on rotundone concentrations

Vineyard mesoclimate influences final rotundone concentration in wine grapes; cool-

climate regions typically produce grapes and wines with greater rotundone concentrations than

warm climate regions (Geffroy et al., 2014; Herderich et al., 2015; Geffroy et al., 2016a). In

addition to mesoclimatic influence, viticultural practices such as fruiting zone leaf removal may

alter rotundone concentration and subsequent ‘black pepper’ aroma in wine by modifying cluster

microclimate conditions (Geffroy et al., 2014; Homich et al., 2017). However, the relationship

between fruiting zone microclimate and rotundone development within the grape berries is still

unclear. Leaf removal at veraison lowered rotundone concentration in Duras wines as compared

16

to an undefoliated control (Geffroy et al., 2014), whereas increasing cluster sunlight exposure

from pea-size berry stage to harvest increased rotundone concentration in Noiret grapes and wine

as compared to vines with highly shaded clusters (Homich et al., 2017). Furthermore, Shiraz

berries from cluster portions naturally shaded by the canopy had higher rotundone concentrations

when compared to more sun exposed cluster portions (Zhang et al., 2015a).

Differences in the timing of leaf removal, as well as in climatic conditions (e.g., heat

accumulation, growing season length) and fruit maturity at harvest amongst experimental sites

may explain contradictory results regarding leaf removal. Shiraz fruit exposed to air zone

temperatures above 25 °C had lower rotundone concentrations as compared to fruit exposed to

cooler temperature (Zhang et al., 2015b). Therefore, it is possible that in hot or warm climates,

leaf removal may expose grape berries for long periods of time to excessively high temperatures

that may inhibit rotundone synthesis or accumulation. Conversely, in cooler climates leaf

removal may increase berries temperature to ranges (e.g., < 25 °C) that may facilitate, or perhaps

not affect, rotundone synthesis and accumulation. This would suggest that there are temperature-

based thresholds at which rotundone accumulation is either facilitated or inhibited.

Rotundone concentration can greatly vary within a single vineyard due to soil

characteristics and topography (Scarlett et al., 2014). These spatial variation trends were stable

across seasons, indicating that soil characteristics and topography might have measurable and

consistent, however indirect, effects on patterns of rotundone accumulation (Bramley et al.,

2017). It is unlikely that differences in grapevine vegetative vigor is driving spatial distribution

of rotundone accumulation, but these studies suggest that future work should include soil- and

topography-related characteristics in analyses of rotundone.

17

Despite the existing research focused on probing the relationships between rotundone and

various environmental factors, and to a lesser extent, viticultural variables, few studies have

sought to evaluate these associations in tandem (Geffroy et al., 2014; Zhang et al., 2015a).

Attempting to do so, and in turn creating a hierarchy of variables by degree of influence, would

be beneficial in further understanding the relationships that drive rotundone accumulation and

concentration in berries at harvest.

Advanced statistical methods are critical to assess the effects of interrelated factors on a

given variable. Indeed, multiple linear regression was applied to analysis of bitter pit incidence in

Honeycrisp apples in order to determine which nutrients and tree characteristics best predict the

disease (Baugher et al., 2017), while partial least squares regression was used to determine which

viticultural and environmental factors drive IBMP in Cabernet franc grapes (Scheiner et al.,

2012). Multiple linear regression was also used to assess the various factors that directly and

indirectly affect grape berry mass (Triolo et al., 2018). Further, applications of comparative

metabolomic and transcriptomic approaches to analysis of berry-derived metabolites typically

utilize multivariate statistics, including principal components analysis and various types of

discriminant analysis (Anesi et al., 2015).

Models have been developed for rotundone concentrations in Shiraz grapes and wines:

the percentage of post-veraison degree-hours above 25 °C (i.e., DH25) was used to construct a

predictive model for wine rotundone concentrations, while the Gompertz function was used to

construct a rotundone accumulation model that could predict berry rotundone concentrations and

illustrate accumulation patterns using calendar days since fruit set or cumulative DH above 25 °C

(Zhang et al., 2015b). However, due to the complexity of the reported relationships between

concentrations of aroma compounds and both climatic and viticultural variables, it is therefore

18

important to include both types of variables to validate the significance of previously reported

associations, as well as identify those that may have been overlooked.

The overarching goal of this thesis and the study discussed in Chapter 2 is to develop a

predictive model that assesses the degrees of influence that both viticultural and climatic factors

have on rotundone production in Noiret wine grapes. Given that both regional and microclimatic

factors might affect rotundone concentrations, this thesis aims to define these relationships at

both the vineyard (i.e., mesoclimate) and the fruiting zone level (i.e., microclimate). Specifically,

this thesis addresses two objectives that are central to the study described in Chapter 2: (i) to

identify the key climatic and viticultural variables that influence rotundone concentration in

Noiret grapes ; and (ii) to investigate the role of berry sunlight exposure and temperature on

rotundone accumulation in Noiret grapes at harvest. Until recently, rotundone-based research

exclusively focused on cold-tender Vitis vinifera varieties (e.g., Shiraz, Duras, etc.) that cannot

be reliably grown in many of the winegrowing regions of the northeastern United States. The

recent extraction of rotundone from Noiret (Vitis spp.), a variety released by Cornell University

in 2006 (Reisch et al., 2006), offers an opportunity to explore the dynamics of rotundone

production in another Vitis species and confirm if previously observed relationships with

varieties of Vitis vinifera parentage grown in warmer climates also exist with one of Vitis hybrid

parentage.

Findings from this work can also provide a basis for developing management

recommendations for Noiret wine grape growers throughout the Northeast and Midwest U.S., so

that growers can produce grapes with rotundone concentrations that match consumer

preferences. Moreover, results from this study could be useful for other economically relevant

varieties grown in the northeastern and midwestern United States where rotundone has been

19

identified (i.e., Grüner Veltliner). Taken altogether, this study seeks to advance understanding of

rotundone accumulation, as well as the dissemination of information related to rotundone

concentration dynamics in Noiret wine grapes that can be used to assist the growth of the wine

industries throughout the cool-climate wine regions of the United States.

20

Chapter 2: Weather conditions during fruit ripening, vine nutrient status, and vine size

collectively influence rotundone concentrations in cool-climate Noiret wine grapes

2.1 Introduction

Many chemical classes of plant secondary metabolites produced within grape berries are

aroma-active (Waterhouse et al., 2016). Amongst these aroma molecules are a few that are

classified as impact compounds for their ability to impart a specific aroma to a wine or drive its

varietal character. Aroma impact compounds and their interactions are an essential component of

wine quality as they can contribute to pleasant or unpleasant sensory wine perception. A well-

documented example is 3-isobutyl-2-methoxypyrazine (IBMP), a secondary metabolite present

in wines made from several wine grape varieties, including Sauvignon blanc and Cabernet Franc

(Scheiner et al., 2012; Robinson et al., 2014). IBMP contributes an aroma of ‘green bell pepper’

which, when present in high concentration, is generally considered an undesirable wine attribute

(Scheiner et al., 2012; Robinson et al., 2014).

In 2008, the sesquiterpene rotundone (C12H22O) was identified as the impact compound

responsible for the key ‘black pepper’ aroma of Shiraz wines (Wood et al., 2008). Since its first

extraction from Australian Shiraz grapes and wine, it has been identified and extracted from

other red-fruited Vitis vinifera varieties across many wine-producing regions (i.e., Duras and

Gamay in France, Mourvèdre and Durif in Australia, and Vespolina from Italy; Wood et al.,

2008; Caputi et al., 2011; Geffroy et al., 2014) and to a lesser extent in white-fruited V. vinifera

varieties (Riesling and Grüner Veltliner in Austria, Italy, and Slovakia; Caputi et al., 2011;

Herderich et al., 2012). Most recently, rotundone was extracted from grapes and wine of a red-

fruited Vitis interspecific hybrid variety, Noiret (Homich et al., 2017).

21

Rotundone is a strong volatile impact compound, with a detection threshold of 16 ng/L in

red wine (Wood et al., 2008). It slowly accumulates mainly within the berry exocarp beginning

at veraison, and within a few weeks accumulates at a quicker rate before reaching a

concentration plateau late in fruit ripening phase (Geffroy et al., 2014; Zhang et al., 2015b;

Zhang et al., 2016). Wine consumers have positive perceptions of rotundone in most cases

(Geffroy et al., 2018). According to Geffroy et al. (2016a), ‘peppery’ wines are mainly

appreciated by wine connoisseurs, who are typically more willing to pay higher prices for a

bottle of wine when compared to consumers with less wine tasting experience and who drink

wine at a lesser rate. Although two sensory studies have reported varying degrees of anosmia at

20% (Wood et al., 2008) and 31% (Geffroy et al., 2018) of panelists, the presence of rotundone

in high concentrations is nonetheless economically important for growers of varieties that

include rotundone, like Shiraz or Duras.

The primary determinant of rotundone concentration in finished wine is the concentration

present in the grapes at harvest (Geffroy et al., 2014; Homich et al., 2017). Therefore, several

studies have focused on identifying factors responsible for rotundone accumulation in the fruit to

predict how seasonal or vineyard environmental conditions as well as grower practices might

influence the ‘peppery’ intensity of the resulting wine (for examples, see: Geffroy et al., 2014;

Zhang et al., 2015a; Zhang et al., 2015b; Bramley et al., 2017; Homich et al., 2017) . Similar to

other aroma impact compounds (e.g., IMBP, monoterpenes, and C13-norisoprenoids), rotundone

accumulation in grapes depends upon climatic factors. Rotundone accumulation in Vitis vinifera

and Vitis hybrid varieties was positively associated with cool temperatures (i.e., cool vintages or

cool sites) (Herderich et al., 2015; Homich et al., 2017). Additional research revealed that the

relationship between rotundone concentration and temperature is spatially structured within an

22

individual grape cluster. Cooler portions of the cluster tend to accumulate higher concentrations

of rotundone than those exposed to higher temperatures (Zhang et al., 2015a).

The relationships between rotundone and other weather parameters, such as precipitation

and solar radiation, were also investigated. Grapes grown in shade, whether due to vineyard row

orientation, position of a cluster within the canopy or of a berry within an individual cluster, had

higher concentration of rotundone when compared to grapes grown with higher solar exposure

(Herderich et al., 2015). The increased rotundone concentrations were attributed to the direct

effect of solar radiation, to the increased temperatures of berries with high sun exposure, or

perhaps their combination. Moreover, within the same grape variety, rotundone concentration at

harvest was higher during a wetter season, as compared to a drier season (Geffroy et al., 2014).

Less clear is the influence of cultural practices on rotundone accumulation. The timing

and severity of fruiting zone leaf removal, a popular canopy management strategy, has been the

most studied, because it influences fruit sun exposure and temperature. However, the

manipulation of the fruiting zone microclimate has yielded contrasting effects with regards to

rotundone accumulation in the fruit, likely because of the different weather conditions of the

experimental sites and phenological timing of leaf removal. Additionally, most studies have

explored relationships between rotundone and a given weather or plant variable (e.g., ambient

temperature, solar radiation, vine water status) without incorporating multiple variables into a

single, cohesive analysis. However, combinations of environmental and viticultural factors might

operate in tandem to determine rotundone concentration in the fruit and ‘peppery’ intensity of the

wine. Understanding the relative importance of these variables on rotundone concentration

within a given system may help clarify which vineyard sites or viticultural management methods

are more conducive to producing wines with a desired level of pepperiness.

23

This study sought to address this knowledge gap and incorporate a multitude of

environmental, viticultural, and physiological data into a single study to assess which variables

had the greatest influence on rotundone concentrations within Noiret wine grapes. Given that

rotundone was only recently extracted from Noiret, this study also aims to further our

understanding of these relationships within the context of cool-climate Vitis hybrid production.

The objectives of this study were twofold: (1) to identify the key climatic and viticultural

variables that influence rotundone concentration in Noiret grapes using 7 vineyards with varying

weather conditions; and (2) to investigate the relationships between fruit sunlight exposure, berry

temperature, and rotundone accumulation in Noiret grapes at harvest.

2.2 Methods & Materials

2.2.1 Experimental Design

The study was conducted in 2016 and 2017 at seven Noiret (Vitis hybrid cross of

NY65.0467.08 and Steuben) vineyards located in Pennsylvania (n = 3) and New York State (n =

4), U.S. (Figure 1). The three Pennsylvania vineyards included three commercial vineyards

located in State College (Site 1), Falls (Site 2), and North East (Site 3; Table 1). In New York

there were two commercial vineyards in Portland (Site 4) and Branchport (Site 5), and two

research vineyards at the Cornell University AgriTech in Geneva (Site 6; Site 7). Sites 5, 6 and 7

were in the Finger Lakes American Viticultural area (AVA).

The vines were trained to two different training systems (Table 1). At site 1 all the vines

were trained to a bilateral high wire cordon (HWC) at a height of 1.8 m, whereas at sites 2, 3,

and 4 all the vines were trained to a bilateral cordon vertical shoot positioned (VSP) system at a

height of 0.9 to 1.1 m. Both training systems were used at sites 5 and 6 in separate vineyard

blocks, with a section of the vineyard trained to HWC and another to VSP. Further information

24

regarding vineyard age, vine and row spacing, rootstock, soil series classification, and basic soil

texture characteristics is summarized in Table 1. Disease, pest, and canopy management

practices (e.g., shoot thinning, shoot training, hedging) were performed by the grower cooperator

in accordance with standard commercial practices for hybrid Vitis cultivars in the eastern U.S.

(Wolf, 2008).

Figure 1. Map of vineyards chosen for the study. A dark red circle was imposed at the geographical coordinates of each study site, with the two circles representing sites 5 and 6 overlapping due to close geographical proximity.

25

Table 1. Location and vineyard information for the Noiret sites used in the multivariate analysis. Site Treatment Location Rootstock Spacing

(m/row x m/vine)

Traininga system

Vineyard ageb

Soil seriesc

1 C State College, PA 101-14 Mgt 1.83 x 2.44 HWC 10 Hublersburg silt loam 1 LR State College, PA 101-14 Mgt 1.83 x 2.44 HWC 10 Hublersburg silt loam 2 C Falls, PA Own-rooted 1.83 x 2.44 VSP 15 Lordstown channery silt loam 2 LR Falls, PA Own-rooted 1.83 x 2.44 VSP 15 Lordstown channery silt loam 3 C North East, PA Own-rooted 1.83 x 2.44 VSP 7 Chenango gravelly silt loam 3 LR North East, PA Own-rooted 1.83 x 2.44 VSP 7 Chenango gravelly silt loam 4 C Portland, NY Own-rooted 1.83 x 2.44 VSP 16 Chenango gravelly loam 4 LR Portland, NY Own-rooted 1.83 x 2.44 VSP 16 Chenango gravelly loam 5 C Branchport, NY 101-14 Mgt 1.83 x 2.44 HWC 7 Valois gravelly silt loam 5 LR Branchport, NY 101-14 Mgt 1.83 x 2.44 HWC 7 Valois gravelly silt loam 5 C Branchport, NY 101-14 Mgt 1.83 x 2.44 VSP 14 Langford-Erie channery silt loam 5 LR Branchport, NY 101-14 Mgt 1.83 x 2.44 VSP 14 Langford-Erie channery silt loam 6 C Geneva-RS, NYd Own-rooted 2.70 x 3.60 HWC 9 Honeoye loam 6 LR Geneva-RS, NY Own-rooted 2.70 x 3.60 HWC 9 Honeoye loam 6 C Geneva-RS, NY Own-rooted 2.70 x 3.60 VSP 9 Honeoye loam 6 LR Geneva-RS, NY Own-rooted 2.70 x 3.60 VSP 9 Honeoye loam 7 C Geneva-CN, NYe 101-14 Mgt 2.70 x 3.60 HWC 10 Honeoye loam 7 LR Geneva-CN, NY 101-14 Mgt 2.70 x 3.60 HWC 10 Honeoye loam

aHWC: High-wire cordon; VSP: Vertical shoot-positioned system. bVineyard age determined as number of years from planting to the beginning of the study (2016). cData sourced from the USDA National Resources Conservation Service (NRCS) Web Soil Survey, https://websoilsurvey.sc.egov.usda.gov. dVineyard located at Cornell University AgriTech Research South (RS) farm. eVineyard located at Cornell University AgriTech Crittenden (CN) farm.

26

At each vineyard, two panels (i.e., two sections of two-post spaces) of 3-4 contiguous

vines (1.83-2.70 m long row each) were selected for data collection. The two experimental units

were randomly assigned to either a control (C; fruiting zone non-defoliated) or fruiting zone leaf

removal treatment (LR). Noiret vines typically exhibit high vegetative growth with highly

shaded clusters (Vanden Heuvel et al., 2013). Fruiting zone leaf removal was used to maximize

the range of temperatures and cluster sun exposure across sites to better assess relationships

between rotundone concentration and these micrometeorological factors. Our goal was not to

assess differences between C and LR treatments, as the treatments were not replicated at any site.

Fruiting zone defoliation was imposed pre-veraison at Eichhorn-Lorenz (E-L) phenological stage

31, defined as “berry pea-size stage” (Coombe, 1995). Leaves were removed from each shoot

within the fruiting zone, from the basal node to that above the distal cluster. Leaves were

removed multiple times during both seasons, as a previous study suggested that pre-veraison leaf

removal coupled with maintained fruiting zone sun exposure may increase rotundone

concentration in Noiret fruit, as compared to highly shaded fruit harvested from vigorous vines

(Homich et al., 2017). Fruiting zone defoliation was implemented on the same experimental

vines during the 2016 and 2017 seasons.

2.2.2 Site-specific weather conditions

Vineyard air temperature, rainfall, and photosynthetically active radiation (PAR) were

recorded at 15-minute intervals with HOBOÒ weather sensors and dataloggers (Onset Computer

Corporation, Bourne, MA) at sites 1, 2, 3, and 4, starting on June 23 and ending on October 31 in

2016, and starting on May 1 and ending on October 31 in 2017. At the other sites, the same data

were obtained from Network for Environment and Weather Applications (NEWA) weather

27

stations (http://newa.cornell.edu). A NEWA weather station was located at site 5 and within 0.71

and 1.57 km from site 6 and 7, respectively.

As HOBOÒ weather stations only measured solar radiation from 400 to 700 nm (PAR),

regression between PAR and solar radiation was calculated. Briefly, a wide-spectrum (measuring

a total wavelength range of 300-1100 nm) silicon pyranometer was added to the HOBOÒ

weather station at site 1 to record solar radiation and PAR concurrently. The 30-minute average

PAR was linearly related to solar radiation (y = 0.5099x – 0.0302; r2= 0.96; n = 477) and used to

convert PAR values to solar radiation (µmol/m2/s to W/m2) for the four HOBOÒ weather

stations. Concurrently, NEWA-sourced solar radiation data was converted from Langley units to

W/m2 to have comparable values across all sites.

Several mesoclimatic (i.e., site specific) parameters were calculated for each site (Table

2). Seasonal growing degree days (GDD) were calculated from May 1 to harvest using 10 °C as a

baseline (GDD = [(maximum temperature + minimum temperature)/2] – 10). Additionally,

cumulative GDD were calculated from the onset of veraison to harvest (GDDv) for each site.

To assess seasonal solar radiation, described here as ‘vineyard solar-hours,’ total solar-hours

(SH800), expressed as the total number of hours that exceed 800 W/m2 (average hourly solar

radiation > 800 W/m2) were calculated for each site from May 1 to harvest for 2016 and 2017.

Rationale for using 800 W/m2 as a threshold is based on the reasoning that the value indicates

full-sun ambient vineyard conditions. Total solar-hours were also calculated from veraison to

harvest (SH800v).

2.2.3 Fruiting zone weather conditions

At each site wireless temperature data loggers (iButton Fob, Model DS9093Fl,

Embedded Data Systems, Lawrenceburg, KY) were used to record air temperature at 20-minute

28

intervals in the fruiting zone of C and LR vines throughout the 2016 and 2017 seasons. Two

sensors were placed within each experimental unit at the trellis wire closest to the fruiting zone,

and averaged data from both sensors in each experimental unit was calculated. Post-veraison

berry flesh temperature was measured at site 1 from September 16 to October 5, 2017, from E-L

36 “berry with intermediate Brix values” to E-L 38 “berries harvest-ripe” on two randomly

chosen clusters from each experimental unit. For each cluster, five 12.7 mm hypodermic

thermocouple probes (Model HYP1-30-1/2-T-G-60-SMP-M, Omega Engineering, Stamford,

CT) were inserted into a berry at different locations within a cluster (top-east, top-west, mid-

west, bottom-east, bottom-west). All 20 thermocouples were connected to a data logger unit

(CR6, Campbell Scientific, Logan, UT) and berry flesh temperature was continuously measured

and logged at 20-minute intervals throughout the measurement period. Linear regression was

used to fit berry flesh temperature data to the air temperature data for both the LR and C

treatments (LR: y = 1.2034x – 2.4302, r2 = 0.98, n = 96; CON: y = 0.6802x + 2.3627, r2 = 0.98, n

= 96). The regression equations were used to estimate berry temperature for all the other sites for

both seasons.

Post-veraison vineyard thermal time (sensu Zhang et al., 2015b) was calculated as degree

hours (DH) index based on estimated berry temperatures. The percentage of post-veraison

average hourly temperatures that fell within 10-15 °C, 15.1-20 °C, 20.1-25 °C, 25.1-30 °C, 30.1-

35 °C, 35.1-40 °C, and >40.00 °C were calculated (DH10, DH15, DH20, DH25, DH30, DH35, and

DH40, respectively). Each DH index was calculated as:

DHx = [(number of hours between T1 – T2 from veraison to harvest / number of hours

between veraison and harvest) *100];

29

where x is the base temperature of the DH range, T1 is the lower threshold temperature, and T2

was the upper threshold temperature. Degree hours were calculated to confirm a previously

reported negative relationship between rotundone concentration in grapes and fruiting zone air

temperatures between 25 and 30 °C, in addition to exploring the nature of this relationship across

other DH ranges (Zhang et al., 2015a). Due to a large period of missing data, it was not possible

to calculate DH ranges for sites 1, 2, 3, 4, and 5 in 2016. Values were calculated for sites 6 and 7

in 2016, and all sites in 2017.

Enhanced point quadrat analysis (EPQA; Meyers and Vanden Heuvel, 2008) was

performed three times per season per site to assess canopy density and fruiting zone sunlight

penetration. Each year, EPQA was measured when leaves were removed for the first time (E-L

31, “berry pea-size stage”), again at 50% veraison (E-L 35, “veraison”), and at post-veraison

between E-L 36 “berries with intermediate Brix values” and E-L 37 “berries not quite ripe”

stages. Point Quadrat Analysis (PQA) was performed by inserting a thin metal rod into the

grapevine fruiting zone at 20 cm intervals perpendicular to the vine row for a total of 36 insertion

points per experimental unit (Smart and Robinson, 1991). PQA analysis was coupled with PAR

measured within 2 hours of solar noon on the same day, given full-sun conditions, using a LI-

250A quantum ceptometer (LI-COR Bioscience, Lincoln, NE). Four PAR measurements were

taken per vine within each experimental unit: ambient PAR was measured first within the row

with the ceptometer raised and facing skyward, followed by three within-canopy measurements

at different orientations (directly upwards, or 0° relative to the vertical canopy, 45° towards the

clusters, and lastly 45° away from the clusters) to capture the full variability of within-canopy

PAR. Within-canopy values were divided by ambient values to calculate the ratio of PAR

penetrating the canopy.

30

Characteristics related to canopy density and fruit sunlight exposure were then analyzed

using Canopy Exposure Mapping Tools (v. 1.7, freeware from J.M. Meyers, Cornell University,

Ithaca, NY; Meyers and Vanden Heuvel, 2008). The software was used to calculate leaf layer

number (LLN), occlusion layer number (OLN), percent interior clusters (PIC), percent interior

leaves (PIL), cluster exposure layer (CEL), leaf exposure layer (LEL), and leaf and cluster flux

availability (LEFA and CEFA, respectively).

2.2.4 Vine vegetative growth and yield components

At each commercial vineyard, harvest date was determined by the grower cooperator. At

harvest, all the clusters were weighed and counted; average cluster weight was calculated as the

total yield divided by the total number of clusters. Twenty clusters were randomly collected from

each experimental unit at harvest, stored at -20 °C, and later used for berry weight, chemical

composition, carbon isotope composition, and rotundone quantification analyses. Pruning weight

was measured during the dormant season, between February and March 2017 and 2018. All yield

and pruning weights were measured using a hanging scale with a 0.01 kg accuracy (Pelouze

7710, Rubbermaid, Inc., Huntersville, NC). Crop load was calculated as yield divided by pruning

weight (Ravaz index). In 2016, an early commercial harvest at site 5 resulted in loss of yield and

related data for the experimental VSP-trained vines; additionally, pruning data were lost for 2017

at site 2 due to mixing of pruned canes between C and LR vines.

2.2.5 Vine nutrient and water status

To assess grapevine nutrient status, 30 leaf petioles were randomly collected from each

experimental unit at veraison in 2016 and 2017. Petioles were sampled from the youngest,

mature, healthy leaves of primary bearing shoots, from both sides of the canopy (Wolf, 2008).

Samples were dried at 60 °C for 48 hours and submitted to The Pennsylvania State University

31

Agricultural Analytical Services Laboratory for macronutrient (N, P, K, Mg, Ca, S) and

micronutrient (Mn, Fe, Cu, B, Zn) analyses by acid digestion and ICP elemental analysis (Huang

et al., 1985).

A 200-berry sample was randomly taken from the frozen clusters collected at harvest for

each experimental unit to assess vine water status via carbon isotope composition (δ13C) analysis

(Gaudillère et al., 2002). Plant tissue δ13C is reported across a gradient of negative values, with

more negative values correlating strongly with higher water availability and pre-dawn water

potentials and less negative values with lower water availability and an increasing likelihood of

drought stress (Bchir et al., 2016). Among the plant tissues analyzed, δ13C of berries at harvest

showed the highest correlation with grapevine water status (Bchir et al., 2016). The 200-berry

sample was split into two subsamples, oven-dried for six days at 60 °C, frozen with N2 gas,

ground into a powder, and submitted to the Cornell University Stable Isotope Laboratory for EA-

IRMS analysis. The results were expressed as ‰ δ13C, or the difference in carbon isotope

composition of the grape sample relative to that of the Pee Dee Belemnite internal standard.

Carbon isotope composition was calculated as:

δ13C = [(Rg – Rpdb) / Rpdb] x 1000;

where Rg = 13C/12C ratio of the grape sample and Rpdb = 13C/12C ratio of the Pee Dee Belemnite

standard.

2.2.6 Fruit chemistry and rotundone analysis

In both years and for each experimental unit, fruit chemical composition (total soluble

solids [TSS], pH, and titratable acidity [TA]) was measured on a 100-berry sample selected from

the frozen clusters collected at harvest. Frozen berry samples were thawed in a water bath at 60

°C and crushed for juice chemistry analysis. Total soluble solids were measured using a hand-

32

held refractometer (Master, Atago USA, Inc., Bellevue, WA) and pH using a benchtop pH-meter

(Orion Star A111, Thermo Fisher Scientific, Waltham, MA). Titratable acidity was assessed

using an autotitrator (G20, Mettler Toledo, Columbus, OH) on a 10 mL juice sample titrated to

an endpoint pH of 8.2 with a 0.1 M NaOH solution. Average berry weight was calculated using a

200-berry sample taken from the frozen harvested clusters.

Berry processing for rotundone extraction and analysis followed the protocol used by

Homich et al. (2017). A berry sample of 125 g was taken from each frozen grape sample per

experimental unit; the berries were deseeded, flash-frozen using N2 gas, and ground into a fine

powder using a kitchen blender (Sunbeam Products, Inc., Boca Raton, FL). Each sample was

then transferred to a glass jar and the headspace purged with argon gas for 30 seconds before

sealing the jar with the cap. Jars were stored at -80 °C until rotundone extraction. Rotundone was

extracted from 25 g of frozen grape powder spiked with 100 µL of d5-rotundone (516 µL/L in

ethanol) internal standard. Fifty mL of acetone were added to the spiked grape powder and the

solution was orbitally shaken at 225 rpm for one hour. Extracts were vacuum-filtered using a

0.10 µm glass fiber filter paper (Pall Corp., Port Washington, NY) and placed into a N blow-

down evaporator (RapidVap Vertex+ Dry Evaporator, Labconco Corp., Kansas City, MO) to

allow for evaporation of the sample solvent at 40 °C until an aqueous residue of ~20 mL

remained. The residue was diluted to 85 mL with model wine (12% ethanol, 5 g/L tartaric acid,

pH 3.2), split into two Teflon fluorinated ethylene propylene centrifuge tubes (Nalgene Nunc

International Corp., Rochester, NY) and centrifuged for 12 minutes at 4000 rpm (Homich et al.,

2017).

Solid phase extraction (SPE) was performed on the supernatant according to a modified

version of Siebert et al. (2008) protocol using a TELOS 12-position manifold (Kinesis Inc.,

33

Vernon Hills, IL) and 6 mL SPE cartridges (Phenomenex Strata styrene-divinylbenzene [SDB-L]

500 mg/6 mL tubes, Phenomenex, Torrance, CA). All SPE tubes were first conditioned with one

cartridge volume (about ~5-6 mL) of n-pentane/ethyl acetate (4:1) solution, one cartridge volume

of methanol, and a final cartridge volume of model wine. Following conditioning, the aqueous

berry sample extracts were loaded onto the SPE tubes and washed with a cartridge full of

ultrapure water and a final wash of n-pentane (2 mL, discarded). Elution was performed using

two 5 mL aliquots of n-pentane/ethyl acetate solution (9:1) per SPE tube and the eluted extracts

were collected in two 10 mL glass culture tubes per sample. A N blowdown evaporator was used

to evaporate the sample solvent until dryness. The dry residues were reconstituted in 0.5 mL of

pure ethanol (0.25 mL per sample tube), followed by an addition of 6.5 mL of ultrapure water

(3.25 mL per sample tube). The reconstituted extracts were transferred to 10 mL GC vials with

magnetic screwcaps and frozen at -80 °C until rotundone analysis.

To achieve high resolution separation between rotundone present in the grape sample and

the added, deuterated internal standard, and avoid peak interferences within chromatographic

analysis, rotundone analysis was conducted via solid phase microextraction multidimensional

gas chromatography-mass spectrometry (SPME-MDGC-MS) by the Australian Wine Research

Institute (AWRI, Glen Osmond, SA) according to the protocol outlined in Geffroy et al. (2014).

2.2.7 Data analysis and multivariate model construction

Data analysis was performed using SAS statistical software (v. 9.4, SAS Institute, Cary,

NC). Relationships between all measured variables (Table 2) were first evaluated visually using

SAS’s PROC GPLOT; PROC CORR was then used to assess linear correlations between

rotundone concentration and the 21 variables presented in Tables 3, 4, and 5. Variables that were

correlated with rotundone (Pearson’s coefficients > 0.5, p < 0.05) were again plotted using

34

PROC GPLOT to assess linearity. PROC REG was used to develop a series of multiple linear

regression models and subsequently identify a subset of variables that could be used for a

predictive model.

Multiple linear regression models were developed with combined data for 2016 and

2017, as the number of observations per year (n =16 for 2016, n =18 for 2017) were insufficient

for regression analysis by year. Models were first constructed using three selection options,

including FORWARD selection (a=0.1), BACKWARD elimination (a=0.1), and STEPWISE

selection (a=0.1). The resulting models were compared, considering the coefficient of

determination (r2), the adjusted r2, Mallow’s conceptual predictive criterion (Cp), and mean

square error (MSE).

The RSQUARE option in PROC REG was used to request all possible regressions, and

all possible combinations of variables were evaluated using r2, adjusted r2, Cp, MSE, Bayesian

information criterion (BIC), and Akaike information criterion (AIC). A set of candidate models

were selected to evaluate model diagnostics with the R, INFLUENCE, VIF, and COLLINOINT

options in PROC REG. Influence statistics generated by the INFLUENCE option (Hat Diagonal

statistic, CovRatio Statistic, DFFITS, DFBETAS, and PRESS) were used to test for influential

observations within the data set. Variance inflation indices were requested using the VIF option.

The COLLINOINT option generated collinearity diagnostics, including eigenvalues, condition

indices, and eigenvalue-associated variance values for each variable. Using this method, a

parsimonious final predictive model (i.e., a predictive model with high explanation and the

fewest necessary variables) was selected for predicting rotundone concentration in Noiret grapes.

The same statistical approach was used to identify the fruiting zone weather variables

listed in Table 2 that had the greatest influence on rotundone concentrations at harvest. Results

35

from this regression analysis are not intended for predictive use, but for determining which

micrometeorological conditions (e.g., continuous fruiting zone berry temperature or fruiting zone

sun exposure measured three times) had the strongest influence on rotundone concentrations.

36

Table 2. Vine and climate measurements recorded during 2016 and 2017 to predict rotundone concentration in the fruit at harvest.

Vine Metrics Climate Production Metrics Nutrient and Water Statusa Mesoclimateb Microclimatec

Yield Nitrogen Temperature Air temperature Cluster number Phosphorous GDD Berry temperature Cluster weight Potassium GDDv CEFA

Berry weight Magnesium Rainfall LEFA

Pruning weight Calcium Rainfallv DH10

Crop load Berry δ13C Solar radiation DH15

Juice soluble solids SH800 DH20

Juice pH SH800v DH25

Juice titratable acidity DH30 DH35 DH40

aBerry δ13C = Ratio of 13C:12C measured in grape berries at harvest. bGDDv = Veraison-to-harvest growing degree days; Rainfallv = Veraison-to-harvest rainfall; SH800 = Seasonal solar-hours above 800 W/m2; SH800v = Veraison-to-harvest solar-hours above 800 W/m2. cCEFA = Cluster exposure flux availability; LEFA = Leaf exposure flux availability; DH10 = Percentage of veraison-to-harvest degree-hours between 10.1 and 15 °C; DH15 = Percentage of veraison-to-harvest degree-hours between 15.1 and 20 °C; DH20 = Percentage of veraison-to-harvest degree-hours between 20.1 and 25 °C; DH25 = Percentage of veraison-to-harvest degree-hours between 25.1 and 30 °C; DH30 = Percentage of veraison-to-harvest degree-hours between 30.1 and 35 °C; DH35 = Percentage of veraison-to-harvest degree-hours between 35.1 and 40 °C; DH40 = Percentage of veraison-to-harvest degree-hours greater than 40 °C.

37

2.3 Results

2.3.1 Site-specific weather conditions

Overall, the 2016 growing season was warmer and drier than the 2017 season (Table 3).

Seasonal heat accumulated from May 1 to the day of grape harvest was higher in 2016 for all the

seven sites evaluated; GDD averaged 1512 ± 41 in 2016 and 1386 ± 64 in 2017. There was less

variation in heat accumulation between sites in 2016 than in 2017: in 2016, GDD varied from

1438 (site 5) to 1551 (site 3), while in 2017 GDD ranged from 1292 (site 1) to 1468 (site 4). In

2016, the sites with the lowest SH800 were sites 6 and 7 (259), while that with the highest SH800

was site 3 (1043; Table 3). In 2017, sites with the lowest and highest SH800 were site 5 (712) and

site 4 (1054), respectively. Most sites exhibited low variation in SH800 between the two years

except for sites 5, 6, and 7, located in the Finger Lakes AVA: site 5 had 37% higher SH800 in

2017 when compared to 2016, while SH800 was almost four times higher at sites 6 and 7 in 2017

compared with 2016.

Although 2017 was overall a cooler season, GDD from veraison to harvest were higher in

2017 as compared to 2016 in six out of the seven sites (Table 3). Inter-site variation in GDDv

was lower in 2016 compared to 2017. GDDv ranged from 249 (site 5) to 360 (site 1) in 2016 and

from 253 (site 1) to 457 (site 4) in 2017. Similarly, the veraison-to-harvest period was sunnier in

2017 than in 2016, except for sites 1 and 2, where site 1 was cloudier in 2017 and site 2 was

similar across both years (Table 3). In 2016 there was greater inter-site variation in SH800v as

well; SH800v ranged from 0 (sites 6 and 7) to 244 (site 1) in 2016, and 120 (site 5) to 284 (site 4)

in 2017.

Cumulative rainfall was higher in 2017 for all sites except for sites 3 and 4 (Table 3).

Seasonal rainfall ranged from 237 (site 6) to 476 mm (site 2) in 2016; in the following year it

38

ranged from 302 (site 3) to 636 mm (sites 6 and 7). The sites located in the Finger Lakes AVA

region (sites 5, 6, and 7) had the lowest rainfall in 2016, but the highest in 2017.

2.3.2 Fruiting zone weather conditions

The DH range with the highest proportion of hours between veraison and harvest was

within the 15.1 to 20 °C range (DH15) for all experimental units except for one LR unit at site 6

(HWC) in 2016, the C experimental unit at site 7 in 2016, and the LR unit at site 4 in 2017. The

DH range with the highest proportion of hours for the first unit was DH10, while DH10 and DH15

had the same percentage of hours for the second unit , and DH20 for the third unit, respectively

(Table 4). All C experimental units were cooler than the LR counterparts at each site, as they had

Table 3. Weather data measured for all the experimental sites in 2016 and 2017. Year Site GDD GDDva Rainfall

(mm) Rainfallvb

(mm) SH800

c SH800vd

2016 1 1526 360 319 86 868 244 2 1474 312 475 120 1026 213 3 1551 278 382 110 1043 169 4 1547 307 395 162 1011 189 5 1438 249 287 73 520 68 6 1523 293 237 96 259 0 7 1528 291 238 97 259 0

2017 1 1292 253 495 35 873 176 2 1417 341 512 81 949 213 3 1401 375 301 16 974 234 4 1468 457 384 197 1054 284 5 1305 271 515 84 712 120 6 1405 320 635 79 995 191 7 1412 327 635 79 996 216

aGDDv = Veraison-to-harvest GDD. bRainfallv = Veraison-to-harvest rainfall. cSH800 = Seasonal solar-hours above 800 W/m2. dSH800v = Veraison-to-harvest solar-hours above 800 W/m2.

39

higher DH10 and DH15 values and lower DH25, DH30, DH35, and DH40 for all the units that have

data (Table 4).

Overall, within-treatment variation tended to be lower at higher temperatures (Table 4).

At low temperatures (DH10) within-treatment variation across sites was greater for C than LR

units (19.41% to 31.50% for C, vs. 14.63% to 24.68% for LR). Similarly, within-treatment

variation at the higher temperature range (DH30) tended to be greater for C than LR units (0 to

4.97% for C, vs. 5.83 to 9.32% for LR).

Across all sites and years, except for the C unit at site 2 in 2017, fruit was exposed for

fewer hours to temperatures above 30 °C as compared to temperatures within the 25.1 - 30 °C

range (DH40 < DH35< DH30 < DH25). DH35 was below 1% for all the C experimental units. DH35

values for the LR experimental units ranged from 3.22% (site 6 VSP) to 3.65% (site 7) for 2016

and 1.37% (site 4) to 4.57% (site 1) for 2017. DH40 was negligible or below 1% across all sites,

indicating that post-veraison berry temperatures rarely exceeded 40 °C.

As expected, the percentage of ambient photon flux intercepted by both clusters (CEFA)

and leaves (LEFA) was greater for the LR units as compared to the C for all sampling dates and

both years (Table 5). During the ripening period, CEFA ranged from 0% (site 1) to 30% (site 3)

for the C, and from 34% (site 5 HWC) to 83% (site 2) for the LR in 2016. In 2017, CEFAr

varied from 0% (site 5 HWC) to 23% (site 4) for the C and from 23% (site 5 HWC) to 82% (site

4) for the LR units (Table 5).

40

Table 4. Fruiting zone temperature (DH) metrics for 2016 and 2017. Year Site Treatmenta DH10

b (%)

DH15 (%)

DH20 (%)

DH25

(%) DH30

(%) DH35

(%) DH40

(%)

2016 1 C NAc NA NA NA NA NA NA 1 LR NA NA NA NA NA NA NA 2 C NA NA NA NA NA NA NA 2 LR NA NA NA NA NA NA NA 3 C NA NA NA NA NA NA NA 3 LR NA NA NA NA NA NA NA 4 C NA NA NA NA NA NA NA 4 LR NA NA NA NA NA NA NA 5 C NA NA NA NA NA NA NA 5 LR NA NA NA NA NA NA NA 5 C NA NA NA NA NA NA NA 5 LR NA NA NA NA NA NA NA 6 C 29.47 31.04 18.85 10.93 4.97 0.62 0.00 6 LR 24.68 23.95 18.95 11.56 7.60 3.33 0.31 6 C 28.85 30.62 20.00 9.68 3.22 0.10 0.00 6 LR 22.70 24.58 17.60 12.50 6.66 3.22 0.20 7 C 31.50 31.50 20.22 9.24 2.84 0.00 0.00 7 LR 24.08 25.40 17.98 10.97 6.91 3.65 0.30

2017 1 C 21.01 38.13 18.30 11.86 0.67 0.00 0.00 1 LR 15.93 29.66 15.93 15.93 7.96 4.57 0.16 2 C 27.78 39.24 17.74 5.82 0.53 0.53 0.00 2 LR 20.87 28.58 18.27 10.39 9.32 2.24 0.53 3 C 19.41 43.49 24.89 10.56 0.00 0.00 0.00 3 LR 14.63 32.92 21.54 16.26 9.14 1.52 0.00 4 C 20.93 44.13 20.85 9.13 0.08 0.00 0.00 4 LR 14.87 21.12 21.50 14.71 8.64 1.37 0.00 5 C 30.04 33.11 16.11 8.44 0.54 0.00 0.00 5 LR 23.35 25.54 14.91 8.99 7.23 1.86 0.00 5 C 29.27 34.10 16.11 7.45 0.43 0.00 0.00 5 LR 23.68 26.09 15.57 9.10 6.03 1.53 0.00 6 C 27.22 36.38 15.46 7.96 0.64 0.00 0.00 6 LR 20.74 28.61 14.35 9.72 5.83 1.85 0.00 6 C 26.11 37.31 16.75 7.87 0.46 0.00 0.00 6 LR 19.07 29.62 16.48 9.25 6.01 1.48 0.00 7 C 27.06 36.01 15.36 9.14 1.09 0.00 0.00 7 LR 19.29 29.07 16.36 10.05 6.48 1.55 0.00

aC: Control; LR: fruiting zone leaf removal. bPost-veraison degree-hours calculated within temperature ranges of 10-15 °C, 15.1-20 °C, 20.1-25 °C, 25.1-30 °C, 30.1-35 °C, 35.1-40 °C, and >40 °C, respectively. cData unavailable due to temperature sensor error.

41

Table 5. Fruiting zone solar exposure metrics (LEFA; CEFA) for 2016 and 2017. Year Site Treatmenta LEFApb

(%) LEFAv

(%) LEFAr

(%) CEFAp

(%) CEFAv

(%) CEFAr

(%)

2016 1 C 33 32 15 3 18 0 1 LR 44 53 55 38 54 54 2 C 34 41 27 10 24 4 2 LR 62 62 62 73 83 73 3 C 41 32 35 14 30 23 3 LR 58 74 57 79 80 64 4 C 32 37 43 12 21 23 4 LR 58 58 48 58 48 48 5 C 21 19 21 4 8 10 5 LR 53 49 43 52 43 44 5 C 33 25 28 4 3 4

5 LR 48 46 47 55 34 39

6 C 27 20 26 8 10 14 6 LR 46 41 44 37 37 45 6 C 27 27 33 23 13 21 6 LR 56 39 45 55 39 41 7 C 35 34 33 26 9 19 7 LR 61 62 51 51 59 42

2017 1 C 23 15 12 3 2 2 1 LR 47 67 56 50 67 57 2 C 38 18 23 12 1 2 2 LR 73 48 53 63 57 60 3 C 28 39 29 15 15 16 3 LR 51 73 65 66 58 67 4 C 28 38 31 23 15 17 4 LR 69 62 60 82 73 74 5 C 20 15 26 0 0 21 5 LR 42 31 47 46 23 45 5 C 29 25 30 12 11 6 5 LR 38 42 45 45 31 36 6 C 24 14 17 5 3 8 6 LR 40 40 44 46 38 37 6 C 24 18 30 13 3 8 6 LR 40 53 48 37 42 56 7 C 24 14 24 9 0 5 7 LR 35 38 53 39 42 57

aC: Control; LR: fruiting zone leaf removal. bLeaf (LEFA) and cluster (CEFA) exposure flux availability, measured at berry pea-size stage (p), veraison (v), and during grape ripening (r).

42

2.3.3 Viticultural and physiological data

Mean values for yield parameters, pruning weight, crop load, and basic juice chemistry

are reported in Appendix A. As expected, there was large variation in production parameters

which was, at least in part, explained by the different management practices (i.e., shoot and

cluster thinning) used by the grower cooperators. Yield, for example, varied between 1.54 (site 2

LR) and 6.72 (site 5 C) kg/m of cordon in 2016 and from 1.88 (site 2 C) to 6.49 (site 5 LR) kg/m

in 2017. Basic juice chemistry (TSS, pH, TA) values were within the range of those reported for

Noiret in previous studies conducted in the northeast U.S. (Vanden Heuvel et al., 2013; Homich

et al., 2017).

Concentrations of the major macronutrients were at deficiency levels for some of the

sites, although visual symptoms of leaf nutrient deficiency were not observed except for Mg

(Appendix B). For example, concentration of leaf petiole N was at deficiency level (<0.80%) for

a few experimental units in 2016 (site 3 C; site 6) and for more sites in 2017 (site 2; site 4; site 6

LR HWC and VSP units; and site 7 LR). Phosphorus concentration for site 7 (2016, 2017) and

for the C unit at site 3 (2016) was in the deficiency range (<0.14%), while K concentration was

low for (<1.20%) sites 2 and 6 in 2017.

Conversely, there were two sites in 2016 (site 2; site 6 C) and more in 2017 (site 2; site 4;

site 6) that exceeded the recommended late-season P leaf petiole concentration (0.14-0.30%).

Potassium exceeded recommended concentrations (1.2-2.0%) at site 1 in both years, while at site

5 three out of the four experimental units had excessive concentration of K in 2016 (HWC C and

VSP) and 2017 (HWC C and VSP). Likely because of excessive K uptake, Mg concentration at

site 1 in both 2016 and 2017 was at a deficiency level (<0.35%), and visual Mg deficiency

symptoms were observed in both seasons.

43

Berry carbon isotope ratio, a proxy for vine water status, exhibited a moderate degree of

variation between both years (Appendix B). On average, δ13C ranged from -24.8, i.e., lower

water status (site 7 LR), to -29.4, i.e., higher water status (site 2 C), in 2016 and from -27.1 (Site

5 C) to -29.7 (site 6 C) in 2017. Large inter-annual variability was observed at sites 6 and 7,

while δ13C values at sites 1 and 2 remained nearly consistent across both years.

Berry rotundone concentration at harvest exhibited both inter-site and inter-annual

variation (Figure 2). Values ranged from 108.9 ng/kg (site 5 LR) to 830.2 ng/kg (site 1 LR) in

2016 and from 246.6 ng/kg (site 3 LR) to 1176.1 ng/kg (site 4 C) in 2017. The LR unit at site 2

was omitted from the analysis due to issues with sample analysis. Sites 5, 6, and 7 displayed

moderate to high variation between years in rotundone concentration for both treatments—on

average a 18.4% reduction for site 5, a 129% increase for site 6, a 148.2% increase for site 7—

rotundone concentration at site 4 was on average more than four times higher in 2017 when

compared to the previous vintage (Figure 2). Conversely, site 1 experienced the highest decrease

in rotundone concentration from the first to the second year, with 2017 concentration being less

than half that of 2016.

All C units except for site 1 had higher rotundone concentrations, between 0.69% and

64.4%, than the respective LR units in 2017; interestingly, site 1 had higher rotundone

concentrations in the LR unit for both years. Trends were less consistent in 2016: in addition to

site 1, rotundone was higher for LR units at site 5 HWC (5.2%) and site 7 (18.6%) as compared

to the C.

44

Figure 2. Berry rotundone concentrations at harvest 2016 and 2017 for each site and treatment.

45

2.3.4 Multiple linear regression analysis and model selection of a rotundone mesoclimatic

model

Scatter plots indicated that rotundone concentration was linearly related to K, Mg, Ca,

average berry weight, average cluster weight, GDD, GDDv, and rainfall, and both linearly and

quadratically related to SH800 and SH800v (data not shown). Pearson correlation coefficients (r)

were used to assess the strength of linear correlations for both 2016 and 2017 data, as well as for

the data pooled across the two years (Table 6).

The production variables most strongly correlated with rotundone in 2016 and for 2016-

2017 were berry weight and TSS, whereas rotundone was correlated with pH in 2017, but not in

2016. For the remaining production variables, except for cluster weight, the relationships in 2016

and 2017 were inconsistent and altogether were poorly correlated with rotundone concentration.

Interestingly, the relationship between δ13C and rotundone concentration was negative in 2016

and positive in 2017. Amongst leaf petiole nutrients, both Mg and Ca showed the highest,

negative correlations with rotundone.

Overall, weather parameters were better correlated with rotundone when measured from

veraison to harvest instead of for the whole growing season (Table 5). Specifically, SH800v2

exhibited the highest positive correlation with rotundone when data from the two years were

combined (r = 0.71, p <0.001), followed by GDDv. Both linear and quadratic terms were

included in the regression analysis for the variables SH800 and SH800v, as the quadratic terms

were better correlated to rotundone concentration.

46

Table 6. Pearson correlation coefficient representing the linear relationships between rotundone, production, vine water and nutrient status, mesoclimate, and microclimate parameters measured in 2016 and 2017. Rotundone Rotundone Variable 2016 2017 2016

& 2017 Variable 2016 2017 2016

& 2017 Production Site weathera

TSS -0.69 -0.21 -0.48 GDD -0.37 0.50 -0.11 pH -0.03 0.42 0.12 Rainfall 0.46 -0.52 0.09 TA 0.24 -0.46 0.11 SH800

0.38 0.35 0.42 Berry wt 0.46 0.72 0.56 GDDv 0.53 0.75 0.70 Cluster wt 0.46 0.21 0.22 Rainfallv -0.15 0.72 0.33 Cluster no. -0.40 0.12 -0.01 SH800v 0.64 0.60 0.62 Yield -0.19 0.17 0.07 Pruning wt 0.23 -0.22 0.02

Crop load -0.20 0.12 0.04 Vine water and nutrient status Fruiting zone weatherb δ13C -0.70 0.39 -0.33 DH10 0.13 -0.14 -0.30 N 0.56 -0.16 0.11 DH15 0.29 0.27 0.40 P 0.28 0.07 0.23 DH20 -0.21 0.58 0.35 K 0.47 0.20 0.28 DH25

-0.03 0.00 -0.05 Mg -0.56 -0.41 -0.50 DH30

-0.02 -0.23 -0.28 Ca -0.84 -0.10 -0.44 DH35

-0.10 -0.25 -0.30 DH40

-0.19 -0.15 -0.27 LEFAp -0.09 0.12 -0.07 LEFAv 0.04 0.12 0.04 LEFAr -0.17 -0.05 -0.08 CEFAp -0.24 0.07 -0.07 CEFAv 0.00 0.03 -0.01 CEFAr -0.21 -0.08 -0.11 aSH800 = Linear relationship for percent of solar-hours above 800 W/m2; GDDv = Veraison-to-harvest GDD; Rainfallv = Veraison-to-harvest rainfall; SH800v = Percent of veraison-to-harvest solar-hours above 800 W/m2. bDHx = Percent of degree-hours between 10.1-15 °C (DH10), 15.1-20 °C (DH15), 20.1-25 °C (DH20), 25.1-30 °C, 30.1-35 °C (DH30), 35.1-40 °C (DH35), and >40 °C (DH40); LEFA and CEFA = Leaf and cluster exposure flux availability, measured at berry pea-size stage (p), veraison (v), and during grape ripening (r).

47

Several candidate regression models were evaluated using different selection options, but

they did not provide the optimal model with any given predictor variables, as they were prone to

overfitting the data (Freund and Littell, 2006). Therefore, the RSQUARE option was used with

PROC REG to better fit the data and aid in model selection. The best three models out of all

models generated for one-, two-, three-, four-, five-, and six-variable models with the RSQUARE

option are shown in Table 7, including values for various statistical parameters used for model

selection. Analysis of r2, adjusted r2, Cp, AIC, BIC, and MSE values suggested that a six-variable

model may be overfitted (Freund and Littell, 2006) and that a lower-variable model may be

better-suited for predictive purposes (Table 7).

As more variables were added to the models, less and less variation was explained by

each additional variable; this is reflected in the 0.064 increase in adjusted r2 when a fourth

variable is added to the model, for example, when compared to the 0.017 increase when a fifth

variable is added (Table 7). A slight decrease in mean square error (MSE) between the best

fourth- and fifth-variable model indicated that each new variable added again explained a

diminishing proportion of variation, and that models with fewer variables may be better suited

for predictive purposes. Therefore, the first four-variable model was determined to be the best

candidate model for predictive purposes.

48

Table 8. Best regression model equation to be used for rotundone prediction. Year Model r2 Adj. r2 2016 & 2017 Rot. = -0.53 + 0.568 * P – 0.336 * Ca + 0.018 * crop load + 0.003 *GDDv 0.853 0.828

Table 7. The best multi-variable models evaluated during model selection for rotundone prediction. No. of Variables

Model variables r2 Adj. r2 Cpa AICb BICc MSEd

1 SH800v2 0.567 0.551 29.0 -95.6 -95.9 0.034 1 GDDv 0.512 0.494 35.8 -92.2 -92.8 0.038 1 SH800v 0.433 0.412 45.7 -87.8 -88.9 0.045 2 GDDv, Ca 0.703 0.680 14.0 -104.6 -104.0 0.024 2 SH800, SH800v 0.651 0.624 20.5 -99.9 -100.1 0.028 2 GDDv, TSS 0.641 0.613 21.7 -99.1 -99.4 0.029 3 GDDv, Ca, crop load 0.789 0.764 5.23 -112.6 -109.8 0.018 3 GDDv, Ca, pH 0.764 0.736 8.38 -109.3 -107.4 0.020 3 GDDv, Ca, pruning wt 0.761 0.732 8.82 -108.9 -107.1 0.020 4 GDDv, Ca, crop load, P 0.853 0.828 -0.68 -121.0 -113.9 0.013 4 GDDv, Ca, pH, pruning wt 0.842 0.815 0.71 -118.9 -112.6 0.014 4 GDDv, Ca, pruning wt, rainfall 0.839 0.812 1.05 -118.4 -112.3 0.014 5 GDDv, Ca, P, pruning wt, pH 0.873 0.845 -1.17 -123.3 -112.7 0.011 5 GDDv, Ca, P, pruning wt, TA 0.873 0.845 -1.16 -123.2 -112.6 0.011 5 GDDv, Ca, P, pruning wt, yield 0.872 0.845 -1.10 -123.1 -112.6 0.011 6 GDDv, Ca, P, pruning wt, rainfall, cluster no. 0.897 0.870 -2.24 -127.5 -110.9 0.010 6 GDDv, Ca, P, pruning wt, rainfall, yield 0.896 0.868 -2.11 -127.2 -110.8 0.010 6 GDDv, Ca, P, pruning wt, pH, cluster no. 0.894 0.865 -1.80 -126.5 -110.6 0.010 aCp = Mallow’s Cp statistic. bAIC = Akaike information criterion. cBIC = Bayesian information criterion. dMSE = Mean square error.

49

The best three- (GDDv, Ca, and crop load), four- (GDDv, Ca, crop load, and P), and five-

(GDDv, Ca, P, pruning weight, and pH) variable models were chosen as candidate models, and

diagnostic analyses were performed for each model. The analyses suggested an absence of

multicollinearity for the three and four-variable models chosen, while there was a low likelihood

of multicollinearity for the five-variable model (VIF value, tolerance threshold, and F-values are

reported in the Appendix C).

The diagnostic analysis indicated that either the three- (GDDv, Ca, and crop load) or the

four-variable model (GDDv, Ca, crop load, and P) were strong candidates for use as a predictive

model. When considering the diagnostic statistics, F-values, adjusted r2 values, and other model-

selection statistical criteria, and the variables included within the models, the 4-variable model

emerged as the strongest candidate for use as a predictive model due to its increased predictive

power and the low added complexity resulting from the inclusion of an additional variable. The

four-variable model including the variables GDDv, Ca, crop load, and P was thus chosen as the

best predictive model generated through SAS analysis (Tables 7 and 8).

2.3.5 Model validation of predictive rotundone mesoclimatic model

The same four-variable model (GDDv, Ca, crop load, and P) was selected by FORWARD

selection and the RSQUARE option as the optimal fit for the first validation data set (n = 20).

The model equation of the four-variable model was then used to generate predicted rotundone

concentrations for the second validation data subset (n = 16). Predicted rotundone concentrations

values of the validation data plotted against the actual observed values of the original data set

yielded a strong linear relationship (Figure 3). This supported the predictive power and accuracy

of the model. Despite the lack of an external data set for model validation, these two methods

50

allowed for the validation of the four-variable model and supported its use as a predictive model

for determining rotundone concentrations.

Figure 3. Relationship between observed and predicted rotundone concentrations (ng/kg) that were generated using SAS’ PROC SCORE.

51

2.3.6 Multiple linear regression analysis and model selection of a rotundone microclimatic

model

Rotundone concentration was poorly correlated with either sunlight or temperature

indices in the fruiting zone (Table 6). Scatter plots indicated weak negative linear relationships

between rotundone concentration and all DH indices except for DH15 and DH20, with rotundone

exhibiting positive relationships with DH15 in both years while exhibiting a negative relationship

with DH20 in 2016 and a positive relationship in 2017. There was not a clear visual linear trend

between rotundone concentration and CEFA or LEFA for any of the three sampling dates.

Pearson correlation coefficients supported these visual interpretations, as the r values for all DH

indices were low and relationships were insignificant except for DH15, and DH10 when

considered a = 0.10 (DH10: p = 0.14; DH15: p = 0.04; DH20: p = 0.08; DH25: p = 0.80; DH30: p =

0.17; DH35: p = 0.15).

Relationships between rotundone concentration and DH indices varied between vines

with highly shaded clusters (C) and vines exposed to fruiting zone leaf removal (LR). The

indices DH10, DH15, and DH30 were strongly correlated with rotundone concentration for C vines

(r = -0.61, p = 0.03; r = 0.83, p = < 0.000; r = -0.60, p = 0.03), whereas the highest correlating

DH index was DH10 for LR vines (r = -0.57, p = 0.05). Rotundone concentrations of C vines

were negatively correlated with all DH indices, except for DH15, DH20, and DH40, which was not

calculated as berry temperature was never above 40 °C for the C units. However, rotundone

concentration was positively correlated with DH20, DH25, and DH30, but negatively with the

remaining DH indices (DH10, DH15. DH35, DH40) for the LR vines.

Rotundone concentration was poorly correlated with the percentage of sunlight reaching

the leaves (LEFA) or the clusters (CEFA) in the fruiting zone (Table 6). The highest correlation

52

between CEFA, LEFA, and rotundone was when EPQA parameters were measured during fruit

ripening. However, relationships were not strong or significant (CEFAr, p = 0.52; LEFAr, p =

0.61). Different trends emerged when the data was split by treatment: C treatment rotundone

concentration was most highly correlated with LEFA at veraison (r = 0.43, p = 0.08) and CEFA

measured before veraison (r = 0.32, p = 0.20); interestingly, for LR vines both LEFA and CEFA

measured during the fruit ripening period correlated the highest with rotundone concentration

(LEFAr: r = 0.37, p = 0.15; CEFAr: r = 0.43, p = 0.09).

A variety of different candidate models were evaluated using PROC REG and the

FORWARD, BACKWARD, and STEPWISE selection options. A three-variable model was

generated using FORWARD selection as a candidate model (DH15, DH30, and CEFAp) with an

r2 of 0.54 and an adjusted r2 of 0.47. Selection using BACKWARDS elimination generated a

three-variable candidate model (DH10, DH30, CEFAp) with an r2 of 0.57 and an adjusted r2 of

0.51; STEPWISE selection generated the same model. Given the discrepancies in model

selection among these three selection options, the RSQUARE option was used to validate these

selections through comparison of r2, adjusted r2, Cp, AIC, BIC, and MSE values. Results

indicated that a three-variable model (DH10, DH30, and CEFAp) was the best candidate model

with an r2 of 0.57 and an adjusted r2 of 0.51. Further diagnostic analyses reaffirmed the strength

of the three-variable model (results from diagnostic analyses are presented in Appendix D).

Further analysis of model residuals supported the three-variable model as the best

candidate model when compared to a two-variable model for explaining the variation in

rotundone concentrations at harvest due to fruiting zone temperature and sunlight conditions.

Therefore, of the fruiting zone-level weather variables, post-veraison DH10, DH30, and CEFA

measured before veraison explained the most variation in rotundone concentrations (Table 9).

53

Table 9. Best model for explaining fruiting zone weather influence on rotundone concentrations. Year Model r2 Adj. r2 2016 & 2017 Rot. = 0.972 – 0.021 * DH10 – 0.114 * DH30 + 1.230 * CEFAp 0.574 0.511

54

2.4 Discussion

Rotundone concentrations are highly correlated to temperatures and solar radiation

during the ripening period

The primary objective of this study was to identify the key climatic and viticultural

variables associated with rotundone concentration in Noiret wine grapes within the northeast

U.S. Multiple linear regression analysis indicated that post-veraison climatic variables at the

vineyard scale were the strongest predictors of rotundone concentration in Noiret wine grapes at

harvest. Specifically, heat accumulated from veraison to harvest (GDDv) and post-veraison

vineyard solar time (SH800v) were most strongly, and positively, correlated with rotundone

concentration. These variables, when compared to season-long climatic variables, are better

predictors of rotundone concentrations and reaffirm the importance of measuring weather

parameters during the ripening period when rotundone is accumulating in the berries (Zhang et

al., 2015b).

Based on previous literature, we expected that in a cool-climate region with a short

growing season, i.e., the northeast U.S., higher temperatures and solar radiation during fruit

ripening would have decreased and not increased rotundone concentrations. Indeed, wine

rotundone concentrations were negatively and exponentially associated with GDDv, DH25, and

vineyard solar exposure for wines made from Australian Shiraz grapes (Zhang et al., 2015b).

However, Zhang et al. (2015b) calculated post-veraison GDD across 15 seasons at the same

vineyard, while we compared 7 sites with variable weather conditions across two years. The

range of GDDv and solar exposure values reported in our study are wider than those reported in

Zhang et al. (2015b). For example, their post-veraison GDD ranged from about 300 to 408 and

post-veraison solar exposure from 14.2 to 22.8 MJ/m2 day across all 15 seasons. At our sites

55

GDDv varied from 249 to 457, while post-veraison average daily solar exposure means had a

wider range too (7.6 to 21.1 MJ/m2). In addition to being cooler, some sites used in this study

also had lower solar exposure during fruit ripening.

A pattern of regional grouping emerged. Sites in the Finger Lakes AVA (i.e., sites 5, 6,

and 7) were consistently grouped as sites with lower post-veraison solar exposure, heat

accumulation, and rotundone concentrations than most of the other sites. Similarly, sites along

the eastern shore of Lake Erie (i.e., sites 3 and 4) generally exhibited high post-veraison solar

exposure and heat accumulation, but had variable rotundone concentrations between years. This

data suggests that though environmental variables may interact in the northeast U.S. to influence

rotundone concentrations, it is unclear if there is a regionality to rotundone concentrations.

Rotundone concentrations at harvest can be affected by the length of the ripening period.

Rotundone can reach peak concentrations around 44 days following veraison and thereafter

decrease slightly, indicating that an optimal time for harvest may exist that maximizes

concentrations (Geffroy et al., 2014). In our study, the sites with the longest ripening periods

(site 1: 47 days in 2016; and site 4: 52 in 2017) also had the highest rotundone concentrations,

while at site 1 a shorter ripening period in 2017 compared to the previous vintage corresponded

with lower rotundone concentration at harvest. This suggests that the length of the ripening

period might have had a positive role in influencing rotundone concentrations in cool climates.

However, other weather parameters still play a major role in determining rotundone

concentration at harvest, as sites with long ripening periods (i.e., 40 to 45 days), like those in the

Finger Lakes AVA, also had low post-veraison heat accumulation and solar exposure when

compared to the other sites.

56

Production variables do not explain variation in rotundone concentrations

While clear climatic trends were observed, there were few strong associations between

production parameters and rotundone concentrations. Berry weight was the viticultural variable

most strongly correlated with rotundone concentrations. Although berry weight was not selected

as a predictor variable for the multiple linear regression model, the positive relationship with

rotundone suggests that conditions that favor heavier berries might also be conducive to

increased rotundone concentrations. Berry weight was, indeed, significantly correlated with

seasonal GDD (r = 0.37, p = 0.020), GDDv (r = 0.55, p < 0.001), and rainfallv (r = 0.69, p = <

0.001). Inter-seasonal differences in rotundone concentrations were not well correlated with

differences in whole skin-to-juice ratio (Geffroy et al., 2014). Since rotundone is mainly

localized within the berry exocarp and not accumulated within the berry mesocarp (Takase et al.,

2016b), it is unlikely that berry weight had a direct effect on rotundone concentrations.

Concentration of soluble sugars in the fruit at harvest was significantly and negatively

correlated with rotundone concentrations. Considered a metric of grape ripeness, TSS is often

responsive to GDD and seasonal temperatures (Keller, 2015). Thus, as expected, TSS was also

significantly and positively correlated with GDD (r = 0.58, p < 0.001). Despite this correlation,

TSS was not correlated with GDDv (r = -0.22, p = 0.194). It is unclear if there is a direct

relationship between rotundone and TSS; seasons that are warmer and apparently conducive to

higher TSS are also more inhibitory to rotundone accumulation, reflecting the trends between

rotundone and season heat that have been reported for Noiret and other grape varieties

(Herderich et al., 2015; Homich et al., 2017).

Beyond berry weight and TSS, rotundone concentration was not significantly correlated

with any other production variables. A few studies have addressed the relationship between

57

rotundone, production parameters, and viticultural management methods with contrasting

conclusions. For example, thinning the crop at mid-veraison by 40% did not significantly impact

rotundone concentrations in Duras wines, when compared to an unthinned control (Geffroy et al.,

2014). Similarly, reducing crop by 50 and 90% did not significantly influence rotundone

concentrations in Shiraz berries, though the phenological timing of crop thinning was not stated

(Logan 2015). The effects of fruiting zone leaf removal are also inconsistent across studies

(Geffroy et al., 2014; Logan, 2015; Homich et al., 2017). Furthermore, vine vegetative growth

was not a major source of direct influence upon rotundone concentrations in Shiraz grapes

(Scarlett et al., 2014). The results from these studies indicate that direct associations between

rotundone and individual production parameters such as yield and crop load are not strong, and

that any influence due to vine vigor may be instead a result of manipulating fruit light

interception and exposure (Geffroy et al., 2015).

Despite the lack of correlation, crop load was included within the final predictive model.

Correlation is performed with single variables and provides information concerning linear

relationships, but multiple linear regression assesses the relationship between rotundone and a set

of variables. The relative importance of any given variable depends on the other variables in the

model. Thus, it is possible that crop load may explain considerable variation in rotundone

concentration not when it is analyzed by itself, but when included in a multiple linear regression

model. The inclusion of crop load nevertheless suggests that there may be a relationship between

crop load and rotundone that is worth further exploring.

58

Vine nutrients are consistently correlated to rotundone concentrations while water status

varies by year

Rotundone concentration was negatively correlated to seasonal vine water status, similar

to previous work (Geffroy et al., 2014; Geffroy et al., 2016b). However, our data implies that this

relationship breaks down when water is highly abundant and non-limiting. When the data are

separated by year, conflicting trends emerge: d13C correlated negatively and significantly (p =

0.004) in 2016, a relatively dry year, and positively and insignificantly in 2017, a relatively wet

year (p = 0.106). Based on our d13C data, it appears that rotundone is more sensitive to vine

water status during seasons with less precipitation, and particularly if vines may be experiencing

water deficit. This was indeed the case in 2016, when d13C reached values that would indicate

weak-to-moderate (-26 to -25‰) and moderate-to-severe (-25 to -24‰) water deficit

(Santesteban et al., 2015). Furthermore, a large part of New York, including the region where

sites 5, 6, and 7 were located (i.e., the Finger Lakes AVA), experienced severe drought

conditions during the 2016 season (Sweet et al., 2017), which suggests that experimental vines

might have experienced water deficit at some point during the season. Despite different annual

trends, the correlation between seasonal precipitation and d13C was moderate, negative, and

significant in both 2016 (r = -0.54, p = 0.028) and 2017 (r = -0.67, p = 0.002), suggesting that

d13C is indeed an effective index for integrated seasonal vine water status.

Our study is the second to evaluate relationships between rotundone concentrations and

grapevine mineral nutrition. A previous study reported inconclusive relationships between

rotundone concentrations and grapevine macro- and micronutrient concentrations when petioles

are sampled at 6 different time points, beginning at bloom and ending at harvest (Geffroy et al.,

2015). Here, rotundone concentrations were variably correlated with several macro-nutrients

59

(e.g., K, Mg, and Ca) measured at veraison. Both P and Ca were identified as significant

predictor variables and were ultimately included within the final rotundone prediction model.

However, it is unclear if these or other nutrients had direct influence on rotundone accumulation

in the berries. Instead, it is possible that seasons with environmental conditions that lead to

greater plant uptake of Ca and Mg, such as decreased precipitation, likely coincided with

decreased rotundone concentrations. Much like how d13C was strongly correlated with rotundone

in 2016, Ca was also correlated with d13C in 2016 (r = 0.77, p = < 0.001), indicating that Ca

concentration was positively correlated with and possibly influenced by decreasing vine water

status. This is a possibility given the role of Ca in plant responses to abiotic stress (Keller, 2015);

or this association could be linked to decreased uptake of K in drier seasons with reduced

precipitation, and a subsequent increase in Ca could simply be due to this decreased uptake of K.

Phosphorus was also positively associated with seasonal rainfall across the two years (r = 0.39, p

= 0.016), indicating that the seasons and sites with higher rainfall altogether had higher

concentrations of tissue P. The sensitivity of both Ca and P to seasonal rainfall and their

inclusion within the final model imply a potential positive relationship between rotundone and

seasonal rainfall, which would agree with previous literature (Zhang et al., 2015b).

The inclusion of Ca and P within the final model suggests a necessity to further

investigate the relationships between grapevine nutrition and rotundone concentrations.

Grapevine nutrient status was analyzed at veraison, as this is the phenological stage when

nutrients are the most stable and most accurately reflect plant nutrient status (Wolf, 2008). Given

the relationships observed between leaf petiole macronutrients and rotundone, it would be

interesting to analyze the nutrient composition of the berries themselves in relation to rotundone

concentration.

60

A four-variable predictive model explained about 82% of rotundone concentration at

harvest

A four-variable multiple regression model with GDDv, Ca, crop load, and P, was the best

candidate model for predicting rotundone concentrations. Including more predictor variables that

explain little additional variation typically increase the r2 values, but overfitting the model causes

the regression coefficients to be unstable. This would result in a complex regression model that

may not have accurate predictive properties. Since a major objective of this study was to identify

the variables that have the strongest relationship with rotundone concentrations, the addition of

statistically unnecessary variables is detrimental because it obscures which variables have a

dominant influence on rotundone concentrations.

Another consideration of this study was that the final predictive model would need to be

simple enough for practical application and field experimentation: the model would need to

include variables that are both significant for explaining rotundone concentration, while also

easily measured by both wine grape growers and other researchers. The four variables selected

satisfy these requirements while maintaining a high degree of predictive power. Analysis of

model residuals and partial model validation further supported the strength of the chosen model,

though to increase confidence in the model it would be necessary to further validate it with an

external data set. This would ideally be done using additional weather, nutritional, production,

and Noiret rotundone data.

61

Analysis of the fruiting zone temperature yielded weak correlations between rotundone

and temperature classes

A second objective of this study was to analyze the influence of micrometeorological

parameters on rotundone concentrations at the fruiting zone level. Rotundone concentrations

were not strongly correlated with any berry temperature class (DH) measured between veraison

and harvest, despite the strong correlation between rotundone and vineyard GDDv. We expected

rotundone concentration to be inhibited by air temperatures exceeding 25 °C, but rotundone was

not well correlated with DH indexes. Previous work reported positive associations between

rotundone concentrations in Shiraz grapes and DH10, DH15, and DH20, but negative associations

with DH25 (Zhang et al., 2015a; Zhang et al., 2015b). Here rotundone was negatively associated

with DH10 and DH25, and positive with DH15 and DH20, though the strength of these relationships

was much weaker compared to previously reported associations. The absence of strong

relationships between rotundone concentrations at harvest and DH indexes do not give a clear

understanding of the influence of these temperature classes, and therefore do not support

previously reported associations between rotundone and specific DH classes (i.e., DH25) (Zhang

et al., 2015a; Zhang et al., 2015b).

It is important to note that there are differences in terms of the ranges of rotundone

concentrations and percentage of hours in each DH class reported here, when compared to

previous reports. The range of rotundone concentrations reported here exceed the range reported

for Shiraz grapes (Zhang et al., 2015a); similarly, the ranges of total percentages of hours

between 10 and 15 °C (DH10) and 15.1 and 20 °C (DH15) also exceed those of Zhang et al.

(2015a). Despite this, previous claims that DH25 represents a critical temperature index for

rotundone concentrations were not corroborated here, and instead the strongest relationships

62

were seen with cooler temperatures (i.e., DH15) although relationships were nevertheless poor.

Whether or not this is reflective of how rotundone accumulates in Noiret specifically is unclear,

but it does suggest that cooler berry temperatures might be especially important for rotundone

accumulation when all temperature classes are compared.

Rotundone was poorly correlated to cluster exposure at all measurement points

Contrary to our expectation, cluster sunlight exposure, measured as CEFA, correlated

poorly with rotundone across the season, from treatment application to harvest. Previous work

suggested that increased fruit sun exposure may decrease rotundone concentrations, but these

studies did not directly assess fruiting zone solar exposure (Scarlett et al., 2014; Zhang et al.,

2015a; Zhang et al., 2015b). To date, a single study found that increasing radiation flux reaching

the clusters from pre-veraison to harvest increased rotundone in Noiret grapes and wines, in one

of the two study years (Homich et al., 2017). Differences in experimental design between studies

assessing defoliation-induced influence upon rotundone make comparison difficult because

treatments were imposed at different phenological stages. Homich et al. (2017) evaluated the

effects of leaf removal at E-L 31 “berry pea-size stage” (pre-veraison) and one week after E-L

35, “50% veraison.” Conversely, other studies applied leaf removal at mid-veraison (Geffroy et

al., 2014) or imposed shading treatments at veraison (Zhang et al., 2015a). Results from these

studies indicate that there is not a clear consensus on the effects of sunlight cluster exposure on

rotundone at the fruiting zone scale, and that more research is necessary using treatments

implemented at stages that are phenologically consistent across studies.

In our study, the sites in the Finger Lakes AVA had amongst the lowest post-veraison

solar exposure and warmth, matched by consistently low rotundone concentrations. Thus, even if

63

canopy defoliation leads to greater canopy porosity and fruiting zone solar exposure (i.e., higher

CEFA), it is possible that low levels of solar exposure and warmth at the vineyard-scale could

prove to be a stronger influence on rotundone concentrations. In other words, higher CEFA does

not necessarily guarantee higher fruiting zone temperature and solar exposure if these factors are

already reduced at the mesoclimatic level. The absence of a relationship between cluster

exposure and rotundone concentration at harvest might also in part be explained by the

methodology used to assess fruiting zone canopy density and sunlight availability. Fruiting zone

PAR was measured only three times throughout the season and only at solar noon, which may

not accurately measure canopy solar exposure.

It is also unclear how fruiting zone microclimate might affect the precursor to rotundone

formation, a-guaiene. Rotundone was positively correlated with a-guaiene concentrations during

the veraison-to-harvest period; moreover, higher a-guaiene concentrations also occurred at

cooler sites (Takase et al., 2016a). Therefore, it is reasonable to assume that factors that affect

the concentrations of a-guaiene would consequently affect rotundone concentrations.

Understanding how a-guaiene is influenced by viticultural or climatic factors may assist in

understanding why rotundone responds strongly to specific variables and not others and may

help explain what conditions within the fruiting zone would be most conducive to rotundone

formation. Incorporating an analysis of a-guaiene into further research on rotundone could

therefore help unravel the relationship between the two, in addition to the relationship between

these compounds and both viticultural management and climatic conditions.

64

Regression analysis indicates DH10, DH30, and early-season exposure are critical for

rotundone

The three fruiting zone microclimate parameters that best explained rotundone

concentration were DH10, DH30, and CEFAp (adjusted r2 = 0.51). This analysis suggests that

rotundone concentrations at harvest are sensitive to multiple berry temperatures ranges, and that

sensitivity to DH25 may not be as important for Noiret rotundone concentrations as they are for

Australian Shiraz, though it is important to note that this reported relationship was between

rotundone and DH25 calculated using fruiting zone temperature, not berry surface temperature

(Zhang et al., 2015a; Zhang et al., 2015b).

Though we have discussed larger, site-wide climatic dynamics that may broadly

influence rotundone concentrations (i.e., GDDv), this model suggests that berry temperatures

also may have an impact on rotundone concentrations. The negative relationships between

rotundone and both DH10 and DH30 suggest that rotundone might be limited by post-veraison

periods that have excessively cool (i.e., between 10 and 15 °C) or hot (above 30 °C)

temperatures. Despite these differences compared to previous literature, the selection of these

indices supports the claim that rotundone is inhibited by high berry temperatures, and instead is

positively correlated with cooler temperatures (DH15 and DH20). It is unclear, however, how this

relates to accumulation and berry concentrations of the chemical precursor, a-guaiene, and the

rate of enzymatic conversion to rotundone.

The selection of CEFAp was unexpected, as pre-veraison solar exposure was not strongly

related to rotundone concentrations at harvest. The underlying mechanism of this positive

association is uncertain but may reflect canopy microclimate conditions better suited to a-

guaiene and rotundone accumulation, as suggested by Homich et al. (2017). Rotundone

65

concentrations were higher in Noiret fruit harvested from vines that had higher pre-veraison

CEFA, when compared to non-defoliated vines (Homich et al., 2017). Indeed, in this study at

most sites the timing of pre-veraison CEFA measurements coincided with the highest levels of

solar exposure in each year (data not shown), with daily accumulated averages of solar exposure

tapering off in the post-veraison period. It is unclear if there is an indirect or direct effect of pre-

veraison solar exposure on rotundone concentrations, but the inclusion of CEFAp and not

CEFAv or CEFAr within this model indicates that there is a temporal or phenological component

to the influence of solar exposure that is critical for consideration and warrants further

inspection.

Although these three variables comprise a simple model with low predictive capabilities,

especially when compared to the linear model developed for site-wide rotundone prediction, it

provided further clarification as to which specific micrometeorological factors had the strongest

influence on rotundone at the fruiting zone-scale. This model also highlights the difficulty of

separating the influence of fruit solar exposure and temperature, and the necessity of further

work. Selecting two berry temperature-related variables and a cluster exposure variable suggests

the interactive and most likely synergistic relationship between these micrometeorological

variables.

Noiret berry rotundone concentrations measured here compare favorably to those reported

in other literature

Within this study, rotundone concentrations exhibited high variation both geographically

and inter-annually ranging from 108 ng/kg (site 6 HWC LR) to 1176 ng/kg (site 4 C). Rotundone

concentrations at one of the experimental sites (site 6 HWC) averaged 293 ng/kg in 2016 and

66

445 ng/kg in 2017. These values were an order of magnitude lower than those measured at the

same site in 2014 and 2015, which ranged from about 1280 ng/kg to 3450 ng/kg (Homich et al.,

2017). Within-site, inter-annual variation in rotundone is not unusual, as up to 40-fold

differences in rotundone concentration across vintages were reported for Australian Shiraz wine

grapes (Bramley et al., 2017).

Additionally, the post-veraison periods were longer in both study years (57 and 54 days,

for 2014 and 2015, respectively) when compared to the post-veraison ripening periods in our

study (40 to 45 days across both 2016 and 2017), suggesting that berries at site 6 might have

been harvested before they reached the late-season spike in rotundone concentration (Homich et

al., 2017). This also suggests that sites with low post-veraison warmth and solar exposure (i.e.,

sites 6 and 7) might require even longer ripening periods to accumulate the highest possible

concentrations of rotundone. More research is necessary in order to assess whether accumulation

patterns for rotundone can be delayed or accelerated if environmental conditions are suboptimal,

as has been mentioned elsewhere (Zhang et al., 2015b).

Overall rotundone concentrations presented (from 108 to 1176 ng/kg) here fall within the

middle range of those reported worldwide, including Australian Shiraz wine grapes. Early

research into rotundone showed concentrations ranging from 10 to 620 ng/kg (Wood et al.,

2008); Scarlett et al. (2014) reported values from 73 to 1082 ng/kg in samples from a single

season, with Bramley et al. (2017) reporting lower values overall in following seasons at the

same site. Across various experiments performed by Zhang et al. (2015a), berry values altogether

ranged from 15 ng/kg to 447 ng/kg. Interestingly, investigations into Japanese Shiraz yielded

rotundone concentrations ranging from an average of 1033 ng/kg to 2342 ng/kg, exceeding those

of Australian Shiraz and the Noiret results reported here (Takase et al., 2015). Conversely,

67

analysis of New Zealand Shiraz yielded a lower average range of rotundone across a three-year

study period (50 ng/kg to 162 ng/kg) that fell below that reported here in Noiret (Logan, 2015).

Despite the widespread focus on Shiraz worldwide, rotundone has been extracted from

grapes of other ‘peppery’ varieties in concentrations that both exceed and approximate Noiret

rotundone concentrations reported here. Noiret concentrations were much lower than those

measured in Italian Vespolina grapes in 2009 and 2010, as average values ranged from about

1420 ng/kg to 5440 ng/kg, while they remained more comparable to a range of average

rotundone concentrations from 540 ng/kg to 1910 ng/kg in Austrian Grüner Veltliner grapes

(Caputi et al., 2011). Takase et al. (2015) report low average concentrations for a handful of

other V. vinifera varieties, including Merlot (62 ng/kg), Sauvignon blanc (27 ng/kg), and

Cabernet Sauvignon (21 ng/kg), in addition to low concentrations for two popular varieties in

Japan, Koshu (60 ng/kg) and Muscat Bailey A (16 ng/kg). Noiret rotundone concentrations

reported here are much higher than these values.

It is of importance to note that rotundone has been extracted from two other interspecific

hybrids, Muscat Bailey A, a cross of V. labruscana cv. Bailey and V. vinifera cv. Muscat

Hamburg, and Koshu, a variety that contains V. vinifera and East Asian Vitis species parentage

(Goto-Yamamoto et al., 2015; Yamada and Sato, 2016). Interestingly, the parentage of Noiret

also contains Muscat Hamburg (i.e., ‘Black Muscat’), but it is currently unknown if Muscat

Hamburg produces rotundone or is the source of this trait in Noiret and Muscat Bailey A

(Reisch, 2006). The low concentrations produced by Muscat Bailey A suggest otherwise. Aside

from these studies, most other studies focusing on rotundone have almost exclusively measured

this aroma compound in wine, and not berries, making direct comparisons of results with prior

findings difficult.

68

Implications for Noiret growers and manipulation of rotundone concentrations

Consumer acceptance of 'peppery’ wines is variable for Duras wines, with wine

connoisseurs preferring ‘peppery’ wines with high rotundone concentrations while amateur,

infrequent wine drinkers did not prefer wines with moderate or high rotundone concentrations

(Geffroy et al., 2018). It is likely that this mixed consumer approval can be extended to other

‘peppery’ varieties, like Noiret. Consequently, it becomes an economic incentive for growers to

be able to manage their grapevines so that the desired levels of rotundone can be reached in their

final wines. Although research efforts have not definitively parsed apart the effects of leaf

removal and other viticultural practices on the final concentrations of rotundone in grapes and

wine, in neither V. vinifera varieties nor Noiret, the utility of a predictive model cannot be

understated as it could be used to estimate rotundone concentrations from year to year.

The selected four-variable model includes Ca, P, crop load, and GDDv, variables that are

easy for a grower to measure, thus making it possible for a grower to calculate predicted

rotundone concentrations within a given year. Vine nutrient status is routinely analyzed by

growers at veraison; therefore, Ca and P can be easily assessed. Crop load cannot be calculated

until pruning weights are measured during the dormant growing season, but historic crop load

ratios can be substituted if growers want to calculate predicted rotundone concentrations prior to

harvest. Growers might also be able to estimate GDDv close to harvest, as GDD accumulate

slowly during the late-season period.

The broad understanding of climatic relationships with rotundone concentrations can

also assist growers with identifying new sites for Noiret production as well, as historic weather

data could be used to infer whether it has the potential to be a low- or high-producing rotundone

site, and relate this to local consumer preferences of Noiret wines. In theory, management

69

methods could then be adapted to maximize or minimize a site’s rotundone-producing potential,

depending on the goal at hand, much like existing strategies for managing IBMP in grapes and

wine (Scheiner et al., 2012). Although further research is necessary to understand how

viticultural practices affect parameters that are related to rotundone accumulation, our data

suggest that rotundone concentration in Noiret is heavily influenced by a few measurable and

well understood factors.

2.5 Conclusion

The aim of this study was to first evaluate the relationships between rotundone, climatic,

and viticultural variables at the vineyard-scale, and secondly between rotundone and

micrometeorological conditions at the fruiting zone. For Noiret vineyards in Pennsylvania and

New York, rotundone concentration was positively correlated with post-veraison vineyard

weather conditions, namely heat accumulation and solar radiation. Rotundone was also

negatively correlated with petiole Ca and Mg concentrations, and vine water status. We

developed a predictive model tailored to grower application with high predictive power. The

model includes four easily measurable variables: post-veraison GDD, Ca and P petiolar

concentrations at veraison, and crop load. Model validation supported the accuracy and fitness of

this model, but external validation is now necessary to further test the model. At the scale of the

fruiting zone, rotundone concentrations were poorly and negatively related to fruiting zone solar

exposure and estimated berry temperature. Research is necessary to further investigate the effects

of, and interplay between, berry temperatures and solar radiation. This study nevertheless

emphasizes the primary importance of specific climatic variables in determining final rotundone

70

concentrations in Noiret wine grapes at harvest, and the interactions that exist between climate,

viticultural, and physiological variables that also can influence rotundone concentrations.

71

Chapter 3: Conclusion

Vineyard and fruiting zone climatic conditions are two important variables influencing

aroma-active concentrations in wine grapes. Fruiting zone microclimatic adjustment is critical

for grape growers in humid and cool climates, as it reduces disease pressure, enhances grape

ripening, and potentially modifies the concentrations of desirable and undesirable aroma-active

compounds. Understanding how these practices, the climatic factors they affect, and larger

regional climatic trends influence rotundone accumulation is essential to produce quality,

‘peppery’ wine grapes, whether of Noiret or another rotundone-producing variety. This study

focused on better defining how rotundone accumulation in Noiret grapes is affected by climatic

and viticultural factors, as well as how these variables interact to influence rotundone.

We confirmed the importance of post-veraison climatic conditions on rotundone

concentrations, as rotundone concentrations in Noiret were positively correlated with both

vineyard heat accumulation (GDD) and solar radiation. Our results indicate warmer weather

following veraison is important to rotundone accumulation in a cool-climate region or vintage.

The positive relationships between rotundone concentrations and both seasonal and post-veraison

solar radiation conflicts with previous research, yet reaffirms the general importance of site-

specific conditions on rotundone accumulation. However, it is possible that this association may

be indicative of an interspecific hybrid-specific species response to solar radiation, though

further research is necessary in the northeast U.S. to assess whether these trends persist for V.

vinifera varieties as well.

Rotundone was poorly correlated with weather parameters measured at the fruiting zone

level. Despite the strong association between rotundone concentrations and site solar radiation, it

was weakly correlated with fruiting zone solar exposure. The strongest relationship was with pre-

72

veraison fruit solar exposure, indicating that the relationship between rotundone and early season

exposure warrants further investigation (Homich, 2016). This might be complemented by

concurrent analysis of the chemical precursor to rotundone, a-guaiene, to assess whether early

season conditions are modulating rotundone accumulation directly or indirectly via a-guaiene

concentrations. Our results indicate that berry temperatures between 10.1 - 15 °C and greater

than 30 °C are most influential on rotundone accumulation in Noiret wine grapes, when

compared to other temperature ranges (DH15, DH20, DH25, DH35, DH40). Specifically, rotundone

concentrations were negatively correlated with berry temperatures above 30 °C (DH30).These

insights into micrometeorological influence suggest that vineyard management practices that

manipulate the canopy microclimate are likely to influence rotundone concentrations in the cool-

climate regions of the Northeast and Midwest U.S.

Aside from the strong climatic influence, we report here for the first time a relationship

between rotundone concentration and grapevine base cation nutrient concentrations, specifically

Mg and Ca. Strong relationships between Mg, Ca, and rotundone might be explained by their

close associations with several environmental and weather parameters, like seasonal rainfall, and

not due to a direct effect of nutrient concentration on rotundone biosynthesis. Nonetheless, these

relationships warrant further investigation to better understand the cause of the relationship.

Similarly, given that the relationship between seasonal vine water status (d13C) and rotundone

varied between the ‘dry’ (2016) and ‘wet’ (2017) year, it would be worthwhile to analyze

rotundone accumulation in regions with inter-annual variation in water availability, such as in

Pennsylvania and New York. Altogether, our results indicate that associations between

rotundone concentrations and grape and vine physiological measurements are most likely due to

73

environmental and weather conditions, but further research is necessary to confirm this

hypothesis.

A major component of the study presented here was the development of a predictive

model for grape growers to use for predicting rotundone concentrations in Noiret wine grapes.

Using multiple linear regression analysis, we selected a four-variable model for predictive use

that included measurable and intelligible variables: Ca concentration, P concentration, crop load

ratio, and veraison-to-harvest GDD. With these four variables growers can use the predictive

model to estimate the concentration of rotundone that is expected in grapes from a given vine or

vineyard with high precision. This predictive model is tailored for use within the regional climate

of the northeast U.S. and for Noiret grapes and could be used by growers to identify vineyard

sites and vintages that have the potential for yielding ‘peppery,’ quality Noiret wines. Though

further model validation with an external data set is necessary to supplement the partial

validation that was reported here, this model shows promise for applied use. This study

emphasizes the importance of post-veraison climatic variables affecting rotundone

concentrations in cool-climate regions and how these variables interact with vine physiological

parameters and management to yield varying degrees of rotundone accumulation.

74

References

Anesi, A., Stocchero, M., Dal Santo, S., Commisso, M., Zenoni, S., Ceoldo, S., et al. (2015). Towards a scientific interpretation of the terroir concept: plasticity of the grape berry metabolome. BMC Plant Biol. 15:191. doi: 10.1186/s12870-015-0584-4.

Alem, H., Rigou, P., Schneider, R., Ojeda, H., and Torregrosa, L. (2018). Impact of agronomic

practices on grape aroma composition: a review. J. Sci. Food Agric. [Preprint]. Available at: https://doi.org/10.1002/jsfa.9327 (Accessed November 11, 2018).

Allen, M.S., Lacey, M.J., Harris, R.L.N., and Brown, W.V. (1991). Contribution of methoxypyrazines to Sauvignon blanc wine aroma. Am. J. Enology Vitic. 42, 109-112.

Baugher, T.A., Marini, R., Schupp, J.R., and Watkins, C.B. (2017). Prediction of bitter pit in ‘Honeycrisp’ apples and best management implications. HortScience. 52, 1368-1374.

Bchir, A., Escalona, J.M., Gallé, A., Hernández-Montes, E., Tortosa, I., Braham, M., et al. (2016). Carbon isotope discrimination (δ13C) as an indicator of vine water status and water use efficiency (WUE): Looking for the most representative sample and sampling time. Agric. Water Manag. 167, 11-20.

Black, C.A., Parker, M., Siebert, T.E., Capone, D.L., and Francis, I.L. (2015). Terpenoids and

their role in wine flavour: recent advances. Aust. J. Grape Wine Res. 21, 582-600.

Bramley, R.G.V., Siebert, T.E., Herderich, M.J., and Krstic, M.P. (2017). Patterns of within- vineyard spatial variation in the ‘pepper’ compound rotundone are temporally stable from year to year. Aust. J. Grape Wine Res. 23, 42-47.

Bureau, S.M., Baumes, R.L., and Razungles, A.J. (2000). Effects of vine or bunch shading on the glycosylated flavor precursors in grapes of Vitis vinifera L. cv. Syrah. J. Agric. Food Chem. 48, 1290-1297.

Capone, D.L., Jeffery, D.W., and Sefton, M.A. (2012). Vineyard and fermentation studies to elucidate the origin of 1,8-cineole in Australian red wine. J. Agric. Food Chem. 60, 2281- 2287.

Caputi, L., Carlin, S., Ghiglieno, I., Stefanini, M., Valenti, L., Vrhovsek, U., et al. (2011). Relationship of changes in rotundone content during ripening and winemaking to manipulation of the ‘peppery’ character of wine. J. Agric. Food Chem. 59, 5565-5571.

75

Chapman, D.M., Roby, G., Ebeler, S.E., Guinard, J.-X., and Matthews, M.A. (2005). Sensory attributes of Cabernet Sauvignon wines made from vines with different water status. Aust. J. Grape Wine Res. 11, 339-347.

Chapman, D.M., Thorngate, J.H., Matthews, M.A., Guinard, J.-X., and Ebeler, S.E. (2004). Yield effects on 2-methoxy-3-isobutylpyrazine concentration in Cabernet Sauvignon using a solid phase microextraction gas chromatography/mass spectrometry method. J. Agric. Food Chem. 52, 5431-5435.

Coombe, B.G. (1995). Adoption of a system for identifying grapevine growth stages. Aust. J. Grape Wine Res. 1, 100-110.

Dami, I., Ferree, D., Prajitna, A., and Scurlock, D. (2006). A five-year study on the effect of cluster thinning on yield and fruit composition of ‘Chambourcin’ grapevines. HortScience. 41, 586-588.

Douglas, D., Cliff, M.A., and Reynolds, A.G. (2001). Canadian terroir: characterization of Riesling wines from the Niagara Peninsula. Food Res. Int. 34, 559-563.

Drew, D.P., Andersen, T.B., Sweetman, C., Møller, B.L., Ford, C., and Simonsen, H.T. (2016). Two key polymorphisms in a newly discovered allele of the Vitis vinifera TPS24 gene are responsible for the production of the rotundone precursor a-guaiene. J. Exp. Botany 67, 799-808.

Feng, H., Yuan, F., Skinkis, P.A., and Qian, M.C. (2014). Influence of cluster zone leaf removal on Pinot noir grape chemical and volatile composition. Food Chem. 173, 414-423.

Fischer, U., Roth, D., and Christmann, M. (1999). The impact of geographic origin, vintage and wine estate on sensory properties of Vitis vinifera cv. Riesling wines. Food Qual. Prefer. 10, 281-288.

Freund, R.J., and Littell, R.C. (2006). SAS System for Regression. 3rd edition. North Carolina: SAS Institute.

Gaudillère, J.-P., Van Leeuwen, C., and Ollat, N. (2002). Carbon isotope composition of sugars in grapevine, an integrated indicator of vine water status. J. Exp. Botany 53, 757-763.

Geffroy, O., Buissière, C., Lempereur, V., and Chatelet, B. (2016a). A sensory, chemical and consumer study of the peppery typicality of French Gamay wines from cool-climate vineyards. J. Int. Sci. Vigne Vin. 50, 35-47.

76

Geffroy, O., Descôtes, J., Serrano, E., Li Calzi, M., Dagan, L., and Schneider, R. (2018). Can a

certain concentration of rotundone be undesirable in Duras red wine? A study to estimate a consumer rejection threshold for the pepper aroma compound. Aust. J. Grape Wine Res. 24, 88-95.

Geffroy, O., Scholasch, T., Dufourcq, T., and Serrano, E. 2015. Understanding and mapping rotundone spatial variability in Vitis vinifera L. c.v Duras. In Proceedings of the 19th International Meeting GiESCO; 2015 May; Gruissan, France. (2016). p. 589-592.

Geffroy, O., Siebert, T., Silvano, A., and Herderich, M. (2017). Impact of winemaking techniques on classical enological parameters and rotundone in red wine at the laboratory scale. Am. J. Enology Vitic. 68, 141-146.

Geffroy, O., Dufourcq, T., Carcenac, D., Siebert, T., Herderich, M., and Serrano, E. (2014). Effect of ripeness and viticultural techniques on the rotundone concentration in red wine made from Vitis vinifera L. cv. Duras. Aust. J. Grape Wine Res. 20, 401-408.

Geffroy, O., Siebert, T., Herderich, M., Mille, B., and Serrano, E. (2016b). On-vine grape drying combined with irrigation allows to produce red wines with enhanced phenolic and rotundone concentrations. Sci. Hort. 207, 208-217.

Genthner, E.R. (2014). Identification of rotundone as an important contributor to the flavor of oak aged spirits. [dissertation]. [Urbana (IL)]: University of Illinois at Urbana-Champaign.

González-Barreiro, C., Rial-Otero, R., Cancho-Grande, B., and Simal-Gándra, J. (2013). Wine aroma compounds in grapes: a critical review. Crit. Rev. Food Sci. Nutr. 55, 202-218.

Goto-Yamamoto, N., Sawler, J., and Myles, S. (2015). Genetic analysis of east Asian grape cultivars suggests hybridization with wild Vitis. PLoS ONE 10(10): e0140841. doi:10.1371/journal.pone.0140841.

Guth, H. (1997). Identification of character impact odorants of different white wine varieties. J. Agric. Food Chem. 45, 3022-3026.

Hed, B., Ngugi, H.K., and Travis, J.W. (2014). Short- and long-term effects of leaf removal and gibberellin on Chardonnay grapes in the Lake Erie region of Pennsylvania. Am. J. Enology Vitic. 66, 22-29.

77

Herderich, M., Barter, S., Black, C.A., Bramley, R., Capone, D., Dry, P., et al. (2015). “Terroir effect on grape and wine aroma compounds,” in Advances in Wine Research, ed. S.E. Ebeler (Washington, DC: ACS), 131-146.

Herderich, M.J., Siebert, T.M., Parker, M., Capone, D.L., Jeffery, D.W., Osidacz, P., et al. (2012). “Spice up your life: analysis of key aroma compounds in Shiraz,” in Flavor Chemistry of Wine and Other Alcoholic Beverages, eds. M.C. Qian, and T.H. Shellhammer (Washington, D.C.: ACS), 3-13.

Hickey, C.C., Kwasniewski, M.T., and Wolf, T.K. (2018). Leaf removal effects on Cabernet franc and Petit Verdot: II. Grape carotenoids, phenolics, and wine sensory analysis. Am. J. Enology Vitic. 69, 231-246.

Homich, L.J. (2016). Effect of cluster sunlight exposure on rotundone concentration in Noiret

grapes and wine. [master’s thesis]. [State College (PA)]: Pennsylvania State University.

Homich, L.J., Elias, R.J., Vanden Heuvel, J.E., and Centinari, M. (2017). Impact of fruit-zone leaf removal on rotundone concentration in Noiret. Am. J. Enology Vitic. 68, 447-457.

Huang, A.C., Burrett, S., Sefton, M.A., and Taylor, D.K. (2014). Production of the pepper aroma compound, (-)-rotundone, by aerial oxidation of alpha-guaiene. J. Agric. Food Chem. 62, 10809-10815.

Huang, A.C., Sefton, M.A., and Taylor, D.K. (2015). Comparison of the formation of peppery

and woody sesquiterpenes derived from alpha-guaiene and alpha-bulnesene under aerial oxidative conditions. J. Agric. Food Chem. 63, 1932-1938.

Huang, C.-Y.L., and Schulte, E.E. (1985). Digestion of plant tissue for analysis by ICP emission

spectroscopy. Commun. In Soil Sci. Plant. Anal. 16, 943-958.

Iland, P., Dry, P., Proffitt, T., and Tyerman, S. (2011). The Grapevine: From the Science to the Practice of Growing Vines for Wine. Adelaide: Patrick Iland Wine Promotions.

Ishihara, M., Tsuneya, T., and Uneyama, K. (1991). Guaiene sesquiterpenes from agarwood.

Phytochemistry. 30, 3343-3347. Ishihara, M., Tsuneya, T., and Uneyama, K. (1993). Components of the volatile concentrate of

agarwood. J. Essent. Oil Res. 5, 283-289.

78

Kapadia, V.H., Naik, V.G., Wadia, M.S., and Dev, S. (1967). Sesquiterpenoids from the essential oil of cyperus rotundus. Tetrahedron Letters. 8, 4661-4667.

Keller, M. (2015). The Science of Grapevines: Anatomy and Physiology. Burlington: Academic Press.

King, E.S., Stoumen, M., Buscema, F., Hjelmeland, A.K., Ebeler, S.E., Heymann, H., et al. (2014). Regional sensory and chemical characteristics of Malbec wines from Mendoza and California. Food Chem. 143, 256-267.

Komm, B.L., and Moyer, M.M. (2015). Effect of early fruit-zone leaf removal on canopy development and fruit quality in Riesling and Sauvignon blanc. Am. J. Enology Vitic. 66, 424-434.

Kwasniewski, M.T., Vanden Heuvel, J.E., Pan, B.S., and Sacks, G.L. (2010). Timing of cluster light environment manipulation during grape development affects C13 norisoprenoid and carotenoid concentrations in Riesling. J. Agric. Food Chem. 58, 6841-6849.

Lacey, M.J., Allen, M.S., Harris, R.L.N., and Brown, W.V. (1991). Methoxypyrazines in Sauvignon blanc grapes and wines. Am. J. Enology Vitic. 42, 103-108.

Lee, S.-H., Seo, M.-J., Riu, M., Cotta, J.P., Block, D.E., Dokoozlian, N.K., et al. (2007). Vine microclimate and norisoprenoid concentration in Cabernet Sauvignon grapes and wines. Am. J. Enology Vitic. 58, 291-301.

Logan, G.A. (2015). Rotundone in New Zealand Vitis vinifera L. Syrah: fruit, fermentation and

functional food chemistry. [master’s thesis]. [Auckland (NZ)]: University of Auckland. Macauley, L.E., and Morris, J.R. (1993). Influence of cluster exposure and winemaking

processes on monoterpenes and wine olfactory evaluation of Golden Muscat. Am. J. Enology Vitic. 44, 198-205.

Mateo, J.J., and Jiménez, M. (2000). Monoterpenes in grape juice and wines. J. Chromatogr. A 881, 557-567.

Mattivi, F., Caputi, L., Carlin, S., Lanza, T., Minozzi, M., Nanni, D., et al. (2011). Effective

analysis of rotundone at below-threshold levels in red and white wines using solid-phase microextraction gas chromatography/tandem mass spectrometry. Rapid Commun. Mass Spectrom. 25, 483-488.

79

Meyers, J.M., and Vanden Heuvel, J.E. (2008). Enhancing the precision and spatial acuity of point quadrat analysis via calibrated exposure mapping. Am. J. Enology Vitic. 59, 425-431.

Naef, R. (2011). The volatile and semi-volatile constituents of agarwood, the infected heartwood of Aquilaria species: a review. Flavour Frag. J. 26, 73-89.

Nakanishi, A., Fukushima, Y., Miyazawa, N., Yoshikawa, K., Maeda, T., and Kurobayashi, Y.

(2017a). Identification of rotundone as a potent odor-active compound of several kinds of fruits. J. Agric. Food Chem. 65, 4464-4471.

Nakanishi, A., Fukushima, Y., Miyazawa, N., Yoshikawa, K., Maeda, T., and Kurobayashi, Y. (2017b). Quantitation of rotundone in grapefruit (Citrus paradisi) peel and juice by stable isotope dilution assay. J. Agric. Food Chem. 65, 5026-5033.

Nauer, S., Brandes, W., Patzl-Fischerleitner, E., Hann, S., and Eder, R. (2018). Analyse von (-)- rotundon mittels SPE-SPME-GC-MS in Österreichischen qualitätsweinen der rebsorte ‘Grüner Veltliner.’ Mitteilungen Klosterneuburg. 68, 107-119.

Niebler, J., Zhuravlova, K., Minceva, M., and Buettner, A. (2016). Fragrant sesquiterpene ketones as trace constituents in frankincense volatile oil of boswellia sacra. J. Nat. Prod. 79, 1160-1164.

Nelson, R.E. (2011). Compositional and sensory analysis of Finger Lakes Riesling. [master’s

thesis]. [Ithaca (NY)]: Cornell University.

Ong, P.K.C., and Acree, T.E. (1999). Similarities in the aroma chemistry of Gewürtztraminer variety wines and lychee (Litchi chinesis Sonn.) fruit. J. Agric. Food Chem. 47, 665-670.

Ou, C., Du, X., Shellie, K., Ross, C., and Qian, M.C. (2010). Volatile compounds and sensory attributes of wine from cv. Merlot (Vitis vinifera L.) grown under differential levels of water deficit with or without a kaolin-based, foliar reflectant particle film. J. Agric. Food Chem. 58, 12890-12898.

Pandey, A.K., and Chowdhury, A.R. (2002). Essential oil of cyperus scariosus R. Br. tubers from

central India. Indian Perfum. 46, 325-328. Pickering, G., Lin, J., Riesen, R., Reynolds, A., Brindle, I., and Soleas, G. (2004). Influence of

Harmonia axyridis on the sensory properties of white and red wine. Am. J. Enology Vitic. 55, 153-159.

80

Reisch, B.I., Luce, R.S., Bordelon, B., and Henick-Kling, T. (2006). “’Noiret’ grape.” In New York’s Food and Life Sciences Bulletin. (Geneva, NY: Cornell University), 1-8.

Reynolds, A.G., Edwards, C.G., Wardle, D.A., Webster, D.R., and Dever, M. (1994). Shoot density affects ‘Riesling’ grapevines. I. Vine performance. J. Am. Soc. Hortic. Sci. 119, 874-880.

Reynolds, A.G., Wardle, D.A., and Dever, M. (1996). Vine performance, fruit composition, and wine sensory attributes of Gewürztraminer in response to vineyard location and canopy manipulation. Am. J. Enology Vitic. 47, 77-92.

Reynolds, A.G., Wardle, D.A., and Naylor, A.P. (1991). Impact of training system, vine spacing, and basal leaf removal on Riesling. Vine performance, berry composition, canopy microclimate, and vineyard labor requirements. Am. J. Enology Vitic. 47, 63-76.

Roberts, R.W., Reynolds, A.G., and De Savigny, C. (2007). Composition and wine sensory attributes of Chardonnay Musqué from different viticultural treatments: Implications for a winegrape quality model. Int. J. Fruit Sci. 7, 57-83.

Robinson, A.L., Boss, P.K., Solomon, P.S., Trengrove, R.D., Heymann, H., and Ebeler, S.E. (2014). Origins of grape and wine aroma. Part 1. Chemical components and viticultural impacts. Am. J. Enology Vitic. 65, 1-24.

Ryona, I., Pan, B.S., Intrigliolo, D.S., Lakso, A.L., and Sacks, G.L. (2008). Effects of cluster light exposure on 3-isobutyl-2-methoxypyrazine accumulation and degradation patterns in red wine grapes (Vitis vinifera L. cv. Cabernet Franc). J. Agric. Food Chem. 56, 10838-10846.

Santesteban, L.G., Miranda, C., Barbarin, I., and Royo, J.B. (2015). Application of the measurement of the natural abundance of stable isotopes in viticulture: a review. Aust. J. Grape Wine Res. 21, 157-167.

Scarlett, N.J., Bramley, R.G.V., and Siebert, T.E. (2014). Within-vineyard variation in the ‘pepper’ compound rotundone is spatially structured and related to variation in the land underlying the vineyard. Aust. J. Grape Wine Res. 20, 214-222.

Scheiner, J.J., Sacks, G.L., Pan, B., Ennahli, S., Tarlton, L., Wise, A., et al. (2010). Impact of severity and timing of basal leaf removal on 3-isobutyl-2-methoxypyrazine concentrations in red winegrapes. Am. J. Enology Vitic. 61, 358-364.

81

Scheiner, J.J., Vanden Heuvel, J.E., Pan, B., and Sacks, G.L. (2012). Modeling impacts of viticultural and environmental factors on 3-isobutyl-2-methoxypyrazine in Cabernet franc grapes. Am. J. Enology Vitic. 63, 94-105.

Siebert, T.E., Wood, C., Elsey, G.M., and Pollnitz, A.P. (2008). Determination of rotundone, the pepper aroma impact compound, in grapes and wine. J. Agric. Food Chem. 56, 3745-3748.

Skinkis, P.A., Bordelon, B.P., and Butz, E.M. (2010). Effects of sunlight exposure on berry and wine monoterpenes and sensory characteristics of Traminette. Am. J. Enology Vitic. 61, 147-156.

Smart, R., and Robinson, M. (1991). Sunlight into Wine: A Handbook for Winegrape Canopy Management. Ashford: Winetitles.

Sweet, S.K., Wolfe, D.W., DeGaetano, A., and Benner, R. (2017). Anatomy of the 2016 drought

in the northeastern United States: Implications for agriculture and water resources in humid climates. Agric. For. Meteorol. 247, 571-581.

Talaverano, I., Ubeda, C., Cáceres-Mella, A., Valdés, M.E., Pastenes, C., and Peña-Neira, A. (2017). Water stress and ripeness effects on the volatile composition of Cabernet Sauvignon wines. J. Sci. Food Agric. 98, 1140-1152.

Takase, H., Sasaki, K., Ikoma, G., Kobayashi, H., Matsuo, H., Suzuki, S., et al. (2016a). Farnesyl diphosphate synthase may determine the accumulation level of (–)-rotundone in ‘Syrah’ grapes. Vitis 55, 99-106.

Takase, H., Sasaki, K., Shinmori, H., Shinohara, A., Mochizuki, C., Kobayashi, H., et al. (2015). Analysis of rotundone in Japanese Syrah grapes and wines using stir bar sorptive extraction (SBSE) with heart-cutting two-dimensional GC-MS. Am. J. Enoloy Vitic. 66, 398-402.

Takase, H., Sasaki, K., Shinmori, H., Shinohara, A., Mochizuki, C., Kobayashi, H., et al. (2016b). Cytochrome P450 CYP71BE5 in grapevine (Vitis vinifera) catalyzes the formation of the spicy aroma compound (–)-rotundone. J. Exp. Botany 67, 787-798.

Triolo, R., Roby, J.P., Plaia, A., Hilbert, G., Buscemi, S., Di Lorenzo, R., et al. (2018). Hierarchy

of factors impacting grape berry mass: separation of direct and indirect effects on major berry metabolites. Am. J. Enology Vitic. 69, 103-112.

82

Vanden Heuvel, J.E., Lerch, S.D., Lenerz, C.C., Meyers, J.M., and Mansfield, A.K. (2013). Training system and vine spacing impact vine growth, yield, and fruit composition in a vigorous young ‘Noiret’ vineyard. Horttechnology. 23, 505-510.

Van Leeuwen, C., and Seguin, G. (2006). The concept of terroir in viticulture. J. Wine Res. 17, 1-10.

Verzera, A., Tripodi, G., Dima, G., Condurso, C., Scacco, A., Cincotta, F., et al. (2014). Leaf

removal and wine composition of Vitis vinifera L. cv. Nero d’Avola: the volatile aroma constituents. J. Sci. Food Agric. doi: 10.1002/jsfa.7075.

Waterhouse, A.L., Sacks, G.L., and Jeffery, D.W. (2016). Understanding Wine Chemistry. West

Sussex: Wiley.

Wolf, T.K. (2008). Wine Grape Production Guide for Eastern North America. Ithaca: Plant and Life Science Publishing.

Wood, C., Siebert, T.E., Parker, M., Capone, D.L., Elsey, G.M., Pollnitz, A.L, et al. (2008). From wine to pepper: Rotundone, an obscure sesquiterpene, is a potent spicy aroma compound. J. Agric. Food Chem. 56, 3738-3744.

Yamada, M., and Sato, A. (2016). Advances in table grape breeding in Japan. Breed Sci. 66, 34-

45.

Zhang, P., Barlow, S., Krstic, M., Herderich, M., Fuentes, S., and Howell, K. (2015a). Within- vineyard, within-vine, and within-bunch variability of the rotundone concentration in berries of Vitis vinifera L. cv. Shiraz. J. Agric. Food Chem. 63, 4276-4283.

Zhang, P., Fuentes, S., Siebert, T., Krstic, M., Herderich, M., Barlow, E.W.R., et al. (2016). Terpene evolution during the development of Vitis vinifera L. cv. Shiraz grapes. Food Chem. 204, 463-474.

Zhang, P., Howell, K., Krstic, M., Herderich, M., Barlow, E.W.R., and Fuentes, S. (2015b). Environmental factors and seasonality affect the concentration of rotundone in Vitis vinifera L. cv. Shiraz wine. PLoS One. 10:7. doi: 10.1371/journal.pone.0133137.

Zhuang, S., Tozzini, T., Green, A., Acimovic, D., Howell, G.S., Castellarin, S.D, et al. (2014). Impact of cluster thinning and basal leaf removal on fruit quality of Cabernet Franc (Vitis vinifera L.) grapevines grown in cool climate conditions. HortScience. 49, 750-756.

83

Appendices

Appendix A. Fruit maturity and production metrics for all sites for 2016 and 2017 seasons. Year Site Treat-

menta TSS

(°Brix) pH TA

(g/L) Berry

wt (g)

Cluster wt (g)

Cluster (no./ vine)

Yield (kg/m)

Prun. wt (kg/m)

Crop load (kg/kg)

2016 1 C 18.4 3.61 6.36 1.99 205.9 14.5 2.98 0.82 3.64 1 LR 19.2 3.62 6.10 1.99 177.1 17.2 3.05 0.57 5.31 2 C 18.2 3.26 6.99 2.14 145.2 12.7 1.85 0.46 4.04 2 LR 20.2 3.33 6.35 1.95 109.5 14.1 1.54 0.67 2.30 3 C 20.0 3.46 6.30 1.97 181.4 21.3 3.87 0.54 7.13 3 LR 19.2 3.45 6.78 1.87 174.1 24.1 4.19 0.58 7.18 4 C 19.8 3.42 7.17 1.98 128.1 26.3 3.36 0.27 12.51 4 LR 19.4 3.43 7.37 1.93 137.4 26.8 3.68 0.40 9.18 5 C 19.8 3.41 7.92 1.92 152.3 44.2 6.72 0.43 15.56 5 LR 19.2 3.38 7.72 1.96 136.1 42.4 5.77 0.30 19.17 5 C NAb NA NA NA NA NA NA 0.70 NA 5 LR NA NA NA NA NA NA NA 0.70 NA 6 C 21.0 3.5 7.07 1.94 133.2 29.3 3.90 0.67 5.80 6 LR 20.4 3.42 6.39 1.98 101.5 34.5 3.50 0.49 7.19 6 C 19.6 3.47 4.97 1.88 124.0 32.5 4.03 0.56 7.23 6 LR 20.2 3.46 5.25 2.08 140.7 28.7 4.04 0.45 8.95 7 C 20.2 3.56 5.25 1.78 99.5 16.1 1.61 0.32 5.02 7 LR 20.8 3.52 5.69 1.61 104.1 25.2 2.62 0.27 9.83

2017 1 C 18.0 3.4 9.57 1.74 159.1 29.0 4.61 0.95 4.83 1 LR 18.7 3.37 6.86 1.64 127.3 29.0 3.69 0.45 8.23 2 C 16.9 3.24 7.75 1.96 112.2 16.8 1.88 NAc NA 2 LR 17.8 3.19 7.49 1.94 115.1 22.4 2.58 NA NA 3 C 19.2 3.46 7.21 1.68 146.9 38.3 5.62 0.81 6.93 3 LR 18.2 3.37 7.36 1.79 124.6 35.2 4.38 0.66 6.59 4 C 18.1 3.53 6.99 2.44 131.7 39.3 5.18 0.42 12.37 4 LR 18.0 3.62 7.62 2.20 136.4 30.1 4.10 0.45 9.18 5 C 17.1 3.45 8.77 1.79 133.2 30.6 4.08 0.39 10.43 5 LR 17.8 3.3 8.89 1.77 99.1 53.0 5.25 0.29 18.30 5 C 18.2 3.23 8.56 1.89 151.2 37.3 5.64 0.36 15.88 5 LR 18.0 3.39 8.99 1.80 125.4 51.8 6.49 0.26 25.28 6 C 19.1 3.39 8.31 2.01 144.5 26.2 3.79 0.93 4.08 6 LR 19.8 3.23 8.23 1.95 115.4 34.2 3.94 0.60 6.53 6 C 19.0 3.33 8.61 1.85 112.1 34.0 3.81 0.92 4.17 6 LR 20.4 3.41 7.84 1.76 112.3 39.9 4.48 0.61 7.35 7 C 19.6 3.38 9.02 2.00 121.4 32.5 3.95 1.04 3.78 7 LR 19.2 3.62 7.98 1.89 142.9 27.3 3.91 0.99 3.95

aC: Control; LR: fruiting zone leaf removal. bData unavailable due to commercial harvest of experimental fruit. cData unavailable due to commercial dormant pruning of experimental vines.

84

Appendix B. Nutrient concentrations and water status (via δ13C) of experimental units across all sites for 2016 and 2017. Year Site Treatmenta N

(%) P

(%) K

(%) Mg (%)

Ca (%)

δ13C (‰)

2016 1 C 1.00 0.19 2.61 0.26 1.06 -28.6 1 LR 1.07 0.24 3.14 0.21 1.08 -29.1 2 C 0.88 0.54 1.10 0.84 1.25 -29.4 2 LR 0.89 0.52 1.25 0.68 1.30 -28.3 3 C 0.75 0.13 0.93 1.32 1.82 -28.0 3 LR 0.85 0.15 1.07 1.03 1.82 -28.4 4 C 0.92 0.23 1.18 0.83 2.47 -26.2 4 LR 1.03 0.26 2.29 0.6 2.40 -26.3 5 C 1.06 0.29 2.74 0.60 1.63 -27.2 5 LR 0.97 0.24 1.78 0.82 1.76 -28.1 5 C 1.04 0.18 2.81 0.59 1.78 NAb

5 LR 1.08 0.23 3.23 0.58 1.91 NA

6 C 0.76 0.27 1.41 0.83 2.06 -26.2 6 LR 0.64 0.32 1.23 0.93 2.41 -27.2 6 C 0.69 0.53 0.70 1.35 2.45 -26.2 6 LR 0.70 0.37 0.82 1.18 2.28 -27.6 7 C 0.86 0.13 1.92 0.68 2.29 -25.2 7 LR 0.83 0.10 1.74 0.69 2.07 -24.8

2017 1 C 1.17 0.34 2.53 0.27 1.19 -28.5 1 LR 0.95 0.35 3.23 0.31 1.68 -28.8 2 C 0.75 0.59 0.79 0.78 1.33 -29.4 2 LR 0.72 0.62 0.82 0.91 1.72 -29.1 3 C 1.12 0.17 0.86 0.99 1.88 -27.3 3 LR 1.17 0.17 1.53 0.72 2.08 -27.9 4 C 0.78 0.38 2.04 0.39 1.99 -27.7 4 LR 0.76 0.46 2.48 0.30 1.99 -27.9 5 C 1.01 0.24 2.34 0.85 1.95 -27.1 5 LR 1.01 0.32 1.79 1.01 1.96 -28.2 5 C 0.91 0.19 2.11 0.54 1.83 -27.9 5 LR 0.92 0.26 2.62 0.58 2.01 -28.5 6 C 0.81 0.33 0.93 0.63 2.02 -28.7 6 LR 0.68 0.48 0.63 1.05 2.83 -29.2 6 C 0.67 0.52 0.83 0.94 2.14 -29.7 6 LR 0.72 0.49 0.53 1.15 2.51 -29.1 7 C 0.82 0.27 1.33 0.66 1.93 -28.4 7 LR 0.74 0.19 1.01 0.72 2.04 -29.6

aC: Control; LR: fruiting zone leaf removal. bData unavailable due to commercial harvest occurring prior to harvest of experimental vines.

85

Appendix C: Supporting details regarding mesoclimatic model diagnostic analysis

Multicollinearity analyses were also performed on the selected four- and five-variable

candidate models. Variance inflation (VIF) values were compared to a tolerance threshold of 6.8

for the four-variable model. All variable VIF were below the threshold (GDDv = 1.10; Ca =

1.06; crop load = 1.11; and P = 1.10) suggesting an absence of collinearity. The overall model

was statistically significant (p < 0.05), and had an F-value of 34.92. The highest condition index

value was for the variable P (1.45), followed by crop load (1.38). Additionally, P had a

moderately high proportion of variation value corresponding to crop load (proportion of variation

value = 0.53), and crop load with GDDv (proportion of variation value = 0.64), suggesting that

these predictor variables may not be well estimated.

Multicollinearity analysis of the five-variable model also suggests a low likelihood of

collinearity. A tolerance threshold of 7.87 was calculated for the five-variable model, and all VIF

values fell below this threshold (GDDv = 1.51; Ca = 1.17; P = 1.35; pruning weight = 1.18; pH =

1.59), suggesting a low likelihood of multicollinearity. The model was statistically significant (p

< 0.05) as well, and had a F-value of 31.71. The five-variable model yielded low condition

indices for all predictor variables but had the widest range of eigenvalues reported (from 1.59 to

0.35) from all three models, suggesting that this model has the highest likelihood of

multicollinearity (Freund and Littell, 2006). Furthermore, pH had the highest condition index

and very high proportion of variation values corresponding to itself and GDDv (0.75 and 0.68,

respectively), and pruning weight had a moderately high proportion of variation value

corresponding to Ca (0.52). These results altogether indicate that collinearity may be an issue for

the five-variable model.

86

Analysis of model residuals for all three candidate models was performed in order to

detect the presence of outliers, influential observations, and any violations of normality or

homogenous variance. PROC REG was used with the INFLUENCE option to analyze the

models for the presence of outliers and influential observations, using values generated for the

following statistics: RStudent, Hat Diagonal H, CovRatio, DFFITS, DFBETAS, and PRESS.

Values for the RStudent statistic, used to assess possible outliers, were compared to a calculated

t-distribution with 24 degrees of freedom (DF) for the three-variable model, 23 DF for the four-

variable model, 22 DF for the five-variable model (DF = n - m - 2, where n = the number of

observations within the model, and m = the number of independent variables within the model).

Hat Diagonal H values were used to assess the leverage of individual observations via comparing

values to a critical threshold of 0.27, 0.34, and 0.41, for the three-, four- , and five-variable

models, respectively (threshold, or hi, = 2*(m + 1) / n).

The CovRatio is used to assess whether the inclusion of a given observation within a

model provides the model with increased precision and minimized variance, or decreased

precision and greater variance. The calculated bounds used to assess CovRatio values were 1.41

for the three-variable model, 1.51 for the four-variable model, and 1.62 for the five-variable

model [Bound = (1 + 3*(m + 1)/n)]. Values generated for the DFFITS statistic are used to also

assess the influence of observations, with values greater than 0.74, 0.83, and 0.90 for the

respective three models indicating a high incidence of influence caused by a specific observation

(DFFITS threshold = 2 Ö[(m + 1)/n]). A threshold for DFBETAS is used to identify specific

observations of individual variables that may be influential, and the lower bound is the same for

all models: 0.371, calculated using the formula 2/Ön.

87

Diagnostic statistics suggest that the three-variable model includes a few outliers and

influential variables. Four observations (9, 15, 20, and 29) were identified as being possible

outliers in the three-variable model via the corresponding residual and RStudent values.

Observations 23, 24, and 30 had respective h-values above the diagnostic threshold of 0.27 (h-

values were 0.37, 0.32, and 0.37, respectively) and were considered to have high leverage.

Observations 9 and 20 had CovRatio values of 0.44 and 0.40, far from 1.41, the calculated

bound. This suggests that these observations may be affecting the precision of the model;

furthermore, observations 9, 15, and 23 had DFFITS values exceeding the 0.74 threshold in

absolute value with values of 1.23, -1.36, and 1.08, indicating that these observations may also

be influential on the model. Analysis of DFBETAS values revealed that crop load as an

independent variable is responsible for the high influence of observations 15 and 23. These

analyses altogether indicate that observations 9 and 23 may be fit for removal from the three-

variable model.

Diagnostic analysis of the four-variable model also suggests the presence of outliers and

influential observations within the model. Residual analysis of RStudent values indicated that

observations 9, 15, 20, 23, and 29 are potential outliers within the data set. Additionally, h-values

for observations 23, 24, and 30 were 0.37, 0.34, and 0.37, and were all near or above the lower

bound of 0.34, indicating that these observations have high leverage. This mirrors the h-value

analysis of the 3-variable model. For the CovRatio analysis, values of observations 9 (0.99), 15

(0.57), 20 (0.51), and 29 (0.45) deviated from the CovRatio bound of 1.51, indicating that

observations 15, 20, and 29 have a large degree of influence on the model, and observation 9 to a

lesser extent. The DFFITS threshold of 0.83 was exceeded by observation values in absolute

value of observation 9 (1.21), while observation 15 appeared near the threshold (-1.37). This

88

again provides further evidence for the high influence of observation 9. Values greater than 0.37

for DFBETAS were also used to identify specific variable observations that are influential: these

include observations 20 (-0.61) especially 23 (0.99) for GDDv, and observation 15 for Ca (0.92),

and observation 9 for P (0.86). These analyses again suggest that observation 9 may be fit for

removal from this model.

The five-variable model has the highest likelihood of containing outliers or influential

observations according to the results of diagnostic analyses. Residuals of the five-variable model

for observations 9, 15, 19, 23, 29 were large, and especially large for observation 20, indicating

that these variables may be outliers. A threshold of 0.41 was used to assess h-values, and only

observation 9 had an h-value that exceeded this threshold (h9 = 0.47), suggesting that it has high

leverage within the model. Analysis of CovRatio values using a bound of 1.62 identified the

same influential observations as the 4-variable CovRatio analysis, as well as some additional

observations: observations 15 (0.85), 19 (0.45), 20 (0.04), and 29 (0.86). Observations 9, 15, 19,

20, 23, 29 violated the DIFFITS threshold of 0.90 with respective values of -1.09, -1.13, 1.33, -

2.84, 1.22, and 1.12, increasing the likelihood that these observations had a high degree of

influence. Further analysis of the DFBETAS values for these observations revealed that

observations 9 and 15 for Ca were influential (0.79 and 0.54, respectively), observations 19, 20,

and 23 were highly influential for GDDv (0.71, -2.03, and 0.99 respectively), pH observations 20

(1.84) and 29 (0.95) were highly influential, and observation 20 (2.05) for P was also highly

influential. These results indicate that many observations and specific variables included within

the five-variable model are problematic and impart a high degree of influence upon the model.

Therefore, the five-variable model may be overfitted and less accurate than the other two

candidate models.

89

Comparatively, the three- and four-variable models have a lower likelihood of containing

outliers and influential variables when compared to the five-variable model. A comparison of the

sum of squared residual (residual SS) values with the predicted residual sum of squares (PRESS)

across the models can be used to further analyze models for the presence of outliers or influential

observations. For example, PRESS statistics that are increasingly larger than the residual SS

suggests the presence of influential observations and outliers. Residual SS values for the three-

variable, four-variable, and five-variable models were 0.45, 0.31, and 0.27, while PRESS statistic

values were 0.64, 0.50, and 0.52. Comparison of these values indicate that the differences

between the actual and predicted SS values within each candidate model were the same for the

three- and four-variable models (PRESS – ResidualSS = 0.19), and larger for the five-variable

model (PRESS – ResidualSS = 0.25). This provides further evidence that the five-variable model

has influential observations or outliers included within the model.

Diagnostic analyses indicate that the five-variable model is not the best model for

predictive purposes. Given the added complexity of the five-variable model when compared to

the three- and four-variable models and the limited increase in r2 and adjusted r2 between the

four- and five-variable models (0.85 and 0.82 for the four-variable model, compared to 0.87 and

0.84 resulting from an added variable in the five-variable model), only the three- and four-

variable models were selected for further analysis as candidate models. Partial regression

leverage plots (PRLP) were created alongside residual plots using PROC REG and the

PARTIAL option to graphically assess potential violations of homogenous variance and

normality assumptions. Generated plots indicated that both the three- and four-variable models

did not violate either assumption and reaffirmed the linearity of the relationships of GDDv, Ca,

cropload, and P with rotundone concentrations when measured at harvest. Thus, no

90

transformation of the data was required to improve the fit of the individual variables or satisfy

the assumptions of normality and homogenous variance.

Appendix D: Supporting details regarding microclimatic model diagnostic analysis

Two models were selected for diagnostic analyses to aid with final microclimatic model

selection: a two-variable model (DH30, and CEFAp), and a three-variable model (DH10, DH30,

and CEFAp). Diagnostic analysis was performed using the same methodology previously

mentioned in Appendix C. Collinearity analysis was first performed, and variance inflation (VIF)

values were analyzed for both models. All variable VIF values for the two-variable model

exceeded the calculated tolerance threshold of 1.97 (DH30 = 4.90; CEFAp = 4.90), indicating a

slight likelihood of collinearity, and the model was statistically significant (p < 0.000) with an F-

value of 10.26. Condition index values for the two variables were low, but the proportion of

variation value corresponding to CEFAp and DH30 was high (proportion of variation value =

0.946) while the eigenvalue for CEFAp was very low (0.107). This suggests that collinearity

does exist within the model.

For the three-variable model, two variables exceeded the tolerance threshold of 2.34

(DH30 = 5.04; CEFAp = 6.60), while the third did not (DH10 = 1.96). Again, this indicates a

slight likelihood of collinearity. The three-variable model had a smaller F-value than the 2-

variable model (F-value = 9.02), and was still statistically significant (p < 0.000). Condition

index numbers for the three variables were low as well, but the proportion of variation value

corresponding to CEFAp and DH30 was high (0.86) while the eigenvalue for CEFAp was very

low (0.090), indicating that this variable may be associated with collinearity within the model.

This is supported by the ranges of eigenvalues, as these values ranged from 0.107 to 1.892 for

91

the two-variable model and 0.090 to 2.43 for the three-variable model, indicating that the three-

variable model has a higher likelihood of including collinear variables (Freund and Littell, 2006).

Overall, these values suggest that collinearity most likely exists within the 3-variable model.

Analysis of model residuals was performed for both models to detect the presence of

outliers, influential observations, and any violations of normality and homogenous variance.

Studentized residual statistics (i.e., RStudent values) were compared to a calculated t-distribution

with 20 degrees of freedom (DF) for the two-variable model, and 19 DF for the three-variable

model (DF = n - m - 2, where n = the number of observations within the model, and m = the

number of independent variables within the model). For both models, observations 11, 23, and

24 were identified as possible outliers using the RStudent statistics.

Comparison of Hat Diagonal H values for all observations in each model with a model-

specific critical threshold (threshold, or hi, = 2*(m + 1) / n) revealed that two observations had

high leverage within each model. For the two-variable model, h-values were compared to a

critical threshold of 0.25 and observations 11 and 24 had values higher than this threshold (h11 =

0.37, and h24 = 0.35). Similarly, for the three-variable model observations 11 and 24 also had

higher h-values (h11 = 0.37, and h24 = 0.35) than the calculated critical threshold of 0.33. These

results suggest that these two observations have high leverage in both candidate models, and that

they may be unfit for retention in the final selected model.

The CovRatio was used to further assess observations and see which observations most

strongly affected the precision of the model and the overall model variance. Values for each

model were compared to model-specific CovRatio bounds that were calculated using the

following equation: critical bounds = [1 + 3*(m + 1)/n]. The CovRatio values for the two-

variable model were compared to 1.37, while the values for the three-variable model were

92

compared to 1.5. Observations 14, 17, and 23 had CovRatio values (0.75, 0.75, and 0.52,

respectively) that deviated the most from the critical threshold of 1.37 for the two-variable

model. Comparatively, for the three-variable model, the CovRatio values for observations 11

(0.60), 14 (0.79), 23 (0.45), and 24 (0.68) were the furthest from the threshold of 1.5. Since these

values were less than the threshold, it is thus likely that the inclusion of these observations in the

dataset would decrease model precision. Moreover, this analysis further supports the removal of

observations 11, 23, and 24 in particular, as multiple diagnostic analyses have identified these

observations as being problematic for model accuracy and precision.

Lastly, the two models were analyzed using DFFITS, DFBETAS, and PRESS statistics,

in order to better understand which observations may be influential and whether any outliers

exist. All DFFITS values for observations within the two-variable model were compared to a

threshold of 0.707, while all DFFITS values for observations within the three-variable model

were compared to a threshold of 0.816; if the absolute value of any DFFITS value exceeded

these thresholds, the observation associated with that value is considered an influential

observation. For the two-variable model, DFFITS values for observations 11 (1.68), 23 (1.56),

and 24 (1.62) all exceeded the calculated threshold of 0.707. Moreover, DFFITS values for

observations 11 (2.00), 21 (0.958) , 23 (1.54), and 24 (1.73) all exceeded the threshold of 0.816,

further supporting the possibility of these observations being influential.

Analysis of DFBETAS provided further clarification of the role of influential

observations in both candidate models, and which independent variables influential observations

were associated with. All DFBETAS values for both models were compared to a threshold of

0.408 (threshold = 2/Ön), with observations and independent variables considered influential if

the respective value exceeded this threshold. Observations 11, 23, and 24 were identified as

93

influential observations for both independent variables in the two-variable model: respective

DFBETAS values for DH30 were 1.46, 1.38, and 0.85, while respective DFBETAS values for

CEFAp were 1.58, 1.11, and 1.33. For the three-variable model, results were more variable:

observation 11 had DFBETAS values exceeding the threshold only for DH30 (1.68) and CEFAp

(1.53); observation 17 had DFBETAS values exceeding the threshold only for DH10 (0.580);

observation 21 had DFBETAS values exceeding the threshold only for DH10 (0.73) and CEFAp

(0.58); observation 23 had DFBETAS values exceeding the threshold for DH30 (1.20), DH10

(0.48), and CEFAp (0.65); and observation 24 had DFBETAS values exceeding the threshold

only for DH30 (0.89) and CEFAp (1.20). These results altogether indicate that observations 11,

23, and 24 are influential observations, and that removal of these observations may increase

model accuracy and precision.

The PRESS statistic was also used to evaluate both candidate models and further assess

for the presence of influential observations. For the two-variable model, the residual SS was

821766 and the PRESS was 1321835; additionally, for the three-variable model, the sum of

squared residuals was 690686, while the PRESS was 1271399. Comparison of these values

indicate that the differences between the actual and predicted SS values differ between models.

There is a larger difference between the actual and predicted SS values for the three-variable

model, which suggests that there is a stronger presence of outliers or influential variables within

the model than in the two-variable model. This reaffirms the findings of the previous diagnostic

analyses, and supports the removal of influential observations from the model.

Overall, diagnostic analyses indicate that there are influential observations in both the

two- and three-variable models. Given that these models were constructed with a low number of

observations per variable, however, removing any of the observations from the models is not

94

advisable. Further, despite the added complexity of the three-variable model due to the inclusion

of a third variable (DH10), this inclusion also helps explain a higher percentage of the overall

variation in rotundone concentrations. Indeed, the adjusted r2 value for the three-variable model

is 0.51 (r2 = 0.57), compared to an adjusted r2 value of 0.44 for the two-variable model (r2 =

0.49). The main objective of model construction using microclimatic variables was to identify

which variables best capture the variation in rotundone concentrations, and the inclusion of DH10

as a third regressor variable helps satisfy this objective. Lastly, though this model identifies a

subset of microclimatic variables that explain about 51% of observed variation in rotundone

concentrations, further data collection is necessary in order to validate this model and support

these associations.