An Abstract of the Dissertation Of

AN ABSTRACT OF THE DISSERTATION OF

Melekşen Akın for the degree of Doctor of Philosophy in Horticulture presented on June 8, 2016.

Title: Statistical Methods for Tissue Culture Medium Optimization and A Multiplexed Fingerprinting Set for Hazelnuts

Abstract approved: ______Barbara M. Reed

Hazelnut is one of the most important nuts in worldwide production and the European hazelnut (Corylus avellana L.) is the most economically important of the 11 recognized hazelnut species. Development of new cultivars is continuous, with emphasis on better nut quality, high yield and disease resistance. Hazelnuts are highly heterozygous, and clonally propagated. Traditional propagation methods in hazelnut are not rapid enough to provide the required nursery stock for newly released hazelnut cultivars, but micropropagation can provide rapid production of hazelnut planting stock. Several growth media are available for specific cultivars, but many are not suitable for the wide range of germplasm used in new cultivars. Micropropagation of hazelnuts remains challenging due to the various responses of diverse genotypes to in vitro growth. Several studies incrimentally improved the growth medium, but determining exact nutrient requirements was difficult. The aim of this study was to determine which statistical methods would make the growth medium optimization process more practical and to develop an optimal micropropagation medium for diverse hazelnuts by testing salts and ions as factors within the experimental design. In addition an SSR fingerprinting set suitable for a diverse group of hazelnuts was developed. The first study was designed to test the effect of salts on three hazelnut genotypes and compare two methods of data analysis. Driver and Kuniyuki Walnut

medium (DKW) macro-salts (NH4NO3, Ca(NO3)2∙4H2O, CaCl2∙2H2O, MgSO4∙7H2O,

KH2PO4 and K2SO4) were varied from 0.5× to 3× DKW concentrations with 42 combinations in a IV-optimal design. Shoot quality, shoot length, multiplication and callus formation were rated and analyzed using Response Surface Methodology (RSM) and the Chi-Squared Automatic Interaction Detection (CHAID) data mining algorithm. Both analyses indicated that NH4NO3 was a predominant nutrient factor. RSM results were genotype dependent while CHAID included genotype as a factor in the analysis, allowing development of a common medium rather than several genotype-specific media. Overall, CHAID results were more specific and easier to interpret than RSM graphs. The optimal growth medium for diverse hazelnut genotypes was formulated as: 0.5× NH4NO3, 3× KH2PO4, 1.5× Ca(NO3)2 and and the rest of the macro salts set at 1× DKW with modified minor nutrients [4× H3BO3, 4×

Na2MoO4∙2H2O, 4× Zn(NO3)2∙6H2O, 0.5× MnSO4∙H2O, 0.5× CuSO4∙5H2O]. The second study was to determine the effects of ions on tissue culture + 2+ 2+ 2- 3- medium optimization. NH4 , Ca , Mg , SO4 and PO4 ions were used as factors in + - a D-optimal design. K and NO3 ions were used to bring the pH level to neutral, and as factors in the statistical analysis. The CHAID data mining algorithm was used to analyze shoot growth responses of three hazelnut genotypes. The algorithm trees revealed significant variables and their interactions, and provided exact cut-off amounts for each of the ions for the related growth response by incorporating genotype as an independent factor. The critical cut-off values for good shoot quality, - elongation, multiplication and medium callus formation were suggested to be: NO3 + 2+ 2+ + <88 mM, NH4 <20 mM, Ca <5 mM, Mg >5 mM and K <46 mM. Another step of the research was to develop a reliable and economical fingerprinting set consisting of high core-repeat SSRs (≥3) for genotype identification of 102 hazelnut accessions. Identification of trueness-to-type by phenotypic observation is very difficult and labeling mistakes during the several steps of micropropagation can result in costly errors. The use of SSRs for plant identification is preferred over other molecular markers because they are reproducible across laboratories, exhibit co-dominant inheritance, have a large number of alleles per locus and are randomly distributed throughout the genome. Twenty SSRs containing repeat

motifs of three or more nucleotides distributed throughout the hazelnut genome were screened on eight genetically diverse cultivars to assess polymorphism, allele size range, and ease of scoring. Six SSRs were discarded after genotyping 96 hazelnut samples, either due to large allele bin widths and/or alleles that do not match the motifs, complicating allele scoring. Fourteen polymorphic, easy-to-score SSRs with non-overlapping alleles were selected and amplified in a single multiplex. The multiplexed set generated the same alleles that were obtained when amplifying each SSR individually in the eight test accessions. SSR primer concentrations were then optimized to generate a clear signal for each locus. This 14-SSR fingerprinting set was used to genotype 102 hazelnut accessions, and distinguished unique accessions mainly according to parentage and in some cases based on geographic origin As a result of these studies, salt- and ion-based optimized tissue culture medium formulations were developed for diverse hazelnuts. The importance of salts and ions as factors within the experimental design and analysis was examined, and using salts as factors results in complexity within the design as the effects of ions can not be determined. Although salt optimization studies are still a powerful tool, and are experimentally easier, optimization at the ionic level provided a clearer evaluation of the ion-based growth responses, because the plants take up minerals as ions of the corresponding salts. Data mining (CHAID) was used to make the tissue culture optimization process more practical compared to analysis with the standard ANOVA, regression and RSM. CHAID delineated specific concentrations that were effective and allowed easier analysis of nutrient content for an improved medium. A reliable and cost-effective multiplexed fingerprinting set of 14 SSR markers was developed for confirming identity and paternity in diverse hazelnut cultivars and species and 102 accessions were fingerprinted.

Statistical Methods for Tissue Culture Medium Optimization and A Multiplexed Fingerprinting Set for Hazelnuts

by Melekşen Akın

A DISSERTATION

submitted to

Oregon State University

in partial fulfillment of the requirements for the degree of

Doctor of Philosophy

Presented June 8, 2016 Commencement June 2017

Doctor of Philosophy dissertation of Melekşen Akın on June 8, 2016

APPROVED:

Major Professor, representing Horticulture

Head of the Department of Horticulture

Dean of the Graduate School

I understand that my dissertation will become part of the permanent collection of Oregon State University libraries. My signature below authorizes release of my dissertation to any reader upon request.

Melekşen Akın, Author

ACKNOWLEDGEMENTS

I would like to thank Dr. Barbara M. Reed for her invaluable guidance, assistance and tremendous encouragement. I greatly appreciate the time and helpful suggestions of my committee members Dr. Nahla V. Bassil, Dr. Shawn A. Mehlenbacher, my statistics minor advisor Dr. Yuan Jiang and my graduate school representative Dr. Christopher C. Mundt. I would like to thank Dr. Ecevit Eyduran for his help and precious suggestions in data analysis. I would like to thank the NCGR tissue culture lab manager Jeanine DeNoma and NCGR plant genetics lab manager April Nyberg for training and sharing valuable experience. I would like to express my deepest gratitude to my family for their love, support and understanding. Last and special thanks to my friend Dr. Peral Eyduran for her invaluable emotional support and encouragement.

CONTRIBUTION OF AUTHORS

Dr. Barbara M. Reed was involved in the planning, design, analysis and interpretation of the data in Chapter 2 and Chapter 3. Dr. Ecevit Eyduran helped to develop and perform CHAID statistical analysis in Chapter 2 and Chapter 3. Dr. Nahla V. Bassil assisted with planning, design, analysis and interpretation of the data in Chapter 4. April Nyberg helped to set up the study in Chapter 4.

TABLE OF CONTENTS

Page

CHAPTER 1: INTRODUCTION AND LITERATURE REVIEW ...... 2 European Hazelnut ...... 2 Hazelnut Micropropagation ...... 3 Mineral Nutrition ...... 3 Experimental Design and Statistical Methods ...... 5 Ions as Factors in Experimental Design and Analysis ...... 7 DNA Markers ...... 9 References ...... 12

CHAPTER 2: CHAID DATA MINING ALGORITHM IS MORE PRACTICAL THAN RESPONSE SURFACE METHODOLOGY FOR TISSUE CULTURE MEDIUM OPTIMIZATION ...... 18 Abstract ...... 19 Introduction ...... 19 Materials and Methods ...... 22 Plant material and in vitro culture conditions ...... 22 Data ...... 23 Statistical Analysis ...... 23 Results and Discussion ...... 24 Quality ...... 24 Shoot length ...... 26 Shoot number...... 27 Callus ...... 28 Conclusions ...... 31 References ...... 31

CHAPTER 3: ION SPECIFIC EFECTS ON TISSUE CULTURE MEDIUM OPTIMIZATION ...... 43 Abstract ...... 44 Introduction ...... 44 Materials and Methods ...... 46 Plant material and in vitro culture conditions ...... 46 Data ...... 47

TABLE OF CONTENTS (Continued)

Page

Statistical Analysis ...... 47 Results and Discussion ...... 48 Quality ...... 49 Shoot length ...... 50 Shoot number...... 51 Callus ...... 51 Conclusions ...... 53 References ...... 53

CHAPTER 4: MULTIPLEXED MICROSATELLITE FINGERPRINTING SET FOR HAZELNUTS ...... 64 Abstract ...... 65 Introduction ...... 65 Materials and Methods ...... 67 Data Analysis...... 68 Results and Discussion ...... 69 Conclusions ...... 72 References ...... 73

CHAPTER 5: SUMMARY...... 86

BIBLIOGRAPHY ...... 89

APPENDICES ...... 95

LIST OF FIGURES

Figure Page

2.1. Response surface graph of mineral nutrient effects on hazelnut shoot quality for a.‘Dorris’, b.‘Wepster’ and c.‘Zeta’. The quality ratings were1=poor, 2=moderate, 3=good and highest (red-yellow) to lowest quality (green-blue)...... 34

2.2. Shoots of a.‘Dorris’, b.‘Wepster’ and c.‘Zeta’grown on the DKW salts control and two treatments which produced higher plant quality...... 35

2.3. The CHAID decision tree diagram for plant quality of ‘Dorris’, ‘Wepster’ and ‘Zeta’. The quality ratings were1=poor, 2=moderate, 3=good. Nodes were determined by the significance of the factors. Salt cut-off values are × DKW...... 36

2.4. Response surface graphs of mineral nutrient effects on shoot length (mm) of a. ‘Dorris’, b.‘Wepster’ and c. ‘Zeta’. The shoot lengths (mm) were color coordinated from longest (red-yellow) to shortest (green-blue). The red dot represents the control with average shoot length of 50 mm...... 37

2.5. The CHAID decision tree diagram for shoot length of ‘Dorris’, ‘Wepster’ and ‘Zeta’ hazelnuts. Nodes were determined by the significance of the factors. Salt cut- off values are × DKW...... 38

2.6. Response surface graphs of mineral nutrient effects on shoot number of a. ‘Dorris’, b. ‘Wepster’ and c. ‘Zeta’. The shoot numbers were color coordinated from most (red-yellow) to fewer shoots (green-blue). The red dot represents the control...39

2.7. The CHAID decision tree diagram for shoot number of ‘Dorris’, ‘Wepster’ and ‘Zeta’. Nodes were determined by the significance of the factors. Salt cut-off values are × DKW...... 40

3.1. The CHAID decision tree diagram for plant quality of hazelnuts ‘Barcelona’, ‘Jefferson’ and ‘Wepster’. Nodes were determined by the significance of the factors. Cut-off values are mM ion concentrations. Mean and predicted values are based on the 1=poor, 2= moderate, 3=good rating given to each shoot...... 55

3.2. The CHAID decision tree diagram for shoot length of ‘Barcelona’, ‘Jefferson’ and ‘Wepster’. Nodes were determined by the significance of the factors. Cut-off values are mM ion concentrations...... 56

LIST OF FIGURES (Continued)

Figure Page

3.3. The CHAID decision tree diagram for shoot number of hazelnuts ‘Barcelona’, ‘Jefferson’ and ‘Wepster’. Nodes were determined by the significance of the factors. Cut-off values are mM ion concentrations...... 57

3.4. The CHAID decision tree diagram for callus formation of hazelnuts ‘Barcelona’, ‘Jefferson’ and ‘Wepster’. Nodes were determined by the significance of the factors. Cut-off values are mM ion concentrations. Mean and predicted values are based on the 1=large callus, 2=moderate callus, 3=no callus rating given to each shoot...... 58

3.5. Shoots of ‘Barcelona’, ‘Jefferson’ and ‘Wepster’ grown on treatments which are near the suggested ion concentration ranges...... 59

4.1. UPGMA cluster analysis of 81 unique genotypes among 87 hazelnut accessions using 14-SSRs multiplex...... 77

LIST OF TABLES

Table Page

2.1. Six factor design including 42 treatment points. Design points 1-40 for investigating the effects of individual factors on mineral nutrition of hazelnut cultivars and DKW medium controls (points 41-42). DKW medium concentrations 1×: NH4NO3 (1416 mg), Ca(NO3)2∙4H2O (1960 mg), CaCl2∙2H2O (147 mg), MgSO4∙7H2O (740 mg), KH2PO4 (259 mg), K2SO4 (1560 mg). All treatments included modified minor nutrients (Hand and Reed 2014)...... 41

2.2. DKW nutrient factors that had significant effects on four growth responses for each hazelnut cultivar at p-value (<0.05)...... 42

+ - 3.1. Five factor D-optimal design with 32 treatments. K and NO3 are not factors in + - the experimental design, but were used to adjust the pH. The values for K and NO3 are the amounts required to produce a neutral charge level of the ions, and were used in the final analysis...... 60

3.2. Salt types and amounts (mg/L) used for the treatments to obtain the specific ion concentrations...... 61

3.3. Macronutrient basal salts (mg/L) and ionic compositions (mM) for DKW and salt based optimal medium in Chapter 2, and optimal ion concentration ranges found in this study...... 63

4.1. List of 102 hazelnut genotypes used in this study. Local inventory number at the NCGR and OSU collection, species, pedigree and origin are indicated...... 78

4.2. Twenty SSRs tested for fingerprinting in hazelnuts. The motif, size range, number of alleles, linkage group (LG) and primer concentrations for the optmized multiplex set are provided...... 82

4.3. Diversity parameters of 14 single-locus hazelnut SSRs used in multiplex. Allele number (A), observed heterozygosity (Ho), expected heterozygosity (He) or gene diversity, polymorphism information content (PIC), and frequency of null alleles (r) in the 87 hazelnuts evaluated...... 83

4.4. 14-SSR fingerprints of some hazelnut accessions and their parents. Alleles transferred from the parents to the progeny are in bold. The accession in bold font is the progeny of the two parents in the section...... 84

LIST OF APPENDICES

Appendix Page

A. Response surface graphs of mineral nutrient effects on callus (1=callus > 2mm, 2=callus ≤ 2 mm, and 3=absent) ratings of a. ‘Wepster’ and b. ‘Zeta’. The callus was color coordinated from less callus (red-yellow) to more callus (green-blue). a. The red dot represents the control with average callus formation of 2...... 95

B. The CHAID decision tree diagram for callus formation of ‘Dorris’, ‘Wepster’ and ‘Zeta’ hazelnuts. Nodes were determined by the significance of the factors. Salt cut- off values are × DKW...... 96

C. Fingerprints of hazelnut genotypes for 14 single-locus hazelnut SSRs used in multiplex...... 97

D. Allele size ranges of six SSRs discarded and 14 SSRs used in the multiplex. Wide bin widths (> 1.4) are underlined; alleles that do not change according to the motif length are in bold...... 104

E. Electropherogram example for BR446 ...... 109

F. Electropherograms of 14-SSR multiplex ...... 110

G. Fingerprints of hazelnut genotypes for six single-locus hazelnut SSRs discarded from the study...... 111

STATISTICAL METHODS FOR TISSUE CULTURE MEDIUM OPTIMIZATION AND A MULTIPLEXED FINGERPRINTING SET FOR HAZELNUTS

CHAPTER 1: INTRODUCTION AND LITERATURE REVIEW

European Hazelnut The genus Corylus (Betulaceae) is native to temperate regions of the northern hemisphere extending from China, Japan, Korea, North America, the Russian Far East to the Caucasus and Turkey (Kasapligil 1972). Hazelnuts grow best in maritime regions with mild, moist winters and cool summers. The major production areas are located close to large water bodies at middle latitudes in the northern hemisphere. There are six species that grow as shrubs (C. avellana, C. americana, C. heterophylla, C. cornuta, C. californica, and C. sieboldiana) and five hazelnut tree species (C. colurna, C. jacquemontii, C. chinensis, C. fargesii and C. ferox) (Mehlenbacher 2009). The European hazelnut Corylus avellana L. is the most economically important with a worldwide production of around 872,000 t of in-shell nuts and a cultivated area of about 604,000 ha (average 2008-2012) (FAOstat 2016). Hazelnut is one of the most important nuts in worldwide production following cashew, walnut, almond, chestnut and pistachio. The major hazelnut producing countries are Turkey (598.158 t), Italy (104.577 t), USA (32.399 t), Azerbaijan (30.035 t) and Georgia (25.020 t) (average 2008-2012) (FAOstat 2016). The Black Sea region of Turkey produces almost 70% of the global hazelnut production. The Willamette Valley in Oregon provides 99% of the U.S. hazelnut production. Ninety percent of all hazelnut production is processed, for use in chocolate or bakery products. The demand for hazelnuts is continuously increasing. Hazelnuts are diploid (2n=2x=22), monoecious, dichogamous and wind- pollinated trees or shrubs with high genetic variability. Cross pollination is enforced due to sporophytic self-incompatibility, and use of pollinizers is required for good yield in a hazelnut orchard (Mehlenbacher 2014). Cultivars are highly heterozygous, and clonally propagated by conventional methods including layering, cutting and grafting, or micropropagated. Traditional propagation methods in hazelnut are not rapid enough to provide the required nursery stock material of the new released

hazelnut cultivars. Micropropagation is an option for rapid propagation of hazelnuts, but still has some drawbacks.

Hazelnut Micropropagation Micropropagation is a technique for rapid propagation of plants under sterile conditions. Year-round application and providing true-to–type, disease-free plantlets are among the advantages of micropropagation. The first in vitro hazelnuts were obtained from embryos (Radojevic et al. 1975). Several protocols and procedures were developed for micropropagation of hazelnuts. Murashige and Skoog (Murashige and Skoog 1962) was modified by changing the iron source (Fe EDTA replaced by Sequestrene 138 Fe), and adding Zuccherelli vitamins and plant growth regulators to promote multiplication (Al Kai et al. 1984). Many studies adjusted plant growth regulators to promote multiplication as well (Bassil et al. 1992; Damiano et al. 2005). Driver and Kuniyuki (DKW) medium (Driver and Kuniyuki 1984) was found superior to woody plant medium (WPM) and Anderson medium (Anderson 1984), and the carbon source of DKW medium was 3 % (w/v) glucose substituted for the standard sucrose, and sequestrene iron was used to increase multiplication of hazelnuts (Yu and Reed 1993; Yu and Reed 1995). Hazelnut shoots were tested for multiplication and quality on DKW, WPM, Perez-Tornero medium (PT) (Perez- Tornero et al. 2000) and half-strength MS medium. DKW medium or a combination of DKW medium and WPM were found superior (Damiano et al. 2005). Hazelnut shoot growth in a bioreactor was improved by adding antioxidants like ascorbic acid, melatonin, acetylsalicylic, or salicylic acid to DKW medium (Jyoti 2013). Most studies did not consider the components of the basal medium.

Mineral Nutrition Growth medium for micropropagation is mainly composed of mineral nutrients, a carbon source, vitamins and growth regulators. There are 14 elements essential for healthy plant growth. Plants need relatively large amounts of the ions of nitrogen, potassium, calcium, phosphorus, magnesium and sulphur, as well as the micro elements in small amounts (Ramage and Williams 2002). Nitrogen is a major

component in mineral nutrition and the primary constituent of various tissue culture media. Nitrogen is effective in a wide range of plant growth responses including callus growth, organogenesis, embryogenesis and multiplication (Niedz and Evens 2008). Calcium is a major cation for balancing anions within the plant, and found to be influential in the alleviation of shoot tip necrosis and hyperhydricity (Machado et al. 2014; Singha et al. 1990). Phosphorous is a component of many macromolecules including nucleic acids, and deficiencies could lead to failure of shoot meristem formation (George and de Klerk 2008; Ramage and Williams 2002). Sulphur deficiency leads to rigid, brittle, thin stemmed plants (Verbruggen and Hermans 2013). In vitro and in vivo plants differ in mineral nutrient uptake. In vitro plants are rootless, thus can take nutrients directly without the limiting effect of roots (Williams 1993). The agar within the tissue culture medium and the carbohydrate concentration are also factors influential on mineral uptake (Adelberg et al. 2010). Mineral nutrition affects overall plant quality, growth and development. Basal salts in growth media supply the macronutrients and micronutrients required for plant mineral nutrition. Although a variety of medium formulations are available that could be used for hazelnut micropropagation, growth responses of diverse cultivars and genotypes range from good to impossible to propagate (Yu and Reed 1995). The general media formulations for hazelnuts are usually slight modifications of traditional tissue culture media. MS medium was modified by reducing KNO3 and NH4NO3 levels, and replacing KH2PO4 with NaH2PO4, with iron doubled, and iodine reduced (Anderson 1984). The in vitro success of hazelnut cultivar Tonda Gentille Delle Langhe was improved by decreasing the nitrate concentration by one half, and doubling the CaCl2 and MgSO4 concentrations compared to MS (Diaz Sala et al. 1990). Another approach was to use the chemical composition of the hazelnut kernel to improve tissue culture medium for hybrid hazelnuts (Nas and Read 2004). This resulted in decreases in ammonium, increases in MgSO4 and KH2PO4, and changes to the micronutrients, especially increased Cu and Mo than DKW or MS media. Although there are many variables affecting plant growth, all of the mentioned optimization studies tested only a few factors at the same time. However, plant

nutrition is a complex phenomenon affected by many variables and their interactions. To make the mineral nutrition optimization process more effective, experimental designs and statistical methods that are able to evaluate many factors simultaneously, and detect their interactions on plant growth response, are required. The first studies to model hazelnut nutrients using response surface methodology (RSM) optimal design tested various combination levels of NH4NO3,

Ca(NO3)2, mesos (MgSO4 and KH2PO4), K2SO4 and minor nutrients in a range of

0.5× to 2× DKW medium (Hand et al. 2014). NH4NO3, Ca(NO3)2, mesos and minors had a significant effect on hazelnut growth and multiplication. The minor nutrient levels of DKW were then tested over a range of 0.5× to 4× DKW using a RSM optimal design, and found that there were many micronutrient interactions. Higher levels of B, Mo, and Zn and lower Mn and Cu resulted in increased overall shoot quality, length, and multiplication in the tested hazelnut genotypes (Hand and Reed 2014).

Experimental Design and Statistical Methods Proper experimental designs and statistical methods are required to make the tissue culture optimization process effective and unbiased. Tissue culture medium optimization studies have traditionally focused on a few factors studied at the same time, with simple ANOVA analysis of the factorial designs. Factorial designs require a very high number of treatment combinations, even when only a few factors are used (Compton and Mize 1999; Ibañez et al. 2003; Mize et al. 1999; Nas et al. 2005). In general, tissue culture plant growth optimization studies are set up as factorial experiments in completely randomized designs, randomized complete block designs or split-plot designs which do not allow testing of many factors at various levels due to the unmanageable number of treatment combinations created. This type of experiment requires a large number of explants and intensive labor which in general is not feasible due to limited number of explants at the initiation stage of an tissue culture optimization study, and limited budget (Compton and Mize 1999). For example, a five-factor traditional factorial design with 4 levels (45) consists of 1024 treatment combinations, and will require 2048 explants if 2 replicates per treatment

are used. Nas et al. (2005) suggested fractional factorial design to reduce the number of treatments. The procedure is similar to the factorial design but decreases the number of treatment combinations by half. For example, a five-factor fractional factorial design with 4 levels (1/2 45) consists of 512 treatments, which is still too high to be done in a typical laboratory. New methodologies for improving in vitro shoot growth by changing mineral nutrient concentrations include using advanced statistical optimization models such as RSM and neuro-fuzzy logic (Alanagh et al. 2014; Gago et al. 2011; Niedz and Evens 2007). Response Surface Methods are designs and models which allow optimization of continuous factors (Bradley 2007; Montgomery 2005). RSM is a statistical technique for modeling and analysis of responses affected by several factors, with the main objective to optimize the response. If the response is a linear function of the factors, it is a first order RSM model (multiple regression). If curvature is present in the response surface, a higher degree polynomial should be used, for example a second order polynomial. Computer generated RSM optimal designs consist of selected treatment points within all possible treatment combinations representing the experimental space, and allow testing of multiple nutrient factors at various levels, reducing treatments, time, labor and explant number tremendously compared to traditional factorial designs (Anderson and Whitcomb 2005; Hand and Reed 2014; Reed et al. 2013). For example, a five-factor traditional factorial design with three levels consists of 243 treatment combinations, whereas an optimal design would consist of 32 treatments representing the same design space. The main drawback of RSM with respect to tissue culture medium optimization is that RSM predicts genotype-specific optimization levels for the nutrients tested. RSM can deal only with continuous independent factors and can not incorporate nominal variables such as genotype. The objective in tissue culture medium optimization is often to develop a common medium for diverse genotypes rather than several genotype-specific media. Other statistical techniques used for defining optimal mineral nutrition are artificial neural network and neurofuzzy logic algorithms. The neurofuzzy logic algorithm is a hybrid technology that models nonlinear relationships between variables and combines artificial neural network and fuzzy logic techniques (Gago et al. 2011).

Data mining algorithms are another option for analysis of plant tissue culture optimization studies set up with RSM computer generated experimental design. The Chi-Squared Automatic Interaction Detection (CHAID) data mining algorithm constructs non-binary decision trees by multi-way node splitting the heterogeneous data set into many homogenous subgroups based on the significance and interactions of explanatory variables affecting the response (Ali et al. 2015). The algorithm is appropriate for nominal, ordinal, and continuous variables. As opposed to regression analysis and RSM, CHAID is a non-parametric technique, and therefore does not require distributional assumptions like normality and linearity. The algorithm is capable of detecting interactions between variables, and non-linear effects which are generally not defined by traditional statistical techniques. The CHAID tree shows in a simple way significant variables and their interactions, as well as the exact cut-off level ranges of the explanatory variables. The algorithm is also successful in overcoming missing data cases for dependent and independent variables (Hébert et al. 2006; Rashidi et al. 2014). The tree-based algorithm applied to the data obtained from an RSM optimal design would help to more easily define the optimum concentration ranges of nutrients in tissue culture media and make the optimization process more practical. CHAID can incorporate nominal variables such as genotype within the analysis, thus allowing growth response evaluation of several genotypes simultaneously, which could contribute to developing a common tissue culture medium rather than several genotype-specific media.

Ions as Factors in Experimental Design and Analysis The literature mainly documents tissue culture medium optimization studies based on salts as factors (Hand et al. 2014; Reed et al. 2013). Another approach to improve mineral nutrition would be to optimize the ions of the salts used. Ions are important in mineral nutrition because plants mainly utilize the concentrations of the + ions of nutrient salts. For example, nitrogen is absorbed by plants in the form of NH4 - and NO3 , thus the effect of nitrogen on plant responses is determined by the ion composition. Ions were found directly correlated with a wide range of in vitro plant growth responses. Nitrogen is reported to be influential on various plant growth

responses including callus growth, organogenesis, embryogenesis and multiplication (Niedz and Evens 2008). Nitrogen effects were determined on diverse pear species + + - using a mixture-amount design. Plant quality was affected by NH4 and K : NO3 - + + proportion. NO3 and NH4 :K ratio had a significant effect on shoot elongation and multiplication. Responses were analyzed with RSM and the concentration requirements were genotype dependent (Wada et al. 2015). Mixture component - design was used to determine nitrogen effects on raspberries. NO3 was the most + + significant factor on plant quality. NH4 and K mixture components and their - interaction with NO3 were effective on many of the examined responses including shoot length, multiplication and leaf color. Genotype-specific concentration requirement models for optimized response were built by RSM (Poothong and Reed 2016). Calcium is absorbed as Ca2+ ions and various concentrations were influential in the alleviation of shoot tip necrosis and hyperhydricity (Machado et al. 2014; Singha 2- et al. 1990). Plants utilize sulphur in the form of SO4 , and deficiency leads to rigid, brittle, thin stemmed plants. Magnesium is absorbed from the medium as the Mg2+ ion and is an essential constituent of chlorophyl, as well as an activator of many enzymes (Verbruggen and Hermans 2013). Phosphorous is a component of many macromolecules including nucleic acids, phospholipids and co-enzymes. It is utilized 3- by plants in the form of PO4 and deficiencies could lead to failure of shoot meristem formation (George and de Klerk 2008; Ramage and Williams 2002). Therefore the types and concentrations of ions within the media are very important in plant mineral nutrition. Although mineral nutrients are utilized in ionic form by the plants, the literature mainly documents optimization of mineral salt concentrations for various in vitro growth responses (Hand et al. 2014; Reed et al. 2013). Ion confounding is of major importance in tissue culture medium optimization when salts are used as predictor variables within experimental design. Salt experiments exhibit significant ion confounding, and do not provide characterization of the basic ion-specific effects on in vitro responses. Ion confounding occurs when the target ions are covaried by ions within the salts used. It is impossible to detect the ionic effect of a design set up with salts as factors. The effect on the plant growth response in an experiment

designed based on salts as factors, is the average effect of its anion and cation. For example the effect of NH4NO3 basal salt on the plant response actually comes from + - + - the effects of its ions and their interaction (NH4 , NO3 , NH4 × NO3 ). For identifying ion confounding, salt amounts and related ions within the tissue culture media should be converted to ion concentrations using pH adjustments (Niedz and Evens 2006).

DNA Markers Traditional methods for identification of cultivars and clones are based on phenotype and not always reliable due to environmental effects (Marinoni et al. 2003). DNA markers are naturally occurring DNA polymorphisms found through the genome, which provide information about the genetic characterization of an organism, and therefore are not dependent on the environment (Gupta et al. 1999). Molecular markers have many applications including marker-assisted selection, linkage map construction, cultivar fingerprinting, genetic diversity studies, and identification of duplicates in germplasm collections. DNA fingerprinting is useful for identity verification, which is important to protect breeders’ rights, and provide a tool for verifying cultivar integrity in propagation systems, or in a germplasm collection. There are two main categories of DNA markers, restriction fragment length polymorphism (RFLP) markers and polymerase chain reaction (PCR)-based markers. PCR-based markers include random amplified polymorphic DNA (RAPD), amplified fragment length polymorphism (AFLP), microsatellite or simple sequence repeat (SSR), inter-simple sequence repeat ISSR, and simple nucleotide polymorphism (SNP) (Testolin and Cipriani 2010). RFLP was the first type of marker developed. The DNA sample is cut by restriction enzymes and the resulting DNA fragments are separated by gel electrophoresis followed by hybridization with a DNA probe using Southern blotting, and visualized by autoradiography. Polymorphism arises from differences in the length of the fragments. RFLPs are co-dominant and different fragment sizes are scored as different alleles (Botstein et al. 1980). This hybridization based technique

was replaced with PCR-based markers, due to disadvantages such as requiring high- quality DNA, probe library construction, difficulties in statistical evaluations and standardization. PCR-based markers have many advantages over RFLP markers. They are faster, require a small amount of DNA, and allow exchange of primer sequence information (Testolin and Cipriani 2010). RAPD was the first type of PCR-based marker developed, which targets random sites in the genome (Williams et al. 1990). DNA fragments are amplified by PCR using arbitrary short primer sequences (mostly 10 base pairs in length). The primers anneal in opposite orientations on two complementary DNA strands to produce amplicons. PCR products are separated by size on agarose gels, stained with ethidium bromide, and visualized under ultraviolet light. PCR products are scored for presence or absence of unique bands. RAPD markers are dominant, easy to score, and inexpensive. The main limitation of RAPD is a lack of reproducibility between labs. RAPD markers were used for marker assisted selection and genetic linkage map construction in hazelnut (Mehlenbacher et al. 2006), and to study genetic diversity of Iranian hazelnut landraces (Mohammadzedeh et al. 2014). AFLPs are PCR-based dominant markers (Vos et al. 1995). DNA is digested into many fragments by two restriction enzymes (a frequent cutter and a rare cutter). Adapters are ligated to the DNA fragments which are then amplified in two steps using the PCR reaction. Fragments are separated by size using acrylamide gel electrophoresis, then stained with silver to visualize fragments as bands. Polymorphism is detected by presence or absence of the amplified PCR fragments. AFLPs are dominant markers, reproducible between labs, and reliable. However, these markers are expensive, complicated and require high-quality DNA (Testolin and Cipriani 2010). AFLP markers linked to eastern filbert blight resistance were identified in two hazelnuts (OSU 408.040 and ‘Ratoli’) (Chen et al. 2005; Sathuvalli et al. 2011). Microsatellites or simple sequence repeats (SSRs) are short tandem repeats of 1-6 bp sequence motifs, that are randomly distributed throughout the genome. SSRs are categorized as mono-, di-, tri-, tetra-, penta-, and hexanucleotides based on the number of nucleotides in the repeat motif. SSRs are multi-allelic, co-dominant, highly

polymorphic, relatively abundant within the genome, transferable to related species and genera, as well as reproducible between laboratories (Powell et al. 1996). The evolution of microsatellites is not clear. Many assume that it is due to slippage of DNA polymerase during DNA replication or unequal crossing-over between homologous chromosomes during recombination (Schlötterer and Tautz 1992). Unique alleles result in different lengths of amplified DNA fragment. More than 700 polymorphic SSR markers were developed in C. avellana (Bassil et al. 2013; Bassil et al. 2005a; Bassil et al. 2005b; Bhattarai 2015; Boccacci et al. 2005; Boccacci et al. 2015; Gürcan and Mehlenbacher 2010a; Gürcan and Mehlenbacher 2010b; Gürcan et al. 2010a; Peterschmidt 2013). These SSRs were used for linkage mapping (Bhattarai 2015; Gürcan et al. 2010a; Mehlenbacher et al. 2006), QTL analysis (Beltramo et al. 2016), for genetic relationship studies (Bassil et al. 2013; Boccacci et al. 2013; Boccacci and Botta 2010; Boccacci et al. 2006; Boccacci et al. 2008; Campa et al. 2011; Gokirmak et al. 2009; Gürcan et al. 2010b), for cultivar identification and parentage confirmation in hazelnut (Bassil et al. 2009; Botta et al. 2005; Gokirmak et al. 2009; Sathuvalli and Mehlenbacher 2012). SSRs with di-nucleotide repeats are the most abundant throughout the genome, but tri- and higher SSR repeats are also present (Hearne et al. 1992). Di- nucleotide containing SSRs usually show stuttering, split peaks and binning errors which lead to genetic profile discrepancies (Baldoni et al. 2009; Testolin and Cipriani 2010). Higher core repeats are preferable for automated fingerprinting, because they are easier to score. The use of different amplicon lengths and fluorescent tags allows multiplexing of several SSR markers in one PCR reaction and accurately scoring the size of each fragment by capillary electrophoresis. Multiplex fingerprinting sets for identity verification are less time consuming and more cost-effective than testing each SSR separately. The first objective of this research is to develop an optimized tissue culture medium for diverse hazelnuts using RSM optimal designs and testing salts and ions as experimental factors. Another objective is to define statistical methods which will make the optimization process more effective and practical. For that purpose we will compare RSM and CHAID statistical methodologies. The last objective is to develop

a reliable and economical fingerprinting set consisting of high core-repeat SSRs. SSRs with repeat motifs ≥ 3 nucleotides and with good genome coverage will be tested, and those that are easy-to-score with non-overlapping alleles will be selected and amplified in a single multiplex. A reliable and cost-effective SSR multiplexed fingerprinting set will be developed to confirm identity and paternity in diverse hazelnut accessions.

References Adelberg JW, Delgado MP & Tomkins JT (2010) Spent medium analysis for liquid culture micropropagation of Hemerocallis on Murashige and Skoog medium. In Vitro Cellular & Developmental Biology - Plant 46(1):95-107 Al Kai H, Salesses G & Mouras A (1984) Multiplication in vitro du noisetier (Corylus avellana L.). Agronomie 4:399-402 Alanagh EN, Garoosi GA, Haddad R, Maleki S, Landín M & Gallego PP (2014) Design of tissue culture media for efficient Prunus rootstock micropropagation using artificial intelligence models. Plant Cell, Tissue and Organ Culture 117(3):349-359 Ali M, Eyduran E, Tariq MM, Tirink C, Abbas F, Bajwa MA, Baloch MH, Nizamani AH, Waheed A, Awan MA, Shah SH, Ahmad Z & Jan S (2015) Comparison of artificial neural network and decision tree algorithms used for predicting live weight at post weaning period from some biometrical characteristics in Harnai sheep. Pakistan Journal of Zoology 47(6):1579-1585 Anderson MJ & Whitcomb PJ (2005) RSM simplified: optimizing processes using response surface methods for design of experiments. New York, NY: Productivity Press Anderson WC (1984) Micropropagation of filberts, Corylus avellana. Comb Proc Int Plant Prop Soc 33:132-137 Baldoni L, Cultrera N, Mariotti R, Ricciolini C, Arcioni S, Vendramin G, Buonamici A, Porceddu A, Sarri V, Ojeda M, Trujillo I, Rallo L, Belaj A, Perri E, Salimonti A, Muzzalupo I, Casagrande A, Lain O, Messina R & Testolin R (2009) A consensus list of microsatellite markers for olive genotyping. Mol. Breed. 24:213-231 Bassil NV, Boccacci P, Botta R, Postman J & Mehlenbacher SA ( 2013) Nuclear and chloroplast microsatellite markers to assess genetic diversity and evolution in hazelnut species, hybrids and cultivars. Genet. Resources Crop. Evol. 60:543- 568 Bassil NV, Botta R & Mehlenbacher SA (2005a) Microsatellite markers in hazelnut: isolation, characterization and cross-species amplification. J. Am. Soc. Hortic. Sci. 130:543-549 Bassil NV, Botta R & Mehlenbacher SA (2005b) Additional microsatellite markers of the European hazelnut. Acta Hort. 686:105-110

Bassil NV, Hummer K, Botu M & Sezer A (2009) SSR fingerprinting panel verifies identities of clones in backup hazelnut collection of USDA genebank. 845:95-102 Bassil NV, Mok D, Mok M & Rebhuhn BJ (1992) Micropropagation of the hazelnut, Corylus avellana. Acta Hortic 300:137-140 Beltramo C, Valentini N, Portis E, Torello Marinoni D, Boccacci P, Sandoval Prando MA & Botta R (2016) Genetic mapping and QTL analysis in European hazelnut (Corylus avellana L.). Molecular Breeding 36(3):1-17 Bhattarai G (2015) Microsatellite marker development, characterization and mapping in European hazelnut (Corylus avellana L.), and investigation of novel sources of eastern filbert blight resistance in Corylus. Dissertation, Oregon State University, Corvallis, USA Boccacci P, Akkak A, Bassil NV, Mehlenbacher SA & Botta R (2005) Characterization and evaluation of microsatellite loci in European hazelnut (Corylus avellana L.) and their transferability to other Corylus species. Molec. Ecol. Notes 5:934-937 Boccacci P, Aramini M, Valentini N, Bacchetta L, Rovira M, Drogoudi P, Silva AP, Solar A, Calizzano F, Erdoğan V, Cristofori V, Ciarmiello LF, Contessa C, Ferreira JJ, Marra FP & Botta R (2013) Molecular and morphological diversity of on-farm hazelnut (Corylus avellana L.) landraces from southern Europe and their role in the origin and diffusion of cultivated germplasm. Tree Genetics & Genomes 9(6):1465-1480 Boccacci P, Beltramo C, Sandoval Prando MA, Lembo A, Sartor C, Mehlenbacher SA, Botta R & Torello Marinoni D (2015) In silico mining, characterization and cross-species transferability of EST-SSR markers for European hazelnut (Corylus avellana L.). Molecular Breeding 35(1):1-14 Boccacci P & Botta R (2010) Microsatellite variability and genetic structure in hazelnut (Corylus avellana L.) cultivars from different growing regions. Scientia Horticulturae 124:128-133 Boccacci P, Botta R & Akkak A (2006) DNA typing and genetic relations among European hazelnut (Corylus avellana L.) cultivars using microsatellite markers. Genome 49:598-611 Boccacci P, Rovira M & Botta R (2008) Genetic diversity of hazelnut (Corylus avellana L.) germplasm in northeastern Spain. HortScience 43:667–672 Botstein D, White RL, Skolnick M & Davis RW (1980) Construction of a genetic linkage map in man using restriction fragment length polymorphisms. American Journal of Human Genetics 32(3):314-331 Botta R, Akkak A & Boccacci P (2005) DNA-typing of hazelnut: a universal methodology for describing cultivars and evaluating genetic relatedness. Acta Hortic 686:117-124 Bradley N (2007) The response surface methodology. Dissertation, Indiana University South Bend, USA Campa A, Trabanco N, Pérez-Vega E, Rovira M & Ferreira JJ (2011) Genetic relationship between cultivated and wild hazelnuts (Corylus avellana L.) collected in northern Spain. Plant Breeding 130(3):360-366

Chen H, Mehlenbacher SA & Smith DC (2005) AFLP markers linked to eastern filbert blight resistance from OSU 408.040 hazelnut. J. Amer. Soc. Hort. Sci. 130(3):412-417 Compton M & Mize C (1999) Statistical considerations for in vitro research: I — Birth of an idea to collecting data. In Vitro Cellular & Developmental Biology - Plant 35(2):115-121 Damiano C, Catenaro E, Giovinazzi J & Frattarelli A (2005) Micropropagation of hazelnut (Corylus avellana L.). Acta Hortic 686:221-225 Diaz Sala C, Rey M & Rodriguez R (1990) In vitro establishment of a cycloclonal chain from nodal segments and apical buds of adult hazel (Corylus avellana L). Plant Cell Tiss Organ Cult 23:151-157 Driver JA & Kuniyuki AH (1984) In vitro propagation of Paradox walnut rootstock HortScience 19:507-509 FAOstat (2016) Agriculture data http://faostat3.fao.org/home/index.html Accessed 25 May 2016 Gago J, Perez-Tornero O, Landin M, Burgos L & Gallego PP (2011) Improving knowledge of plant tissue culture and media formulation by neurofuzzy logic: a practical case of data mining using apricot databases. Journal of plant physiology 168(15):1858-65 George E & de Klerk G-J (2008) The components of plant tissue culture media I: macro- and micro-nutrients. In Plant propagation by tissue culture 3rd edition. Edited by: George EF, Hall MA, de Klerk G-J. Dordrecht, The Netherlands: Springer; 2008:65-113 Gokirmak T, Mehlenbacher SA & Bassil NV (2009) Characterization of European hazelnut (Corylus avellana) cultivars using SSR markers Genet. Resour. Crop. Ev 56:147-172 Gupta PK, Varshney RK, Sharma PC & Ramesh B (1999) Molecular markers and their applications in wheat breeding. Plant Breeding 118(5):369-390 Gürcan K & Mehlenbacher SA (2010a) Development of microsatellite marker loci for European hazelnut (Corylus avellana L.) from ISSR fragments Mol. Breeding 26:551-559 Gürcan K & Mehlenbacher SA (2010b) Transferability of microsatellite markers in the Betulaceae J. Amer. Soc. Hort. Sci. 135:159-173 Gürcan K, Mehlenbacher SA, Botta R & Boccacci P (2010a) Development, characterization, segregation, and mapping of microsatellite markers for European hazelnut (Corylus avellana L.) from enriched genomic libraries and usefulness in genetic diversity studies. Tree Genetics and Genomes 6:513-531 Gürcan K, Mehlenbacher SA & Erdogan V (2010b) Genetic diversity in hazelnut cultivars from Black Sea countries assessed using SSR markers. Plant Breeding 129:422-434 Hand C, Maki S & Reed B (2014) Modeling optimal mineral nutrition for hazelnut micropropagation. Plant Cell, Tissue and Organ Culture 119(2):411-425 Hand C & Reed BM (2014) Minor nutrients are critical for the improved growth of Corylus avellana shoot cultures. Plant Cell, Tissue and Organ Culture 119(2):427-439

Hearne CM, Ghosh S & Todd JA (1992) Microsatellites for linkage analysis of genetic traits. Trends in Genetics 8(8):288-294 Hébert M, Collin-Vézina D, Daigneault I, Parent N & Tremblay C (2006) Factors linked to outcomes in sexually abused girls: a regression tree analysis. Compr Psychiatry 47:443-455 Ibañez MA, Martin C & Pérez C (2003) Alternative statistical analyses for micropropagation: A practical case of proliferation and rooting phases in Viburnum opulus. In Vitro Cellular & Developmental Biology - Plant 39(5):429-436 Jyoti J (2013) Micropropagation of hazelnut (Corylus species). Master of Science, Plant Agriculture, University of Guelph, Ontario, Canada. Kasapligil B (1972) A bibliography on Corylus (Betulaceae) with annotations. Ann Rep North Nut Grow Assoc 63:107–162 Machado M, da Silva A, Biasi L, Deschamps C, Filho J & Zanette F (2014) Influence of calcium content of tissue on hyperhydricity and shoot tip necrosis of in vitro regenerated shoots of Lavandula angustifolia Mill Brazilian Archives of Biology and Technology 57(5):636-643 Marinoni D, Akkak A, Bounous G, Edwards KJ & Botta R (2003) Development and characterization of microsatellite markers in Castanea sativa (Mill.). Molecular Breeding 11(2):127-136 Mehlenbacher SA (2009) Genetic resources for hazelnut: state of the art and future perspectives. Acta Hort 845:33-38 Mehlenbacher SA (2014) Geographic distribution of incompatibility alleles in cultivars and selections of European hazelnut. J. Amer. Soc. Hort. Sci. 139:191-212 Mehlenbacher SA, Brown RN, Nouhra ER, Gokirmak T, Bassil NV & Kubisiak TL (2006) A genetic linkage map for hazelnut (Corylus avellana L.) based on RAPD and SSR markers. Genome 49:122-133 Mize C, Koehler K & Compton M (1999) Statistical considerations for in vitro research: II — Data to presentation. In Vitro Cellular & Developmental Biology - Plant 35(2):122-126 Mohammadzedeh M, Fattahi R, Zamani Z & Khadivi-Khub A (2014) Genetic identity and relationships of hazelnut (Corylus avellana L.) landraces as revealed by morphological characteristics and molecular markers. Scientia Horticulturae 167:17-26 Montgomery DC (2005) Design and analysis of experiments: response surface method and designs. New Jersey: John Wiley and Sons, Inc. Murashige T & Skoog F (1962) A revised medium for rapid growth and bio assays with tobacco tissue cultures. Physiol Plant 15:473-497 Nas MN, Eskridge K & Read P (2005) Experimental designs suitable for testing many factors with limited number of explants in tissue culture. Plant Cell, Tissue and Organ Culture 81(2):213-220 Nas MN & Read PE (2004) A hypothesis for the development of a defined tissue culture medium of higher plants and micropropagation of hazelnuts. Scientia Horticulturae 101(1-2):189-200

Niedz RP & Evens TJ (2006) A solution to the problem of ion confounding in experimental biology. 3(6):417 Niedz RP & Evens TJ (2007) Regulating plant tissue growth by mineral nutrition. In Vitro Cellular & Developmental Biology - Plant 43(4):370-381 Niedz RP & Evens TJ (2008) The effects of nitrogen and potassium nutrition on the growth of nonembryogenic and embryogenic tissue of sweet orange (Citrus sinensis (L.) Osbeck). BMC Plant Biol 8:126 Perez-Tornero O, Lopez JM, Egea J & Burgos L (2000) Effect of basal media and growth regulators on the in vitro propagation of apricot (Prunus armenica L.) cv. Canino. J. Hortic. Sci. Biotech. 75: 283-286 Peterschmidt B (2013) DNA markers and characterization of novel sources of eastern filbert blight resistance in European hazelnut (Corylus avellana L.). Dissertation, Oregon State University, Corvallis, USA Poothong S & Reed BM (2016) Optimizing shoot culture media for Rubus + − germplasm: the effects of NH4 , NO3 , and total nitrogen. In Vitro Cellular & Developmental Biology - Plant:1-11 Powell W, Machray GC & Provan J (1996) Polymorphism revealed by simple sequence repeats. Trends in Plant Science 1(7):215-222 Radojevic N, Vujieie R & Nesrovie M (1975) Embryogenesis in tissue culture of Corylus avellana L. Z Pflanzenphysiol Bd. 77: 33-41. Ramage C & Williams R (2002) Mineral nutrition and plant morphogenesis Cell. Dev. Biol. Plant 38:116–124 Rashidi S, Ranjitkar P & Hadas Y (2014) Modeling bus dwell time with decision tree-based methods. Transportation research record: Journal of the Transportation Research Board 2418:74-83 Reed BM, Wada S, DeNoma J & Niedz RP (2013) Improving in vitro mineral nutrition for diverse pear germplasm. In Vitro Cellular and Developmental Biology - Plant 49:343-355 Sathuvalli VR, Chen H, Mehlenbacher SA & Smith DC (2011) DNA markers linked to eastern filbert blight resistance in “Ratoli” hazelnut (Corylus avellana L.). Tree Genetics & Genomes 7(2):337-345 Sathuvalli VR & Mehlenbacher SA (2012) Characterization of American hazelnut (Corylus americana) accessions and Corylus americana x Corylus avellana hybrids using microsatellite markers Genetic Resources and Crop Evolution 59(6):1055-1075 Schlötterer C & Tautz D (1992) Slippage synthesis of simple sequence DNA. Nucleic Acids Research 20(2):211-215 Singha S, Townsend EC & Oberly GH (1990) Relationship between calcium and agar on vitrification and shoot-tip necrosis of quince (Cydonia oblonga Mill.) shoots in vitro. Plant Cell, Tissue and Organ Culture 23(2):135-142 Testolin R & Cipriani G (2010) Molecular markers for germplasm identification and characterization. Acta Hort 859:59-72 Verbruggen N & Hermans C (2013) Physiological and molecular responses to magnesium nutritional imbalance in plants. Plant and Soil 368(1):87-99

Vos P, Hogers R, Bleeker M, Reijans M, van de Lee T, Hornes M, Frijters A, Pot J, Peleman J & Kuiper M (1995) AFLP: a new technique for DNA fingerprinting. Nucleic Acids Research 23(21):4407-4414 Wada S, Niedz RP & Reed BM (2015) Determining nitrate and ammonium requirements for optimal in vitro response of diverse pear species. In Vitro Cellular & Developmental Biology - Plant 51(1):19-27 Williams JG, Kubelik AR, Livak KJ, Rafalski JA & Tingey SV (1990) DNA polymorphisms amplified by arbitrary primers are useful as genetic markers. Nucleic Acids Research 18(22):6531-6535 Williams RR (1993) Mineral nutrition in vitro—a mechanistic approach. Aust J Bot 41:237-251 Yu X & Reed BM (1993) Improved shoot multiplication of mature hazelnut (Corylus avellana L.) in vitro using glucose as a carbon source. Plant Cell Rep 12:256– 259 Yu X & Reed BM (1995) Micropropagation system for hazelnuts (Corylus species). HortScience 30:120–123

CHAPTER 2: CHAID DATA MINING ALGORITHM IS MORE PRACTICAL THAN RESPONSE SURFACE METHODOLOGY FOR TISSUE CULTURE MEDIUM OPTIMIZATION

Melekşen Akın1, Ecevit Eyduran2 and Barbara M. Reed3

1Oregon State University, Department of Horticulture, ALS 4017, Corvallis, OR 97331, USA 2Igdir University, Agricultural Faculty, Department of Animal Science, Biometry Genetics Unit, Igdir-Turkey 3USDA-ARS, National Clonal Germplasm Repository, 33447 Peoria Rd, Corvallis, OR 97333, USA

Submitted to Plant Cell Tissue and Organ Culture

Abstract Defining optimal mineral-salt concentrations for in vitro plant development is challenging, due to the many chemical interactions in growth media and genotype variability among plants. Statistical approaches that are easier to interpret are needed to make the optimization processes practical. Response Surface Methodology (RSM) and the Chi-Squared Automatic Interaction Detection (CHAID) data mining algorithm were used to analyze the growth of shoots in a hazelnut tissue-culture medium optimization experiment. Driver and Kuniyuki Walnut medium (DKW) salts

(NH4NO3, Ca(NO3)2∙4H2O, CaCl2∙2H2O, MgSO4∙7H2O, KH2PO4 and K2SO4) were varied from 0.5× to 3× DKW concentrations with 42 combinations in a IV-optimal design. Shoot quality, shoot length, multiplication and callus formation were evaluated and analyzed using the two methods. Both analyses indicated that NH4NO3 was a predominant nutrient factor. RSM projected that low NH4NO3 and high

KH2PO4 concentrations were important for quality, shoot length, multiplication and callus formation in some of the hazelnut genotypes. CHAID analysis indicated that

NH4NO3 at < 1.701× DKW and KH2PO4 at > 2.012× DKW were the most critical factors for shoot quality. NH4NO3 at < 0.5× DKW and Ca(NO3)2 at < 1.725× DKW were essential for good multiplication. RSM results were genotype dependent while CHAID included genotype as a factor in the analysis, allowing development of a common medium rather than several genotype-specific media. Overall, CHAID results were more specific and easier to interpret than RSM graphs. The optimal growth medium for Corylus avellana L. cultivars should include: 0.5× NH4NO3, 3×

KH2PO4 and 1.5× Ca(NO3)2. Keywords: Hazelnut, Medium optimization, Micropropagation, Mineral nutrition, Statistical analysis

Introduction Growth medium salts, plant growth regulators, temperature and lighting are all key factors for improving in vitro plant growth. Tissue culture medium optimization studies have traditionally focused on a few factors studied at the same time, and based on simple ANOVA analysis or classical factorial designs. Factorial designs require a

large number of treatments, even when only a few factors are included (Compton and Mize 1999; Ibañez et al. 2003; Mize et al. 1999; Nas et al. 2005). Experimental designs and statistical analyses that are able to evaluate the effect of many factors with various levels and their interactions on mineral nutrition of in vitro plants are required for better optimization process. New methodologies for improving in vitro shoot growth by changing mineral nutrient concentrations include using advanced statistical optimization models such as response surface methodology (RSM) and neuro-fuzzy logic (Alanagh et al. 2014; Gago et al. 2011; Niedz and Evens 2007). RSM is a statistical technique for modeling and analysis of responses affected by several factors, with the main objective to optimize the response. If the response is a linear function of the factors, it is a first order RSM model (multiple regression). If curvature is present in the response surface, a higher degree polynomial should be used, for example a second order polynomial. The parameters within the polynomials are estimated according to the Method of Least Squares. Response Surface Methods are designs and models which allow optimization of continuous factors (Bradley 2007; Montgomery 2005). Therefore, the main drawback with respect to tissue culture medium optimization is that nominal variables such as genotype cannot be included as a factor in the analysis. RSM simultaneously evaluates polynomial relationships between several independent variables and the dependent variable, and provides genotype specific models showing general areas of optimal plant growth. Computer generated RSM optimal designs allow testing of multiple nutrient factors at once, reducing treatments, time, labor and explant number compared to traditional factorial designs (Anderson and Whitcomb 2005; Hand and Reed 2014; Reed et al. 2013). Other statistical techniques used for defining optimal mineral nutrition are artificial neural network and neurofuzzy logic algorithms. The neurofuzzy logic algorithm is a hybrid technology that models nonlinear relationships between variables and combines artificial neural network and fuzzy logic techniques (Gago et al. 2011). Another option for data analysis that has not yet been used for plant tissue culture optimization is the Chi-Squared Automatic Interaction Detection (CHAID) data mining algorithm. CHAID constructs a visual (non-binary) decision tree that

contains many homogeneous subsets with multi-way node splitting from a heterogeneous data set, by selecting explanatory variables that significantly interact with a response variable (Ali et al. 2015). The algorithm is appropriate for nominal, ordinal, and continuous data. The decision tree building process consists of merging, splitting and stopping steps. Optimal splits are identified by chi-square statistics when the dependent variable is categorical, and an F-test is used when the response is continuous. Continuous variables are split into categories with similar numbers of observations. CHAID splits the explanatory variables based on their smallest Bonferroni adjusted p-value. The categories of the independent variables are used for calculating p-values to check whether the mean responses are same for different categories of the predictors. If the p-values are not significant, the pair is merged into a single group. A group with three or more categories is evaluated to define the most significant binary split. The splitting process continues until the node size is less than the predefined minimum node size value. The decision tree components are root node (containing the dependent variable), parent node (the first two or more categories after the data is split), child nodes (explanatory variable categories below the parent categories) and the terminal node, which is the last category. The most significant category on the dependent variable is at the top of the tree and the least important one (terminal node) is at the bottom (Rashidi et al. 2014; Statistics-Solutions 2016). As opposed to regression analysis, CHAID is a non-parametric technique, and therefore does not require distributional assumptions like normality and linearity. The algorithm can project interactions between variables, and non-linear effects which are generally missed by traditional statistical techniques (Hébert et al. 2006). The tree- based algorithm applied to the data obtained from an optimal design would help to more precisely define the optimum concentrations of salts in tissue culture media and to better understand any interactions. CHAID allows the analysis of responses of several genotypes simultaneously, which could contribute to developing a common tissue culture medium rather than several genotype-specific media. This study was designed to compare the RSM and CHAID analyses for in vitro culture data, and to provide practical approaches for tissue culture medium

optimization. Optimal shoot quality, shoot length, multiplication, and callus formation were determined with each statistical technique.

Materials and Methods

Plant material and in vitro culture conditions

Shoots of C. avellana L. hazelnuts ‘Dorris’, ‘Wepster’ and ‘Zeta’ were micropropagated on COR-2013 medium consisting of modified DKW (Driver and

Kuniyuki 1984) medium salts [1.5× Ca(NO3)2∙4H2O, 1.5× MgSO4∙7H2O and 1.5×

KH2PO4; 4× H3BO3, 4× Na2MoO4∙2H2O, 4× Zn(NO3)2∙6H2O, 0.5× MnSO4∙H2O, 0.5× -1 -1 CuSO4∙5H2O] with 30 g∙L glucose, 200 mg∙L sequestrene 138 Fe EDDHA, 2 mg∙L-1 thiamine, 2 mg∙L-1 nicotinic acid, 2 mg∙L-1 glycine, 1 g∙L-1 myo-inositol, 22.2 µM N6benzyladenine (BA), and 0.5% (w/v) agar (PhytoTechnology Laboratories A1111). Each vessel (Magenta GA7, Magenta, Chicago, IL) held 40 ml medium, and was autoclaved for 20 min at 121 0C. The growth room conditions consisted of 80 µmol∙m2s-1 light intensity with a 16-h photoperiod of half warm-white and half cool- white fluorescent lamps (Ecolux Starcoat, 32 watt; GE, Fairfield, CT) at 25±2 0C. A IV-optimal experimental design was set up by Design-Expert software (Design-Expert 2010) using the standard salt composition of DKW medium (1×) with the modified micronutrients listed above (Hand and Reed 2014). A six factor design with NH4NO3, Ca(NO3)2∙4H2O, CaCl2∙2H2O, MgSO4∙7H2O, KH2PO4 and K2SO4 salts were tested in a range of 0.5-3.0× DKW. Forty two treatments were assigned with the last two treatments as controls of DKW salts with modified micronutrients (Table 2.1). Shoots were cut to 3 cm with the apical meristem removed. During the first transfer the callus was removed and shoots were transferred to a new medium. In the second transfer shoots were reduced to 3 cm by removing the base and the apical meristem. For each treatment two boxes with five shoots for each cultivar were used (n=10). Boxes were randomized on the growth room shelf. Shoots were grown on each treatment medium for 10 wk with the first and second transfers for 3 wk, and the last transfer for 4 wk.

Data Data was recorded on individual plants. Three shoots collected on a diagonal (to include two corners and the central plant) from each culture vessel (n=6) were evaluated as follows: shoot quality, a subjective visual assessment of shoot vigor and form was evaluated as 1=poor, 2=moderate and 3=good. Shoots longer than 5 mm were counted. The longest shoots were measured in millimeters. Callus formation was evaluated as: 1=callus ≥ 2 mm, 2=callus < 2 mm, and 3=absent (no callus present).

Statistical Analysis The mean response (quality, shoot length, shoot number and callus) of six shoots per treatment for each genotype was assessed by RSM using Design Expert 8 software (Design-Expert 2010). The factors and their polynomial relationships and interactions which affected plant growth responses were determined at p<0.05. RSM does not provide an option for defining a variable structure; all factors are treated as continuous. Graphical models of RSM were produced based on the six salt-factor design space. Genotype was not included as a factor. Separate models were built for each genotype. The factors with highest statistical impact were assigned as axes of the graphs. CHAID analyzed quality, shoot length, shoot number and callus as dependent variables based on the six basal salts and the three genotypes (‘Dorris’, ‘Wepster’, and ‘Zeta’) as independent variables. The mean response of six shoots per treatment was calculated for each genotype. Therefore, the dependent variables and the basal salts were treated as continuous variables, only genotype was selected as a nominal variable within the algorithm. To prune the redundant structuring of the tree diagram, the ideal minimum plant numbers for parent and child nodes were assigned based on the best tree diagram with highest Pearson correlation and no overlapping of the factors. Overlapping is when an independent factor is significant on more than one node within the same branch, and causes objects to be included in more than one node (Díaz-Pérez and Bethencourt-Cejas 2016). The minimum plant numbers for quality were assigned as 28:14, shoot length 6:3, shoot number 24:12 and callus 12:6. The first number of each ratio is the minimum sample size for the last parent node (28 for

quality) and the second number is the minimum sample size which could be assigned for its child nodes (14 for quality). Statistical analyses of the tree-based CHAID algorithm were performed with IBM SPSS Statistics for Windows (Version 22.0, Armonk, NY) software (SPSS 2013).

Results and Discussion This study was designed to compare statistical methods for improving, and making more practical, the optimization process for mineral nutrients of tissue culture medium using RSM and the CHAID data mining methodology. RSM is a computer assisted technique which models optimal growth areas using selected treatment points within the range of all possible treatment combinations, and therefore reduces the number of treatments required for traditional factorial designs (Anderson and Whitcomb 2005). CHAID is a data mining algorithm used for constructing decision trees with homogeneous sub-groups. It is useful for detecting non-linear and interaction effects without requiring linearity and normality assumptions (Hébert et al. 2006).

Quality Plant quality rating is a subjective evaluation of general plant growth consisting of leaf and shoot health, as well as multiplication (Niedz et al. 2007). The RSM models for quality were significant (p<0.05) for all of the genotypes tested

(Table 2.2). Compared to DKW medium, increased KH2PO4 and lower K2SO4 concentrations affected the quality of all three cultivars (p<0.05) and the NH4NO3 requirement was also low (Fig. 2.1; Fig. 2.2). For 'Dorris' there were interactions (p<0.05) of several factors that impacted the shoot quality (Table 2.2). 'Dorris' required very high KH2PO4 and low K2SO4 concentrations and moderate to high

Ca(NO3)2 for higher quality shoots (Fig. 2.1a). Low NH4NO3 was a significant factor for quality of ‘Wepster’, as were very high KH2PO4 and low K2SO4 concentrations

(Fig. 2.1b). 'Zeta' quality models indicated that low to medium KH2PO4, low K2SO4 and low NH4NO3 were all required for the best growth. Improved growth was seen with several of the treatment combinations (Fig. 2.2).

The CHAID data mining algorithm indicated that the most significant factor for plant quality was NH4NO3, followed by KH2PO4 and K2SO4 (Fig. 2.3). The genotype effect in the CHAID algorithm was determined to be an insignificant source of variation for plant quality. The Pearson correlation coefficient between actual and predicted values for quality was 0.661 (p<0.01), indicating a medium to high predictive ability of the algorithm. All the plants used in the analysis (Node 0) were split into two nodes (Nodes 1 and 2) based on the response to NH4NO3 concentration.

The decisive cut-off value of NH4NO3 was 1.701×. The group of shoots with the best response to NH4NO3 < 1.701× (Node 1) was statistically different (adj. p<0.001) from the group of shoots responding best at NH4NO3 > 1.701× (Node 2). The quality of shoots grown on NH4NO3 < 1.701× was also influenced by KH2PO4, while the quality of shoots on NH4NO3 > 1.701× was significantly affected by K2SO4 (adj. p<0.001). Node 4, the group of plants exposed to NH4NO3 < 1.701× and KH2PO4 > 2.012× had significantly higher quality (2.037) than the other nodes (Fig. 2.3). This indicates that the resulting optimal medium for overall shoot quality would be

NH4NO3 < 1.701×, KH2PO4 > 2.012× for all three genotypes.

The differences in KH2PO4 and NH4NO3 requirement of plants noted in earlier studies could be related to genotype, the statistical methods applied, and the concentration ranges of variables used (Hand et al. 2014; Nas and Read 2004; Reed et al. 2013). Hand et al. (2014) linked quality with nitrogen factors, mesos and potassium sulfate, but the response was greatly dependent on genotype. All of the hazelnut genotypes in the current study required very low K2SO4 for better quality according to RSM graphs. K2SO4 was also significant in CHAID but only when

NH4NO3 > 1.701×. Ca(NO3)2 was significant for quality of ‘Dorris’ in RSM, but wasn’t significant in CHAID (Fig. 2.3). This difference could be due to the fact that CHAID algorithm simultaneously evaluated all three hazelnut cultivars. The cultivars were analyzed separately with RSM because genotype is not a continuous variable. Genotype was also evaluated separately in previous research using RSM (Hand et al. 2014; Reed et al. 2013). RSM is generally used to define the relationships between continuous (measurable) independent and dependent (response) variables, whereas CHAID is able to deal with categorical as well as continuous variables, and this

allows evaluation of genotype as a factor, which could help to develop a general medium for diverse hazelnuts.

Shoot length The RSM models for shoot length were statistically significant for all the hazelnut cultivars (p<0.05) (Table 2.2). Interactions of several factors influenced

(p<0.05) shoot length of ‘Dorris’ in RSM (Table 2.2). ‘Dorris’ required high KH2PO4 and Ca(NO3)2, but low NH4NO3 for an ideal shoot length of 40 mm (Fig. 2.4a).

‘Wepster’ had the same requirements of KH2PO4 and NH4NO3 for longer shoots (40 mm), but did not require high Ca(NO3)2 concentrations (Fig. 2.4b). Shoots of 'Zeta' were typically long and the low-medium NH4NO3 and low-medium K2SO4 concentrations found in DKW were sufficient for good (40-50 mm) shoot length (Fig. 2.4c). A CHAID decision tree diagram was constructed to predict shoot length from several significant factors: genotype, NH4NO3, KH2PO4, CaCl2, and K2SO4 (Fig. 2.5). There was a very high Pearson correlation of 0.853 between actual and predicted shoot length values (p<0.01). The main factor affecting shoot length was NH4NO3 (adj. p<0.001), and the second most important factor was genotype (adj. p<0.001).

KH2PO4, CaCl2, and K2SO4 were also influential for shoot length (adj. p<0.001). All of the plants in Node 0 at the top of the tree diagram were split into two nodes (Nodes

1 and 2) with respect to NH4NO3. Node 1 had significantly longer shoots (39.34 mm) than Node 2 (25.98 mm) (adj. p <0.0001). Genotype affected shoot length within both

Nodes 1 and 2 (adj. p <0.001). Node 1 (NH4NO3 < 1.701×) branched into two nodes by genotype. Node 3 (‘Dorris’ and ‘Wepster’) was divided into three nodes (Nodes 7,

8 and 9) based on KH2PO4. The greatest shoot length for these genotypes (39.44 mm) was with NH4NO3 < 1.701× and KH2PO4 > 2.75× (Node 9). Node 4 (‘Zeta’) on treatments with NH4NO3 < 1.701× had an average shoot length of 50.79 mm, and was divided into three new nodes (Nodes 10, 11 and 12), based on Ca(NO3)2. The longest shoots (60.389 mm) for ‘Zeta’ were recorded from Node 12 with NH4NO3 < 1.701× and CaCl2 > 1.8×. A common salt concentration for the tested genotypes with ideal shoot length of 40 mm is suggested to be: NH4NO3 < 1.701×, KH2PO4 > 2.75×, CaCl2 < 1.738×.

Shoot length results varied between the two analyses. The CHAID data mining algorithm determined that KH2PO4 > 2.75× and NH4NO3 < 1.701× were critical cut-off values for the longest shoots of ‘Dorris’ and ‘Wepster’. This was similar to RSM results. However, for ‘Zeta’ RSM indicated only NH4NO3 was important (Fig. 2.4; Table 2.2), while CHAID found CaCl2 also to be significant (Fig. 2.5). Nas and Read (2004) defined a micropropagation medium for hybrid hazelnuts (C. avellana x C. americana) based on the chemical composition of the hazelnut kernel. Better shoot elongation was observed on lower NH4NO3, Ca(NO3)2 and

CaCl2, as well as high MgSO4 and KH2PO4 compared to DKW. The differences in salt requirements could be attributed mainly to genotype (C. avellana vs. hybrid hazelnuts), experimental design and the various statistical methods applied. Hand et al. (2014) found shoot length quite variable by genotype, however increased nitrogen factors and mesos were commonly involved.

Shoot number RSM models for shoot number were significant (p<0.01) for all of the hazelnut cultivars (Table 2.2). NH4NO3 was significant for all of the genotypes (p<0.01), and low concentrations were required for high shoot number (Fig. 2.6).

‘Dorris’ produced high shoot multiplication (4.5) at the highest KH2PO4 and the lowest MgSO4 amounts (p<0.05). Low amounts of K2SO4 and very low Ca(NO3)2 increased shoot numbers for ‘Dorris’, depending on the amounts of NH4NO3 and

MgSO4 (p<0.05) (Fig. 2.6a). A negative interaction of NH4NO3 × CaCl2 resulted in the highest shoot number (2.6) for ‘Wepster’ (p<0.001), with the lowest NH4NO3 and the highest CaCl2 and a low amount of K2SO4 (Fig. 2.6b). ‘Zeta’ showed very low

Ca(NO3)2 and NH4NO3 requirements for high shoot number (2.4) (p<0.0001) (Fig.

2.6c). Low Ca(NO3)2 and NH4NO3 were both important for increased shoot production in other C. avellana genotypes (Hand et al. 2014). Nas and Read (2004) suggested lower NH4NO3, Ca(NO3)2 and CaCl2, as well as higher MgSO4 and

KH2PO4 for better multiplication compared to DKW. The CHAID algorithm indicated that the most significant factor for shoot number was NH4NO3 (adj. p<0.001), followed by Ca(NO3)2 (adj. p<0.01) (Fig. 2.7). Genotype was an insignificant source of variation for multiplication. The Pearson

correlation coefficient for shoot number was 0.653 (p<0.01). All the shoots (Node 0) were split into three nodes (Nodes 1, 2 and 3) based on the response to NH4NO3 concentration. Node 1 (NH4NO3 < 0.5×) shoots were affected by the Ca(NO3)2 concentration (average shoot number 2.197). Node 2 was those grown within the range of 0.5× < NH4NO3 < 2.6× (average shoot number 1.773), were also influenced by Ca(NO3)2 and KH2PO4 (adj. p<0.01). Shoot numbers of plants exposed to NH4NO3 > 2.6× were not affected by any other basal salt factor or genotype (adj. p<0.001).

The highest multiplication (2.46) was obtained in Node 4 with NH4NO3 < 0.5× and

Ca(NO3)2 < 1.725×) (Fig. 2.7). Salt requirements for good multiplication varied significantly depending on genotype in RSM (Fig. 2.6), whereas clear cut-off values of NH4NO3 < 0.5× and

Ca(NO3)2 < 1.725× were suggested by the CHAID for best multiplication for the all tested genotypes (Fig. 2.7). Hand et al. (2014) using RSM suggested low amounts of

NH4NO3 and Ca(NO3)2 for a higher multiplication rate of five Corylus avellana cultivars. Our RSM results showed higher shoot number (4.5 versus 2.5) for ‘Dorris’ than Hand et al. (2014), which could be attributed to the higher concentration range of

KH2PO4 within the experimental design (3× versus 1.5×), as well as to the differences in micro nutrient concentrations used in both studies.

Callus Responses of ‘Wepster’ and ‘Zeta’ were significant for callus formation in

RSM (p<0.05), and the most influential factor was NH4NO3 (p<0.001) (Table 2.2).

KH2PO4, K2SO4, NH4NO3 and MgSO4 were considered significant for callus formation on ‘Zeta’ (p<0.05). An interaction of NH4NO3 × K2SO4 affected callus formation on ‘Wepster’ (p<0.05) (Table 2.2). The highest concentrations of NH4NO3 and K2SO4 resulted in the least callus (rating of 2.6) for ‘Wepster’ (Appendix A).

‘Zeta’ required the highest amounts of NH4NO3, K2SO4 and the lowest concentrations of KH2PO4 and MgSO4 for low callus (rating of 2.8) (Appendix A). These requirements contradict those needed for good plant quality (Fig. 2.1).

Mineral nutrient factors NH4NO3, MgSO4, K2SO4 along with genotype were statistically defined as affecting callus formation using a CHAID decision tree (Appendix B). The Pearson correlation coefficient was strongly significant at 0.70.

At the top of the decision tree diagram, the root node showed an average callus rating of 2.18 for all the analyzed shoots. NH4NO3 had a dominant effect on callus formation for all of the hazelnut genotypes (adj. p<0.001) and the shoots were classified into three Nodes (Appendix B). Node 1 was the group of hazelnut shoots exposed to NH4NO3 < 0.5× (mean rating 1.986); Node 2 within the range of 0.5× <

NH4NO3 < 1.701× (mean rating 2.136), and Node 3 exposed to NH4NO3 > 1.701×

(mean rating 2.364). Callus formation of shoots within Node 2 (0.500× < NH4NO3 < 1.701×) was not affected by any other basal salt factor or genotype. Genotype was an important factor for callus formation in Node 3 (NH4NO3 > 1.701×) (adj. p<0.01), which branched into Nodes 7 and 8. Callus of ‘Dorris’ exposed to NH4NO3 > 1.701× (Node 7) was not affected by any other salt factor. The cut-off values for callus of

‘Wepster’ and ‘Zeta’ hazelnuts were estimated as 1.701× NH4NO3 and 0.5× K2SO4.

Callus production of ‘Wepster’ and ‘Zeta’ within group NH4NO3 > 1.701× (Node 8), was affected by K2SO4 with a mean rating of 2.482 (adj. p<0.05). The least callus (2.585) was obtained in Node 10, which was the group of ‘Wepster’ and ‘Zeta’ exposed to NH4NO3 > 1.701× and K2SO4 > 0.5×. In addition, Node 9 (‘Wepster’ and

‘Zeta’ in the range of NH4NO3 > 1.701× and K2SO4 < 0.5×) produced the mean callus rating of 2.21. Despite resulting in less callus, both of those nodes were poor for overall shoot quality. Terminal node 5 (average callus 2.2) which is within the range of NH4NO3 < 0.5× and 0.95× < MgSO4 < 1.762×, is consistent with the requirements for good plant quality. For callus, RSM was generally consistent with the results of CHAID, but didn’t provide a significant model for ‘Dorris’ (Table 2.2). Hand et al. (2014) suggested high amounts of NH4NO3 and Ca(NO3)2 for less callus formation on

‘Dorris’. In the current study, very high concentrations of K2SO4 and very low amounts of KH2PO4 and MgSO4 were required for less callus formation of ‘Zeta’ (Appendix A). These differences could be attributed to differences in the statistical background of RSM and CHAID methodologies. All the cultivars were simultaneously analyzed using CHAID algorithm, which provided special information about the classification of plants giving similar or different responses to the combinations defined by RSM design.

In the RSM analysis, great variation was apparent between hazelnut cultivars in terms of quality, shoot length, shoot number and callus (Hand et al. 2014; Hand and Reed 2014) where cultivar was not included as a factor, but the CHAID analysis revealed the similarities and differences between the cultivars. In earlier RSM based studies, cultivar variation made it extremely difficult to formulate a common hazelnut micropropagation medium (Hand et al. 2014; Hand and Reed 2014). The results of the tree-based CHAID algorithm in the current study closely mirrored the graphical results of RSM, but were easier to interpret. Alteration of the concentration ranges of nutrient factors within the RSM design might also improve the optimization process. Previous studies showed that a neuro-fuzzy logic data mining algorithm was useful for characterizing predictive nutrient factors that directly correlated with plant responses to tissue culture medium (Alanagh et al. 2014; Gago et al. 2011), but the exact cut-off amounts of nutrients, and how their significance on the response could change based on the factor concentration range, were not determined. Nas and Read (2004) hypothesized that the seed mineral and organic composition could be ideal starting point for tissue culture medium optimization. However, the seed nitrogen content was found toxic for the in-vitro hazelnuts. Overall shoot quality, shoot length and multiplication improved with lower

NH4NO3 concentrations. In the CHAID analysis low NH4NO3 provided good shoot quality, shoot length, multiplication and reduced callus (Fig. 2.3, 2.5, 2.7; Appendix B). In RSM it was more complicated to determine a common formula for high quality, shoot length and multiplication versus a low callus model. We can more easily conclude from the prediction trees of the CHAID algorithm (Fig. 2.3, 2.5, 2.7;

Appendix B) that NH4NO3 < 0.5×, KH2PO4 > 2.75×, Ca(NO3)2 < 1.725× are the required critical cut-offs for optimum growth medium of the three hazelnut genotypes evaluated. Based on this analysis the new medium amounts were set at: 0.5×

NH4NO3, 3× KH2PO4, 1.5× Ca(NO3)2. The other salt factors that were analyzed could be set at the standard DKW concentrations (1×).

Conclusions Computer generated optimal designs, like RSM, are an excellent tool for reducing treatment numbers compared to traditional factorial designs, and analyzing the resulting data with CHAID provides clear direction for developing a single optimal medium or a small number of suitable media for a range of genotypes. These advanced data mining approaches could be implemented to deduce optimum cut off- values of nutrient factors from mineral nutrition studies. CHAID is a novel and promising approach in tissue culture medium optimization. It provides a visual tree and exact cut-offs of the significant nutrients which makes it easier to define optimal concentrations of the nutrient salts. Evaluating in vitro culture data with the CHAID algorithm would provide clearer evaluation of the responses of the in vitro plants to the tested variables. The property of CHAID allowing the analysis of all genotypes together would contribute to developing one or a few optimal media for many genotypes rather than many cultivar-specific formulations.

Acknowledgements Funding for this study was provided by the U.S. Department of Agriculture, Agricultural Research Service CRIS project 5358-21000-033D. M. Akin was supported by a Higher Education Scholarship from the government of Turkey.

References Alanagh EN, Garoosi GA, Haddad R, Maleki S, Landín M & Gallego PP (2014) Design of tissue culture media for efficient Prunus rootstock micropropagation using artificial intelligence models. Plant Cell, Tissue and Organ Culture 117(3):349-359 Ali M, Eyduran E, Tariq MM, Tirink C, Abbas F, Bajwa MA, Baloch MH, Nizamani AH, Waheed A, Awan MA, Shah SH, Ahmad Z & Jan S (2015) Comparison of artificial neural network and decision tree algorithms used for predicting live weight at post weaning period from some biometrical characteristics in Harnai sheep. Pakistan Journal of Zoology 47(6):1579-1585 Anderson MJ & Whitcomb PJ (2005) RSM simplified: optimizing processes using response surface methods for design of experiments. New York, NY: Productivity Press Bradley N (2007) The response surface methodology. Dissertation, Indiana University South Bend, USA

Compton M & Mize C (1999) Statistical considerations for in vitro research: I — Birth of an idea to collecting data. In Vitro Cellular & Developmental Biology - Plant 35(2):115-121 Design-Expert (2010) Stat-Ease, Inc., Minneapolis, MN. Díaz-Pérez FM & Bethencourt-Cejas M (2016) CHAID algorithm as an appropriate analytical method for tourism market segmentation. Journal of Destination Marketing & Management Driver JA & Kuniyuki AH (1984) In vitro propagation of Paradox walnut rootstock. HortScience 19(4):507-509 Gago J, Perez-Tornero O, Landin M, Burgos L & Gallego PP (2011) Improving knowledge of plant tissue culture and media formulation by neurofuzzy logic: a practical case of data mining using apricot databases. Journal of plant physiology 168(15):1858-65 Hand C, Maki S & Reed B (2014) Modeling optimal mineral nutrition for hazelnut micropropagation. Plant Cell, Tissue and Organ Culture 119(2):411-425 Hand C & Reed BM (2014) Minor nutrients are critical for the improved growth of Corylus avellana shoot cultures. Plant Cell, Tissue and Organ Culture 119(2):427-439 Hébert M, Collin-Vézina D, Daigneault I, Parent N & Tremblay C (2006) Factors linked to outcomes in sexually abused girls: a regression tree analysis. Compr Psychiatry 47:443-455 Ibañez MA, Martin C & Pérez C (2003) Alternative statistical analyses for micropropagation: A practical case of proliferation and rooting phases in Viburnum opulus. In Vitro Cellular & Developmental Biology - Plant 39(5):429-436 Mize C, Koehler K & Compton M (1999) Statistical considerations for in vitro research: II — Data to presentation. In Vitro Cellular & Developmental Biology - Plant 35(2):122-126 Montgomery DC (2005) Design and analysis of experiments: response surface method and designs. New Jersey: John Wiley and Sons, Inc. Nas MN, Eskridge K & Read P (2005) Experimental designs suitable for testing many factors with limited number of explants in tissue culture. Plant Cell, Tissue and Organ Culture 81(2):213-220 Nas MN & Read PE (2004) A hypothesis for the development of a defined tissue culture medium of higher plants and micropropagation of hazelnuts. Scientia Horticulturae 101(1-2):189-200 Niedz RP & Evens TJ (2007) Regulating plant tissue growth by mineral nutrition. In Vitro Cellular & Developmental Biology - Plant 43(4):370-381 Niedz RP, Hyndman SE & Evens TJ (2007) Using a gestalt to measure the quality of in vitro responses. Scientia Horticulturae 112(3):349-359 Rashidi S, Ranjitkar P & Hadas Y (2014) Modeling bus dwell time with decision tree-based methods. Transportation Research Record: Journal of the Transportation Research Board 2418:74-83 Reed BM, Wada S, DeNoma J & Niedz RP (2013) Improving in vitro mineral nutrition for diverse pear germplasm. In Vitro Cellular and Developmental Biology - Plant 49:343-355

SPSS (2013) Statistics for Windows, Version 22.0. Armonk, NY: IBM Corp. Statistics-Solutions (2016) CHAID. http://www.statisticssolutions.com/non- parametric-analysis-chaid/

Quality-Dorris Design-Expert® Software 3.00 2.20 a Factor Coding: Actual Quality 2.30 2.50 2.20

1.00 2.00

X1 = B: NH4NO3 2.40 X2 = C: Ca(NO3)2.4H2O 1.50

Actual Factors 2.30 A: CaCl2.2H2O = 1.00 1.00 C: Ca(NO3)2.4H2O 2.20 D: MgSO4.7H2O = 1.00 2.10 2.00 E: KH2PO4 = 3.00 0.50 F: K2SO4 = 0.50 0.50 1.00 1.50 2.00 2.50 3.00 B: NH4NO3 Quality-Wepster Design-Expert® Software 3.00 b Factor Coding: Actual Quality 2.50 2.20 2.00

1.00 2.00 1.90

X1 = B: NH4NO3 1.80 1.70

X2 = E: KH2PO4 1.50 E : K H 2 P O 4 Actual Factors A: CaCl2.2H2O = 1.00 1.00 C: Ca(NO3)2.4H2O = 1.00 D: MgSO4.7H2O = 1.00 0.50 F: K2SO4 = 0.50 0.50 1.00 1.50 2.00 2.50 3.00 B: NH4NO3 Quality-Zeta Design-Expert® Software 3.00 c Factor Coding: Actual Quality 2.50 2.20

1.00 2.00 1.80 X1 = B: NH4NO3 X2 = E: KH2PO4 1.50

1.90 E : K H 2 P O 4 Actual Factors 1.70 A: CaCl2.2H2O = 1.00 1.00 C: Ca(NO3)2.4H2O = 1.00 D: MgSO4.7H2O = 1.00 0.50 F: K2SO4 = 0.50 0.50 1.00 1.50 2.00 2.50 3.00 B: NH4NO3

Fig. 2.1. Response surface graph of mineral nutrient effects on hazelnut shoot quality for a.‘Dorris’, b.‘Wepster’ and c.‘Zeta’. The quality ratings were1=poor, 2=moderate, 3=good and highest (red-yellow) to lowest quality (green-blue).

Dorris Control Treatment 6 Treatment 15

Wepster Control Treatment 15 Treatment 36

Zeta Control Treatment 15 Treatment 24

Fig. 2.2. Shoots of a.‘Dorris’, b.‘Wepster’ and c.‘Zeta’grown on the DKW salts control and two treatments which produced higher plant quality.

Quality

Fig. 2.3. The CHAID decision tree diagram for plant quality of ‘Dorris’, ‘Wepster’ and ‘Zeta’. The quality ratings were1=poor, 2=moderate, 3=good. Nodes were determined by the significance of the factors. Salt cut-off values are × DKW.

Shoot length-Dorris Design-Expert® Software 3.00 a Factor Coding: Actual Shoot length 2.50 40.00 50.00 20.00 2.00 35.00 X1 = B: NH4NO3 1.50 X2 = E: KH2PO4 E : K H 2 P O 4 30.00 Actual Factors A: CaCl2.2H2O = 1.00 1.00 25.00 C: Ca(NO3)2.4H2O = 2.50 D: MgSO4.7H2O = 1.00 0.50 F: K2SO4 = 1.00 0.50 1.00 1.50 2.00 2.50 3.00

B: NH4NO3 Shoot length-Wepster Design-Expert® Software 3.00 b Factor Coding: Actual Shoot length 40.00 2.50 Design Points 50.00 35.00 2.00 20.00

1.50 30.00 X1 = B: NH4NO3

E : K H 2 P O 4 X2 = E: KH2PO4 1.00 2 25.00 Actual Factors A: CaCl2.2H2O = 1.00 0.50 C: Ca(NO3)2.4H2O = 1.00 0.50 1.00 1.50 2.00 2.50 3.00 D: MgSO4.7H2O = 1.00 F: K2SO4 = 1.00 B: NH4NO3

Shoot length-Zeta Design-Expert® Software 3.00 c Factor Coding: Actual 30.00 Shoot length 2.50 Design Points 50.00 2.00 40.00 20.00 50.00 1.50

X1 = B: NH4NO3 F : K 2 S O 4 X2 = F: K2SO4 1.00 2 Actual Factors A: CaCl2.2H2O = 1.00 0.50 C: Ca(NO3)2.4H2O = 1.00 0.50 1.00 1.50 2.00 2.50 3.00 D: MgSO4.7H2O = 1.00 E: KH2PO4 = 1.00 B: NH4NO3

Fig. 2.4. Response surface graphs of mineral nutrient effects on shoot length (mm) of a. ‘Dorris’, b.‘Wepster’ and c. ‘Zeta’. The shoot lengths (mm) were color coordinated from longest (red-yellow) to shortest (green-blue). The red dot represents the control with average shoot length of 50 mm.

Shoot length

< 1.701 x

Fig. 2.5. The CHAID decision tree diagram for shoot length of ‘Dorris’, ‘Wepster’ and ‘Zeta’ hazelnuts. Nodes were determined by the significance of the factors. Salt cut-off values are × DKW.

Shoot number-Dorris Design-Expert® Software 3.00 a Factor Coding: Actual Shoot number 2.50 2.50 3.50

1.00 2.00 3.00 3.50 X1 = B: NH4NO3 X2 = C: Ca(NO3)2.4H2O 1.50 4.00 Actual Factors

A: CaCl2.2H2O = 1.00 1.00 C: Ca(NO3)2.4H2O D: MgSO4.7H2O = 0.50 4.50 E: KH2PO4 = 3.00 0.50 F: K2SO4 = 0.50 0.50 1.00 1.50 2.00 2.50 3.00 B: NH4NO3 Shoot number-Wepster Design-Expert® Software 3.00 b Factor Coding: Actual 1.60 Shoot number 2.50 3.50

1.00 2.00 1.80

X1 = A: CaCl2.2H2O 1.50

X2 = B: NH4NO3 2.00 B : N H 4 N O 3 Actual Factors 2.20 C: Ca(NO3)2.4H2O = 1.00 1.00 2.40 D: MgSO4.7H2O = 1.00 E: KH2PO4 = 1.00 0.50 F: K2SO4 = 0.50 0.50 1.00 1.50 2.00 2.50 3.00 A: CaCl2.2H2O Shoot number-Zeta Design-Expert® Software 3.00 c 1.20 Factor Coding: Actual Shoot number 2.50 Design Points 1.40 3.50 2.00 1.60 1.00 1.80

1.50 2.00 X1 = B: NH4NO3 X2 = C: Ca(NO3)2.4H2O 2.20

1.00 2 C: Ca(NO3)2.4H2O Actual Factors 2.40 A: CaCl2.2H2O = 1.00 0.50 D: MgSO4.7H2O = 1.00 0.50 1.00 1.50 2.00 2.50 3.00 E: KH2PO4 = 1.00 F: K2SO4 = 1.00 B: NH4NO3

Fig. 2.6. Response surface graphs of mineral nutrient effects on shoot number of a. ‘Dorris’, b. ‘Wepster’ and c. ‘Zeta’. The shoot numbers were color coordinated from most (red-yellow) to fewer shoots (green-blue). The red dot represents the control.

Shoot Number

Fig. 2.7 . The CHAID decision tree diagram for shoot number of ‘Dorris’, ‘Wepster’ and ‘Zeta’. Nodes were determined by the significance of the factors. Salt cut-off values are × DKW.

Table 2.1. Six factor design including 42 treatment points. Design points 1-40 for investigating the effects of individual factors on mineral nutrition of hazelnut cultivars and DKW medium controls (points 41-42). DKW medium concentrations 1×: NH4NO3 (1416 mg), Ca(NO3)2∙4H2O (1960 mg), CaCl2∙2H2O (147 mg), MgSO4∙7H2O (740 mg), KH2PO4 (259 mg), K2SO4 (1560 mg). All treatments included modified minor nutrients (Hand and Reed 2014). Factor 1 Factor 2 Factor 3 Factor 4 Factor 5 Factor 6 Treatments NH4NO3 Ca(NO3)2∙4H2O CaCl2∙2H2O MgSO4∙7H2O KH2PO4 K2SO4 1 0.50 3.00 3.00 3.00 3.00 0.50 2 0.50 0.93 3.00 3.00 0.50 0.50 3 3.00 2.65 2.30 3.00 1.28 2.19 4 3.00 0.50 3.00 1.70 3.00 0.51 5 3.00 2.65 2.30 3.00 1.28 2.19 6 0.50 1.71 3.00 0.50 3.00 3.00 7 0.50 3.00 3.00 0.93 0.50 0.50 8 1.75 0.50 0.50 1.74 0.50 0.50 9 3.00 3.00 0.50 1.73 0.50 1.77 10 0.50 1.71 3.00 0.50 3.00 3.00 11 1.81 3.00 0.50 0.50 3.00 0.50 12 3.00 0.50 1.75 3.00 0.50 1.76 13 1.76 1.73 0.50 3.00 0.50 3.00 14 0.50 0.50 0.50 3.00 3.00 1.80 15 0.50 3.00 3.00 3.00 3.00 0.50 16 3.00 3.00 3.00 0.50 2.00 1.75 17 3.00 1.75 0.50 3.00 2.00 0.50 18 0.50 3.00 0.50 1.76 2.00 3.00 19 1.70 1.79 1.79 1.80 2.13 1.71 20 1.70 1.79 1.79 1.80 2.13 1.71 21 0.50 3.00 3.00 3.00 0.50 2.58 22 1.70 1.80 1.80 1.80 2.08 1.70 23 3.00 3.00 1.70 3.00 3.00 3.00 24 0.50 1.00 3.00 2.50 2.75 0.63 25 0.50 3.00 0.93 3.00 0.50 0.50 26 0.50 0.50 1.75 0.50 2.01 0.50 27 0.50 1.74 0.50 0.50 0.50 1.75 28 1.75 0.50 3.00 0.50 0.50 1.77 29 0.75 1.30 0.50 3.00 1.21 1.30 30 3.00 0.50 0.50 0.50 2.70 3.00 31 1.70 1.80 1.80 1.80 2.08 1.70 32 3.00 1.74 1.75 0.50 0.50 0.50 33 3.00 1.75 3.00 1.74 0.50 3.00 34 1.75 3.00 1.74 0.50 0.50 3.00 35 0.95 3.00 0.50 2.58 0.80 2.10 36 0.50 3.00 3.00 0.95 3.00 1.18 37 1.70 1.80 1.80 1.80 2.08 1.70 38 1.75 0.50 3.00 3.00 2.01 3.00 39 0.50 0.50 1.74 1.73 0.50 3.00 40 2.60 3.00 3.00 3.00 0.50 0.50 41 1.00 1.00 1.00 1.00 1.00 1.00 42 1.00 1.00 1.00 1.00 1.00 1.00

Table 2.2. DKW nutrient factors that had significant effects on four growth responses for each hazelnut cultivar at p-value (<0.05).

Measured Responses Cultivars Quality (p-value) Shoot length (p-value) Shoot number (p-value) Callus (p-value) Model (0.0217) Model (0.0255) Model (0.0026) Model (NS*)

KH PO (0.0029) NH NO (0.0021) NH NO (0.0004) 2 4 4 3 4 3 K SO (0.0412) KH PO (0.0008) Ca(NO ) ∙4H O (0.0001) 2 4 2 4 3 2 2 NH NO ×Ca(NO ) ∙4H O Dorris 4 3 3 2 2 NH NO × Ca(NO ) ∙4H O (0.0157) MgSO ∙7H O (0.0275) (0.0020) 4 3 3 2 2 4 2 Ca(NO ) ∙4H O×MgSO ∙7H O 3 2 2 4 2 Ca(NO ) ∙4H O×MgSO ∙7H O (0.0142) KH PO (0.0013) (0.0331) 3 2 2 4 2 2 4 NH NO K SO (0.0348) 4 3 2 4 MgSO ∙7H O×K SO (0.0135) 4 2 2 4

Model (0.0070) Model (0.0070) Model (0.0015) Model (0.0195)

NH4NO3 (< 0.0001) NH4NO3 (0.0001) NH4NO3 (0.0022) NH4NO3 (< 0.0001) Wepster KH2PO4 (0.0011) KH2PO4 (0.0003) K2SO4 (0.0165) NH4NO3×K2SO4 (0.0417) K SO 2 4 NH NO ×CaCl ∙2H O (0.0004) (0.0171) 4 3 2 2

Model (0.0153) Model (0.003) Model (0.0006) Model (0.014)

NH4NO3 (0.002) NH4NO3 (< 0.0001) NH4NO3 (< 0.0001) NH4NO3 (0.0002)

KH2PO4 (0.0187) K2SO4 (0.0342) Ca(NO3)2∙4H2O (< 0.0001) KH2PO4 (0.0399) Zeta K SO (0.0067) K SO (0.0139) 2 4 2 4 NH4NO3×MgSO4∙7H2O (0.0344)

* “NS” stands for not significant.

CHAPTER 3: ION SPECIFIC EFECTS ON TISSUE CULTURE MEDIUM OPTIMIZATION

Melekşen Akın1, Ecevit Eyduran2 and Barbara M. Reed3

Abstract The general approach for tissue culture medium optimization is to use salts as factors in experimental design and analysis. However, using salts as factors leads to ion confounding, making ion effects on particular growth responses difficult to detect. This study focuses on testing ions as factors for the hazelnut + 2+ 2+ 2- 3- medium optimization process. NH4 , Ca , Mg , SO4 and PO4 ions were used + - as factors in a D-optimal design. K and NO3 ions were used to bring the pH level to neutral, and as factors in the statistical analysis. The Chi-Squared Automatic Interaction Detection (CHAID) data mining algorithm was used to analyze shoot growth responses of ‘Barcelona’, ‘Jefferson’ and ‘Wepster’ hazelnuts. The CHAID trees revealed significant variables and their interactions, and provided + - + exact cut-off amounts for each of the ions. K , NO3 , genotype, and NH4 had + significant effects on shoot quality. NH4 was of primary significance for shoot 2+ - 2+ length followed by Mg , NO3 and Ca . Multiplication was mainly affected by 2+ + Ca . A concentration range of NH4 > 33. 3 mM was required for the least callus formation, but this range was not appropriate for good shoot quality or elongation. The critical cut-off values for good shoot quality, elongation, multiplication and - + 2+ medium callus formation are suggested to be: NO3 <88 mM, NH4 <20 mM, Ca <5 mM, Mg2+ >5 mM and K+ <46 mM. Keywords: CHAID algorithm, Corylus avellana, Hazelnut, Ions, Micropropagation

Introduction The success of micropropagation greatly depends on the nature and concentration of mineral nutrients in the medium. There are 14 elements essential for healthy plant growth. Plants need relatively large amounts of the ions of nitrogen, potassium, calcium, phosphorus, magnesium and sulphur, as well as the micro elements in small amounts (Ramage and Williams 2002). Nitrogen is a major component in mineral nutrition and the primary constituent of various tissue culture media. It is an essential element of proteins, nucleic acids and chlorophyl. + - Plants absorb nitrogen in the form of NH4 and NO3 , thus the effect of nitrogen on plant responses is determined by the ion composition. Nitrogen is effective in a wide range of plant growth responses including callus growth, organogenesis,

embryogenesis and multiplication (Niedz and Evens 2008). Calcium is a major cation for balancing anions within the plant, and varying levels of Ca2+ are influential in the alleviation of shoot tip necrosis and hyperhydricity (Machado et 2- al. 2014; Singha et al. 1990). Plants utilize sulphur in the form of SO4 , and deficiency leads to rigid, brittle, thin stemmed plants. Magnesium is absorbed from the medium as the Mg2+ ion and is an essential constituent of chlorophyl, as well as an activator of many enzymes (Verbruggen and Hermans 2013). Phosphorous is a component of many macromolecules including nucleic acids, 3- phospholipids and co-enzymes. It is utilized by plants in the form of PO4 and deficiencies could lead to failure of shoot meristem formation (George and de Klerk 2008; Ramage and Williams 2002). Although mineral nutrients are utilized in ionic form by the plants, the literature mainly documents mineral salt optimization for various in vitro growth responses (Hand et al. 2014; Reed et al. 2013). Ion confounding is of major importance in tissue culture medium optimization when salts are used as predictor variables within an experimental design (Niedz and Evens 2008). Salt experiments exhibit significant ion confounding, and do not provide characterization of the basic ion-specific effects on in vitro responses. For example, a one factor experiment run with a range of NH4NO3 concentrations could have a variation in + - + - the response due to NH4 , NO3 , or their interaction (NH4 × NO3 ). When salts are used as independent variables within a design, by varying the salt concentration, the ion amounts constituting the salts are also changed. Therefore the effects of individual ions are confounded with each other, and it is impossible + - + - to detect the significance of NH4 , NO3 , or NH4 × NO3 on the measured response. The effect of the dependent variable in an experiment designed based on + - salt factors, eventually is the average effect of its anion and cation (NH4 and NO3 ) (Niedz and Evens 2008). Ion confounding could be overcome by using ions as factors in an experimental design, by calculating ionic formulation of the salts (Niedz and Evens 2006). Response Surface Methodology (RSM) optimal designs allow testing of multiple predictor variables at once, and greatly reduce the number of treatments, thus minimizing time, labor and explant number compared to traditional factorial designs (Anderson and Whitcomb 2005; Hand and Reed 2014; Reed et al. 2013).

The Chi-Squared Automatic Interaction Detection (CHAID) is a data mining algorithm which has various advantages over standard analysis of variance (ANOVA) and multiple regression. CHAID is a non-parametric approach, which doesn’t require assumptions like normality and constant variance. It can be applied to continuous, nominal and ordinal variables. The algorithm is successful in overcoming missing data cases for both dependent and independent variables. The CHAID decision tree shows the significant variables and interactions and is easy to interpret (Hébert et al. 2006; Rashidi et al. 2014). The objective of the current study was to predict the required ion concentrations for optimal growth of diverse hazelnuts from a D-optimal design using the CHAID data mining algorithm. To our knowledge, this is the first broad study using most of the required macro nutrient ions for tissue culture medium optimization, which is free of ion confounding.

Materials and Methods

Plant material and in vitro culture conditions

Shoots of C. avellana L. hazelnuts ‘Barcelona’, ‘Jefferson’ and ‘Wepster’ were micropropagated on COR-2013 medium (Hand 2013)consisting of modified

DKW (Driver and Kuniyuki 1984) medium salts [1.5× Ca(NO3)2∙4H2O, 1.5×

MgSO4∙7H2O and 1.5× KH2PO4; 4× H3BO3, 4× Na2MoO4∙2H2O, 4× -1 Zn(NO3)2∙6H2O, 0.5× MnSO4∙H2O, 0.5× CuSO4∙5H2O] with 30 g∙L glucose, 200 mg∙L-1 sequestrene 138 Fe EDDHA, 2 mg∙L-1 thiamine, 2 mg∙L-1 nicotinic acid, 2 mg∙L-1 glycine, 1 g∙L-1 myo-inositol, 22.2 µM N6benzyladenine (BA), and 0.5% (w/v) agar (PhytoTechnology Laboratories A1111). All plant growth regulators and vitamins were obtained from PhytoTechnology Laboratories and all stock solutions were prepared in house. Each vessel (Magenta GA7, Magenta, Chicago, IL) held 40 ml medium, and was autoclaved for 20 min at 121 0C (118 kPa). Stock plants were transferred at three-wk intervals. The growth room conditions consisted of 80 µmol∙m2s-1 light intensity with a 16-h photoperiod of half warm-white and half cool-white fluorescent lamps (Ecolux Starcoat, 32 watt; GE, Fairfield, CT) at 25±2 0C.

A RSM D-optimal experimental design was set up with Design-Expert software (Design-Expert 2010) using ions as factors to represent those of the major salts. The modified micronutrients listed above were used without alteration + +2 +2 3- 2- (Hand and Reed 2014). NH4 , Ca , Mg , PO4 and SO4 ions were tested in a range of mM concentrations with 32 treatment combinations (Table 3.1).

NH4NO3, Ca(NO3)2∙4H2O, Mg(NO3)2∙6H2O, KNO3, MgSO4∙7H2O, K2SO4,

(NH4)2SO4, (NH4)2HPO4, NH4H2PO4 and KH2PO4 salts were converted to the + - corresponding ion type and concentrations (Table 3.2). K and NO3 ions derived from KOH and HNO3 were used to adjust the pH to produce a neutral charge level of the ions, and were not included as factors within the design. The charge balance (pH) was equivalent for all the treatment combinations within the design. Shoots were cut to 3 cm with the apical meristem removed. During the first transfer the callus was removed and shoots were transferred to new medium. In the second transfer shoots were reduced to 3 cm by removing the base and the apical meristem. For each treatment two boxes with five shoots were used (n=10) for each cultivar. Boxes were randomized on the growth room shelf. Shoots were grown on each treatment medium for 10 wk with the first and second transfers for 3 wk, and the last transfer for 4 wk.

Data After 10 weeks of growth, three shoots selected on a diagonal from the corner of each vessel (n=6), were evaluated as 1=poor, 2=moderate and 3=good. Shoots longer than 5 mm were counted. The longest shoot was measured in millimeters. Callus formation was evaluated as: 1=callus ≥ 2 mm, 2=callus < 2 mm, and 3=absent (no callus present). The remaining four shoots were photographed.

Statistical Analysis The mean response (quality, shoot length, shoot number and callus) of six shoots per treatment for each genotype was analyzed using the CHAID data mining algorithm. Quality, shoot length, shoot number and callus were treated as continuous variables and a separate tree was generated for each response including genotype (‘Barcelona’, ‘Jefferson’ and ‘Wepster’) as a nominal - + independent variable. Besides the five ions within the design, NO3 and K were also included within the analysis, and assigned as continuous independent

variables. Minimum plant numbers for parent and child nodes were assigned based on the tree diagram with the highest Pearson correlation and no overlapping. Overlapping occurs when an independent factor is significant on more than one node within the same branch, and results in objects to be included in more than one node (Díaz-Pérez and Bethencourt-Cejas 2016). The minimum plant numbers required to prune the redundant structuring of the trees were assigned as 6 for parent (node that is split into smaller child nodes) and 3 for child nodes for all of the evaluated responses for highest predictive performance of the algorithm. Adjusted p-values were derived from Bonferroni correction. Predictive performance was measured with the Pearson correlation coefficient between the actual and predicted values for each growth response. CHAID algorithm regression trees were generated with IBM SPSS Statistics for Windows (Version 23.0, Armonk, NY) software (SPSS 2013).

Results and Discussion This study was conducted to optimize micropropagation medium for diverse hazelnuts by using ions as predictor variables within the experimental design, thus preventing ion confounding. Computer generated optimal designs consist of selected treatment points within all possible treatment combinations representing the experimental space, thus greatly minimizing the treatment combinations with respect to traditional factorial designs (Anderson and Whitcomb 2005). For example, a five-factor traditional factorial design with three levels consists of 243 treatment combinations, whereas the RSM optimal design used in this study consists of 32 treatments representing the same design space. CHAID is a multivariate analysis tool that generates visual, easy to interpret regression trees by splitting the independent variables into groups that best differentiate the response. The groups within the tree are internally homogeneous and externally heterogeneous. The data is split into groups according to the significance of the predictors in reducing the total variation within the response, thus the most significant factor is at the top and the least significant one at the bottom of the tree (Hébert et al. 2006).

Quality + - The CHAID data mining algorithm indicated K , NO3 , genotype, and + NH4 as significant factors in plant quality (Fig. 3.1). The Pearson correlation coefficient, which is the predictive ability of the algorithm was 0.673 (p<0.01). K+ was the most important factor for shoot quality, which also affected the significance of the other ions based on its concentration range. Node 0 constituted all the shoots used in the analysis. Node 1 was the group of shoots exposed to K+ <46 mM with mean quality of 1.86. Node 1 was split in two groups (Node 4 and - Node 5) based on the concentration range of NO3 (Fig. 3.1). Node 4 was - representative of the shoots exposed to NO3 <88 mM. The shoots in Node 4 were + subgrouped as Node 8 and Node 9 based on the amount of NH4 . Node 8 was the + group of shoots at NH4 <20 mM and showed the highest quality (2.6 of 3). The shoots treated with 46 mM< K+ <66 mM were assigned in Node 2, and no other factors were significant. In Node 3 the shoots were exposed to 66 mM< K+ , and genotype was significant at this concentration range. The group including ‘Barcelona’ and ‘Wepster’ had the lowest quality (1.1). The required ion concentration ranges for best shoot quality for ‘Barcelona’, ‘Jefferson’ and + - + ‘Wepster’ were: K <46 mM, NO3 <88 mM and NH4 <20 mM. DKW values of + - + K 19.8 mM, NO3 34.3 mM, NH4 17.7 mM are within this range (Table 3.3). Hand et al. (2014) defined hazelnut quality as affected by nitrogen factors, mesos and potassium sulfate, with great variation by genotype when analyzed using RSM. Increased calcium nitrate was an important factor in quality for 'Jefferson' in that study. We were able to define common ion concentrations for hazelnut quality by using the CHAID algorithm. CHAID incorporates nominal variables such as genotype within the analysis, whereas RSM only analyzes continuous variables and generates separate models for each genotype. Hazelnut mineral nutrition optimization using salts (Chapter 2) found that NH4NO3 and

KH2PO4 were critical for ‘Dorris’, ‘Wepster’, and ‘Zeta’ hazelnut genotypes. The requirement for NH4NO3 was low to medium (0.5-1.7× DKW), and for KH2PO4 higher than 2.012× DKW. In the current study, plant quality was greatly impacted + - + by the various concentrations of K , NO3 and NH4 , and higher quality was obtained (2.6) for ions compared to the salt study, (2.0). Some of the variation in plant quality could also be attributed to the difference in genotypes used. A nitrate-ammonium study of pear genotypes revealed a range of requirements from

- + low to high NO3 (20-60 mM) and NH4 (5-45 mM) for higher quality response in + - the diverse species according to RSM results, and indicated significant K : NO3 interaction for some of the genotypes (Wada et al. 2015).

Shoot length The Pearson correlation coefficient between actual and predicted shoot length was estimated as 0.62 (p<0.01). The most significant ion affecting shoot + length was NH4 with a cut-off value of 33.33 mM (Fig. 3.2). Shoots in the + 2+ concentration range of NH4 <33.33 mM (Node 1) were affected by Mg and subgrouped as Node 3 and Node 4. Node 4 was representative of the shoots exposed to Mg2+ > 5 mM, which showed the highest predicted shoot length (45.6 mm). Two other nodes (Node 7, 41 mm and Node 9, 39 mm) produced shoots that were shorter, but of acceptable length (Fig. 3.2). The cut-off values for ideal shoot + length for ‘Barcelona’, ‘Jefferson’ and ‘Wepster’ are suggested to be: NH4 <33.33 mM, and Mg2+ >5 mM. DKW value of Mg2+ 3 mM is out of this range (Table 3.3). Hand et al. (2014) found that shoot elongation was variable by genotype, however in general, increased calcium nitrate, MgSO4 and KH2PO4 (up to 1.5× DKW) were involved for the five genotypes. Only increased calcium nitrate was significant for 'Jefferson'. Nas and Read (2004) medium recommended lower

NH4NO3, Ca(NO3)2 and CaCl2, as well as high MgSO4 and KH2PO4 with comparison to DKW for better elongation of hybrid hazelnuts (C. avellana x C. americana) (Nas and Read 2004). Low to medium NH4NO3 (0.5-1.7× DKW) and high KH2PO4 (3× DKW) were better for shoot length for C. avellana ‘Dorris’, ‘Wepster’, and ‘Zeta’ when tested with salts (Chapter 2). In the current study K+ 3- and PO4 were not significant for shoot length. This discrepancy could be attributed to the genotype, experimental design set up and the type of the factors used (salt versus ion). It is also a good example of ion confounding because in the current study only the cation Mg2+ was effective on shoot elongation, but not any - 2- of the corresponding anions (in this case NO3 and SO4 ) for the related salts

(Mg(NO3)2, MgSO4) used in the study. Wada et al. (2015) reported medium to - + + high NO3 (20-60 mM), as well as a range of NH4 :K ratios for better shoot elongation on diverse pear species.

Shoot number The Pearson correlation coefficient between actual and predicted shoot number was 0.63 (p<0.01). Ca2+ was the most significant ion for shoot multiplication response (Fig. 3.3). Node 0 was split into two groups (Node 1and Node 2) based on the amount of Ca2+. Node 1 represented the shoots exposed to Ca2+ <5 mM and was subgrouped to Node 3 and Node 4 according to genotype. The group of ‘Barcelona’ and ‘Jefferson’ (Node 3) had the most shoots (1.7). Node 4 was the best for ‘Wepster’, although still with a low multiplication (1.308). An alternative split with slightly lower results was based on Ca2+ >5 mM, 2- SO4 > 8 mM and genotype with Node 7 (‘Barcelona’ 1.5 shoots) and Node 8 (‘Jefferson’ and ‘Wepster’ 1.2 shoots) (Fig. 3.3). The suggested ion concentration range for a better multiplication rate for ‘Barcelona’, ‘Jefferson’ and ‘Wepster’ is: Ca2+ <5 mM, which is lower than the 9.3 mM found in DKW (Table 3.3).

Hand et al. (2014) RSM results suggested low levels of NH4NO3 and

Ca(NO3)2 (0.5× DKW), and sometimes higher MgSO4 and KH2PO4 for a higher shoot number. The earlier salt optimization study CHAID results also suggested low NH4NO3 (0.5× DKW) and low to medium Ca(NO3)2 (0.5 to1.5× DKW) for good multiplication of ‘Dorris’, ‘Wepster’, and ‘Zeta’ hazelnuts (Chapter 2). In 2+ 2- the current study only Ca and SO4 were the main driving factors in multiplication. Since these are common ions in growth medium the difference in result from the salt study could be attributed mainly to the ion confounding phenomenon.

Callus The CHAID tree indicated that the main significant factor in callus + 2- formation was NH4 , followed by genotype and SO4 (Fig. 3.4). The Pearson correlation coefficient between actual and predicted callus production was significant at 0.682 (p<0.01). All analyzed hazelnut shoots (Node 0) were split in + two groups based on the response to NH4 concentration with decisive cut-off + value estimated as 33.33 mM. Node 1 (NH4 <33.33 mM) was affected by genotype and subgrouped as Node 3 (‘Barcelona’ and ‘Wepster’) and Node 4 (‘Jefferson’). Node 4 showed the most callus formation (rated as 1.94 out of 3). + The best response for callus formation was at NH4 >33.33 mM. Node 2 was split according to genotype (Fig. 3.4). Shoots of ‘Wepster’ (Node 6) showed the least

callus formation (2.98). Shoots of ‘Barcelona’ and ‘Jefferson’ (Node 5) were 2- subgrouped based on the concentration range of SO4 . Node 9 demonstrated the least callus formation for the shoots of ‘Barcelona’ and ‘Jefferson’ (2.789) at 2- + SO4 >13 mM. NH4 >33.33 mM resulted in less callus formation, but this range was not appropriate for good shoot quality and length (Fig. 3.1, 3.2, 3.4).

Hand et al. (2014) also suggested high levels of NH4NO3 and Ca(NO3)2 (1.5× DKW) for reducing the amount of callus formation of hazelnuts. The recent salts optimization study (Chapter 2) also recommended high amounts of NH4NO3

(3× DKW) and low to high (0.5-3× DKW) K2SO4 for reduced callus formation in + 2- hazelnuts. In the current study only NH4 and SO4 were significant, which indicates ion confounding in the salt based experiments. The corresponding ions + - of the salts (in this situation K and NO3 ) in the previous salt-based studies were not significant in the current study, also indicating ion confounding. Besides ion confounding, the importance of ion-based tissue culture experiments can be attributed to the fact that many plant responses like hyperhydricity, shoot tip necrosis, and embryogenesis are attributed to ionic concentrations within the growth medium (Machado et al. 2014; Ramage and Williams 2002; Singha et al. 1990). When we consider all the responses together, the ion concentration ranges suggested for good shoot quality, elongation and multiplication, as well as - reduced callus formation for these three hazelnut genotypes are: NO3 <88 mM, + 2+ 2+ + NH4 <20 mM, Ca <5 mM, Mg >5 mM and K <46 mM (Table 3.3). There are no treatment points within the experimental design that align with the suggested cut-off values (Table 3.1). Treatments with many (but not all) of the suggested ion values had relatively good plant growth (Fig. 3.5). Chapter 2 Ca2+ and Mg2+ ion mM ranges in the salt experiement are not within the recommended ion range of the current study (Table 3.3). These discrepancies could be mainly attributed to the ion confounding, the nature of the experimental designs used and the pH factor. In the current study the pH was uniform within the design space for the combinations of ions in each treatment, but that was not the case in the salt based experiment (Chapter 2). The medium pH depends on the type and concentration of the ions, and can also affect mineral nutrition due to the effects on uptake and interaction of ions (Evens and Niedz 2008).

Conclusions This study provides an evaluation of plant responses at the ionic level by removing ion confounding from the experimental design. Ion-specific effects and their interactions can only be determined when ions are used as predictors within a multivariate experimental design. Characterization of basic ion-specific effects is important for proper designation of plant responses to the specific nutrient levels, as well as regulating in vitro growth and development. Using salts as factors results in complexity within the design as the effects of ions are confounded. This study provides clear evaluation of the ion effects on the growth responses of three hazelnuts, thus clarifying our understanding of the effects of these mineral nutrients on in vitro plant growth. Based on this analysis of ions on shoot growth, - + a culture medium for hazelnuts should include NO3 <88 mM, NH4 <20 mM, Ca2+ <5 mM, Mg2+ >5 mM and K+ <46 mM.

References Anderson MJ & Whitcomb PJ (2005) RSM simplified: optimizing processes using response surface methods for design of experiments. New York, NY: Productivity Press Design-Expert (2010) Stat-Ease, Inc., Minneapolis, MN. Díaz-Pérez FM & Bethencourt-Cejas M (2016) CHAID algorithm as an appropriate analytical method for tourism market segmentation. Journal of Destination Marketing & Management Driver JA & Kuniyuki AH (1984) In vitro propagation of Paradox walnut rootstock. HortScience 19(4):507-509 Evens TJ & Niedz RP (2008) Are hofmeister series relevant to modern ion- specific effects research? Scholarly Research Exchange 2008:1-9 George E & de Klerk G-J (2008) The components of plant tissue culture media I: macro- and micro-nutrients. In plant propagation by tissue culture 3rd edition. Edited by: George EF, Hall MA, de Klerk G-J. Dordrecht, The Netherlands: Springer; 2008:65-113 Hand C (2013) Improving initiation and mineral nutrition for hazelnut (Corylus avellana) micropropagation. Dissertation, Oregon State University, Corvallis, USA Hand C, Maki S & Reed B (2014) Modeling optimal mineral nutrition for hazelnut micropropagation. Plant Cell, Tissue and Organ Culture 119(2):411-425 Hand C & Reed BM (2014) Minor nutrients are critical for the improved growth of Corylus avellana shoot cultures. Plant Cell, Tissue and Organ Culture 119(2):427-439 Hébert M, Collin-Vézina D, Daigneault I, Parent N & Tremblay C (2006) Factors linked to outcomes in sexually abused girls: a regression tree analysis. Compr Psychiatry 47:443-455

Fig. 3.1. The CHAID decision tree diagram for plant quality of hazelnuts ‘Barcelona’, ‘Jefferson’ and ‘Wepster’. Nodes were determined by the significance of the factors. Cut- off values are mM ion concentrations. Mean and predicted values are based on the 1=poor, 2= moderate, 3=good rating given to each shoot.

Fig. 3.2. The CHAID decision tree diagram for shoot length of ‘Barcelona’, ‘Jefferson’ and ‘Wepster’. Nodes were determined by the significance of the factors. Cut-off values are mM ion concentrations.

Fig. 3.3. The CHAID decision tree diagram for shoot number of hazelnuts ‘Barcelona’,

‘Jefferson’ and ‘Wepster’. Nodes were determined by the significance of the factors. Cut-off values are mM ion concentrations.

Fig. 3.4. The CHAID decision tree diagram for callus formation of hazelnuts ‘Barcelona’, ‘Jefferson’ and ‘Wepster’. Nodes were determined by the significance of the factors. Cut- off values are mM ion concentrations. Mean and predicted values are based on the 1=large callus, 2=moderate callus, 3=no callus rating given to each shoot.

Fig. 3.5. Shoots of ‘Barcelona’, ‘Jefferson’ and ‘Wepster’ grown on treatments which are near the suggested ion concentration ranges.

+ - Table 3.1. Five factor D-optimal design with 32 treatments. K and NO3 are not factors in the experimental design, but were used to adjust the pH. The values for + - K and NO3 are the amounts required to produce a neutral charge level of the ions, and were used in the final analysis.

+ 2+ 2+ + 3- 2- - Treatments NH4 Ca Mg K PO4 SO4 NO3 1 20.00 12.00 10.00 55.00 15.00 8.00 88.00 2 40.00 6.80 10.00 45.50 3.00 13.00 90.00 3 20.00 12.00 5.00 65.00 3.00 8.00 100.00 4 60.00 1.50 5.00 46.00 9.00 18.00 74.00 5 20.00 1.50 15.00 66.00 3.00 8.00 100.00 6 20.00 12.00 15.00 45.00 3.00 18.00 80.00 7 20.00 1.50 5.00 86.00 3.00 8.00 100.00 8 60.00 1.50 15.00 26.00 15.00 8.00 88.00 9 20.00 12.00 5.00 65.00 15.00 18.00 68.00 10 60.00 12.00 15.00 5.00 3.00 8.00 100.00 11 60.00 12.00 15.00 5.00 15.00 13.00 78.00 12 20.00 5.00 15.00 59.00 15.00 18.00 68.00 13 40.00 6.80 10.00 45.50 3.00 13.00 90.00 14 60.00 12.00 5.00 25.00 3.00 18.00 80.00 15 60.00 12.00 5.00 25.00 15.00 8.00 88.00 16 20.00 12.00 15.00 45.00 3.00 8.00 100.00 17 60.00 1.50 8.30 39.30 3.00 8.00 100.00 18 20.00 12.00 5.00 65.00 3.00 18.00 80.00 19 60.00 12.00 15.00 5.00 9.00 18.00 74.00 20 60.00 5.00 5.00 39.00 15.00 18.00 68.00 21 60.00 12.00 5.00 25.00 3.00 8.00 100.00 22 33.30 1.50 11.70 59.30 15.00 18.00 68.00 23 60.00 1.50 15.00 26.00 3.00 18.00 80.00 24 60.00 12.00 15.00 5.00 9.00 18.00 74.00 25 60.00 1.50 5.00 46.00 15.00 8.00 88.00 26 33.30 12.00 15.00 31.70 15.00 8.00 88.00 27 33.30 1.50 11.70 59.30 15.00 18.00 68.00 28 60.00 1.50 15.00 26.00 3.00 18.00 80.00 29 20.00 1.50 5.00 86.00 3.00 18.00 80.00 30 20.00 1.50 5.00 86.00 15.00 8.00 88.00 31 20.00 1.50 15.00 66.00 11.00 11.30 85.30 32 33.30 12.00 15.00 31.70 15.00 8.00 88.00

Table 3.2. Salt types and amounts (mg/L) used for the treatments to obtain the specific ion concentrations.

Treatments NH4NO3 Ca(NO3)2∙4H2O Mg(NO3)2∙6H2O KNO3 MgSO4∙7H2O K2SO4 (NH4)2SO4 (NH4)2HPO4 NH4H2PO4 KH2PO4 1 1601 2834 2564 2426 - 1394 - - - 2041 2 2705 1606 - 4307 2465 - 396 13 - 395 3 1121 2834 - 6268 1232 - 396 - - 408 4 2721 354 - 3741 1232 - 1718 - - 1225 5 1601 354 3846 4752 - 1394 - - - 408 6 1121 2834 - 4246 3697 - 396 - - 408 7 1121 354 - 8392 1232 - 396 - - 408 8 4402 354 3846 - - 958 330 - - 2041 9 - 2834 - 4449 1232 523 1321 - - 2041 10 3682 2834 3846 - - 174 925 - - 408 11 1921 2834 3846 - - - 1718 - 1150 680 12 1121 1181 - 4449 3697 - 396 - - 2041 13 2705 1606 - 4307 2465 - 396 13 - 395 14 2721 2834 - 2224 1232 - 1718 - - 408 15 4803 2834 - 404 1232 523 - - - 2041 16 1601 2834 3846 2629 - 1394 - - - 408 17 4803 354 2137 2056 - 1394 - - - 408 18 - 2834 - 5662 1232 523 1321 - - 408 19 4002 2834 - - 3697 - 396 - 460 680 20 2721 1181 - 2426 1232 - 1718 - - 2041 21 4322 2834 - 2224 1232 - 396 - - 408 22 1654 354 - 4482 2875 - 837 - - 2041 23 4322 354 - 2325 3697 - 396 - - 408 24 4002 2834 - - 3697 - 396 - 460 680 25 4322 354 - 3134 1232 - 396 - - 2041 26 2668 2834 3846 67 - 1394 - - - 2041 27 1654 354 - 4482 2875 - 837 - - 2041

28 4322 354 - 2325 3697 - 396 - - 408 29 - 354 - 7785 1232 523 1321 - - 408 30 - 354 - 7785 1232 174 264 1057 - 953 31 - 354 3846 5288 - 238 1317 4 - 1492 32 2668 2834 3846 67 - 1394 - - - 2041

Table 3.3. Macronutrient basal salts (mg/L) and ionic compositions (mM) for DKW and salt based optimal medium in Chapter 2, and optimal ion concentration ranges found in this study.

Optimal ion DKW Salt based optimal medium concentration ranges Macronutrients (mg/L)

NH4NO3 1416 708

Ca(NO3)2∙4H2O 1960 2940

CaCl2∙2H2O 147 147

MgSO4∙7H2O 740 740

KH2PO4 259 777

K2SO4 1560 1560 Ion composition of macronutrients (mM) NH4+ 17.7 8.8 < 20 - NO3 34.3 33.7 < 88 Ca2+ 9.3 13.4 < 5 Mg2+ 3 3 > 5 K+ 19.8 23.6 < 46 3- PO4 1.9 5.7 2- SO4 12 12 Cl- 2 2 H+ 3.81 11.42

CHAPTER 4: MULTIPLEXED MICROSATELLITE FINGERPRINTING SET FOR HAZELNUTS

Melekşen Akın1, April Nyberg2 and Nahla V. Bassil2

1Oregon State University, Department of Horticulture, ALS 4017, Corvallis, OR 97331, USA 2USDA-ARS, National Clonal Germplasm Repository, 33447 Peoria Rd, Corvallis, OR 97333, USA

Abstract The objective of this study was to develop a robust and cost-effective fingerprinting set for hazelnuts using microsatellite (SSR) markers. Twenty SSRs containing repeat motifs of ≥ three nucleotides distributed throughout the hazelnut genome were screened on eight genetically diverse cultivars to assess polymorphism, allele size range, and ease of scoring. Six SSRs were discarded after genotyping 96 hazelnut samples either due to large allele bin widths and/or alleles that do not match the motifs complicating allele scoring. Fourteen polymorphic, easy-to-score SSRs with non-overlapping alleles were selected and amplified in a single multiplex. The multiplexed set generated the same alleles that were obtained when amplifying each SSR individually in the eight test accessions. SSR primer concentrations were then optimized to generate a clear signal for each locus. This 14-SSR fingerprinting set was used to genotype 102 hazelnut accessions from different origins. The fingerprinting set distinguished unique accessions mainly according to parentage and in some cases based on geographic origin. Tools for DNA fingerprinting of clonally propagated horticultural crops like hazelnut are in demand and this multiplexed set constitutes a reliable, less-time consuming and cost-effective procedure for identity and paternity confirmation in hazelnut.

Introduction The genus Corylus (Betulaceae) includes 11 species that are native to temperate regions of the northern hemisphere extending from Japan, Korea, China, Russian Far East to the Caucasus and Turkey (Kasapligil 1972). . The European hazelnut Corylus avellana L. is the most economically important with a worldwide production of around 872,000 t of in-shell nuts and a cultivated area of about 604,000 ha (average 2008-2012) (FAOstat 2016). Hazelnut is one of the most important nuts in worldwide production following cashew, walnut, almond, chestnut and pistachio. The major hazelnut producing countries are Turkey (598.158 t), Italy (104,577 t), USA (32,399 t), Azerbaijan (30,035 t) and Georgia (25,020 t) (average 2008-2012) (FAOstat 2016). Ninety percent of the production is processed. Hazelnuts are diploid (2n=2x=22), monoecious, dichogamous and wind-pollinated trees or shrubs with high genetic variability. Cross pollination is

enforced due to sporophytic self-incompatibility, and use of pollenizer is required for good yield in a hazelnut orchard (Mehlenbacher 2014). Cultivars are highly heterozygous, and clonally propagated by conventional methods including layering, cutting, grafting, or micropropagation. Identification of trueness-to-type by phenotypic observation is very difficult and mistakes during the several steps of nursery plant propagation are costly. Therefore, developing a reliable DNA fngerprinting set for identity verification of hazelnuts is crucial to protect breeders’ rights, and provide a tool for verifying cultivar integrity in propagation systems, or in a hazelnut collection.

The US Department of Agriculture (USDA), Agricultural Research Service (ARS), National Clonal Germplasm Repository (NCGR), in Corvallis, Oregon maintains the largest collection of Corylus in the world with more than 800 hazelnut accessions representing different Corylus species (Bassil et al. 2013). A multiplexed DNA fingerprinting test that can be amplified in a single polymerase chain reaction (PCR) would provide a quick and economical method for identification of hazelnut accessions and parentage confirmation, suitable for use in field and tissue culture collections.

Microsatellites or simple sequence repeats (SSRs) are tandemly repeated 1-6 bp sequence motifs that are randomly distributed throughout the genome. SSRs are multi-allelic, co-dominant, highly polymorphic, relatively abundant within the genome, transferable to related species and genera, as well as reproducible between laboratories (Powell et al. 1996). These characteristics led to their extensive use in fingerprinting, paternity testing and identity certification. More than 700 polymorphic SSR markers were developed in C. avellana (Bassil et al. 2013; Bassil et al. 2005a; Bassil et al. 2005b; Bhattarai 2015; Boccacci et al. 2005; Boccacci et al. 2015; Gürcan and Mehlenbacher 2010a; Gürcan and Mehlenbacher 2010b; Gürcan et al. 2010a; Peterschmidt 2013). They have been used for linkage construction (Bhattarai 2015; Gürcan et al. 2010a; Mehlenbacher et al. 2006), QTL analysis (Beltramo et al. 2016), assessment of genetic relationships among cultivars (Bassil et al. 2013; Boccacci et al. 2013; Boccacci and Botta 2010; Boccacci et al. 2006; Boccacci et al. 2008; Campa et al. 2011; Gokirmak et al. 2009; Gürcan et al. 2010b), and for identification and parentage

confirmation (Bassil et al. 2009; Botta et al. 2005; Gokirmak et al. 2009; Sathuvalli and Mehlenbacher 2012). Although di-nucleotides are the most abundant throughout the genome, tri- and higher SSR repeats are also present (Hearne et al. 1992). However, di-nucleotide containing SSRs often show stuttering, split peaks and binning errors which lead to genetic profile discrepancies (Baldoni et al. 2009; Testolin and Cipriani 2010). Therefore, SSRs with longer repeat motifs are easier to score and are preferred for automated fingerprinting. The objective of this study was to develop a robust and cost-effective multiplexed fingerprinting set consisting of high core repeat SSRs (≥3), and to test them in a diverse set of field grown C. avellana accessions and some hybrids from the USDA-ARS and the OSU collections. We also used this set to confirm identity of in vitro cultures of nine commercially important cultivars or pollinizers by comparing their fingerprints to those obtained from field-grown trees. Materials and Methods DNA was extracted from actively growing leaves of 93 field-planted (70 from the USA-ARS and 23 from the OSU collection) and nine tissue culture grown hazelnut plants (Table 4.1) in spring 2015. Up to 50 mg of leaf tissue from each sample were frozen in liquid nitrogen and homogenized with an MM 301 Mixer Mill (Retsch International, Haan, Germany). DNA was extracted according to the modified Puregene (Gentra Systems Inc. Minneapolis, MN, USA) protocol (Gilmore et al. 2011). Fourteen SSR primer pairs were first evaluated (Table 4.2) in a test panel of eight diverse hazelnut accessions (Table 4.1) for polymorphism and ease of scoring indicated by lack of PCR artifacts (stuttering or split peaks). Each of these SSRs was amplified individually using the Type-it® Microsatellite PCR Kit (Qiagen, Inc., Valencia, CA, USA). Each reaction consisted of a 15 µL volume of 1× Type-It Multiplex PCR Master Mix containing D2, D3 or D4 fluorescent dye labeled primers (WellRED Beckman Coulter, Inc., CA, USA) and 10.5 ng genomic DNA. The cycling protocol began with an initial denaturation at 95 °C for 5 min, followed by ten cycles of 30 sec at 95 °C; 90 sec at 62 °C, decreasing 1 °C each cycle; and 30 sec at 72 °C. PCR continued for 29 cycles of 30 sec at 95 °C, 90 sec at 52 °C, and 30 sec at 72 °C. The protocol was terminated with a final

elongation step at 60 °C for 30 min. PCR success was confirmed by agarose gel electrophoresis and PCR products were separated by capillary electrophoresis using the Beckman CEQ 8000 (Beckman Coulter, Inc.). Alleles were scored and visualized with the fragment analysis module of the CEQ 8000 software (Beckman Coulter, Inc.). Alleles were scored by fitting peaks into bins less than 1 nucleotide wide. Each SSR was evaluated for presence of well separated alleles that form narrow peaks (bin widths ≤ 1) that match the motif and for amplification of a single locus as indicated by the presence of a maximum of two products. Preliminary analysis of these 14 SSRs (Table 4.2) in 96 hazelnut accessions (Table 4.1) identified 6 SSRs that generated alleles with wide bins and/or whose consecutive alleles differed from expectation based on the repeat motif. These six SSRs were replaced with other six SSRs and were tested again in simplex and multiplex PCR as described above to confirm lack of primer interaction.

Fourteen polymorphic, easy to score SSRs were selected to meet the following criteria: no PCR artifacts (low to no stuttering, low to no split peaks); single locus (amplifying up to two products); and with alleles that form narrow peaks (bin widths < 1.4) and that differ according to SSR motif. They were amplified in a single multiplex using the above PCR conditions. SSR primer concentrations were optimized in the eight hazelnut accession test panel to generate a clear signal for each locus and the optimal concentrations are shown in Table 4.2. The 14 SSRs were then amplified in multiplex in the remaining 94 DNA samples using the optimized primer concentrations (Table 4.2). Product separation, allele sizing and visualization were as described above.

Data Analysis PowerMarker (Version 3.0) (Liu and Muse 2004) was used to calculate diversity parameters of the 14 SSRs including the number of alleles (A), observed heterozygosity (Ho), gene diversity or expected heterozygosity (He) and polymorphism information content (PIC) (4.3). Ho is the frequency of heterozygous genotypes per locus and is calculated by dividing the number of heterozygous genotypes by the total number of genotypes at each locus; He is the probability that a randomly chosen individual is heterozygous at a given locus;

PIC is a measure of marker informativeness and useful for probability estimation of polymorphism between genotypes (Botstein et al. 1980). The dendrogram was constructed (Fig. 4.1) with PowerMarker (Version 3.0) (Liu and Muse 2004) using the Unweighted Pair Group Method of Arithmetic Average (UPGMA) cluster analysis based on the proportion of shared alleles distance (Dsa) (Bowcock et al. 1994): 1 m aj Dsa   min( pij, qij) m j1 i1 where m is the number of loci, pij and qij are the frequencies of the ith allele at the jth locus, and aj is the number of alleles at the jth locus.

Results and Discussion Twenty SSRs with repeat motifs ≥ 3 nucleotides that were previously reported to be polymorphic and easy to score (Bhattarai 2015; Peterschmidt 2013; Sathuvalli and Mehlenbacher 2013) were evaluated in a diverse panel of eight accessions (Table 4.1) for ease of scoring determined by lack of artifacts (stuttering, split peaks) on the Beckman CEQ capillary electrophoresis platform, and for lack of primer interaction when amplified in multiplex PCR. Comparison of alleles generated when each of the fourteen SSRs was amplified individually in the test panel as opposed to in multiplex PCR generated the same alleles, indicating no primer interaction. Amplification of the first 14 SSR multiplex set (Table 4.2) in 96 hazelnut accessions (Table 4.1) allowed us to assess SSRs for the presence of well separated alleles that form narrow peaks (bin widths ≤ 1) that match the motif. Six SSRs did not match this criterion (BR249, BR359, BR427, BR446, LG631 and LG 639) (Table 4.2). Some of the alleles generated by LG639 had wide bins and consecutive alleles did not differ by the tetra-nucleotide motif size (Appendix D). The 225 allele for example ranged from a minimum of 225.8 to 228.44 with a bin width of 2.64 bps (Appendix D). At BR249 only two of the six alleles (302 and 296) differed by the expected 6 bp motif size; the 304 allele had a wide bin exceeding 2 bp; and the distance between the unique alleles varied and was less than 6 bp (Appendix D). At BR427 the 315, 317 and 319 alleles differed by two bps instead of the expected motif size of three and the bin width for the 319 allele

was 1.65 (Appendix D). Six of the eight alleles amplified by BR359 were two base pairs apart and allele widths of 387 and 393 > 2.0 (Appendix D). At LG631, the 432 and 441 alleles were easy to score but alleles that ranged from 435.07 to 439.84 were continuous and could not be clearly scored (Appendix D). BR446 amplified more than two alleles in many accessions and was difficult to score consistently because of split peaks (Appendix D, E). These six SSRs were subsequently replaced by the following six SSRs: BR343, BR438, GB395, GB875, GB949, and GB950 (Table 4.2). After again confirming that primer interaction did not occur between any of the SSRs in this new 14 SSR fingerprinting set, primer concentrations were optimized to yield a clear signal and it was used to genotype all the hazelnut accessions included in this study (Appendix C). Each of the 11 linkage groups was represented by one SSR in these 14 SSRs except for LG1 with two SSRs and LG5 with 3 SSRs (Table 4.2). On LG1, BR 270 mapped at 62.2 cM distance, while BR343 was at 137.6 cM, and thus were distant from each other (Peterschmidt 2013). The S-locus controls pollen- stigma incompatibility and is placed on LG5 where three of our SSRs were located: GB875 (0 cM) BR259 (64.9 cM), and GB673 (88.2 cM) (Bhattarai 2015; Mehlenbacher 2014). LG688 is located on LG6, where the ‘Gasaway’ eastern filbert blight resistance was mapped (Sathuvalli and Mehlenbacher 2013). This fingerprinting set contains two markers that are linked to two important traits in hazelnut: incompatibility (GB259) and eastern filbert blight resistance (LG688). These two SSRs will be evaluated in the future for marker assisted selection for these traits (Bhattarai 2015). The number of alleles per locus ranged from five in BR270, BR322, GB949, GB950, BR464 and LG688 to nine at BR259, CAC-C008 and GB395, with an average of 6.36 (Table 4.3). The PIC values ranged from 0.41 to 0.81 with an average of 0.59. The most polymorphic loci were BR259, CAC-C008 and GB395 with PIC values of 0.82, 0.77 and 0.76, respectively. The least polymorphic locus was BR438 with PIC value of 0.41. The observed heterozygosity (Ho) for individual loci ranged from 0.44 to 0.75 and averaged

0.61, while expected heterozygosity (He) ranged from 0.45 to 0.84 and averaged 0.64 (Table 4.3). The number of alleles reported for CAC-C008 by Bassil et al. (2013) was higher than that observed in the current study (21 versus 9 alleles). A

larger number of alleles is not surprising given that Bassil et al. (2013) examined 11 hazelnut species while the current study consisted mostly of C. avellana genotypes and four hybrid accessions. The genetic diversity parameters for the remaining 13 SSRs were comparable to previous reports (Bhattarai 2015; Peterschmidt 2013; Sathuvalli and Mehlenbacher 2013). UPGMA cluster analysis grouped accessions based on the proportion of shared alleles distance and presented the genetic relationships among the accessions in a dendrogram (Fig. 4.1). Eighty-one unique genotypes were observed with this 14-SSR multiplex in the 93 field-grown accessions. ‘Contorta’ was discarded from the UPGMA cluster analysis because it amplified three alleles at GB950. Two trees of ‘Red Majestic’ had different alleles at nine (BR414, GB949, GB950, CAC-C008, Gb673, BR259, BR464, LG688 and GB395) of the fourteen SSRs. Three sets of trees with the same name but obtained by the NCGR from different sources had the same profile and included four trees of ‘Tonda Gentile delle Langhe’ and two trees each of ‘Yamhill’ and ‘Imperiale de Trebizonde’ (Fig. 4.1). Genetic profiles of ‘Dorris’, ‘Dundee’, ‘Lewis’, ‘Jefferson’ and ‘Zeta’ from the OSU collection were identical to those from NCGR collection (Table 4.1) and were not included in the dendrogram. Genotypes of the French cultivars Purple Aveline and Pellicule Rouge were identical, as previously found by Gokirmak et al. (2009) based on 21 SSRs. Both of these cultivars have purple pellicles. However, ‘Purple Aveline’ has dark red leaves and the leaves of ‘Pellicule Rouge’ are green with a trace of red pigment. Differences in pellicle and leaf color are attributed to mutation (Gokirmak et al. 2009). Fingerprints of each of the nine tissue-culture grown plants (Table 4.1) were identical to those of field-grown trees of the same cultivar (Appendix C). Therefore, these tissue cultured plants appear true-to-type and this fingerprinting set is useful for detection of micropropagation error. In the UPGMA dendrogram the cultivars and selections mostly grouped according to pedigree. The two interspecific hybrids with C. colurna in their pedigrees (‘Grand Traverse’ and ‘Newberg’) were most distant from the remaining accessions and grouped together. Although the OSU release ‘Dundee’ is a hybrid between C. colurna and C. avellana, it grouped within the C. avellana group. The remaining accessions were all cultivars of C. avellana except for two

C. americana x C. avellana hybrids (‘NY398’ and ‘NY616’) that grouped together. OSU releases ‘Clark’ and ‘Lewis’, grouped with one of their parents ‘Tombul Ghiaghli’, and close to the other parent ‘Willamette’. ‘Gamma’ clustered together with its Spanish parent ‘Casina’. OSU release ‘Jefferson’ clustered with its pollen parent ‘OSU 414.062’. The pollinizer ‘Epsilon’ clustered close to its grandparent ‘Tonda Romana’ which was in the same cluster as the Spanish cultivar ‘Artellet’ as previously observed by Gokirmak et al. (2009) and Gurcan et al. (2010a). ‘Gem’ and ‘Brixnut’ are selections from the Pacific Northwest (USA) and resulted from the ‘Barcelona’ × ‘DuChilly’ cross. ‘Gem’ clustered together with ‘Barcelona’, and ‘Brixnut’ was in the same group. Other cultivars that clustered with one of their parents included ‘Ennis’ with its parent ‘Barcelona’; ‘Wepster’ with ‘Tonda Pacifica’; and ‘Tonda Pacifica’ with ‘Tonda Gentile delle Langhe’. In some cases, accessions also grouped according to geographical origin. For example, a Spanish Italian group contained the Italian cultivars lannusa Racinante, Tonda Bianca, Montebello and its offspring Willamette, in addition to ‘Yamhill’ with its grandparent ‘Montebello’ (Mehlenbacher et al. 2009). A similar Spanish-Italian group was previously observed by Gokirmak et al. (2009) and Gurcan et al. (2010a). French cultivars DuChilly, Purple Aveline and Pellicule Rouge grouped together. The Turkish cultivar Palaz clustered together with OSU 54.039 which is a selection from a Turkish seedlot. ‘Barcelloner Zellernuss’ and ‘Pendula’ selections from England clustered together. As observed in Table 4.4, each of the 14 SSRs exhibited Mendelian inheritance. The cultivars Brixnut, Gem, Clark and Jefferson inherited one allele from each parent at each of the fourteen SSRs.

Conclusions This fingerprinting DNA test of 14 multiplexed SSRs from all 11 hazelnut linkage groups developed in this study is a reliable and cost-effective method for confirming identity and paternity in hazelnut. This test will be useful for breeders, germplasm collection curators, propagators and growers for verification of trueness-to-type, and to facilitate comparison of accessions from different germplasm collections and identify possible duplications. Genetic diversity

studies are also helpful for choosing suitable parents for breeding especially in heterozygous crops like hazelnut in which inbreeding depression is of concern, and necessitating selection of unrelated parents.

References Baldoni L, Cultrera N, Mariotti R, Ricciolini C, Arcioni S, Vendramin G, Buonamici A, Porceddu A, Sarri V, Ojeda M, Trujillo I, Rallo L, Belaj A, Perri E, Salimonti A, Muzzalupo I, Casagrande A, Lain O, Messina R & Testolin R (2009) A consensus list of microsatellite markers for olive genotyping. Mol. Breed. 24:213-231 Bassil NV, Boccacci P, Botta R, Postman J & Mehlenbacher SA ( 2013) Nuclear and chloroplast microsatellite markers to assess genetic diversity and evolution in hazelnut species, hybrids and cultivars. Genet. Resources Crop. Evol. 60:543-568 Bassil NV, Botta R & Mehlenbacher SA (2005a) Microsatellite markers in hazelnut: isolation, characterization and cross-species amplification. J. Am. Soc. Hortic. Sci. 130:543-549 Bassil NV, Botta R & Mehlenbacher SA (2005b) Additional microsatellite markers of the European hazelnut. Acta Hort. 686:105-110 Bassil NV, Hummer K, Botu M & Sezer A (2009) SSR fingerprinting panel verifies identities of clones in backup hazelnut collection of USDA genebank. 845:95-102 Beltramo C, Valentini N, Portis E, Torello Marinoni D, Boccacci P, Sandoval Prando MA & Botta R (2016) Genetic mapping and QTL analysis in European hazelnut (Corylus avellana L.). Molecular Breeding 36(3):1-17 Bhattarai G (2015) Microsatellite marker development, characterization and mapping in European hazelnut (Corylus avellana L.), and investigation of novel sources of eastern filbert blight resistance in Corylus. Dissertation Oregon State University, Corvallis, USA Boccacci P, Akkak A, Bassil NV, Mehlenbacher SA & Botta R (2005) Characterization and evaluation of microsatellite loci in European hazelnut (Corylus avellana L.) and their transferability to other Corylus species. Molec. Ecol. Notes 5:934-937 Boccacci P, Aramini M, Valentini N, Bacchetta L, Rovira M, Drogoudi P, Silva AP, Solar A, Calizzano F, Erdoğan V, Cristofori V, Ciarmiello LF, Contessa C, Ferreira JJ, Marra FP & Botta R (2013) Molecular and morphological diversity of on-farm hazelnut (Corylus avellana L.) landraces from southern Europe and their role in the origin and diffusion of cultivated germplasm. Tree Genetics & Genomes 9(6):1465-1480 Boccacci P, Beltramo C, Sandoval Prando MA, Lembo A, Sartor C, Mehlenbacher SA, Botta R & Torello Marinoni D (2015) In silico mining, characterization and cross-species transferability of EST-SSR markers for European hazelnut (Corylus avellana L.). Molecular Breeding 35(1):1-14 Boccacci P & Botta R (2010) Microsatellite variability and genetic structure in hazelnut (Corylus avellana L.) cultivars from different growing regions. Scientia Horticulturae 124:128-133

Boccacci P, Botta R & Akkak A (2006) DNA typing and genetic relations among European hazelnut (Corylus avellana L.) cultivars using microsatellite markers. Genome 49:598-611 Boccacci P, Rovira M & Botta R (2008) Genetic diversity of hazelnut (Corylus avellana L.) germplasm in northeastern Spain. HortScience 43:667–672 Botstein D, White RL, Skolnick M & Davis RW (1980) Construction of a genetic linkage map in man using restriction fragment length polymorphisms. American Journal of Human Genetics 32(3):314-331 Botta R, Akkak A & Boccacci P (2005) DNA-typing of hazelnut: a universal methodology for describing cultivars and evaluating genetic relatedness. Acta Hortic 686:117-124 Bowcock AM, Ruiz-Linares A, Tomfohrde J, Minch E, Kidd JR & Cavalli-Sforza LL (1994) High resolution of human evolutionary trees with polymorphic microsatellites. Nature 368(6470):455-457 Campa A, Trabanco N, Pérez-Vega E, Rovira M & Ferreira JJ (2011) Genetic relationship between cultivated and wild hazelnuts (Corylus avellana L.) collected in northern Spain. Plant Breeding 130(3):360-366 FAOstat (2016) Agriculture data http://faostat3.fao.org/home/index.html Accessed 25 May 2016 Gilmore BS, Bassil NV & Hummer KE (2011) DNA extraction protocols from dormant buds of twelve woody plant genera J. Amer. Pom. Soc. 65(4):201-207 Gokirmak T, Mehlenbacher SA & Bassil NV (2009) Characterization of European hazelnut (Corylus avellana) cultivars using SSR markers Genet. Resour. Crop. Ev 56:147-172 Gürcan K & Mehlenbacher SA (2010a) Development of microsatellite marker loci for European hazelnut (Corylus avellana L.) from ISSR fragments Mol. Breeding 26:551-559 Gürcan K & Mehlenbacher SA (2010b) Transferability of microsatellite markers in the Betulaceae J. Amer. Soc. Hort. Sci. 135:159-173 Gürcan K, Mehlenbacher SA, Botta R & Boccacci P (2010a) Development, characterization, segregation, and mapping of microsatellite markers for European hazelnut (Corylus avellana L.) from enriched genomic libraries and usefulness in genetic diversity studies. Tree Genetics and Genomes 6:513-531 Gürcan K, Mehlenbacher SA & Erdogan V (2010b) Genetic diversity in hazelnut cultivars from Black Sea countries assessed using SSR markers. Plant Breeding 129:422-434 Hearne CM, Ghosh S & Todd JA (1992) Microsatellites for linkage analysis of genetic traits. Trends in Genetics 8(8):288-294 Kasapligil B (1972) A bibliography on Corylus (Betulaceae) with annotations. Ann Rep North Nut Grow Assoc 63:107–162 Liu K & Muse S (2004) PowerMarker: new genetic data analysis software. Version 3.0. http://www.powermarker.net. Mehlenbacher SA (2014) Geographic distribution of incompatibility alleles in cultivars and selections of European hazelnut. J. Amer. Soc. Hort. Sci. 139:191-212 Mehlenbacher SA, Brown RN, Nouhra ER, Gokirmak T, Bassil NV & Kubisiak TL (2006) A genetic linkage map for hazelnut (Corylus avellana L.) based on RAPD and SSR markers. Genome 49:122-133

Mehlenbacher SA, Smith DC & McCluskey RL (2009) ‘Yamhill’ hazelnut HortScience 44:845-847 Peterschmidt B (2013) DNA markers and characterization of novel sources of eastern filbert blight resistance in European hazelnut (Corylus avellana L.). Dissertation, Oregon State University, Corvallis, USA Powell W, Machray GC & Provan J (1996) Polymorphism revealed by simple sequence repeats. Trends in Plant Science 1(7):215-222 Sathuvalli VR & Mehlenbacher SA (2012) Characterization of American hazelnut (Corylus americana) accessions and Corylus americana x Corylus avellana hybrids using microsatellite markers Genetic Resources and Crop Evolution 59(6):1055-1075 Sathuvalli VR & Mehlenbacher SA (2013) De novo sequencing of hazelnut bacterial artificial chromosomes (BACs) using multiplex Illumina sequencing and targeted marker development for eastern filbert blight resistance. Tree Genetics & Genomes 9(4):1109-1118 Testolin R & Cipriani G (2010) Molecular markers for germplasm identification and characterization. Acta Hort 859:59-72

Fig. 4.1. UPGMA cluster analysis of 81 unique genotypes among 87 hazelnut accessions using 14-SSRs multiplex.

Table 4.1. List of 102 hazelnut genotypes used in this study. Local inventory number at the NCGR and OSU collection, species, pedigree and origin are indicated.

Cultivar name Local number Species Pedigree Origin Tissue culture plants Barcelona C. avellana Chance seedling Spain Dorris C. avellana OSU 309.074 × Delta Oregon, USA

Dundee C. hybrid C. colurna × C. avellana Oregon, USA Felix C. avellana OSU 384.095 × Delta Oregon, USA Jefferson C. avellana OSU 252.146 × OSU 414.062 Oregon, USA

Lewisb C. avellana OSU 17.028 × Willamette Oregon, USA

Wepster C. avellana Tonda Pacifica × OSU 440.005 Oregon, USA York C. avellana OSU 479.027 × OSU 504.065 Oregon, USA

Zeta C. avellana OSU 342.019 × Zimmerman Oregon, USA

NCGR field collection Ala-Kieri 187.001 C. avellana Selection Finland Alli 999.001 C. avellana ‘Kaiserin Eugenie’ × wild European hazelnut Estonia Anglais 481.001 C. avellana Selection Uncertain Artellet 256.003 C. avellana Selection Spain Aurea 126.001 C. avellana Clonal selection with yellow foliage England United Kingdom B 3 273.001 C. avellana Selection Macedonia Barcelloner Zellernuss 331.001 C. avellana Selection England United Kingdom Barcelonaa 36.001 C. avellana Chance seedling Spain Barr's Zellernuss 333.001 C. avellana Selection England United Kingdom Bergeri 262.001 C. avellana Selection Italy Brixnut 26.001 C. avellana Barcelona × DuChilly Oregon, USA Burgundy Lace 955.07 C. avellana OSU 562.034 × OSU 562.062 Oregon, USA Butler 116.001 C. avellana Barcelona × Daviana Oregon United States

Albania 55 625.001 C. avellana Selection Albania OSU 495.045 427.001 C. avellana Selection, seeds from Russia Russian Federation OSU 556.011 707.001 C. avellana From seeds purchased at a market in Istanbul Istanbul, Turkey Casina 28.001 C. avellana Selection Asturias, Spain Clarka 705.001 C. avellana Tombul Ghiaghli × Willamette Oregon, USA Contorta 50.001 C. avellana Clonal selection with contorted stems and leaves England Cosford 41.001 C. avellana Selection England Cutleaf 18.001 C. avellana Selection with cut-leaves England Daviana 42.001 C. avellana Selection England Dorris 890.001 C. avellana OSU 309.074 × Delta Oregon, USA Du Chilly 232.001 C. avellana Selection (Kentish Cob) Kent, England Dundee 165.001 C. hybrid C. colurna × C. avellana Oregon, USA Ennisa 11.001 C. avellana Barcelona × Daviana Washington, USA Jefferson 894.001 C. avellana OSU 252.146 × OSU 414.062 Oregon, USA Gamma 776.001 C. avellana Casina × VR 6-28 Oregon, USA Gasaway 54.001 C. avellana Selection Washington, USA Gema 23.001 C. avellana Barcelona × DuChilly United States. WA Georgian 759.010 955.04 C. avellana Selection Republic of Georgia OSU 54.039 88.001 C. avellana Open-pollinated seedling selection from Turkish seedlot Oregon, USA Grand Traversea 559.001 C. hybrid Faroka (C.colurna × C.avellana) × OSU 18.114 Michigan, USA Gunslebener Zellernuss 382.001 C. avellana Selection Germany Gustav's Zellernuss 206.001 C. avellana Selection Germany Hall's Gianta 16.001 C. avellana Selection Germany Iannusa Racinante 368.001 C. avellana Selection Sicily, Italy Imperial de Trebizonde 81.001 C. avellana Selection Turkey Imperial de Trebizonde 484.001 C. avellana Selection Turkey Kadetten Zellernuss 323.001 C. avellana Selection Germany

Lewis 633.001 C. avellana OSU 17.028 (Barcelona × Tombul Ghiaghli) × Willamette Oregon, USA Montebelloa 17.001 C. avellana Selection Italy Mortarella 268.007 C. avellana Selection Italy Negret 8.001 C. avellana Selection Spain Nottinghama 297.001 C. avellana Selection England NY 398 193.001 C. hybrid Rush × Red Lambert (C. americana × C. avellana) New York, USA NY 616 104.001 C. hybrid Rush × Barcelona (C. americana × C. avellana) New York, USA OSU 26.072 388.001 C. avellana Seedling selection from seedlot B-122 Oregon, USA Palaz 265.002 C. avellana Selection Turkey Payrone 458.001 C. avellana Selection Torino, Italy Pellicule Rouge 38.001 C. maxima Selection France Purple Aveline 832.001 C. maxima Selection France Ratoli 344.001 C. avellana Selection Spain Rote Zellernus 13.001 C. avellana Selection Netherlands Sacajawea 859.001 C. avellana OSU 43.091 (self-pollination of 'Montebello'?) × Sant Pere Oregon, USA Sant Jaume 249.001 C. avellana Barcelona × Pinyolenc #2 Spain Simon 343.002 C. avellana Negret × Garrofi Spain Tapparona di Mezzanego 750.001 C. avellana Selection Liguria, Italy Tombul Ghiaghli 55.001 C. avellana Selection Turkey Tonda Bianca 21.003 C. avellana Selection Italy Tonda di Giffoni 22.001 C. avellana Selection Italy Tonda Gentile delle Langhe 31.001 C. avellana Selection Italy Tonda Gentile delle Langhe 110.001 C. avellana Selection Italy Tonda Gentile delle Langhe 113.001 C. avellana Selection Italy Tonda Gentile delle Langhe 114.001 C. avellana Selection Italy Tonda Romana 5.001 C. avellana Selection Italy Willamette 500.001 C. avellana Montebello × Compton Oregon, USA Yamhill 738.001 C. avellana OSU 296.082 × VR 8-32 Oregon, USA

Yamhill 898.001 C. avellana OSU 296.082 × VR 8-32 Oregon, USA Zeta 779.001 C. avellana OSU 342.019 × Zimmerman Oregon, USA OSU field Dundeeb LB07.26.05 C. hybrid C. colurna × C. avellana Oregon, USA Lewisb LB07.38 C. avellana OSU 17.028 (Barcelona × Tombul Ghiaghli) × Willamette Oregon, USA Jeffersonb LB15.06 C. avellana OSU 252.146 × OSU 414.062 Oregon, USA Zetab LB11.40 C. avellana OSU 342.019 × Zimmerman Oregon, USA Dorrisb LB40.22 C. avellana OSU 309.074 × Delta Oregon, USA Fusco Rubra 372.007 C. avellana Selection Germany Tonda Pacifica V04 C. avellana Tonda Gentile delle Langhe × OSU 23.024 Oregon, USA Pendula USDA Shop C. avellana Selection England OSU 408.040 LB07.34 C. avellana C. avellana University of Minnesota, USA Epsilon LB11.35 C. avellana OSU 350.089 × Zimmerman Oregon, USA OSU 681.078 Russian LB14.09 C. avellana Open pollinated seedling Moscow, Russia Red Majestic #2 LB15.01 C. avellana Selection Germany Red Majestic #1 LB15.02 C. avellana Selection Germany Red Dragon LB20.20 C. avellana OSU 487.055 × OSU 367.039 Oregon, USA York LB22.04 C. avellana OSU 479.027 × OSU 504.065 Oregon, USA McDonald LB22.07 C. avellana Tonda Pacifica × Santiam Oregon, USA Wepster LB23.04 C. avellana Tonda Pacifica × OSU 440.005 Oregon, USA Felix LB24.11 C. avellana OSU 384.095 × Delta Oregon, USA Eta LB26.20 C. avellana OSU 581.039 × OSU 553.090 Oregon, USA Theta LB27.11 C. avellana OSU 561.184 × Delta Oregon, USA Newberg LB07.25a C. hybrid C. colurna × C. avellana Oregon, USA OSU 414.062 MP03.07 C. avellana OSU 23.017 × VR-1127 Oregon, USA OSU 252.146 MP03.20 C.avellana OSU 41.083 × OSU 17.028 Oregon, USA aTest panel accessions b Six hazelnut accessions not evaluated with the first 14-SSR fingerprinting set

Table 4.2. Twenty SSRs tested for fingerprinting in hazelnuts. The motif, size range, number of alleles, linkage group (LG) and primer concentrations for the optmized multiplex set are provided.

Primer Size range Number of Name Motif LG Concentration References (bp) alleles (A) (mM) Primers removed from the study

LG639 (ATGT)6 224-236 10 6 - (Sathuvalli and Mehlenbacher 2013)

BR249 (AACAGA)5 287-305 7 10S - (Peterschmidt 2013)

BR427 (CCA)5 314-317 5 5S - (Peterschmidt 2013)

BR359 (TCT)5 387-402 13 4S, 4R - (Peterschmidt 2013)

LG631 (TCT)6 355-439 4 6 - (Sathuvalli and Mehlenbacher 2013)

BR446 (CAA)5 155-164 4 11S, 11R - (Peterschmidt 2013) 14 SSR Multiplex

BR270 (CTG)6 90-102 5 1S 0.75 (Peterschmidt 2013)

BR322 (ACT)7 102-114 5 8S 0.375 (Peterschmidt 2013)

BR414 (AAT)6 116-164 10 9S, 9R 0.25 (Peterschmidt 2013) GB949a (TGG)7 154-166 4 10S 3 (Bhattarai 2015) GB950a (TGG)7 162-178 6 7S 1 (Bhattarai 2015) a BR438 (TCA)8 187-211 4 11S 0.25 (Peterschmidt 2013)

CAC C008 (AAG)11(AAG)3 200-253 21 4 4 (Bassil et al. 2013)

BR259 (TCA)10 226-250 9 5S 2 (Peterschmidt 2013)

BR464 (ATC)7 274-292 5 3S 2 (Peterschmidt 2013) a GB875 (GGA)9 330-355 8 5S 3 (Bhattarai 2015)

GB673 (TCACCA)5 355-379 7 5S 2 (Bhattarai 2015)

LG688 (TTC)5 360-372 4 6b 1 (Sathuvalli and Mehlenbacher 2013) a GB395 (CTC)6 377-408 7 2S, 2R 3 (Bhattarai 2015) a BR343 (TGC)6 393-407 3 1S,1R 1 (Peterschmidt 2013) aSix SSRs used to replace the six discarded SSRs.

Table 4.3. Diversity parameters of 14 single-locus hazelnut SSRs used in multiplex. Allele number (A), observed heterozygosity (Ho), expected heterozygosity (He) or gene diversity, polymorphism information content (PIC), and frequency of null alleles (r) in the 87 hazelnuts evaluated.

Marker Allele number (A) Heterozygosity (Ho) Gene diversity (He) PIC r BR270 5 0.48 0.57 0.48 0.057 BR322 5 0.55 0.61 0.56 0.034 BR414 7 0.63 0.59 0.53 -0.030 GB949 5 0.44 0.57 0.51 0.084 GB950 5 0.63 0.68 0.64 0.031 BR438 6 0.48 0.45 0.41 -0.023 CAC-C008 9 0.70 0.80 0.77 0.053 BR259 9 0.72 0.84 0.82 0.063 BR464 5 0.55 0.54 0.46 -0.008 GB875 8 0.67 0.68 0.63 0.007 GB673 6 0.75 0.67 0.63 -0.044 LG688 5 0.61 0.61 0.57 -0.001 GB395 9 0.64 0.79 0.76 0.083 BR343 5 0.66 0.59 0.52 -0.041 Mean 6.36 0.61 0.64 0.59

Table 4.4. 14-SSR fingerprints of some hazelnut accessions and their parents. Alleles transferred from the parents to the progeny are in bold. The accession in bold font is the progeny of the two parents in the section.

Accessions BR270 BR270 BR322 BR322 BR414 BR414 GB949 GB949 GB950 GB950 CAC-C008 CAC-C008 GB875 GB875 Barcelona 93 105 108 122 125 154 160 165 171 200 218 346 DuChilly 93 99 105 122 134 154 160 165 168 200 218 349 Brixnut 93 105 122 125 154 165 171 200 218 346 349 Barcelona 93 105 108 122 125 154 160 165 171 200 218 346 DuChilly 93 99 105 122 134 154 160 165 168 200 218 349 Gem 93 105 108 122 125 154 160 165 171 200 218 346 349 Barcelona 93 105 108 122 125 154 160 165 171 200 218 346 Daviana 93 99 105 108 125 160 162 168 200 218 346 349 Ennis 93 105 108 122 125 160 162 171 200 218 346 349 TombulGhiaghli 99 105 108 122 125 160 163 165 200 209 346 349 Willamette 93 99 108 122 125 154 160 165 168 218 221 346 349 Clark 99 105 108 122 125 160 165 200 218 346 OSU 252.146 93 99 105 108 122 125 154 160 165 168 209 218 346 349 OSU 414.062 93 105 122 125 154 165 168 209 241 346 Jefferson 93 105 122 154 168 209 346

Continued Accessions GB673F GB673F BR438 BR438 BR259 BR259 BR464 BR464 LG688 LG688 GB395 GB395 BR343 BR343 Barcelona 355 379 197 199 226 244 280 286 363 369 380 383 396 DuChilly 355 199 201 226 238 280 286 366 369 380 396 402 Brixnut 355 199 201 226 244 280 286 363 366 383 396 402 Barcelona 355 379 197 199 226 244 280 286 363 369 380 383 396 DuChilly 355 199 201 226 238 280 286 366 369 380 396 402 Gem 355 379 197 199 226 280 369 383 396 402 Barcelona 355 379 197 199 226 244 280 286 363 369 380 383 396 Daviana 355 373 199 235 286 369 380 402 Ennis 355 197 199 244 280 286 363 369 380 396 402 TombulGhiaghli 355 199 201 250 235 280 286 369 377 383 396 Willamette 355 373 197 199 235 244 280 369 383 386 393 396 Clark 355 373 199 235 250 280 286 369 383 396 OSU 252.146 355 373 197 199 235 250 280 286 369 380 389 396 402 OSU 414.062 355 199 244 280 369 377 383 396 402 Jefferson 355 373 199 235 244 280 369 377 380 396 402

CHAPTER 5: SUMMARY Hazelnut is one of the most important nuts in worldwide production. Development of new cultivars is a continuous process, with emphasis on better nut quality, high yield and disease resistance. Hazelnuts are highly heterozygous, and clonally propagated. Traditional propagation methods are not rapid enough to provide the required nursery stock for newly released hazelnut cultivars, but micropropagation can provide rapid production of planting stock. Although a variety of medium formulations are available for hazelnut micropropagation, the diversity of cultivars and genotypes provide growth responses ranging from good to impossible to propagate. Therefore, an optimized tissue culture medium or a small number of suitable media for a range of hazelnut genotypes is required. One of the main factors in the growth of in vitro cultures is the nutrient medium and the amount and type minerals available to the plant. While mineral nutrition is a very important part of in vitro culture, the mineral nutrition optimization process is a complex phenomenon affected by many variables and their interactions. Experimental designs and statistical methods are required that allow the evaluation of many factors at the same time. They also should detect any interactions that affect the plant growth response, in order to make the optimization process effective. In this project, two statistical methods were compared to identify which would make the growth medium optimization process more practical. Optimal micropropagation media for diverse hazelnuts were developed by testing salts and ions as factors within the experimental design. The first optimization study was designed with salts as factors within an optimal design that encompassed all possible combinations of five mineral salts from Driver and Kuniyuki (DKW) medium (Driver and Kuniyuki 1984) and compared response surface methodology (RSM) with a Chi-Squared Automatic Interaction Detection (CHAID) data mining algorithm. Overall, CHAID results were more specific and easier to interpret than RSM graphs. Computer generated optimal design, like RSM, is an excellent tool for reducing treatment numbers compared to traditional factorial designs, but the interpretation of the data is often difficult. CHAID is a novel and promising approach in tissue culture medium optimization. It provides a visual tree and exact cut-offs of the concentrations of significant nutrients which makes it easier to define optimal concentrations of the

nutrient salts. CHAID includes the option to analyze all of the genotypes together which results in development of one or a few optimal media for many genotypes rather than many cultivar-specific formulations. Based on this analysis the new hazelnut medium amounts for better plant quality, shoot length, multiplication and reduced callus were set at: 0.5× NH4NO3, 3× KH2PO4, 1.5× Ca(NO3)2. The other salt factors that were analyzed could remain at the standard DKW concentrations (1×). The ion based optimization study was evaluated with the CHAID data mining algorithm. The CHAID trees revealed significant variables and their interactions, and provided exact cut-off amounts for each of the ions. The critical cut-off values for good shoot quality, elongation, multiplication and moderate - + 2+ callus formation were suggested to be: NO3 <88 mM, NH4 <20 mM, Ca <5 mM, Mg2+ >5 mM and K+ <46 mM. As a result of these studies, salt- and ion-based optimized tissue culture medium nutrient requirement were predicted for several hazelnut cultivars. The importance of salts and ions as factors within the experimental design and analysis was examined, and using salts as factors resulted in complexity within the design as the effects of ions were confounded. Although salt optimization studies are a powerful tool, and are experimentally easier, optimization at the ionic level provided a clearer evaluation of the growth responses, because the plants take up minerals as ions of the corresponding salts. The last step of the research was to develop a reliable and cost-effective SSR multiplexed fingerprinting set to confirm identity and paternity in diverse hazelnut cultivars and species. The multiplex consisted of fourteen polymorphic, easy-to-score SSRs with non-overlapping alleles. The fingerprinting set identified 81 unique accessions of the 102 samples tested and distinguished genotypes mainly according to parentage, and in some cases based on geographic origin. The fingerprinting test of 14 multiplexed SSRs representing all 11 hazelnut linkage groups is a reliable and cost-effective method for confirming identity and paternity in hazelnut. This test will be useful for breeders, germplasm collection curators, propagators and growers for verification of trueness-to-type, and to facilitate comparison of different germplasm collections and identify possible duplications.

Future Research Further research could be done to determine ion-specific effects related to their proportion in the growth medium. For example, the nitrogen effect is highly + dependent on both the total amount of nitrogen and on the proportion of NH4 and - NO3 . Mixture and mixture-amount designs could be used to characterize the amount and proportion effects of ions on growth responses. This type of study could more efficiently screen multiple ions with greatest effect on particular plant growth response and utilize the inherent geometric properties of experimental design free of the ion confounding concept to simplify the ‘best’ ion amounts and proportions required for optimal plant growth. In addition this optimized medium should be tested on a wider range of hazelnut genotypes as a proof of concept.

BIBLIOGRAPHY

Adelberg JW, Delgado MP & Tomkins JT (2010) Spent medium analysis for liquid culture micropropagation of Hemerocallis on Murashige and Skoog medium. In Vitro Cellular & Developmental Biology - Plant 46(1):95-107 Alanagh EN, Garoosi GA, Haddad R, Maleki S, Landín M & Gallego PP (2014) Design of tissue culture media for efficient Prunus rootstock micropropagation using artificial intelligence models. Plant Cell, Tissue and Organ Culture 117(3):349-359 Anderson WC (1984) Micropropagation of filberts, Corylus avellana. Comb Proc Int Plant Prop Soc 33:132-137 Anderson MJ & Whitcomb PJ (2005) RSM simplified: optimizing processes using response surface methods for design of experiments. New York, NY: Productivity Press Ali M, Eyduran E, Tariq MM, Tirink C, Abbas F, Bajwa MA, Baloch MH, Nizamani AH, Waheed A, Awan MA, Shah SH, Ahmad Z & Jan S (2015) Comparison of artificial neural network and decision tree algorithms used for predicting live weight at post weaning period from some biometrical characteristics in Harnai sheep. Pakistan Journal of Zoology 47(6):1579- 1585 Al Kai H, Salesses G & Mouras A (1984) Multiplication in vitro du noisetier (Corylus avellana L.). Agronomie 4:399-402 Baldoni L, Cultrera N, Mariotti R, Ricciolini C, Arcioni S, Vendramin G, Buonamici A, Porceddu A, Sarri V, Ojeda M, Trujillo I, Rallo L, Belaj A, Perri E, Salimonti A, Muzzalupo I, Casagrande A, Lain O, Messina R & Testolin R (2009) A consensus list of microsatellite markers for olive genotyping. Mol. Breed. 24:213-231 Bassil NV, Boccacci P, Botta R, Postman J & Mehlenbacher SA ( 2013) Nuclear and chloroplast microsatellite markers to assess genetic diversity and evolution in hazelnut species, hybrids and cultivars. Genet. Resources Crop. Evol. 60:543-568 Bassil NV, Botta R & Mehlenbacher SA (2005a) Microsatellite markers in hazelnut: isolation, characterization and cross-species amplification. J. Am. Soc. Hortic. Sci. 130:543-549 Bassil NV, Botta R & Mehlenbacher SA (2005b) Additional microsatellite markers of the European hazelnut. Acta Hort. 686:105-110 Bassil NV, Hummer K, Botu M & Sezer A (2009) SSR fingerprinting panel verifies identities of clones in backup hazelnut collection of USDA genebank. 845:95-102 Bassil NV, Mok D, Mok M & Rebhuhn BJ (1992) Micropropagation of the hazelnut, Corylus avellana. Acta Hortic 300:137-140 Beltramo C, Valentini N, Portis E, Torello Marinoni D, Boccacci P, Sandoval Prando MA & Botta R (2016) Genetic mapping and QTL analysis in European hazelnut (Corylus avellana L.). Molecular Breeding 36(3):1-17 Bhattarai G (2015) Microsatellite marker development, characterization and mapping in European hazelnut (Corylus avellana L.), and investigation of novel sources of eastern filbert blight resistance in Corylus. Dissertation, Oregon State University, Corvallis, USA Boccacci P, Akkak A, Bassil NV, Mehlenbacher SA & Botta R (2005) Characterization and evaluation of microsatellite loci in European hazelnut

(Corylus avellana L.) and their transferability to other Corylus species. Molec. Ecol. Notes 5:934-937 Boccacci P, Aramini M, Valentini N, Bacchetta L, Rovira M, Drogoudi P, Silva AP, Solar A, Calizzano F, Erdoğan V, Cristofori V, Ciarmiello LF, Contessa C, Ferreira JJ, Marra FP & Botta R (2013) Molecular and morphological diversity of on-farm hazelnut (Corylus avellana L.) landraces from southern Europe and their role in the origin and diffusion of cultivated germplasm. Tree Genetics & Genomes 9(6):1465-1480 Boccacci P, Beltramo C, Sandoval Prando MA, Lembo A, Sartor C, Mehlenbacher SA, Botta R & Torello Marinoni D (2015) In silico mining, characterization and cross-species transferability of EST-SSR markers for European hazelnut (Corylus avellana L.). Molecular Breeding 35(1):1-14 Boccacci P & Botta R (2010) Microsatellite variability and genetic structure in hazelnut (Corylus avellana L.) cultivars from different growing regions. Scientia Horticulturae 124:128-133 Boccacci P, Botta R & Akkak A (2006) DNA typing and genetic relations among European hazelnut (Corylus avellana L.) cultivars using microsatellite markers. Genome 49:598-611 Boccacci P, Rovira M & Botta R (2008) Genetic diversity of hazelnut (Corylus avellana L.) germplasm in northeastern Spain. HortScience 43:667–672 Botstein D, White RL, Skolnick M & Davis RW (1980) Construction of a genetic linkage map in man using restriction fragment length polymorphisms. American Journal of Human Genetics 32(3):314-331 Botta R, Akkak A & Boccacci P (2005) DNA-typing of hazelnut: a universal methodology for describing cultivars and evaluating genetic relatedness. Acta Hortic 686:117-124 Bradley N (2007) The response surface methodology. Dissertation, Indiana University South Bend, USA Chen H, Mehlenbacher SA & Smith DC (2005) AFLP markers linked to eastern filbert blight resistance from OSU 408.040 hazelnut. J. Amer. Soc. Hort. Sci. 130(3):412-417 Compton M & Mize C (1999) Statistical considerations for in vitro research: I — Birth of an idea to collecting data. In Vitro Cellular & Developmental Biology - Plant 35(2):115-121 Damiano C, Catenaro E, Giovinazzi J & Frattarelli A (2005) Micropropagation of hazelnut (Corylus avellana L.). Acta Hortic 686:221-225 Design-Expert (2010) Stat-Ease, Inc., Minneapolis, MN. Díaz-Pérez FM & Bethencourt-Cejas M (2016) CHAID algorithm as an appropriate analytical method for tourism market segmentation. Journal of Destination Marketing & Management Diaz Sala C, Rey M & Rodriguez R (1990) In vitro establishment of a cycloclonal chain from nodal segments and apical buds of adult hazel (Corylus avellana L). Plant Cell Tiss Organ Cult 23:151-157 Driver JA & Kuniyuki AH (1984) In vitro propagation of Paradox walnut rootstock HortScience 19:507-509 Evens TJ & Niedz RP (2008) Are hofmeister series relevant to modern ion- specific effects research? Scholarly Research Exchange 2008:1-9 FAOstat (2016) Agriculture data http://faostat3.fao.org/home/index.html Accessed 25 May 2016

Gago J, Perez-Tornero O, Landin M, Burgos L & Gallego PP (2011) Improving knowledge of plant tissue culture and media formulation by neurofuzzy logic: a practical case of data mining using apricot databases. Journal of plant physiology 168(15):1858-65 George E & de Klerk G-J (2008) The components of plant tissue culture media I: macro- and micro-nutrients. In Plant propagation by tissue culture 3rd edition. Edited by: George EF, Hall MA, de Klerk G-J. Dordrecht, The Netherlands: Springer; 2008:65-113 Gokirmak T, Mehlenbacher SA & Bassil NV (2009) Characterization of European hazelnut (Corylus avellana) cultivars using SSR markers Genet. Resour. Crop. Ev 56:147-172 Gupta PK, Varshney RK, Sharma PC & Ramesh B (1999) Molecular markers and their applications in wheat breeding. Plant Breeding 118(5):369-390 Gürcan K & Mehlenbacher SA (2010a) Development of microsatellite marker loci for European hazelnut (Corylus avellana L.) from ISSR fragments Mol. Breeding 26:551-559 Gürcan K & Mehlenbacher SA (2010b) Transferability of microsatellite markers in the Betulaceae J. Amer. Soc. Hort. Sci. 135:159-173 Gürcan K, Mehlenbacher SA, Botta R & Boccacci P (2010a) Development, characterization, segregation, and mapping of microsatellite markers for European hazelnut (Corylus avellana L.) from enriched genomic libraries and usefulness in genetic diversity studies. Tree Genetics and Genomes 6:513-531 Gürcan K, Mehlenbacher SA & Erdogan V (2010b) Genetic diversity in hazelnut cultivars from Black Sea countries assessed using SSR markers. Plant Breeding 129:422-434 Hand C (2013) Improving initiation and mineral nutrition for hazelnut (Corylus avellana) micropropagation. Dissertation, Oregon State University, Corvallis, USA Hand C, Maki S & Reed B (2014) Modeling optimal mineral nutrition for hazelnut micropropagation. Plant Cell, Tissue and Organ Culture 119(2):411-425 Hand C & Reed BM (2014) Minor nutrients are critical for the improved growth of Corylus avellana shoot cultures. Plant Cell, Tissue and Organ Culture 119(2):427-439 Hearne CM, Ghosh S & Todd JA (1992) Microsatellites for linkage analysis of genetic traits. Trends in Genetics 8(8):288-294 Hébert M, Collin-Vézina D, Daigneault I, Parent N & Tremblay C (2006) Factors linked to outcomes in sexually abused girls: a regression tree analysis. Compr Psychiatry 47:443-455 Ibañez MA, Martin C & Pérez C (2003) Alternative statistical analyses for micropropagation: A practical case of proliferation and rooting phases in Viburnum opulus. In Vitro Cellular & Developmental Biology - Plant 39(5):429-436 Jyoti J (2013) Micropropagation of Hazelnut (Corylus Species). Master of Science, Plant Agriculture, University of Guelph, Ontario, Canada Kasapligil B (1972) A bibliography on Corylus (Betulaceae) with annotations. Ann Rep North Nut Grow Assoc 63:107–162 Liu K & Muse S (2004) PowerMarker: new genetic data analysis software. Version 3.0. http://www.powermarker.net.

Machado M, da Silva A, Biasi L, Deschamps C, Filho J & Zanette F (2014) Influence of calcium content of tissue on hyperhydricity and shoot tip necrosis of in vitro regenerated shoots of Lavandula angustifolia Mill Brazilian Archives of Biology and Technology 57(5):636-643 Marinoni D, Akkak A, Bounous G, Edwards KJ & Botta R (2003) Development and characterization of microsatellite markers in Castanea sativa (Mill.). Molecular Breeding 11(2):127-136 Mehlenbacher SA (2009) Genetic resources for hazelnut: state of the art and future perspectives. Acta Hort 845:33-38 Mehlenbacher SA (2014) Geographic distribution of incompatibility alleles in cultivars and selections of European hazelnut. J. Amer. Soc. Hort. Sci. 139:191-212 Mehlenbacher SA, Brown RN, Nouhra ER, Gokirmak T, Bassil NV & Kubisiak TL (2006) A genetic linkage map for hazelnut (Corylus avellana L.) based on RAPD and SSR markers. Genome 49:122-133 Mehlenbacher SA, Smith DC & McCluskey RL (2009) ‘Yamhill’ hazelnut HortScience 44:845-847 Mize C, Koehler K & Compton M (1999) Statistical considerations for in vitro research: II — Data to presentation. In Vitro Cellular & Developmental Biology - Plant 35(2):122-126 Mohammadzedeh M, Fattahi R, Zamani Z & Khadivi-Khub A (2014) Genetic identity and relationships of hazelnut (Corylus avellana L.) landraces as revealed by morphological characteristics and molecular markers. Scientia Horticulturae 167:17-26 Montgomery DC (2005) Design and analysis of experiments: Response surface method and designs. New Jersey: John Wiley and Sons, Inc. Murashige T & Skoog F (1962) A revised medium for rapid growth and bio assays with tobacco tissue cultures. Physiol Plant 15:473-497 Nas MN, Eskridge K & Read P (2005) Experimental designs suitable for testing many factors with limited number of explants in tissue culture. Plant Cell, Tissue and Organ Culture 81(2):213-220 Nas MN & Read PE (2004) A hypothesis for the development of a defined tissue culture medium of higher plants and micropropagation of hazelnuts. Scientia Horticulturae 101(1-2):189-200 Niedz RP & Evens TJ (2006) A solution to the problem of ion confounding in experimental biology. 3(6):417 Niedz RP & Evens TJ (2007) Regulating plant tissue growth by mineral nutrition. In Vitro Cellular & Developmental Biology - Plant 43(4):370-381 Niedz RP, Hyndman SE & Evens TJ (2007) Using a gestalt to measure the quality of in vitro responses. Scientia Horticulturae 112(3):349-359 Niedz RP & Evens TJ (2008) The effects of nitrogen and potassium nutrition on the growth of nonembryogenic and embryogenic tissue of sweet orange (Citrus sinensis (L.) Osbeck). BMC Plant Biol 8:126 Perez-Tornero O, Lopez JM, Egea J & Burgos L (2000) Effect of basal media and growth regulators on the in vitro propagation of apricot (Prunus armenica L.) cv. Canino. J. Hortic. Sci. Biotech. 75: 283-286 Peterschmidt B (2013) DNA markers and characterization of novel sources of eastern filbert blight resistance in European hazelnut (Corylus avellana L.). Dissertation Oregon State University, Corvallis, USA

Poothong S & Reed BM (2016) Optimizing shoot culture media for Rubus + − germplasm: the effects of NH4 , NO3 , and total nitrogen. In Vitro Cellular & Developmental Biology - Plant:1-11 Powell W, Machray GC & Provan J (1996) Polymorphism revealed by simple sequence repeats. Trends in Plant Science 1(7):215-222 Radojevic N, Vujieie R & Nesrovie M (1975) Embryogenesis in tissue culture of Corylus avellana L. Z Pflanzenphysiol Bd. 77: 33-41. Ramage C & Williams R (2002) Mineral nutrition and plant morphogenesis Cell. Dev. Biol. Plant 38:116–124 Rashidi S, Ranjitkar P & Hadas Y (2014) Modeling bus dwell time with decision tree-based methods. Transportation Research Record: Journal of the Transportation Research Board 2418:74-83 Reed BM, Wada S, DeNoma J & Niedz RP (2013) Improving in vitro mineral nutrition for diverse pear germplasm. In Vitro Cellular and Developmental Biology - Plant 49:343-355 Sathuvalli VR, Chen H, Mehlenbacher SA & Smith DC (2011) DNA markers linked to eastern filbert blight resistance in “Ratoli” hazelnut (Corylus avellana L.). Tree Genetics & Genomes 7(2):337-345 Sathuvalli VR & Mehlenbacher SA (2012) Characterization of American hazelnut (Corylus americana) accessions and Corylus americana x Corylus avellana hybrids using microsatellite markers. Genetic Resources and Crop Evolution 59(6):1055-1075 Sathuvalli VR, Mehlenbacher SA & Smith DC (2012) Identification and mapping of DNA markers linked to eastern filbert blight resistance from OSU 408.040 hazelnut. HortScience 47:570-573 Singha S, Townsend EC & Oberly GH (1990) Relationship between calcium and agar on vitrification and shoot-tip necrosis of quince (Cydonia oblonga Mill.) shoots in vitro. Plant Cell, Tissue and Organ Culture 23(2):135-142 SPSS (2013) Statistics for Windows, Version 22.0. Armonk, NY: IBM Corp. Statistics-Solutions (2016) CHAID. http://www.statisticssolutions.com/non- parametric-analysis-chaid/ Testolin R & Cipriani G (2010) Molecular markers for germplasm identification and characterization. Acta Hort 859:59-72 Verbruggen N & Hermans C (2013) Physiological and molecular responses to magnesium nutritional imbalance in plants. Plant and Soil 368(1):87-99 Vos P, Hogers R, Bleeker M, Reijans M, van de Lee T, Hornes M, Frijters A, Pot J, Peleman J & Kuiper M (1995) AFLP: a new technique for DNA fingerprinting. Nucleic Acids Research 23(21):4407-4414 Wada S, Niedz RP & Reed BM (2015) Determining nitrate and ammonium requirements for optimal in vitro response of diverse pear species. In Vitro Cellular & Developmental Biology - Plant 51(1):19-27 Williams JG, Kubelik AR, Livak KJ, Rafalski JA & Tingey SV (1990) DNA polymorphisms amplified by arbitrary primers are useful as genetic markers. Nucleic Acids Research 18(22):6531-6535 Williams RR (1993) Mineral nutrition in vitro—a mechanistic approach. Aust J Bot 41:237-251 Yu X & Reed BM (1993) Improved shoot multiplication of mature hazelnut (Corylus avellana L.) in vitro using glucose as a carbon source. Plant Cell Rep 12:256–259

Yu X & Reed BM (1995) Micropropagation system for hazelnuts (Corylus species). HortScience 30:120–123

APPENDICES

Callus-Wepster Design-Expert® Software 3.00 a Factor Coding: Actual Callus 2.50 Design Points 2.40 2.50 2.00 1.00 2.20 1.50

X1 = A: CaCl2.2H2O B : N H 4 N O 3 X2 = B: NH4NO3 1.00 2 Actual Factors 2.00 C: Ca(NO3)2.4H2O = 1.00 0.50 D: MgSO4.7H2O = 1.00 0.50 1.00 1.50 2.00 2.50 3.00 E: KH2PO4 = 1.00 F: K2SO4 = 1.00 A: CaCl2.2H2O Callus-Zeta Design-Expert® Software 3.00 b Factor Coding: Actual Callus 2.50 2.80 2.50

1.00 2.00 2.60

X1 = B: NH4NO3 2.20 2.40

X2 = F: K2SO4 1.50 F : K 2 S O 4 Actual Factors A: CaCl2.2H2O = 1.00 1.00 C: Ca(NO3)2.4H2O = 1.00 D: MgSO4.7H2O = 0.50 0.50 E: KH2PO4 = 0.50 0.50 1.00 1.50 2.00 2.50 3.00 B: NH4NO3

Appendix A. Response surface graphs of mineral nutrient effects on callus (1=callus > 2mm, 2=callus ≤ 2 mm, and 3=absent) ratings of a. ‘Wepster’ and b. ‘Zeta’. The callus was color coordinated from less callus (red-yellow) to more callus (green-blue). a. The red dot represents the control with average callus formation of 2.

Appendix B. The CHAID decision tree diagram for callus formation of ‘Dorris’, ‘Wepster’ and ‘Zeta’ hazelnuts. Nodes were determined by the significance of the factors. Salt cut-off values are × DKW.

Callus

Appendix C. Fingerprints of hazelnut genotypes for 14 single-locus hazelnut SSRs used in multiplex.

Genotypesa BR270 BR322 BR414 GB949 GB950 CACC008 GB875 Ala Kieri 187.001 93/99 102/105 125/125 163/163 162/162 241/241 346/349 Albania55 625.001 99/99 102/102 140/140 160/160 162/165 218/218 336/346 Alli 999.001 90/93 102/105 125/125 160/160 165/165 241/241 336/349 Anglais 481.001 93/93 105/105 125/125 160/160 165/165 212/218 340/349 Artellet 256.003 93/93 105/108 122/125 154/160 165/165 218/218 346/355 Aurea 126.001 99/99 105/108 125/125 160/163 162/168 241/241 340/340 B 3 273.001 93/99 105/105 125/140 154/160 168/171 200/241 346/349 Barcelloner Zellernuss 331.001 93/93 108/108 125/125 154/163 162/165 218/241 346/349 Barcelona 36.001 93/93 105/108 122/125 154/160 165/171 200/218 346/346 Barcelona TC 93/93 105/108 122/125 154/160 165/171 200/218 346/346 Barr’s Zellernuss 333.001 93/93 105/105 125/125 163/163 165/171 203/218 340/349 Bergeri 262.001 93/93 105/105 125/140 154/154 162/171 200/241 349/355 Brixnut 26.001 93/93 105/105 122/125 154/154 165/171 200/218 346/349 Burgundy Lace 955.070SM 99/99 105/108 125/137 160/160 168/171 241/241 336/349 Butler 116.001 93/93 105/108 125/125 154/160 162/165 218/218 346/346 Casina 28.001 99/99 105/108 122/125 160/160 165/168 200/221 346/349 Clark 705.001 99/99 105/108 122/125 160/160 165/165 200/218 346/346 Cosford 41.001 93/99 105/108 125/125 160/160 162/165 218/241 340/349 Cutleaf 18.001 93/99 105/105 125/137 160/160 168/171 241/241 336/349 Daviana 42.001 93/99 105/108 125/125 160/160 162/168 218/200 346/349 Dorris 890.001 93/99 102/105 122/125 154/160 165/171 218/241 346/355 Dorris LB40.22SM 93/99 102/105 122/125 154/160 165/171 218/241 346/355 Dorris TC 93/99 102/105 122/125 154/160 165/171 218/241 346/355 Du Chilly 232.001 93/99 105/105 122/134 154/160 165/168 200/218 349/349 Dundee 165.001 96/99 105/105 122/122 157/160 168/168 218/218 346/349 Dundee 07.26.05SM 96/99 105/105 122/122 157/160 168/168 218/218 346/349 Dundee TC 96/99 105/105 122/122 157/160 168/168 218/218 346/349

Ennis 11.001 93/93 105/108 122/125 160/160 162/171 200/218 346/349 Epsilon LB11.35SM 93/93 105/108 125/125 160/160 165/165 221/221 349/355 Eta LB26.20SM 93/99 105/108 122/125 154/154 165/165 209/241 343/346 Felix LB24.11SM 93/99 102/105 125/134 160/160 165/171 221/241 346/349 Felix TC 93/99 102/105 125/134 160/160 165/171 221/241 346/349 Fusco Rubra 372.007SM 90/93 105/105 125/137 163/163 165/165 241/241 340/349 Gamma 776.001 93/99 105/108 125/125 160/160 165/168 221/221 349/349 Gasaway 54.001 93/93 105/105 125/134 160/160 162/165 244/241 349/355 Gem 23.001 93/93 105/108 122/125 154/160 165/171 200/218 346/349 Georgian 759.010 955.040 99/102 105/105 116/125 160/166 168/171 209/218 346/349 Grand Traverse 559.001 96/99 102/108 125/125 160/160 162/162 218/241 349/349 Gunslebener Zellernuss 382.001 99/99 102/108 125/125 154/154 168/171 221/241 346/349 Gustav’s Zellernuss 206.001 90/99 102/102 125/125 160/160 165/171 241/241 349/349 Hall’s Giant 16.001 99/99 102/105 125/125 154/160 171/171 218/241 349/355 Iannusa Racinante 368.001 99/99 105/105 125/137 154/160 165/168 218/221 346/346 Imperial de Trebizonde 484.001 93/99 108/111 122/137 160/166 165/165 218/244 334/346 Imperial de Trebizonde 81.001 93/99 108/111 122/137 160/166 165/165 218/244 334/346 Jefferson 894.001 93/93 105/105 122/122 154/154 168/168 209/209 346/346 Jefferson LB15.06SM 93/93 105/105 122/122 154/154 168/168 209/209 346/346 Jefferson TC 93/93 105/105 122/122 154/154 168/168 209/209 346/346 Kadetten Zellernuss 323.001 90/99 102/102 125/125 154/154 162/165 241/241 349/349 Lewis 633.001 93/99 105/108 125/125 160/163 165/168 209/218 346/349 Lewis LB07.38SM 93/99 105/108 125/125 160/163 165/168 209/218 346/349 Lewis TC 93/99 105/108 125/125 160/163 165/168 209/218 346/349 McDonald LB22.07SM 99/99 105/105 122/134 160/160 165/168 200/221 343/355 Montebello 17.001 99/99 105/108 122/125 154/154 165/168 200/221 346/346 Mortarella 268.007 99/99 105/108 125/140 154/160 165/171 209/241 346/349 Negret 8.001 93/99 105/108 122/125 160/160 165/168 241/209 346/346 Newberg LB07.25aSM 96/99 102/105 125/125 157/160 171/171 218/218 349/349 Nottingham 297.001 99/99 102/105 122/125 160/160 165/168 200/241 346/349

NY 398 193.001 93/99 102/114 125/146 154/166 165/171 218/241 349/349 NY 616 104.001 93/93 108/114 125/146 160/166 165/171 241/200 346/346 OSU 252.146 MP03.20SM 93/99 105/108 122/125 154/160 165/168 209/218 346/349 OSU 26.072 388.001 96/99 105/105 116/125 154/160 165/171 200/241 346/346 OSU 408.040 LB07.34SM 93/99 102/105 125/125 160/160 165/178 218/241 340/340 OSU 414.062 MP03.07SM 93/93 105/105 122/125 154/154 165/168 209/241 346/346 OSU 54.039 88.001 93/99 105/111 125/125 160/166 165/168 218/218 346/349 OSU 681.078 LB14.09SM 90/93 102/105 125/140 154/163 165/171 244/253 340/346 OSU 495.045 427.001 90/102 105/105 125/125 154/163 168/171 218/241 349/355 OSU 556.011 707.001 99/102 105/108 122/122 154/160 165/168 209/241 346/346 Palaz 265.002 93/99 105/111 122/122 160/166 165/165 209/218 346/349 Payrone 458.001 93/99 105/108 122/125 160/160 162/165 200/221 346/349 Pellicule Rouge 38.001 93/99 111/111 122/125 154/160 165/168 200/218 346/349 Pendula SM 90/93 102/108 125/137 160/163 162/162 241/241 330/330 Purple Aveline 93/99 111/111 122/125 154/160 165/168 200/218 346/349 Ratoli 344.001 93/93 105/108 122/125 160/160 165/165 200/209 340/346 Red Dragon LB20.20SM 93/99 102/102 134/137 160/160 162/171 212/218 349/349 Red Majestic #1 LB15.02SM 99/99 102/102 125/140 160/160 168/168 212/212 340/346 RedMajestic #2 LB15.01SM 99/99 102/102 125/125 154/160 162/171 241/241 340/346 Rote Zellernus 13.001 99/99 108/111 125/137 160/160 162/162 218/241 349/349 Sacajawea 859.001 99/99 105/105 122/122 160/160 168/168 209/218 346/346 Sant Jaume 249.001 93/93 105/105 125/125 154/160 165/165 200/241 330/346 Simon 343.002 93/99 105/105 122/122 160/160 168/171 209/218 346/346 Tapparona di Mezanego750.001 93/99 105/108 122/125 154/160 165/168 200/241 346/346 Theta LB27.11SM 93/93 105/105 125/125 160/160 168/171 221/241 346/355 Tombul Ghiaghli 55.001 99/99 105/108 122/125 160/163 165/165 200/209 346/349 TondaBianca 21.003 93/99 105/105 125/125 154/160 168/168 221/241 346/349 Tonda di Giffoni 22.001 93/99 105/105 122/125 160/160 165/168 200/241 346/346 Tonda Gentile delle Langhe 110001 99/99 105/108 122/125 160/160 165/165 200/200 343/346 Tonda Gentile delle Langhe 113.001 99/99 105/108 122/125 160/160 165/165 200/200 343/346

100

Tonda Gentile delle Langhe 114.001 99/99 105/108 122/125 160/160 165/165 200/200 343/346 Tonda Gentile delle Langhe 31.001 99/99 105/108 122/125 160/160 165/165 200/200 343/346 Tonda Pacifica V04SM 99/99 105/105 122/125 160/166 165/165 209/209 343/346 Tonda Romana 5.001 93/93 105/108 122/125 160/163 165/171 200/221 346/355 Wepster LB23.04SM 99/99 105/105 122/125 166/166 162/165 209/218 346/346 Wepster TC 99/99 105/105 122/125 166/166 162/165 209/218 346/346 Willamette 500.001 93/99 108/108 122/125 154/160 165/168 218/221 346/349 Yamhill 738.001 93/93 105/105 122/125 154/154 165/165 221/241 346/349 Yamhill 898.001 93/93 105/105 122/125 154/154 165/165 221/241 346/349 York LB22.04SM 99/99 105/105 122/125 160/160 168/168 221/221 346/355 Zeta 779.001 93/99 105/105 122/125 160/160 165/168 200/221 346/346 Zeta LB11.40SM 93/99 105/105 122/125 160/160 165/168 200/221 346/346 Zeta TC 93/99 105/105 122/125 160/160 165/168 200/221 346/346 aNumbers following the name indicate inventory number of accession at the NCGR; SM indicates the OSU collection of Shawn Mehlenbacher; OSU designations are row and tree (row.tree); TC indicates tissue culture propagated plant.

Appendix C continued Genotypesa GB673F BR438 BR259 BR464 LG688 67L9 GB395 BR343 Ala Kieri 187.001 355/367 199/199 229/247 280/280 363/363 386/386 402/402 Albania55 625.001 367/379 197/199 235/235 280/286 363/369 377/383 402/402 Alli 999.001 355/355 199/199 229/244 280/280 363/369 395/395 393/402 Anglais 481.001 373/379 199/199 235/235 280/286 363/369 380/383 396/402 Artellet 256.003 355/367 197/199 247/247 280/292 363/366 383/386 402/402 Aurea 126.001 373/373 199/199 229/241 280/280 360/363 380/380 402/402 B 3 273.001 367/379 197/201 226/244 280/292 366/363 377/377 396/402 Barcelloner Zellernuss 331.001 355/367 197/199 238/247 286/292 360/369 383/389 402/402 Barcelona 36.001 355/379 197/199 226/244 280/286 363/369 380/383 396/396 Barcelona TC 355/379 197/199 226/244 280/286 363/369 380/383 396/396 Barr’s Zellernuss 333.001 355/379 199/199 235/235 280/286 360/366 383/395 396/396 Bergeri 262.001 355/379 197/199 247/247 286/292 363/369 380/395 393/402

101

Brixnut 26.001 355/355 199/201 226/244 280/286 363/366 383/383 396/402 Burgundy Lace 955.070SM 355/373 199/199 241/241 280/286 369/369 395/395 393/402 Butler 116.001 373/379 199/199 226/235 286/286 369/369 380/380 396/402 Casina 28.001 355/355 199/199 238/247 280/286 366/369 383/386 396/402 Clark 705.001 355/373 199/199 235/250 280/286 369/369 383/383 396/396 Cosford 41.001 355/373 197/199 235/235 286/286 369/369 380/389 402/402 Cutleaf 18.001 355/373 199/199 241/241 280/280 360/369 395/395 393/402 Daviana 42.001 355/373 199/199 235/235 286/286 369/369 380/380 402/402 Dorris 890.001 355/367 199/199 241/244 280/280 363/369 383/395 396/402 Dorris LB40.22SM 355/367 199/199 241/244 280/280 363/369 383/395 396/402 Dorris TC 355/367 199/199 241/244 280/280 363/369 383/395 396/402 Du Chilly 232.001 355/355 199/201 226/238 280/286 366/369 380/380 396/402 Dundee 165.001 355/370 187/199 226/238 274/280 369/369 395/408 396/407 Dundee 07.26.05SM 355/370 187/199 226/238 274/280 369/369 395/408 396/407 Dundee TC 355/370 187/199 226/238 274/280 369/369 395/408 396/407 Ennis 11.001 355/355 197/199 244/244 280/286 363/369 380/380 396/402 Epsilon LB11.35SM 355/355 199/201 235/244 280/286 366/369 383/386 396/396 Eta LB26.20SM 355/367 199/199 235/235 280/280 366/366 383/383 396/402 Felix LB24.11SM 355/367 199/199 229/247 280/280 369/369 377/383 396/399 Felix TC 355/367 199/199 229/247 280/280 369/369 377/383 396/399 Fusco Rubra 372.007SM 355/367 199/201 241/241 274/280 360/363 395/395 402/402 Gamma 776.001 355/379 199/199 226/238 280/286 369/369 383/386 396/402 Gasaway 54.001 367/367 199/201 235/235 280/280 369/369 386/395 402/402 Gem 23.001 355/379 197/199 226/226 280/280 369/369 383/383 396/402 Georgian 759.010 955.040 355/373 199/203 235/250 280/286 369/369 386/386 393/402 Grand Traverse 559.001 367/370 197/201 247/247 274/286 369/369 404/404 402/402 Gunslebener Zellernuss 382.001 355/355 197/199 235/235 280/280 363/366 383/395 402/402 Gustav’s Zellernuss 206.001 355/367 197/199 229/244 280/280 360/369 395/395 402/402 Hall’s Giant 16.001 367/367 197/199 229/235 280/280 360/369 395/395 393/402 Iannusa Racinante 368.001 355/373 199/199 229/244 280/280 363/369 383/386 393/396

102

Imperial de Trebizonde 484.001 355/379 197/199 226/238 280/286 369/369 383/386 396/402 Imperial de Trebizonde 81.001 355/379 197/199 226/238 280/286 369/369 383/386 396/402 Jefferson 894.001 355/373 199/199 235/244 280/280 369/369 377/380 396/402 Jefferson LB15.06SM 355/373 199/199 235/244 280/280 369/369 377/380 396/402 Jefferson TC 355/373 199/199 235/244 280/280 369/369 377/380 396/402 Kadetten Zellernuss 323.001 355/367 199/199 244/247 280/286 363/369 395/395 393/402 Lewis 633.001 355/373 197/201 250/235 280/280 369/369 380/383 396/396 Lewis LB07.38SM 355/373 197/201 250/235 280/280 369/369 380/383 396/396 Lewis TC 355/373 197/201 250/235 280/280 369/369 380/383 396/396 McDonald LB22.07SM 367/379 199/199 226/229 280/280 360/369 377/383 396/402 Montebello 17.001 355/379 199/199 226/244 280/280 363/369 383/386 393/396 Mortarella 268.007 379/379 199/199 226/241 280/286 363/369 380/383 396/402 Negret 8.001 355/355 199/199 235/238 280/280 369/369 383/395 396/402 Newberg LB07.25aSM 367/376 187/201 238/229 280/283 369/366 395/395 396/402 Nottingham 297.001 355/355 199/199 229/238 280/286 360/369 395/395 402/402 NY 398 193.001 367/379 199/211 229/235 280/292 372/363 395/398 393/402 NY 616 104.001 355/379 199/211 244/235 280/292 372/369 383/383 396/402 OSU 252.146 MP03.20SM 355/373 197/199 235/250 280/286 369/369 380/389 396/402 OSU 26.072 388.001 355/367 199/203 226/238 286/286 369/369 383/389 399/402 OSU 408.040 LB07.34SM 355/367 199/199 235/235 286/292 360/369 380/386 402/402 OSU 414.062 MP03.07SM 355/355 199/199 244/244 280/280 369/369 377/383 396/402 OSU 54.039 88.001 355/367 199/201 241/241 280/280 363/366 377/377 399/402 OSU 681.078 LB14.09SM 355/355 197/197 226/241 280/292 360/363 389/395 396/402 OSU 495.045 427.001 367/373 199/199 235/241 280/286 369/369 395/395 393/402 OSU 556.011 707.001 355/355 201/203 229/235 274/286 369/369 377/380 396/402 Palaz 265.002 355/373 199/201 226/235 280/280 366/369 377/383 399/402 Payrone 458.001 355/373 197/199 238/244 280/280 360/369 383/389 393/396 Pellicule Rouge 38.001 355/379 199/199 235/238 286/286 366/369 383/395 396/402 Pendula SM 355/367 197/199 232/232 280/292 363/369 383/395 402/402 Purple Aveline 355/379 199/199 235/238 286/286 366/369 383/395 396/402

103

Ratoli 344.001 355/379 197/199 226/238 280/280 363/369 377/377 396/396 Red Dragon LB20.20SM 355/367 199/199 241/235 280/280 360/369 386/386 402/402 Red Majestic #1 LB15.02SM 355/355 199/199 241/241 280/286 369/369 386/395 402/402 RedMajestic #2 LB15.01SM 355/379 199/199 235/241 280/280 366/366 386/386 402/402 Rote Zellernus 13.001 355/373 199/199 241/241 286/286 360/369 389/395 402/402 Sacajawea 859.001 355/355 199/199 235/244 280/286 369/369 383/395 393/402 Sant Jaume 249.001 355/367 199/199 241/244 280/280 363/369 383/395 396/402 Simon 343.002 355/379 197/199 235/241 280/286 369/369 383/395 402/402 Tapparona di Mezanego750.001 367/379 199/199 235/235 286/286 360/369 380/380 393/402 Theta LB27.11SM 367/379 197/199 229/235 280/280 369/369 380/386 396/402 Tombul Ghiaghli 55.001 355/355 199/201 250/235 280/286 369/369 377/383 396/396 TondaBianca 21.003 355/355 197/199 235/244 280/286 363/369 380/386 393/402 Tonda di Giffoni 22.001 355/379 197/199 226/235 280/280 369/369 380/383 396/402 Tonda Gentile delle Langhe 110001 373/379 199/199 226/241 280/286 360/366 383/395 396/402 Tonda Gentile delle Langhe 113.001 373/379 199/199 226/241 280/286 360/366 383/395 396/402 Tonda Gentile delle Langhe 114.001 373/379 199/199 226/241 280/286 360/366 383/395 396/402 Tonda Gentile delle Langhe 31.001 373/379 199/199 241/226 280/286 360/366 383/395 396/402 Tonda Pacifica V04SM 355/379 199/199 226/244 280/286 360/369 377/377 402/402 Tonda Romana 5.001 355/367 197/199 235/238 280/286 366/369 383/386 396/402 Wepster LB23.04SM 355/379 199/199 226/244 286/286 360/369 383/395 393/402 Wepster TC 355/379 199/199 226/244 286/286 360/369 383/395 393/402 Willamette 500.001 355/373 197/199 235/244 280/280 369/369 383/386 393/396 Yamhill 738.001 355/367 199/199 235/250 280/286 369/369 383/386 393/402 Yamhill 898.001 355/367 199/199 235/250 280/286 369/369 383/386 393/402 York LB22.04SM 355/379 197/199 226/247 280/280 366/369 386/395 396/396 Zeta 779.001 355/355 199/199 244/244 280/286 366/369 383/383 402/402 Zeta LB11.40SM 355/355 199/199 244/244 280/286 366/369 383/383 402/402 Zeta TC 355/355 199/199 244/244 280/286 366/369 383/383 402/402 aNumbers following the name indicate inventroy number of accession at the NCGR; SM indicates the OSU collection of Shawn Mehlenbacher; TC indicates tissue culture propagated plant.

104

Appendix D. Allele size ranges of six SSRs discarded and 14 SSRs used in the multiplex. Wide bin widths (> 1.4) are underlined; alleles that do not change according to the motif length are in bold.

SSRs Allele ID Minimum Maximum Bin Width Basepairs between Alleles Removed from the study 219 219.57 219.90 0.33 222 222.03 224.16 2.13 2.1 225 225.80 228.44 2.64 1.6 LG639 228 229.72 230.20 0.48 1.3 231 231.98 232.24 0.26 1.8 234 233.84 236.24 2.40 1.6 237 237.44 237.89 0.45 1.2 284 284.77 284.89 0.12 287 287.16 287.26 0.10 2.3 292 292.01 292.01 0.00 4.8 BR249 296 296.02 296.02 0.00 4.0 302 302.29 303.60 1.31 6.3 304 304.11 306.13 2.02 0.5 305 305.53 306.54 1.01 312 312.12 312.84 0.72 5.58 BR427 315 315.38 315.60 0.22 2.54 317 316.67 317.07 0.40 1.07 319 317.91 319.56 1.65 0.84 383 383.51 383.51 0.00 387 386.09 388.24 2.15 2.6 389 390.02 390.02 0.00 1.8 BR359 393 393.01 395.10 2.09 3.0 397 396.93 397.97 1.04 1.8 399 399.05 399.05 0.00 1.1

105

401 400.98 401.88 0.90 1.9 403 402.96 403.16 0.20 1.1 432 432.31 433.81 1.50 LG631 435 435.07 439.84 4.77 1.3 441 441.79 442.99 1.20 2.0 154 154.01 154.58 0.57 157 157.31 157.76 0.45 2.7 BR446 160 160.50 160.77 0.27 2.7 163 163.22 163.83 0.61 2.4 166 166.43 166.43 0.00 2.6 14 SSRs Multiplex 90 90.05 90.11 0.06 93 92.93 93.14 0.21 2.8 BR270 96 96.06 96.13 0.07 2.9 99 99.04 99.18 0.14 2.9 102 102.03 102.15 0.12 2.8 102 101.91 102.50 0.59 105 104.88 105.25 0.37 2.4 BR322 108 107.98 108.36 0.38 2.7 111 111.22 111.33 0.11 2.9 114 114.47 114.49 0.02 3.1 116 115.93 115.96 0.03 122 121.96 122.31 0.35 6.0 125 125.05 125.45 0.40 2.7 134 134.55 134.80 0.25 9.1 BR414 137 137.63 137.77 0.14 2.8 140 140.76 140.94 0.18 3.0 146 147.05 147.20 0.15 6.1 164 163.94 163.94 0.00 16.7

106

154 154.24 154.50 0.26 157 157.63 157.71 0.08 3.1 GB949 160 160.56 160.85 0.29 2.8 163 163.79 163.96 0.17 2.9 166 166.93 167.20 0.27 3.0 162 162.18 162.45 0.27 165 165.22 165.58 0.36 2.8 GB950 168 168.41 168.72 0.31 2.8 171 171.56 171.85 0.29 2.8 178 178.17 178.17 0.00 6.3 187 187.36 187.44 0.08 197 197.28 197.70 0.42 9.8 199 199.23 199.75 0.52 1.5 BR438 201 201.42 201.72 0.30 1.7 203 203.72 204.57 0.85 2.0 211 211.60 211.66 0.06 7.0 200 200.78 201.24 0.46 203 203.67 203.67 0.00 2.4 209 209.37 209.76 0.39 5.7 212 212.21 212.32 0.11 2.5 CAC C008 218 217.93 218.38 0.45 5.6 221 220.88 221.06 0.18 2.5 241 241.37 241.86 0.49 20.3 244 244.44 244.54 0.10 2.6 253 253.17 253.17 0.00 8.6 226 226.01 226.27 0.26 229 229.06 229.21 0.15 2.8 BR259 232 232.28 232.28 0.00 3.1 235 234.96 235.29 0.33 2.7

107

238 237.86 238.11 0.25 2.6 241 240.96 241.18 0.22 2.8 244 244.01 244.23 0.22 2.8 247 247.02 247.18 0.16 2.8 250 249.95 250.11 0.16 2.8 274 274.27 274.49 0.22 280 280.02 280.56 0.54 5.5 BR464 283 283.22 283.22 0.00 2.7 286 285.97 286.38 0.41 2.8 292 291.93 292.34 0.41 5.6 330 330.89 330.90 0.01 334 334.23 334.24 0.01 3.3 336 336.73 337.09 0.36 2.5 340 340.07 340.18 0.11 3.0 GB875 343 342.99 343.14 0.15 2.8 346 345.85 346.09 0.24 2.7 349 348.80 349.02 0.22 2.7 352 352.62 352.62 0.00 3.6 355 354.62 355.69 1.07 2.0 355 355.36 355.79 0.43 367 367.11 367.61 0.50 11.3 370 370.34 370.54 0.20 2.7 GB673 373 373.20 373.60 0.40 2.7 376 376.32 376.32 0.00 2.7 379 379.15 379.52 0.37 2.8 360 360.47 360.73 0.26 363 363.27 363.63 0.36 2.5 LG688 366 367.18 367.61 0.43 3.6 369 369.39 370.21 0.82 1.8

108

372 372.89 373.07 0.18 2.7 377 377.11 377.42 0.31 380 380.25 380.66 0.41 2.8 383 383.15 383.52 0.37 2.5 386 386.39 386.78 0.39 2.9 GB395 389 389.38 389.83 0.45 2.6 395 395.34 396.66 1.32 5.5 398 398.88 398.88 0.00 2.2 404 404.32 404.32 0.00 5.4 408 408.70 408.70 0.00 4.4 393 392.55 392.97 0.42 396 395.57 395.96 0.39 2.6 BR343 399 398.72 398.89 0.17 2.8 402 401.38 402.23 0.85 2.5 407 407.70 407.73 0.03 5.5

109

Appendix E. Electropherogram example for BR446.

OSU 26.072

110

Appendix F. Electropherograms of 14-SSR multiplex.

111

Appendix G. Fingerprints of hazelnut genotypes for six single-locus hazelnut SSRs discarded from the study.

LG639- LG639- BR249- BR249- BR427- BR427- BR359- BR359- Local number Accessions 1 2 1 2 1 2 1 2 17.001 Montebello 225 228 302 304 311 387 23.001 Gem 225 231 284 304 311 317 387 393 297.001 Nottingham 222 225 302 311 317 387 393 559.001 Grand Traverse 225 228 302 317 393 36.001 Barcelona 222 225 302 304 311 387 401 11.001 Ennis 222 228 302 311 317 387 393 705.001 Clark 222 225 302 311 317 397 401 16.001 Hall’s Giant 225 234 - - 311 317 393 54.001 Gasaway 222 234 302 304 317 387 403 104.001 NY 616 222 225 287 302 305 311 387 26.001 Brixnut 222 237 304 311 317 387 393 193.001 NY 398 225 234 287 304 305 317 387 393 116.001 Butler 225 228 302 304 311 317 393 401 776.001 Gamma 234 302 304 311 397 403 42.001 Daviana 225 302 317 393 232.001 Du Chilly 231 237 284 304 311 317 393 633.001 Lewis 222 234 302 304 - - 832.001 Purple Aveline 225 234 304 311 387 738.001 Yamhill 225 234 304 311 317 387 779.001 Zeta 234 237 304 311 317 387 397 894.001 Jefferson 225 234 304 311 387 898.001 Yamhill 225 237 302 304 311 317 387 890.001 Dorris 225 228 302 304 311 393 397 165.001 Dundee 225 302 304 311 317 387 401 500.001 Willamette 228 234 302 311 317 387 401

112

859.001 Sacajawea 234 237 304 311 319 387 779.001 Zeta TC 228 234 302 304 311 387 LB23.04-SM Wepster TC 225 237 302 304 311 387 393 38.001 Pellicule Rouge 225 228 302 311 393 397 41.001 Cosford 219 228 304 317 387 393 262.001 Bergeri 222 237 304 311 317 387 393 625.001 Albania55 234 234 302 - - 50.001 Contorta 225 234 302 311 387 484.001 Imperial de Trebizonde 225 234 302 304 311 393 397 22.001 Tonda di Giffoni 225 228 302 311 317 387 397 343.002 Simon 228 237 302 319 397 401 268.007 Mortarella 234 237 302 304 311 317 387 31.001 Tonda Gentile delle Langhe 225 234 302 304 311 387 5.001 Tonda Romana 222 225 302 311 317 387 393 750.001 Tapparona 225 234 302 304 311 317 387 393 707.001 OSU 556.011 225 228 302 311 319 387 401 8.001 Negret 234 237 302 304 311 319 387 397 481.001 Anglais 234 237 302 311 387 393 28.001 Casina 222 228 302 311 317 397 265.002 Palaz 225 234 302 304 311 387 55.001 Tombul Ghiaghli 228 237 302 311 387 397 126.001 Aurea 231 234 302 304 317 387 113.001 Tonda Gentile delle Langhe 234 234 302 311 387 187.001 Ala Kieri 222 237 302 304 311 317 387 399 81.001 Imperial de Trebizonde 234 237 302 304 311 393 397 249.001 Sant Jaume 219 225 302 311 317 387 256.003 Artellet 225 234 304 311 317 387 401 344.001 Ratoli 228 237 302 311 387 397 323.001 Kadetten Zellernuss 225 228 302 317 387 110.001 Tonda Gentile delle Langhe 231 234 302 304 311 387

113

458.001 Payrone 228 237 302 311 317 387 18.001 Cutleaf 225 237 302 304 317 387 114.001 Tonda Gentile dele Langhe 228 234 304 311 387 333.001 Barr’s Zellernuss 225 234 304 304 311 317 387 393 427.001 OSU495.045 222 234 302 311 319 383 393 206.001 Gustav’s Zellernuss 228 231 302 304 317 387 393 21.003 Tonda Bianca 228 237 302 311 317 387 393 368.001 Iannusa Racinante 219 237 302 304 311 317 387 382.001 Gunslebener Zellernuss 228 237 302 311 317 387 393 273.001 B 3 225 234 302 311 317 387 88.001 OSU 54.039 225 225 292 304 311 387 389 13.001 Rode Zeller 234 302 304 317 387 393 388.001 OSU 26.072 225 234 304 311 317 387 401 999.001 Alli 222 234 302 304 317 387 331.001 Barcelloner Zellernuss 222 228 284 304 - - 894.001 Jefferson TC 234 237 302 304 311 317 387 397 890.001 Dorris TC 225 234 284 304 311 387 LB07.34-SM OSU 408.040 237 304 317 319 387 LB11.35-SM Epsilon 222 304 311 387 LB14.09-SM OSU 681.078 234 304 305 317 387 LB15.01-SM Red Majestic #2 234 237 302 311 317 387 LB15.02-SM Red Majestic #1 228 234 302 317 387 LB20.20-SM Red Dragon 228 234 302 304 317 387 LB22.04-SM York 234 302 304 311 317 387 397 LB22.07-SM McDonald 234 237 304 311 317 387 LB23.04-SM Wepster 222 234 302 304 311 387 393 LB24.11-SM Felix 234 237 296 304 317 387 LB26.20-SM Eta 228 234 304 317 387 LB27.11-SM Theta 225 228 302 304 311 317 387 403 LB07.25a-SM Newberg 231 302 314 393

114

MP03.07-SM OSU 414.062 228 302 311 387 403 MP03.20-SM OSU 252.146 234 287 302 311 317 387 397 USDA Shop-SM Pendula 222 225 302 317 387 V04-SM Tonda Pacifica 231 234 302 311 387 372.007-SM Fusco Rubra 231 284 302 317 387 955.040 Georgian 759.010 234 237 302 311 314 387 403 955.070-SM Burgundy Lace 222 225 302 304 317 387 403 165.001 Dundee TC 234 302 304 314 393 LB22.04-SM York TC - - - - 311 317 387 397 36.001 Barcelona TC - - - - 311 387 401 LB24.11-SM Felix TC - - - - 317 387

Appendix G continued LG631- LG631- BR446- BR446- BR446- BR446- Local number Accessions 1 2 1 2 3 4 17.001 Montebello 432 154 157 23.001 Gem 432 154 157 297.001 Nottingham 432 435 154 157 559.001 Grand Traverse 432 154 36.001 Barcelona 432 435 154 157 11.001 Ennis 435 154 157 705.001 Clark 432 151 154 163 166 16.001 Hall’s Giant 432 435 157 163 54.001 Gasaway 435 154 104.001 NY 616 435 440 154 26.001 Brixnut 435 435 157 163 193.001 NY 398 435 440 154 157 116.001 Butler 432 435 154 157 163 166 776.001 Gamma 435 154

115

42.001 Daviana 432 435 154 157 163 166 232.001 Du Chilly 432 435 157 163 633.001 Lewis 435 - - - - 832.001 Purple Aveline 432 435 154 157 160 163 738.001 Yamhill 432 435 154 779.001 Zeta 435 154 157 894.001 Jefferson 432 435 157 898.001 Yamhill 432 435 154 890.001 Dorris 432 435 157 163 165.001 Dundee 432 154 157 500.001 Willamette 432 435 154 157 163 166 859.001 Sacajawea 435 154 157 779.001 Zeta TC 435 157 LB23.04-SM Wepster TC 432 435 154 157 160 163 38.001 Pellicule Rouge 432 435 157 163 41.001 Cosford 432 154 157 163 166 262.001 Bergeri 435 157 160 625.001 Albania55 435 435 - - - - 50.001 Contorta 432 435 154 157 160 163 484.001 Imperial de Trebizonde 432 157 163 22.001 Tonda di Giffoni 432 157 163 343.002 Simon 435 435 154 157 268.007 Mortarella 432 435 154 157 31.001 Tonda Gentile delle Langhe 432 154 157 5.001 Tonda Romana 432 435 154 157 750.001 Tapparona 432 157 160 707.001 OSU 556.011 432 154 8.001 Negret 435 435 154 157 481.001 Anglais 432 435 154 157 28.001 Casina 432 435 154 157

116

265.002 Palaz 432 435 157 163 55.001 Tombul Ghiaghli 435 435 154 157 160 163 126.001 Aurea 432 435 154 113.001 Tonda Gentile delle Langhe 435 435 154 157 187.001 Ala Kieri 432 435 154 157 160 163 81.001 Imperial de Trebizonde 435 157 163 249.001 Sant Jaume 432 154 157 256.003 Artellet 432 157 344.001 Ratoli 435 435 157 323.001 Kadetten Zellernuss 432 154 110.001 Tonda Gentile delle Langhe 432 435 154 157 458.001 Payrone 435 435 157 160 18.001 Cutleaf 435 435 154 157 160 163 114.001 Tonda Gentile dele Langhe 435 154 157 333.001 Barr’s Zellernuss 432 435 154 157 160 163 427.001 OSU495.045 435 154 157 206.001 Gustav’s Zellernuss 432 435 154 157 21.003 Tonda Bianca 432 435 154 157 160 163 368.001 Iannusa Racinante 435 154 157 382.001 Gunslebener Zellernuss 432 435 154 273.001 B 3 432 154 157 88.001 OSU 54.039 432 435 154 157 13.001 Rode Zeller 432 435 157 163 388.001 OSU 26.072 432 435 154 157 163 166 999.001 Alli 435 435 154 157 160 163 331.001 Barcelloner Zellernuss 432 435 - - - - 894.001 Jefferson TC 435 154 157 890.001 Dorris TC 432 435 154 157 160 163 LB07.34-SM OSU 408.040 435 154 157 160 LB11.35-SM Epsilon 435 154 157

117

LB14.09-SM OSU 681.078 432 435 154 157 160 163 LB15.01-SM Red Majestic #2 435 163 LB15.02-SM Red Majestic #1 435 163 LB20.20-SM Red Dragon 435 154 157 LB22.04-SM York 435 154 163 LB22.07-SM McDonald 435 163 LB23.04-SM Wepster 435 154 157 160 163 LB24.11-SM Felix 435 154 157 163 166 LB26.20-SM Eta 435 435 154 157 160 163 LB27.11-SM Theta 432 435 151 154 163 166 LB07.25a-SM Newberg 432 435 154 157 160 163 MP03.07-SM OSU 414.062 435 154 157 MP03.20-SM OSU 252.146 432 435 154 157 USDA Shop-SM Pendula 432 435 154 V04-SM Tonda Pacifica 435 154 157 160 163 372.007-SM Fusco Rubra 432 154 955.040 Georgian 759.010 435 157 160 163 955.070-SM Burgundy Lace 432 435 154 157 160 163 165.001 Dundee TC 435 154 157 160 163 LB22.04-SM York TC - - 151 154 163 166 36.001 Barcelona TC - - 154 157 LB24.11-SM Felix TC - - 154 157 163 166 a’-‘ indicates missing data.