Relationship between taste perception, density and soluble solids concentration in kiwifruit (Actinidia deliciosa)

Transcription

1 New Zealand Journal of Crop and Horticultural Science ISSN: (Print) (Online) Journal homepage: Relationship between taste perception, density and soluble solids concentration in kiwifruit (Actinidia deliciosa) ROBERT B. JORDAN & RICHARD J. SEELYE To cite this article: ROBERT B. JORDAN & RICHARD J. SEELYE (2009) Relationship between taste perception, density and soluble solids concentration in kiwifruit (Actinidia deliciosa), New Zealand Journal of Crop and Horticultural Science, 37:4, , DOI: / To link to this article: Published online: 01 Aug Submit your article to this journal Article views: 279 View related articles Citing articles: 5 View citing articles Full Terms & Conditions of access and use can be found at Download by: [ ] Date: 10 December 2017, At: 16:04

2 Jordan New Zealand & Seelye Taste Journal of relationships Crop and Horticultural of kiwifruitscience, 2009, Vol. 37: (Online); (Print)/09/ The Royal Society of New Zealand 2009 Relationship between taste perception, density and soluble solids concentration in kiwifruit (Actinidia deliciosa) Robert B. Jordan Richard J. Seelye Biosensors and Biomeasurement Plant and Food Research Private Bag 3123 Hamilton 3240, New Zealand Abstract The relationship between taste, and the measurable properties dry matter (DM), ripened soluble solids concentration (rssc), and fruit density of kiwifruit (Actinidia deliciosa var. deliciosa Hayward ) is briefly reviewed. The results of sensory trials involving 480 untrained consumer volunteers and 36 trained descriptive sensory panellists undertaken in 1994 and 1995 using Plant & Food Research s purpose-built sensory facility are presented. These show strong evidence that kiwifruit with higher fruit density or higher rssc (and by inference higher DM) are liked more by consumers regardless of where in New Zealand the fruit were sourced, harvest time, fruit size, curing time, storage duration, or season. The basis for the difference is that higher rssc fruit taste sweeter. Using DM (which is highly correlated with rssc) as a taste indicator has the major advantage that DM remains essentially constant from harvest time to eating time, making it a robust measure throughout the fruit life. Keywords kiwifruit; density; ripe soluble solids concentration; sensory; non-destructive testing; dry matter H09032; Online publication date 5 November 2009 Received 25 May 2009; accepted 21 September 2009 Introduction There has been debate within New Zealand over the last two decades whether measures of the ripened soluble solids concentration (rssc) of green kiwifruit (Actinidia deliciosa (A. Chev.) C.F. Liang et A.R. Ferguson var. deliciosa Hayward ) had any bearing on taste. This paper provides: (1) a brief review of the evidence for and against that relationship, and (2) results from commercial research in the mid 1990s showing a strong correlation between fruit taste perception and both harvest time fruit density and rssc. Because rssc is strongly correlated (Jordan et al. 1997) with both harvest and eating time dry matter (DM), there is a concomitant correlation between taste perception and DM. This relationship has now become accepted, and harvest DM has become a major component of the payment system used by the New Zealand Marketers ZESPRI. Predicting kiwifruit taste Understanding the taste and sensory aspects of kiwifruit is complex and the literature abounds with reports of factors that influence the senses. DM, rssc, sugar and acid concentrations, ph, oxalate levels, volatiles, and many textural properties all have strong sensory influences. Although fruit acidity, firmness, and volatile levels are clearly important, it is difficult to predict at harvest, months in advance, what these variables will be when the fruit are eaten. The studies here do not include fruit acid or volatile measurements, although acid perceptions of consumers are recorded. The firmness of fruit presented to consumers is controlled to minimise its affect. Some of these variables are considered in more depth in studies completed subsequent to this work (Harker et al. 2009; Rossiter et al. 2000; Crisosto & Crisosto 2001). A direct relationship between harvest-density and rssc has been recognised in Japan for some time (Ninomiya et al. 1991; Asami et al. 1988, 1996). Harvest DM is highly correlated with rssc (Jordan et al. 1997) and can be reliably estimated non-destructively at harvest using either density

3 304 New Zealand Journal of Crop and Horticultural Science, 2009, Vol. 37 (Ninomiya et al. 1991; Jordan et al. 2000) or near infrared (NIR) (Slaughter & Crisosto 1996; Jordan et al. 1997; McGlone et al. 2002). DM is thus a strong taste indicator, and is probably the single most important harvest-time property of kiwifruit in the consumer taste experience. Kiwifruit quality properties (e.g., DM) are extremely variable (Smith et al. 1997). Although knowledge of the average values taken from representative samples within single maturity blocks helps to pre-sort fruit lines for quality, the large variations found on the vine, and the lack of external visual clues means that each fruit must be individually tested to guarantee taste. During the 1980s and early 1990s, a number of studies relating taste to measurable kiwifruit properties had appeared in the literature, but none offered definitive evidence, and often showed divergent results. Harman (1981) reported a small study showing a trend for higher rssc kiwifruit to have higher taste acceptability. Crisosto et al. (1984) found that sensory sweetness corresponded well with rssc. Scott et al. (1986) found a highly significant relationship between flavour quality and both rssc and DM, and noted that for flavour prediction it mattered little whether DM was measured at harvest or when fruit were ripe. Stec et al. (1989) proposed that firmness had a major influence on many taste parameters including acceptability, sweetness, and acidity as assessed in both consumer and trained panel studies and that caution was needed in interpreting kiwifruit sensory data. Although Stec et al. concluded that rssc never correlated with sensory panel sweetness or tanginess, they noted that kiwifruit were more acceptable from treatments where the fruit had rssc exceeding 12%. MacRae et al. (1989) showed a strong correlation between rssc and consumer acceptability, but the strong correlation between rssc and harvest date made cause and effect conclusions difficult. Gorini et al. (1990) and Pailly et al. (1990) both noted that high levels of rssc were needed for high taste scores. By contrast, MacRae et al. (1990) found that perceived sweetness showed no relationship to sugar content, rssc, or sugar-acid ratios of kiwifruit. Paterson et al. (1991) also showed that apparent sweetness did not correlate with rssc. Mitchell et al. (1991) reported a limited taste trial in which 94% of the panellists considered fruit with rssc above 14% as acceptable, whereas 46% considered fruit with rssc lower than 13% unacceptable. McMath et al. (1991) found that rssc was a significant component in their model of taste scores, particularly when lower than about 11%. Rossiter et al. (2000) showed an increase in overall acceptability with rssc levels in artificially modified kiwifruit pulps, particularly as rssc rose above 11%. Crisosto & Crisosto (2001) showed an increase in taste parameters with rssc of early harvest kiwifruit. They proposed a minimum rssc of 12.5% for this early-marketed fruit although the major loss of taste appeared in fruit below 11.6%. Other research undertaken at Plant & Food Research over the last 10 years, mostly unpublished until recently (Harker et al. 2009), confirms that the majority of consumers respond positively to increasing DM. Density as a taste segregator Early in the 1990s, Richard Seelye and the late Bruce Stowell of the Te Puke Research Centre suggested that the density of kiwifruit at harvest was a strong indicator of the taste of the fruit, and thus density might provide a practical means at harvest time of differentiating fruit for ultimate eating quality. Small-scale sensory trials they completed showed real differences in acceptability between low- and high-density fruit (M. Stec, Plant & Food Research pers. comm.). These results were reported widely within the New Zealand kiwifruit industry (Jordan et al. 1994). The trials discussed in this paper were initiated to specifically ask the question Do denser kiwifruit taste nicer? As well as investigating the underlying reasons for the relationship, the trials were designed to determine whether or not the relationship held across all kiwifruit or was just a single season, orchard, pack-house, region, fruit-size, storageduration, or harvest-date phenomenon. Until now, for commercial sensitivity reasons, these studies have not been presented in the general literature. Asami et al. (1988) and Ninomiya et al. (1991) had demonstrated relationships between kiwifruit density and either DM or rssc, although unfortunately these Japanese language papers have not been widely circulated. Asami et al. (1996) reported on studies where density was used to separate fruit into rssc groups and Richardson et al. (1997) reported strong correlations between density and both DM and rssc over a wide fruit maturity range. Richardson et al. (1997) also discussed the mechanism for the relationships by showing strong relationships between density and both DM and the sum of the soluble solids and starch concentrations (the major components of kiwifruit DM) as fruit developed. From this they derived equations to predict rssc

4 Jordan & Seelye Taste relationships of kiwifruit from the density of fruit harvested over an extended season. Jordan et al. (2000) further investigated the relationship between density and the sum of starch and soluble solid concentrations using pairs of kiwifruit at different points in their ripening cycle. One member of each pair was analysed immediately and the other after ripening, which showed that total carbohydrates and DM each stayed essentially constant over ripening. Jordan et al. (2000) provide further support for the closeness of the density to rssc and DM relationships (R 2 values of 83% and 85%, respectively) and McGlone et al. (2002) report similar or higher R 2 values. There is a scientific rationale for the rssc to harvest-density relationship. Kiwifruit are c. 83% water, 12% sugars and starch, and perhaps 2% acid (Lintas et al. 1991). Most of the remaining 3% is structural material, leaving a long list of minor components, which do not collectively contribute significantly to fruit mass. At typical harvest times, sugar and starch are about equal in proportions, but as fruit ripen the starch converts to sugar (Given 1993), reaching zero at eating time. The density of fruit sugars, starch, and acids are about the same (Lide 1998) and contribute similarly to the overall fruit density. Harvest density is thus totally dominated by the concentration of the three major non-water components of the fruit: sugars, starch, and acids, potentially making it a good measure of the total concentration of these components. Because the three are not significantly converted into components other than sugar or acids (or to components with very different densities), harvestdensity also provides an estimate of rssc (mainly sugar plus acids). We note that starch hydrolysis will use water in the fruit equivalent to 11.1% of the starch mass on creating sugar 9 g of starch plus 1 g of water produces 10 g of sugar. This creates a slight elevation of the DM content as fruit ripen. However, this effect is small unless the kiwifruit are harvested early in the season when starch may exceed 8% of fruit mass. The presence of gases in the fruit provides a slight complication. Undissolved gases contribute to fruit volume but not significantly to fruit mass, thus reducing overall fruit density. Extremely low DM fruit may be observed to float in water, indicating the presence of materials with densities less than that of water. Hallet et al. (1992) provide estimates of the size of this gas pool (c. 3 5%), but a better understanding is required of fruit gas concentration, particularly of its variability, and the changes occurring throughout storage. 305 The major advantages of using density as a fruit quality selection criterion is that it can be performed non-destructively and can be scaled to operate at high speed using flotation tanks whose density has been adjusted by adding edible salts. Fruit that sink in these tanks have higher density and potentially higher rssc than those that float. Alternatively, the strong relationship between fruit density and terminal velocity as fruit falls in water has potential as a fruit separation mechanism (Jordan & Clark 2004; Jordan et al. 2009); these techniques may allow large quantities of fruit to be continuously sorted for their potential eating properties. Materials and Methods Trials were conducted over two consecutive years. In 1994, a sensory investigation of consumer liking and sensory preference for density-separated fruit using an untrained consumer panel was undertaken with fruit from a single region (Te Puke). This investigation was across a wide range of management and fruit characteristic regimes, including orchard, orchard altitude, fruit size, curing time (see below), and harvest date. In 1995, a similar untrained consumer panel study was conducted, using fruit collected across five regions of New Zealand and stored for different periods to investigate location and storage effects. A third study was undertaken in 1995 using a trained descriptive sensory panel to gain provisional estimates of the magnitudes of a number of sensory attributes (sweetness, acidity) and to show how they varied across density-differentiated fruit. An organisation map of the three trials is provided in Fig. 1. Throughout the remainder of this paper, the term penetrometer denotes physical firmness measurements, to distinguish it from sensory firmness assessed by panellists. We have also hyphenated the names of variables used in the models and assessments (e.g., overall-liking) consumer panel fruit To ensure fruit came from a wide range of sources within the Te Puke region, three orchards were chosen at each of three altitudes, and fruit were collected from each on three occasions to represent early, mid, and late harvests. Export quality fruit were hand picked from four or five randomlyselected, non-adjacent, healthy vines or pergola bays away from shelter belts and row ends within each orchard. Fruit were placed directly into trays

5 306 New Zealand Journal of Crop and Horticultural Science, 2009, Vol. 37 containing appropriately sized plastic inserts to allow pickers to select two trays of 25 fruit from each of three target fruit sizes (industry size 30, 36, and 42; c. 122 g, 103 g, and 84 g, respectively) without weighing. After picking, fruit were transported to the Te Puke Research Centre where they were cured (Pennycook & Manning 1992) by holding the fruit at air temperature for one of three durations (0, 3, or 7 days). To minimise fruit numbers, one random curing-time was assigned to each tray collected, ensuring that the combinations used were balanced over all variables in the trial (Fig. 1). After curing, the 25 fruit from each tray were density sorted using the techniques described below to give four fruit from each of three density groups with boundaries as in Table 1. Using fixed density boundaries for all fruit caused some difficulties in the 1994 study with insufficient low- or high-density fruit being available from some sets. As a result, a small proportion of the fruit presented to panellists was from other groups, almost always from the same orchard, and always having the correct density. This allowed for two replicate sensory evaluations of fruit of three densities in a reduced set of three sizes by three curing-time combinations, over three harvest-dates out of three orchards from each of three altitudes in the Te Puke area (i.e., 2*3*(3*3/3)*3*3*3 = 486 fruit assessed) consumer panel fruit In 1995, fruit were sourced from across New Zealand, and underwent sensory evaluation after two storage durations. Rather than being hand picked, fruit were collected from export pack-houses after normal commercial grading and packing. Four orchards in each of the five major growing regions (Kerikeri, Katikati, Te Puke, Opotiki, and Nelson), were selected to include two average, one poor, and one excellent orchard relating to their storage and export quality history. In this study, although fruit are identified from named regions, the small numbers of orchards used in each instance and that they were not selected randomly, prevents definitive conclusions being drawn about relative regional fruit properties. Twenty trays of export fruit in standard packaging (industry size 36, c. 103 g per fruit) from each orchard were transported to the Te Puke Research Station for density grading as described below. The density groups used in 1995 were reduced in separation slightly from 1994 (1.25% cf. 2%), and fruit from lower or higher latitudes used adjusted density thresholds to ensure sufficient fruit were available (see Table 1). These changes were felt to have minimal effect, particularly because panellists did not generally assess fruit from more than one orchard. Fruit were then stored for 8 or 16 weeks before sensory analysis. This gave eight replicate sensory evaluations of fruit of three densities from each of four orchards in five regions stored for two different storage-durations (i.e., 8*3*4*5*2 = 960 fruit assessed) descriptive panel fruit Fruit for the trained descriptive sensory panel in 1995 were collected in parallel to those for the 1995 consumer panel, although only three of the four orchards (judged as good, average, and poor in quality) from each of two (Nelson and Te Puke) of the five regions were used. Ten trays of fruit (industry size 36, c. 103 g per fruit) were collected from each orchard and were processed as for the consumer panels. This gave 12 replicate sensory evaluations of fruit of three densities from each of three orchards in two regions (i.e., 12*3*3*2 = 216 fruit assessed). Fruit preparation procedures Density sorting was undertaken shortly after harvest when fruit were very firm and was performed manually to minimise damage. Up to 100 fruit were placed, one fruit group at a time, into wire baskets and were lowered into each of two 700-litre Table 1 Target density thresholds (kg/m 3 ) and corresponding estimates of rssc (%) (ripened soluble solids concentration) for 1994 and L low High Estimated Estimated rssc Year Region density density low rssc high rssc separation 1994 Te Puke Nelson Te Puke, Opotiki, Katikati Kerikeri

6 Jordan & Seelye Taste relationships of kiwifruit 307 Fig. 1 Organisation chart for the three trials showing quantities of fruit in each category and the way they are distributed between the geographical variables of orchard, altitudes, and region, and the fruit properties of harvest date, size, and curing time. Larger quantities of fruit were collected to allow sufficient fruit of the appropriate density. In turn, larger quantities of fruit were stored to allow selection of fruit with matched penetrometer values for sensory evaluation.

7 308 New Zealand Journal of Crop and Horticultural Science, 2009, Vol. 37 plastic tanks containing water and technical grade magnesium sulfate made up to the target densities in Table 1. This food compound was chosen for its solubility, cost, and ease of disposal. Care was taken to remove surface air bubbles from each fruit to minimise false density indications. Fruit that floated in the low-density tank were identified as low-density fruit. Fruit that sank in this tank were then lowered into the second tank to separate medium and high-density fruit. In 1994, four fruit were selected at random from each density-group category and after drying were placed in labelled trays. In 1995, one tray of 36 fruit in each density/storage-duration combination was recovered. The 1995 fruit were also water sprayed to reduce residual salt on the fruit, before drying and being placed in labelled trays. Fruit were then stored at 0 C at the Te Puke Research Station until the week before sensory evaluation when they were removed from storage to ripen. Sensory evaluations were undertaken in mid August 1994 and in August or October The trained descriptive evaluations were undertaken in late August To ripen the fruit, the closed trays were placed separately on racks in a warm room (25 C) and were inspected daily by hand for firmness. As each fruit reached a point slightly firmer than desired (6 to 8 N as estimated using a penetrometer on spare fruit) they were moved to a duplicate set of storage trays with identical labels and placed in cool storage. In 1994, this procedure yielded fruit that were slightly below optimal firmness, particularly in some groups. With the experience gained in 1994, the 1995 fruit were closer to optimal firmness. Fruit were then transported to the Mt Albert Research Centre for sensory evaluation. Consumer sensory panel procedures Participants for the untrained consumer sensory panels were male and female volunteers with a liking for kiwifruit covering a range of ages, occupations, and ethnicities. Panellist groups were usually arranged through community organisations such as schools, sports teams, and the like who were asked to provide up to 10 volunteers in reward for payments made directly to their organisations. No previous experience with consumer sensory panels was required: panellists were given a short group introduction to the terminology and the methods used just before the sensory sessions. All assessments took place in a purpose-built laboratory at the Plant & Food Research Sensory Science Facility at Mt Albert in which lighting, temperature, and humidity were controlled. In 1994, the trial was designed for 162 consumer panellists, whereas in 1995, 160 panellists were proposed for each of the two storage-duration assessments (8 and 16 weeks). Each panellist was presented with fruit from each of the three densitygroups in random order, all (where possible) from a single orchard, harvest-date, and curing-time. The fruit within each triplet were closely matched for physical firmness by taking penetrometer measurements on all fruit available and selecting sets of three within each treatment across the three density categories that matched typically to better than 0.3 N within the target range 6 8 N. This produced for each panellist a low-, a medium-, and a high-density fruit with similar penetrometer values. Each fruit to be assessed had the 15 mm caps removed from stem and blossom ends to determine the rssc by squeezing their juice onto a digital refractometer (Atago, Japan) and averaging the two readings. Physical fruit acid measurements were not made in these studies. Panellists were not assessing appearance, and had been warned that the fruit would be presented cut in the shape of a barrel with penetrometer holes in the sides. All three fruit were presented at one time to panellists, who were advised of the order (a predetermined randomisation designed for balance across treatments) to evaluate the fruit to eliminate effects of sensory fatigue and carryover effects. Panellists were provided with a sharp knife and a teaspoon, and with water crackers and water purified by reverse osmosis to cleanse the palate between samples. The consumer panellists were asked to report on: their degree-of-liking of the flavour, texture, and the overall-liking (dislike very much to like very much); the strength (very weak to very strong) of the sweetness, acidity, and (in 1995 only) the characteristic-kiwifruit-flavour; the firmness (very soft to very hard being sensory rather than physical firmness); the juiciness (not juicy to very juicy) of each kiwifruit. Scores were recorded as pencil marks at a position along a 150 mm unstructured line with anchors at each end denoted with phrases such as dislike very much and like very much. The code number of the fruit being assessed was written by the pencil mark. Descriptive sensory panel procedures The descriptive sensory panel evaluations were performed using trained panellists to further quantify the density/sensory relationship by studying the fundamental sensory factors across the three densities. The trained descriptive panel did not enumerate their

8 Jordan & Seelye Taste relationships of kiwifruit liking of the fruit, but were instructed to score the intensities of the overall-flavour, the characteristickiwifruit-flavour, sweetness, acidity, juiciness, firmness, and various aromas (not reported here). The descriptive panel was selected during August 1995 from members of the public and from the research facility. It comprised 36 male and female volunteers with a liking for kiwifruit covering a range of ages, occupations, and ethnicities. Panellists were trained for 3 days before the 2 test-day sessions where they assessed fruit from the two regions. Training comprised group discussions and sensory evaluations using small quantities of fruit put aside from the consumer studies, and familiarisation with the automated data capture systems in use. Panellists were presented with fruit one at a time in randomised order with the stem upwards, and were given a prescribed time interval between each assessment for palate cleansing. As before, penetrometer and rssc measurements were made on the fruit before the sensory evaluation. The stem ends of the fruit were to be evaluated for aroma attributes (not reported here), the middle sections for flavour, and the blossom ends for texture. For the assessment of juiciness and firmness, panellists were instructed to evaluate the flesh only. All results were recorded in computer-equipped sensory booths using Plant & Food Research s proprietary computerised sensory evaluation data capture system. Data analysis Statistical analysis of each of the three sets of sensory scores (1994 and 1995 consumer, and 1995 descriptive panels) were completed by modelling separately the data using a linear mixed model (REML) within the GenStat 5 statistical package (Payne 1994). Because each panellist assessed 3 fruit (one of each density but of similar firmness) from a single orchard, the mixed model was specified with orchard and panellist as random effects. The 1994 consumer panel data were modelled for each of the panellist assessments (overall-liking, flavour, sweetness, acidity, firmness, juiciness, and texture). The effects of density-group (low, medium, or high), orchard-altitude (low, mid, or high) and harvest-date (early, normal, or late) and their interactions, fruit-size, curing-time (0, 3, or 7 days), and age and gender of the panellists were estimated. Significant effects are reported. The 1995 consumer panel data were modelled for each of the panellist assessments (overall-liking, flavour, characteristic-kiwifruit-flavour, sweetness, acidity, and firmness). The effects of density-group 309 (low, medium, or high), storage-duration (8 or 16 weeks), growing region and their interactions, and the order-of-evaluation were estimated. Significant effects are reported. The 1995 descriptive panel data were modelled for each of the panellist assessments (overall-flavour, characteristic-kiwifruit-flavour, sweetness, acidity, juiciness, and firmness). The effects of density-group (low, medium, or high), growing region, and orderof-evaluation were estimated. Significant effects are reported. To investigate the source of the sensory effect, a more extensive modelling study was undertaken on the 1995 data sets. This involved modelling the six sensory responses from the panels using linear mixed models of varying complexity. The original model was first extended by adding rssc to the mix of densitygroup, region, and order-of-evaluation, giving the most complex model. Next, the simplest model, rssc alone as an independent variable, was tried to see if this removed the need for a regional component. The final model used only rssc and order-of-evaluation to see if this further reduced residual SD. Results Fruit sorting efficiency The outcomes of this trial depended on our ability to sort the fruit into meaningful groups by their density. With only two density thresholds to separate the three groups it is not possible to eliminate overlap between sorted fruit densities. This creates a small probability that panellists will receive fruit from different density groups than intended. In the distribution histograms of the fruit for the 1995 consumer panellists, there is considerable rssc overlap between adjacent density groups (Fig. 2). The overlap characteristics were estimated by fitting inverted cumulative normal distributions to the sorting data and gave thresholds in 1995 of 11.6% for the lower and 13.0% for the upper threshold (cf. targets of 12.5% and 13.75% respectively). The SD of these distributions was 0.88% which quantifies the sorting error. The 1994 data had a slightly higher sorting SD (0.95%) but the thresholds (11.7% and 13.0%) were closer to target (12.0% and 14.0%). Note that a nominal relationship between density and rssc was used to set sorting tank density levels, and density sorting took place 2 5 months before rssc was measured. Thus the above SDs incorporate errors from the original density sorting, from the model that density sorts fruit for rssc, from the small differences in thresholds used

9 310 New Zealand Journal of Crop and Horticultural Science, 2009, Vol. 37 for different regions, as well as from the variability expected in the change of internal properties of the fruit over their storage-duration untrained consumer panel The 1994 consumer panel addressed the question: do denser kiwifruit taste nicer independent of orchard, altitude, fruit size, curing time, or harvesttime. Panels comprised 126 females and 26 males with age groupings of 29 (<25 years), 57 (25 39), 44 (40 54), and 22 (>55 years). Ten sample sets were removed where evaluations were deemed to be incomplete or invalid. Six of the seven sensory attributes evaluated by the panel varied significantly between densitygroups, averaged over the five controlled variables (Fig. 3). Overall-liking, flavour, sweetness, acidity, firmness, and texture of low-density fruit all have significantly lower average scores (P < 0.05) than medium and high-density fruit (as indicated in Fig. 3 by non-overlapping error bars). Medium and highdensity fruit are also significantly different (P < 0.05) for flavour and texture, but these differences are not as large as those for low to medium groups. The density sorting process has produced mean rssc values separated by c. 1.5% between adjacent groups (Fig. 3) that are significantly different (P < 0.001). Mean penetrometer values for each of the groups are lower than the target (6 N), and are significantly different from each other, reflecting the difficulty matching firmness in This may have had an effect on the sensory scores of some fruit. Using the REML model, the panel assessment of overall-liking, flavour, and sweetness all showed Fig. 2 Distribution histograms for rssc (ripened soluble solids concentration) across the three density-groups for the 1995 consumer panel. Overlaid to show the extent of the overlap between groups are the efficiencies of the original sorting procedures as gauged by the percentage of fruit that is correctly placed either side of the fitted rssc thresholds of 11.6% and 13%. Diamonds represent the calculated values and the smooth lines are cumulative inverted normal distribution fits to this data (µ 1 =11.6, σ 1 =0.88, µ 2 =13.0, σ 2 =0.88). Fruit for the 1994 consumer and 1995 descriptive panels follow similar patterns. highly significant effects for density (P < 0.001). The Wald statistic table for overall-liking (Table 2) shows no significant effects for any of the controlled variables. Flavour and sweetness show similar patterns, although flavour shows an unexplained significance (P < 0.01) involving the interaction between fruit-size and curing-time. The REML modelling of the sensory evaluations of juiciness shows no significant differences between fruit of the different density groups, although it does show a small effect of harvest date (P < 0.05), which cannot be explained. The sensory evaluations acidity, firmness and texture all show significant effects of density (P < 0.001), perhaps partially owing to the poor quality of the penetrometer matching achieved. Three of these last four sensory attributes also showed unexplained and significant effects from other controlled variables. Acidity had significant fruit-size, gender, and altitude/harvest interactions (P < 0.05), firmness showed a highly significant gender effect (P < 0.001), and juiciness showed a harvest time effect (P < 0.05). When rssc concentrations are viewed against the different controlled variables there are no significant differences other than for density, which showed a highly significant separation (P < 0.001) into three bands corresponding to the three densities untrained consumer panel The 1995 untrained consumer panel performed a number of sensory evaluations on density-segregated fruit sourced from five regions and following 8 and 16 weeks of storage. Panels over the two evaluations comprised 221 female and 94 male participants with

10 Jordan & Seelye Taste relationships of kiwifruit 311 Fig. 3 Results from the 1994 consumer panel averaged over orchard, altitude, fruit-size, curing-time, and harvest-date. The graph at left shows the mean (±1 SED) scores for each attribute. Middle and right-hand plots show the mean (±1 SED) of the rssc and penetrometer values of the samples as presented to the panellists, respectively. age groupings of 81 (<25 years), 88 (25 39), 67 (40 54), and 79 (>55 years), considerably more balanced than in Five sample sets were removed as incomplete or invalid. The 1995 sensory scores are almost identical to those obtained in 1994 for overall-liking, flavour, characteristic-kiwifruit-flavour, and sweetness (Fig. 4), but with highly significant differences (P < 0.001) between all three density-groups. In the 1995 trial, fruit penetrometer values were better controlled (cf. right hand graphs in Fig. 3 and Fig. 4) and by contrast to 1994, the acidity and firmness sensory assessments are not significantly different. Mean rssc levels for the three density-groups were 11.28%, 12.48%, and 13.77%, respectively, and showed a linear trend with highly significantly differences (P < cf. middle graph in Fig. 3). In the results of the detailed REML analysis of the overall-liking scores from the 1995 panel (Table 3) density has a highly significant effect on panel scores. There is a significant region effect, with Kerikeri and Opotiki regions showing significantly higher overall-liking scores than Te Puke. There is also a significant interaction between density-group and storage-duration. This resulted from very low panel scores by low-density fruit that had been stored for 20 weeks, which may have a valid physiological basis. The REML analyses of the sensory assessments for flavour, characteristic-kiwifruit-flavour, and sweetness (full Wald statistic probability tables not Table 2 P values for fixed effects when sequentially adding terms to the model for sensory assessment of overall-liking in the 1994 untrained consumer panels. The one significant effect is asterisked. Fixed term added to the model P value Density <0.001*** Altitude 0.29 Harvest 0.41 Altitude Harvest 0.77 Size 0.80 Curing-time 0.39 Size Curing-time 0.15 Age 0.31 Gender 0.21 Age Gender 0.88 presented here) showed no other significant effects of the controlled variables other than the presence of significantly higher characteristic-kiwifruit-flavour scores (P < 0.05) from Kerikeri fruit compared with some other regions. Acidity scores did not differ between the three density-groups, but did show small but significantly higher (P < 0.05) levels from fruit that had been stored longer (which would seem to go against the trend of acidity reducing with time), and some significant region by density-group interactions (P < 0.05), with low density fruit from Nelson scoring significantly low compared with its cohorts.

11 312 New Zealand Journal of Crop and Horticultural Science, 2009, Vol. 37 Fig. 4 Results from the 1995 consumer panel. Figure at left shows the mean scores for each attribute and the ±1 SED error bars. If the error bars overlap, scores are not significantly different (P > 0.05). Right hand plots show the mean and standard errors of the rssc and penetrometer values of the samples as presented to panellists. Density sorting has separated the rssc values of the three density-groups, and penetrometer values are well matched and within the 6 8 N target range. First four panel assessments show highly significant separation of the responses while two of the last three show none. In 1995, the juiciness scores showed significantly low (P < 0.001) values for low density fruit, and fruit that had been stored for 16 weeks had significantly higher scores (P < 0.05). Firmness assessed by panellists, although not showing any tendency to relate to density-group, did show some regional differences (P < 0.01), but these regional patterns are expected because the Nelson region scored consistently low for both penetrometer and panel assessed firmness, perhaps reflecting the transport difficulties for fruit from this region. There were also significant firmness differences reported depending on gender (P < 0.05) and age (P < 0.01) of the panellists trained descriptive panel In results from the 1995 trained descriptive sensory panel, the right hand graphs show good separation between mean rssc values in the three densitygroups, and very balanced penetrometer values across the density-groups all centred in the target 6 8 N range (Fig. 5). The mean scores and ±1 SED error bars from the descriptive sensory panel assessments of each density-group are shown in the left hand graph of Fig. 5. The values reported in Fig. 5 should not be compared directly with those in Fig. 3 and Fig. 4 because they estimate intensities rather than sensory preferences or liking. Table 3 P values for fixed effects when sequentially adding terms to the model for sensory assessment of overall-liking in the 1995 untrained consumer panels. Significant effects are asterisked. Fixed term added to the model P value Density <0.001*** Storage-duration 0.92 Region 0.01** Density Storage 0.03* Density Region 0.97 Storage Region 0.06 Density Storage Region 0.35 Gender 0.71 Age 0.36 Gender Age 0.88 Order-of-evaluation 0.58 Table 4 P values for fixed effects when sequentially adding terms to the model for sensory assessment of sweetness in the 1995 trained descriptive panels. Significant effects are asterisked. Fixed term added to the model P value Density <0.001*** Region 0.43 Order-of-evaluation 0.03**

12 Jordan & Seelye Taste relationships of kiwifruit 313 Fig. 5 Results from the 1995 descriptive panel. Figure at left shows the mean scores for each attribute and the ±1 SED error bars. If the error bars overlap, scores are not significantly different (P > 0.05). Note that the scale for firmness has been shifted for clarity. Right hand two plots show the mean and standard errors of the rssc (ripened soluble solids concentration) and penetrometer values of the samples as presented to the panellists. Density sorting has separated the rssc values of the three density-groups, and penetrometer values are well matched and are centred in the 6 8 N target range. Sweetness shows highly significant separation of all three density-groups. Table 5 Residual standard deviation for six sensory attributes assessed by the 1995 trained descriptive panel estimated using models of varying complexity. Right hand column shows the P values for the rssc (ripened soluble solids concentration) model, with significant effects asterisked. Model: Density + Density + P value for Region + Region + rssc rssc + model with Residual SD Order Order + rssc alone Order rssc alone Overall flavour <0.001*** Char.-kiwifruit-flavour <0.001*** Sweetness <0.001*** Acidity Juiciness Firmness The descriptive panel assessments of overallflavour and characteristic-kiwifruit-flavour showed significantly higher scores for high-density fruit and no significant differences between low- and mediumdensity fruit. Sweetness showed a progressive and highly significant difference between all three densitygroups. Fruit showed no differences in sensory firmness, acidity, or juiciness across density-groups. There are no significant effects of region on sweetness assessment, but order-of-evaluation does have a significant effect (P < 0.05) (Table 4). This might be expected from the sensory panellists as a result of sensory fatigue as they progress from sample to sample. Density has a highly significant effect on overall-flavour, characteristic-kiwifruit-flavour, and sweetness (P < 0.001; full Wald statistic probability tables not presented here) but no effect on acidity or firmness (P > 0.3). There was a significant region effect on overall-flavour (P < 0.05) which was also observed in the measured rssc values, suggesting that the regional differences in density targets may be the cause. Results of the modelling study of the six sensory attributes assessed by the 1995 descriptive panel show the residual SD (RSD) for each modelled assessment using four different models (Table 5). The first model (column 2) is that reported in Fig. 5 and

13 314 New Zealand Journal of Crop and Horticultural Science, 2009, Vol. 37 predicts the response based on sample density-group, growing region, and order-of-evaluation. This set of RSDs is used as the basis for comparing the other models. When rssc is added to the model, the three sensory assessments with significant density effects (overall-flavour, characteristic-kiwifruit-flavour, and sweetness) all show some reduction in their RSD. The remaining three variates (acidity, juiciness, and firmness probably not density-related) underwent some minor shifts up and down in their RSDs. When the model was reduced to rssc as the single independent variable, there was essentially no change in the RSD of overall-flavour or characteristic-kiwifruit-flavour and an increase in that for sweetness. The final model incorporating the two variates rssc and order-of-evaluation gave the lowest RSD of all four models for the three first sensory measures, and had little effect on the remaining three measures (acidity, juiciness, and firmness). Discussion Untrained consumer panels In the consumer panel studies of both years, there is a strong relationship between harvestdensity, and sensory attributes of kiwifruit in all four of the major sensory categories: sweetness, characteristic-kiwifruit-flavour, flavour, and overallliking. Consumers clearly awarded higher scores for higher density fruit. The strength of this relationship may be higher, considering that there was no clear separation between the density groups presented to the panellists and some may have compared fruit with opposite density values to that intended, thus increasing the scatter of results and reducing relationship strength. Asami et al. (1988) showed that acidity of individual fruit is strongly correlated with its penetrometer firmness. In our study these two measures showed patterns were consistent with this correlation over the 2 years. In 1994, the penetrometer values were poorly matched and low-density fruit had significantly lower penetrometer firmness and gained significantly lower firmness and acidity scores. It seems likely that this spread of penetrometer values has caused a spread in acidity scores. In 1995, the penetrometer values and both firmness and acidity scores were closely matched and showed no differences for the three density-groups. Thus the 1995 results should be more reliable without this firmness confusion; and they support the conclusion that density separation does not influence perceived acidity or firmness by the panellists, nor the penetrometer firmness of the fruit. Reasons for the 1995 juiciness scores being significantly lower (P < 0.001) for low density and longer storage fruit remain unclear. Trained descriptive panel A trained descriptive sensory panel study, used to investigate the cause of the taste differences, showed progressive and significant score differences in assessed sensory sweetness between all three density-groups. No other sensory score showed such consistent separation. We conclude that the major sensory difference between the three density-groups is the difference in their sweetness. Further, we propose that it is this difference in sweetness, i.e., higher concentrations of sugars present in higher density fruit, which drives the consumer preference for higher density fruit. We also extrapolate this to suggest that there would be a similar result if fruit were selected on the basis of their rssc concentration or their DM rather than their density, particularly considering the clear ordering of the rssc values in the three density-groups, and the physiologically strong basis for the relationship between rssc and DM, discussed further below. The descriptive panel assessments of the intensity of overall-flavour and characteristic-kiwifruit-flavour show large and highly significant separation between the high-density and the two lower density groups, but the lower groups almost overlap each other. No explanation can be postulated for this pattern. Fruit showed no such differences between the densitygroups in the trained panel assessments of firmness (as would be expected given that equally firm fruit were given to panellists), acidity (possibly through its strong relationship with physical firmness), nor of juiciness. RSSC relationship A number of the 1995 panel results reported regional differences in panel-assessed sensory values that could not be accounted for by density-group alone. This might be expected because density boundaries were different for each region (Table 1). In addition, the density separation boundaries are blurred by errors in the density/rssc relationship. Both effects might introduce additional variability to panellist scores. Additionally, scores for 1994 were a little higher than those for 1995, but with a concomitant elevation of the rssc values. If rssc is the major driver of the sensory response, then rssc itself

14 Jordan & Seelye Taste relationships of kiwifruit might provide a better indicator of both consumer and descriptive panel sensory responses. This was investigated further by the modelling study of the 1995 descriptive panel (Table 5). Using rssc alone as an independent variable has given an improved model for the three sensory measures: overall-flavour, characteristic-kiwifruit-flavour, and sweetness when compared with the model developed above that incorporates density, region, and order-of-evaluation. This suggests that the variability introduced by: (1) using different density boundaries for different regions, and (2) the sorting errors across the density boundaries themselves, has been considerably reduced by using fruit rssc alone as the driver of the sensory model equation. The best overall model obtained in this study was gained by adding orderof-evaluation to rssc. This density versus rssc study was repeated on the 1995 consumer trial with similar results. Models of overall-liking, characteristic-kiwifruitflavour, flavour, and sweetness, all showed lower residual SDs when modelled with rssc alone than with models that incorporated density, storageduration, and region and all their second and third order interactions. One possible conclusion from these studies is that rssc is the true driver of the sensory response in kiwifruit. Additionally, it seems necessary to take order-of-evaluation into account when panellists study multiple fruit at a sitting. There is an element of fatigue in fruit tasting and we saw average sweetness scores dropping by over 4 mm (3% of the 150 mm score line) for each subsequent fruit assessed. To put this figure in perspective, it was noted that the slope of the rssc versus sweetness response in the rssc-only model was 5.2 mm/rssc%. This defines the fatigue effect as 0.77 rssc% per subsequent fruit assessed. The third fruit assessed of a set of three thus sees a penalty that is about the same size as the separation of two adjacent density groups as used in this trial. Conclusions The trials reported here confirm the rationale for density sorting: higher density fruit at harvest time when ripened have higher rssc and are observed to have better sensory scores on a number of sensory attributes assessed. There is evidence that any increase of the fruit density results in an increase of the sensory properties of the fruit right across the range of fruit observed, although there appears to be a reduction of the response slope at higher densities. 315 This preference for higher density fruit holds for all kiwifruit studied, regardless of their source; all regions, orchards, management regimes, and fruit sizes tested show the effect, and it was observed over two seasons and at two storage durations. We conclude that it appears to be a fundamental property of kiwifruit. Evidence presented here supports the suggestion that the ripe soluble solids concentration (rssc) of the fruit is the fundamental driver of this taste separation and that density is a useful tool to separate fruit non-destructively on the basis of their rssc. From the reported relationship between rssc and DM, we are confident that DM is an equally good basis on which to sort fruit for taste, although we note that starch hydrolysis can cause a small but measurable shift in DM. Density-based sorting procedures clearly have the potential to separate kiwifruit into different consumer taste categories, and thus different value to marketers. Thus, we have tested prototype density sorting equipment that is capable of separating kiwifruit at rates exceeding 2 t/h and that could be economically scaled to full commercial rates (perhaps 10 t or fruit/h). This result also has implications for other nondestructive measurement techniques capable of sorting fruit on their DM concentration at harvest, and near infrared sorting technology is a strong contender for this task. Acknowledgments We acknowledge the efforts of the late Bruce Stowell for his initial forays into the use of density and his subsequent support, and to the late Margaret Stec who demonstrated the initial taste response. We thank researchers Annette Richardson, Ted Dawson, Jill Stanley, Glen Hopkirk, and Elspeth MacRae, for scientific and experimental input into the project. To the many orchardists and ZESPRI staff who assisted with fruit supply, and the team at the Mt Albert Research Centre sensory laboratory and associated statistical groups, Shona Murray, Janine Gilbert, Nicki Hall, Tracey Phelps, Susan McDermott, Rod Ball, and Anne Gunson go our sincere thanks. Harold Henderson undertook the overall statistical analysis and was invaluable in unifying this project. This research was funded by the NZ Kiwifruit Marketing Board (now ZESPRI Group Limited) and by the New Zealand Foundation for Research, Science & Technology (Contract numbers CO 6330 and CO 6624).