Biosystems Engineering (2002) 83 (3), 339–347 t l 6 t e o o incidence efficiently even for late-picked fruit. Together with a proper delay of CA, a sufficiently high O2 concentration during CA was most important. The model was validated with data of 16 orchards gathered over five harvest seasons in two countries which gives it a wide validity range and a high practical 1 c fl u b b e c s I r n C 0 li d t a m in b 1 relevance. # 2002 Silsoe Research Institute. Published by Elsevier Science Ltd. All rights reserved. . Introduction Core breakdown or brown heart in pears (Pyrus ommunis L. cv. Conference) is the browning of the esh, especially around the core region. Other names in se for the same disorder are brown core and internal reakdown, although some authors make a distinction etween core breakdown or brown heart (Larrigaudiere t al., 1998). Eventually, the brown tissue may develop avities because of dehydration. Affected pears are not uitable for consumption even with minor symptoms. nappropriate controlled atmosphere (CA) storage may esult in disordered pears and cause substantial eco- omical losses. Bertolini et al. (1997) reported optimal A conditions for Conference pears to be 1�5% O2 and �8% CO2 which gave best eating quality and very polyphenolic substances from the vacuole, resulting in brown discoloration of the tissue. Decompartmentation might be due to insufficient repair capacity because of lack of energy. Saquet et al. (2000) hypothesised that low energy levels and damage to cell structures associated with an increase in fermentative products and membrane permeability are related to the develop- ment of disorders in pear fruit stored under decreased O2 and/or increased CO2 concentrations. Respiration, delivering the needed maintenance energy, is affected by CA conditions and by the diffusivity of O2 and CO2 in the pear tissue (Lammertyn et al., 2001a). The causes that underlie the development of the disorder are not well known, although there was a lot of correlative knowledge between external parameters and the development of core breakdown (Lammertyn et al., doi:10.1016/S1537-5110(02)00194-0, available online at h PH}Postharvest Technology Effect of Harvest and delaying Control Core Breakdown Inciden Bert E. Verlinden1; Anton de Jager2; Jeroen Lam 1Flanders Centre/Laboratory of Postharvest Technology, Katholieke e-mail of corresponding author: b 2Applied Plant Research, Fruit Research Station, P.O. Box 200, (Received 1 December 2001; accep A logistic regression model was built to describe the eff applied storage factors on the incidence of core breakd statistical analysis showed that the probability of core complicated way than assumed before. In general, m concentrations, at a higher temperature and for lon However, delaying the controlled atmosphere (CA) c mited incidence of disorders, while 0�5% O2 resulted in isorders and poor flavour. Chervin et al. (2000) found hat low O2 storage conditions significantly decreased roma compounds in ‘Packham Triumph’ pears. Velt- an et al. (1999) hypothesised that core browning was duced by decompartmentation of intracellular mem- rane structures allowing polyphenoloxidase to oxidise 537-5110/02/$35.00 33 tp://www.idealibrary.com on ed Atmosphere Storage Conditions on ce in ‘Conference’ Pears mertyn1; Wendy Schotsmans1; Bart M. Nicola.ıı1 Universiteit Leuven, Willem de Croylaan 42, 3001 Leuven, Belgium;
[email protected] 670 AE Zetten, The Netherlands; e-mail:
[email protected] ed in revised form 8 August 2002) ct of picking time and the most relevant commercially wn in pears (Pyrus communis L. cv. Conference). The breakdown depended on several variables in a more re mature fruit, stored at lower O2 and higher CO2 ger times are more susceptible to core breakdown. onditions for 21 days decreased the core breakdown 2000). The first objective of this paper is to combine this knowledge in a comprehensive model in such a way that quantitative results in terms of probabilities of the occurrence of the disorder were obtained, accounting for the substantial biological variation including variation due to orchard and season effects. The second objective was to demonstrate the power and possibilities of the 9 # 2002 Silsoe Research Institute. Published by Elsevier Science Ltd. All rights reserved. B.E. VERLINDEN ET AL.340 appropriate statistical technique to analyse categorical data by logistic regression, which is successfully used in other fields of research, especially in epidemiology (Rothman & Greenland, 1998), but only rarely applied in the field of postharvest. 2. Materials and methods 2.1. Harvest and storage conditions During five seasons, pears (Pyrus communis L. cv. Conference) were harvested in 16 orchards (13 in The Netherlands and three in Belgium) with different susceptibilities for core breakdown, known from pre- vious experience. In Belgium, the commercial picking time for long-time storage was determined based on the comparison of refractometer values of the juice, starch index, acidity and Magness-Taylor firmness with histor- ical data. In The Netherlands, the commercial picking time was determined by Magness-Taylor firmness ð6�3 kg=0�5 cm2) only (de Jager et al., 1996). Detailed information about the orchards, picking date, storage conditions and evaluation time is given below for seven sets of pears used. Set 1. Season 1995. Pears were picked at eight orchards in The Netherlands: (1) on the commercial picking date, (2) 1 and 2 weeks before and (3) 1 and 2 weeks after the commercial picking date. Pears were stored at �0�58C and kept in regular air (RA) for either 0, 2, 10 or 21 days before they were transferred to CA conditions of 2% O2 and 0�5% CO2. Moreover, CA conditions of 2% O2 combined with CO2 concentrations of 1�5 and 3% were applied directly after picking. Set 2. Season 1996. Pears were picked at seven orchards in The Netherlands on the commercial picking date, 1 week before and 1 and 2 weeks after the commercial picking date. Pears were stored at �0�58C and kept in RA for either 2, 10, 21 or 50 days before they were transferred to CA conditions of 2% O2 and 0�5% CO2. Moreover, CA conditions of 0�5% CO2 combined with 2, 4 and 7% O2 were applied directly after picking. Set 3. Season 1997. Pears were picked at 11 orchards in The Netherlands on 1 and 2 weeks after the commercial picking date. Pears were stored at �0�58C and kept in RA for either 0, 2, 7, 21 or 50 days before they were transferred to CA conditions of 2% O2 and 0�5% CO2. Additionally, CA conditions of 0�5% CO2, combined with 2, 4 and 7% O2, were applied directly after picking. Pears picked at 1 and 2 weeks before commercial harvest date, on the commercial harvest date, and 1 and 2 weeks after the commercial harvest date, were stored at �0�58C in RA for 7 days before they were transferred to CA conditions of 2% O2 with either 0�5 or 5% CO2. Set 4. Season 1997. Pears were picked at three orchards in Belgium on, 1 week before and 1 week after the commercial picking date. Pears were stored at �0�5 or 18C and kept in RA for 14, 7 or 0 days, respectively, for pears picked 1 week before, on, and 1 week after commercial harvest date before they were transferred to CA conditions of 2 or 0�5% O2 combined with 0�7% or 5% CO2. Set 5. Season 1998. Pears were picked at seven orchards in The Netherlands on and 1 and 2 weeks after the commercial picking date. They were stored at �0�5 8C and kept in RA for either 7, 21, 33, 50 or 80 days before transferring them to CA conditions of 2% O2 and 0�5% CO2. Moreover, CA conditions of 0�5% CO2 combined with 4 and 7% O2 were applied after 21 days in RA. Pears picked at 1 and 2 weeks before, on, and 1 and 2 weeks after commercial harvest date were kept for 7 days at �0�58C in RA before transferring them to CA conditions of 2% O2 with either 0�5 or 5% CO2. Set 6. Season 1998. Pears were picked at two orchards in Belgium on, 1 week before and 1 week after the commercial picking date. They were stored at �0�58C and kept in RA for 7 days before they were transferred to CA conditions of 2 or 0�5% O2 and 0�7% CO2. Set 7. Season 1999. Pears were picked at two orchards in Belgium on, 1 week before and 1 week after the commercial picking date. They were stored at �0�5 or 18C and kept in RA for either 7 or 28 days before they were transferred to CA conditions of 2% O2 and 0�7% CO2 or 0�5% O2 and 5% CO2. Furthermore, pears stored 7 days at �0�58C in RA were transferred to CA conditions of 2% O2 and 5% CO2 or 0�5% O2 and 0�7% CO2. 2.2. Evaluation of core breakdown The occurrence of core breakdown disorder was evaluated at the end of the storage, between 150 and 230 days after picking, in the pears of sets 1, 2, 3 and 5. The pears of sets 4, 6 and 7 were evaluated after several storage periods. Sets 4 and 6 were evaluated after approximately 2, 4, 6 and 8 months of storage while set 7 was evaluated after approximately 4 and 8 months. Pears were cut into two from stem to calyx end. Any browning regardless of its severity was noted as a disordered pear event. The proportion of disordered pears in a sample of pears that were picked at the same date of the same orchard and subsequently were stored at the same conditions (i.e. temperature, RA/CA conditions, storage time) was calculated. Proportions EFFECT OF HARVEST AND DELAYING CA STORAGE CONDITIONS 341 were based on samples ranging from 16 to 100 pears with a median of 25 pears. The seven data sets together resulted in 1825 calculated proportions of core break- down incidence representing 45 000 pears. A statistical analysis was carried out on the calculated proportions taking into account the number of pears in each sample. 2.3. Modelling Fifteen per cent of the calculated proportions were chosen randomly and kept for validation purposes. The remaining 85% was used to develop the model. Pooling all seven data sets would result in an unbalanced data set. In the initial step, submodels were constructed on the individual sets to study the behaviour of the different harvest and storage factors separately. Then, when the important factors were identified and the way they appear in the model structure was known, a model was fit to all the data sets. This overall model was then validated. 2.3.1. Model development using logistic regression analysis The goal of the model was to find a relation between the incidence of core breakdown and the explanatory variables such as picking time, storage conditions and storage time. Furthermore, the variability due to orchard and seasonal effects need to be quantified. The probability of the occurrence of core breakdown was estimated as the proportion of pears that developed the disorder. Proportions are binomially distributed and, hence, bounded between 0 and 1. The relation between the probability of the disorder incidence and the explanatory variables was modelled by using logistic regression as this technique takes into account a binomial distribution (Agresti, 1996): logitðpÞ ¼ log p 1� p � � ¼ aþ Xn i¼1 bixi: ð1Þ The right-hand side of Eq. (1) represents the contribu- tion of each explanatory variable xi to the probability of core breakdown incidence p. The probability p that core breakdown develops in a pear was linked to the explanatory variables xi with the logit transformation which is the logarithm of the odds. The odds are defined as the ratio of probability that the disorder occurs and of the probability that the disorder does not occur. The logit function transforms the probability scale from the range ½0; 1� to the scale of the explanatory variables which spans in principle ½�1;þ1�. Other transforma- tions are possible, but the logit transformation leads to coefficients, bi, interpretable in terms of odds ratios which is a measure of association that is widely used, especially in epidemiology (Rothman & Greenland, 1998). The symbol a denotes an intercept parameter. The explanatory variables that were considered in this study were the experimental factors such as (1) picking time, measured as deviation from the commercial picking time in weeks, (2) delayed CA, measured as the time in weeks of storage in RA before transfer to CA conditions, (3) O2 and CO2 concentration during CA storage, (4) storage temperature and (5) total storage time measured in weeks. Besides these main effects, their one by one products (statistical interactions), quadratic and cubic terms were considered as well. The orchard and season factors gave rise to unordered class variables. A set of dummy variables for both season ðxs;iÞ and orchard ðxo;iÞ was used to model their corresponding effects through the intercept a in Eqn (1) by means of the following model: a ¼ a0 þ X4 i¼1 bs;ixs;i þ X14 i¼1 bo;ixo;i: ð2Þ The dummy variables xs;i for the season factor in Eqn (2) were defined as follows. First, 1996 was chosen as the reference season. All dummy variables xs;i were set to zero for this reference season. For every other season a different dummy is set to one. Hence, four dummies were needed to code five seasons. The same strategy was followed to set the 14 dummies, xo;i, in Eqn (2) to code the 15 orchards. In this way, the intercept a in Eqn (1) is free to take a different value for each season and orchard combination. The intercept had the value a0 for the reference season and reference orchard. Interactions between season, orchard and the other experimental factors were assumed to be negligible. The intercept parameter a0 and the slope parameters, bi, bs;i and bo;i were estimated using the maximum likelihood criterion (Lindsey, 1997). To develop the different (sub)models, the model described by Eqns (1) and (2) was reduced by deleting insignificant terms. Terms were deleted from the model based on their significance (the probability P), the �2 Log Likelihood statistic (�2 Log L) and the Akaike information criterion (AIC) (Akaike, 1973). However, all dummy variable terms were kept in the model as well as insignificant lower order terms when higher order of the same factor were significant using the hierarchical principle (Lindsey, 1997). For both statistics, �2 Log L and AIC, lower values indicated a more desirable model. Since �2 Log L can only be decreased by including more explanatory variables, it tended to select models that overfit the data. The AIC, however, includes a penalty term for model complexity (the number of parameters) and is, therefore, more reliable to select good models (Mc Cullagh & Nelder, 1989). All logistic regression analyses were carried out using SAS/STAT software version 8 (SAS Institute Inc., 1999). 2.3.2. Model parameter interpretation using odds ratios The odds are defined as the ratio of the probability a pear will develop the disorder and the probability that a pear will not develop the disorder. An odds ratio Ro is then a ratio of two odds: Ro ¼ p1=ð1� p1Þ p2=ð1� p2Þ ð3Þ in which p1 is the probability of core breakdown 3. Results 3.1. Submodel 1: effect of picking time and delayed controlled atmosphere storage Data sets 1–3 and 5 were identified as having balanced information on the effect of the picking time DP in weeks after commercial harvest and time delay in controlled atmosphere (DCA), in weeks on the incidence of core breakdown. The sets were pooled together and 85% of the observations were randomly selected. Data from pears stored only at standard conditions (0�5% CO , 2% O and �0�58C) were used. The pears were l o m B.E. VERLINDEN ET AL.342 incidence for condition 1 defined by a certain set of values for the explanatory variables xðj;Þi and p2 is the probability of core breakdown incidence for another condition 2. When p1 and p2 are small, the odds ratio can be interpreted as the ratio of two probabilities (a risk factor, p1=p2) that core breakdown develops on the two conditions. In the special case that an explanatory variable xi has no interactions with other explanatory variables, then expðbiÞ equals the odds ratio for increasing the explanatory variable xi by one unit and keeping all other explanatory variables constant. 2.3.3. Model validation using principal component ana- lysis The 15% of randomly selected measured proportions which were kept for validation purposes were predicted with the developed logistic regression model. The mean centred values of measured and predicted proportions were subjected to principal component analysis (PCA) to separate the accordance between measured and model values from the measurement as well as modelling errors (Johnson &Wichern, 1992). Means were weighted based on the proportion sample sizes. The analysis was carried out using the Matlab Programming Environment version 5.3 (The MathWorks Inc., Natick, US). Tab Estimates of the reference intercept and slope parameters for subm picking ti Explanatory variable Reference intercept Delayed CA, weeks Deviation from commercial picking, weeks ðDelayed CAÞ � ðDeviation from commercial picking), weeks2 ðDelayed CAÞ2; weeks2 ðDelayed CAÞ3; weeks3 2 2 evaluated between 150 and 230 days of storage. It was assumed that the incidence of core breakdown would not significantly increase between 150 and 230 days. Only the severeness of the disorder was reported to increase in the period between 150 and 230 days of storage (Roelofs & de Jager, 1997; Lammertyn et al., 2001b). Since only the incidence, regardless of its severity, was used in this analysis, the above-stated assumption was justified. The modelling procedure resulted in the following equation: log p 1� p � � ¼ aþ bDCAxDCA þ bDPxDP þ bDCADPxDCAxDP þ bDCA2x 2 DCA þ bDCA3x 3 DCA ð4Þ showing significant effects of both DCA and deviation from commercial picking time. The estimates of the model parameters are listed in Table 1. The proportions of disordered pears stored for approximately 23 weeks in 2% O2, 0�5% CO2 at �0�58 C from one orchard in one season were plotted together with the modelled probabilities of core breakdown incidence as a function of delayed CA for three picking dates in Fig. 1. For illustrative reasons, a particular susceptible orchard and harvest season was chosen ða ¼ �1�54Þ. It was readily evident that picking too late, even for 1 week resulted in a marked increase of core breakdown incidence. However, delaying the CA con- e 1 del 1 describing the time delay in controlled atmosphere (CA) and e effects Parameter Estimate 95% Wald confidence limits a0 �3�72 0�19 bDCA 0�49 0�15 bDP 1�16 0�07 bDCADP 0�14 0�05 bDCA2 �0�36 0�05 bDCA3 0�025 0�004 The estimates of the model parameters are listed in Table 2. All main effects except the CO2 concentration were significant ðP50�05Þ. However, the CO2 concen- tration is indeed important as its interaction with O2 is significant. Rearranging the two terms involving O2 concentration in Eqn (5) yields log p 1� p � � ¼ aþ ðbO2 þ bO2CO2xCO2 ÞxO2 þ bCO2xCO2 þ bT xT þ btxt þ bDPxDP: ð6Þ The combined coefficient describing the O2 effect, ðbO2 þ bO2CO2xCO2 Þ, depended apparently on the CO2 EFFECT OF HARVEST AND DELAYING CA STORAGE CONDITIONS 343 ditions for 3 weeks prevented core breakdown incidence to a great extent. Note that a delayed CA treatment of 1 week resulted in maximum core breakdown. 3.2. Submodel 2: effect of storage time, O2, CO2 and temperature Fig. 1. Measured proportions of disordered pears (symbols) from a particular orchard of the season 1997 (intercept a ¼ �1�54) and stored for approximately 23 weeks in 2% O2, 0�5% CO2 at �0�58C plotted together with the modelled probabilities of core breakdown incidence (lines) as a function of delayed CA for three picking dates DP: — , }, DP ¼ 0 weeks; - - - , *; DP ¼ 1 week; . , &, DP ¼ 2 weeks. Bars indicate 95% confidence limits of the measured proportions Data set 4 provided additional information about storage time t, storage temperature T and CA condition ðO2; CO2Þ effects on the incidence of core breakdown. However, data set 4 did not contain information about delayed CA. Using a similar modelling procedure as for submodel 1, the following model was obtained: log p 1� p � � ¼ aþ bO2xO2 þ bCO2xCO2 þ bO2CO2xO2xCO2 þ bT xT þ btxt þ bDPxDP: ð5Þ Tabl Estimates of the reference intercept and slope parameters for subm time ef Explanatory variable Para Reference intercept O2, % CO2, % O2 � CO2, %% bO Storage temperature, 8C Storage time, weeks Deviation from commercial picking, weeks concentration because of the interaction. The main effect parameter of O2, bO2 , has a negative value. It means that CA storage with higher O2 concentration will result in a lower incidence of core breakdown. However, the parameter of the interaction of O2 with CO2, bO2CO2 , had a positive value and described a counteraction of CO2 to the positive O2 effect. A high value of CO2 concentration will result in a less negative value of the combined coefficient for O2. Hence, for a certain O2 concentration, a higher CO2 concentration will result in more core breakdown. Storage temperature, deviation from optimal picking time and storage time have positive slope parameters indicating that later picked fruits and stored at higher temperature for longer periods will have more core breakdown. The effects of O2, CO2 and storage time on the probability of core breakdown are graphically illustrated in Fig. 2. 3.3. Overall model Before fitting an overall model to all the data, the importance of additional interaction effects between factors of submodels 1 and 2 were investigated. Submodel 1 was extended with O2 and CO2 as explanatory variables. Instead of limiting the data sets 1–3 and 5 to standard CA conditions, all data of these sets were pooled to study possible interaction effects e 2 odel 2 describing the storage time, storage condition and picking fects meter Estimate 95% Wald confidence limits a0 �0�24 0�49 bO2 �1�48 0�26 bCO2 �0�004 0�011 2CO2 0�085 0�067 bT 0�22 0�17 bt 0�029 0�011 bDP 1�62 0�15 0.2 0.4 0.6 0.8 1 B ro w n co re O2 = 0 .5% CO2 = 0 .7% 0 0.2 0.4 0.6 0.8 1 B ro w n co re O2 = 2% CO2 = 0 .7% O2 = 0 .5% CO2 = 5% O2 = 2% CO2 = 5% e e B.E. VERLINDEN ET AL.344 0 10 20 30 0 Storage time, weeks Fig. 2. Measured proportions of disordered pears (symbols) pick season 1997 (intercept a ¼ �0�24) and stored at 18C, plotted tog (lines) as a function of storage time; bars indicate between O2, CO2, delayed CA and picking time. The interaction between O2 and delayed CA was significant. Extending submodel 2 with delayed CA and pooling data sets 4, 6 and 7 yielded a significant interaction between storage time and delayed CA. An overall model was then fitted to 85% randomly selected data from the pool of all seven data sets. The model structure was defined by including all the terms of submodels 1 and 2 plus the two additional interaction terms identified from the extended submodels. The Tabl Estimates of the reference intercept and slope paramet Explanatory variable Reference intercept Delayed CA, weeks ðDelayed CAÞ2; weeks2 ðDelayed CAÞ3; weeks3 Deviation from commercial picking, weeks ðDelayed CAÞ � ðDeviation from commercial pickingÞ; weeks2 Storage time, weeks ðDelayed CAÞ � ðStorage timeÞ; weeks2 O2, % CO2, % O2 � CO2, %% ðDelayed CAÞ � ðO2Þ, weeks, % Storage temperature, 8C 40 0 10 20 30 40 Storage time, weeks d at the commercial harvest date from a particular orchard of the ther with the modelled probabilities of core breakdown incidence 95% confidence limits of the measured proportions model parameters of the harvest and storage effects are listed in Table 3. For each orchard–season combination in the data, a was calculated using Eqn (2). The histogram of a is shown in Fig. 3. 3.4. Model validation The 15% randomly selected measured proportions from the seven data sets were predicted using the e 3 ers for the overall model; CA, controlled atmosphere Parameter Estimate 95% Wald confidence limits a0 �1�50 0�33 bDCA 0�74 0�24 bDCA2 �0�35 0�04 bDCA3 0�025 0�003 bDP 0�93 0�04 bDCADP 0�14 0�03 bt �0�020 0�009 bDCAt 0�026 0�006 bO2 �0�57 0�07 bCO2 0�15 0�05 bO2CO2 0�12 0�03 bDCAO2 �0�52 0�06 bT 0�69 0�10 EFFECT OF HARVEST AND DELAYING CA STORAGE CONDITIONS 345 overall model. The predicted versus the measured proportions are shown in Fig. 4. A principal components analysis was carried out by factorising the predicted and the measured values in two principal components. The first principal component carried 92% of the total variance while the second principal component carried 8%. The axis of the first principal component was plotted in Fig. 4 as well. Graphically, this axis represents the line for which the distances perpendicular to the points representing the measured and predicted couples are minimised. This axis yields the direction of the 10 9 8 7 6 5 4 3 2 1 0 Fr eq ue nc y − 2.4 − 2.1 − 1.8 − 1.5 − 1.2 − 0.9 − 0.3− 0.6 0 0.3 0.6 Intercept α Fig. 3. Histogram of all values for the intercept a in the overall model calculated using Eqn (2) accordance between the predicted and measured proportions while the axis of the second principal component, which is perpendicular on the first, yields the direction of the error. The angle between the horizontal and the first principal axis was 41�48 and its intercept with the vertical axis was 0�023. Their deviation from 458 and 0, respectively, were inter- preted as a measure of model bias. Measured propor- tions with values below 0�25 wherein the first principal component axis crossed the bisector were slightly overpredicted by the model, while proportions above 0�25 were underpredicted. Caution was, however, in order as the position of the first principle axis was determined by the combination of model error and measurement error. Hence, the model bias was over- estimated. 4. Discussion Roelofs (1995) reported that delaying the transfer to cooling conditions after picking in 2 days increased the incidence of core breakdown by 10%. Drake (1994) found that a 10-day delay of CA conditions resulted in increased breakdown in ‘Anjou’ pears compared to pears put in CA immediately after harvest. This worsening effect in the first week of delayed CA was found by Roelofs and de Jager (1997) as well. However, 0 0.2 0.4 0.6 0.8 1.0 0 0.2 0.4 0.6 0.8 1.0 Measured proportion brown core Pr ed ic te d pr op or tio n br ow n co re PC 1 Fig. 4. Measured versus predicted proportions of the validation data set; PC1 indicates the direction of the first principal component which describes the accordance between predicted and measured proportions; the dashed line denotes the direction for an unbiased accordance they found a beneficial effect of 21 and 50 days of delayed CA storage which was confirmed by the current data. To enable the model to describe maximum core breakdown incidence as a function of delayed CA, the square and cubic term for delayed CA were included in submodel 1. A significant interaction between delayed CA and picking date was found. The slope parameter, bDCADP, had a positive value indicating additional core breakdown when both CA were delayed and pears were picked late. It means that a beneficial effect of delayed CA can be expected but it will be less effective for late- picked pears. If CA is delayed too long, quality attributes like firmness and background colour are suffering progressively, since pears will senesce more in RA storage during the long delay. The results of submodel 2 yielded additional impor- tant factors for inclusion in the overall model. In a more detailed study, the effects of O2, CO2, temperature, picking date and storage time were investigated on data from one orchard in set 4 by Lammertyn et al. (2000). Besides these external parameters, intrinsic properties of B.E. VERLINDEN ET AL.346 the pears were taken into account. Lammertyn et al. (2000) concluded that heavy pears were more suscep- tible. It was found that severity of core breakdown increased during the first 4 months after which it slightly decreased in favour of the development of cavities that were frequently related to the development of core breakdown. Roelofs and de Jager (1997) showed that after about 2 months of storage the total number of affected pears did not increase further, but that the number of pears with cavities increased while that with browning decreased. Their general findings on the external parameters were confirmed in the current study which was carried out on several orchards and spanning five seasons, hence generalising the validity of the findings. Little is known about the relation between storage factors and the physiology in the pears and the mechanisms that induce core breakdown. Nanos et al. (1992) found that preclimacteric ‘Bartlett’ pears seemed less stressed and had greater potential for posthypoxia recovery than more mature pears. Later, they attributed this to a short-lived repair response of physiologically more mature pears (Nanos et al., 1994). Veltman et al. (2000) found that elevated CO2 concentration in storage decreased ascorbic acid levels in ‘Conference’ pears. Below a certain threshold core breakdown developed. However, late-picked fruit showed browning at rela- tively high levels of ascorbic acid. Mechanisms that can explain core breakdown incidence and susceptibility might be related to ways in which the fruit copes with the different stresses it is confronted with when picked and put in storage. The fruit should be able to generate enough energy to sustain its repair mechanisms and regenerate its antioxidantia, e.g. ascorbic acid. When pears are stored in lower O2 and higher CO2 concentra- tions, respiration metabolism and energy generation are retarded or less effective when fermentation takes over in favour of the oxidative metabolism (Lammertyn et al., 2001c). Delaying CA might give the fruit the possibility to cope with the ‘picking’ and cooling stress first, diminishing its energy needs gradually before having to deal with the ‘CA stress’ when energy production is lowered again. Extension of the two submodels resulted in two additional interaction terms. Delayed CA interacts with storage time having a positive slope parameter value (Table 3). It means that eventually the beneficial effect of delayed CA will vanish when pears are stored too long. The second interaction term was delayed CA with O2 concentration having a negative slope value, suggesting that delayed CA works better when pears are subse- quently stored at higher O2 levels or that the beneficial effect of delayed CA is decreased when the fruit is stored at lower O2 concentrations. Considerable variability in core breakdown incidence was due to seasonal and orchard effects, as is illustrated in Fig. 3. More negative a values indicate orchard– season combinations with less core breakdown. Using the concept of odds ratios, this can be quantified. Hence, if an orchard–season combination has an a value that is one less than another orchard–season combination, it will have 37% only of its susceptibility to develop the disorder, interpreted in terms of odds ratios. Because storage temperature had no interactions with other explanatory variables, expðbT Þ can be interpreted as an odds ratio and its value equals 2. This means that for a given set of storage conditions and picking time, the probability of core breakdown development is twice as high when the storage temperature is increased by 18C. Taking into account the substantial variability that exists between different orchards and seasons, the validation results shown in Fig. 4 are in agreement with the needed accuracy for the set goals. 5. Conclusion A logistic regression model was built to describe the effect on the probability of developing core breakdown in Conference pears of picking date as well as most relevant storage factors used in practice. Scattered and partial knowledge of separate factors were taken together in an overall model in which interactions between factors were quantified as well. Apparent contradictions due to limited experimental range and inclusion of a limited number of factors by other authors were clarified. The statistical analysis showed that the probability of core breakdown depended on several variables in a more complicated way than assumed before. In general, more mature fruit, stored at lower O2 and higher CO2 concentrations and at a higher temperature for longer times, are more susceptible to core breakdown. However, delaying the controlled atmosphere (CA) conditions for 21 days would decrease the probability of core breakdown incidence efficiently even for late-picked fruit. Together with a proper delay of CA, a sufficiently high O2 concentration during CA was most important. The model was validated with data of 16 orchards gathered over five seasons, which gave it a wide validity range and a high practical usefulness. For North West Europe, depending on the picking time, an O2 concentration between 2 and 3% with a maximum CO2 concentration of 0�7% after a delayed CA of 20 to 30 days are advised. Acknowledgements The Belgian Ministry of Small Enterprises, Traders and Agriculture (project S-5901 and D1/2-5771) and the European Commission (project FAIR1-CT-96-1903) are gratefully acknowledged for financial support. J. Lammertyn is a Research Assistant of the Fund for Scientific Research-Flanders (Belgium) (F.W.O.-Vlaan- deren). References Agresti A (1996). An Introduction to Categorical Data Analysis. John Wiley & Sons, Inc., New York Akaike H (1973). Information theory, an extension of the respiration between ‘Conference’ pear cells in suspension and intact pears. Journal of Experimental Botany, 52(362), 1769–1777 Larrigaudiere C; Lentheric I; Vendrell M (1998). Relationship between enzymatic browning and internal disorders in controlled-atmosphere stored pears. Journal of the Science of Food and Agriculture, 78, 232–236 Lindsey J K (1997). Applying Generalized Linear Models. Springer-Verlag, Berlin McCullagh P; Nelder J A (1989). Generalized Linear Models. Chapman & Hall, London Nanos G D; Romani R J; Kader A A (1992). Metabolic and other responses of ‘Bartlett’ pear fruit and suspension- EFFECT OF HARVEST AND DELAYING CA STORAGE CONDITIONS 347 maximum likelihood principle. Second International Sym- posium on Inference Theory, Akad!eemiai Kiado, Budapest, pp 267–281 Bertolini P; Bottardi S; Dalla Rosa M; Folchi A (1997). Effect of controlled atmosphere storage on the physiological disorders and quality of Conference pears. Italian Journal of Food Science, 9(4), 303–312 Chervin C; Speirs J; Loveys B; Patterson B D (2000). Influence of low oxygen storage on aroma compounds of whole pears and crushed pear flesh. Postharvest Biology and Technol- ogy, 19, 279–285 de Jager A; Johnson D; Hohn E (1996). Determination and prediction of optimal harvest date of apples and pears. Office for Official Publications of the European Commu- nities, Luxembourg Drake S R (1994). Elevated carbon dioxide storage of ‘Anjou’ pears using purge-controlled atmosphere. HortScience, 29(4), 299–301 Johnson R A; Wichern D W (1992). Applied Multivariate Statistical Analysis. Prentice-Hall, Englewood Cliffs, NJ Lammertyn J; Aerts M; Verlinden B E; Schotsmans W; Nicola.ıı B M (2000). Logistic regression analysis of factors influen- cing core breakdown in ‘Conference’ pears. Postharvest Biology and Technology, 20, 25–37 Lammertyn J; Scheerlinck N; Verlinden B E; Schotsmans W; Nicola.ıı B M (2001a). Simultaneous determination of oxygen diffusivity and respiration in pear skin and tissue. Post- harvest Biology and Technology, 23(2), 93–104 Lammertyn J; Dresselaers T; Van Hecke P; Wevers M; Verlinden B E; Jancs!ook P; Franck C; Schotsmans W; Nicola.ıı B M (2001b). Core breakdown in ‘Conference’ pears: a problem of respiration and diffusion. Presented at: 8th International Controlled Atmospheric Research Con- ference, 8–13 July, 2001, Rotterdam, The Netherlands. Acta Horticulturae, in press. Lammertyn J; Franck C; Verlinden B E; Nicola.ıı B M (2001c). Comparative study of the O2, CO2 and temperature effect on cultured ‘Passe Crassane’ pear fruit cells held in 0�25% O2. Journal of the American Society for Horticultural Science, 117(6), 934–940 Nanos G D; Romani R J; Kader A A (1994). Respiratory metabolism of pear fruit and cultured cells exposed to hypoxic atmospheres: associated change in activities of key enzymes. Journal of the American Society for Horticultural Science, 119(2), 288–294 Roelofs F P M M (1995). Annual report 1994. Research Station for Fruit Growing, Whilhelminadorp, The Nether- lands, pp 96–97 Roelofs F P MM; de Jager A (1997). Reduction of brownheart in Conference pears. Proceedings of the seventh Interna- tional Controlled Atmosphere Research Conference, Department of Pomology, University of California, Davis, CA Rothman K J; Greenland S (1998). Modern Epidemiology. Lippincott-Raven Publishers, Philadelphia Saquet A A; Streif J; Bangerth F (2000). Changes in ATP, ADP and pyridine nucleotide levels related to the incidence of physiological disorders in ‘Conference’ pears and ‘Jonagold’ apples during controlled atmosphere storage. Journal of Horticultural Science & Biotechnology, 75(2), 243–249 SAS Institute Inc. (1999). SAS/STAT User’s Guide, Version 8. SAS Institute Inc., Carry, NC, USA Veltman R H; Kho R M; van Schaik A C R; Sanders M G; Oosterhaven J (2000). Ascorbic acid and tissue browning in pears (Pyrus communis L. cvs Rocha and Conference) under controlled atmosphere conditions. Postharvest Biology and Technology, 19, 129–137 Veltman R H; Larrigaudiere C; Wichers H J; Van Schaik A C R; Van der Plas L H W; Oosterhaven J (1999). PPO activity and polyphenol content are not limiting factors during brown core development in pears (Pyrus communis L - cv. Conference). Journal of Plant Physiology, 154(5–6), 697–702 1. Introduction 2. Materials and methods 3. Results Table 1 Figure 1 Table 2 Figure 2 Table 3 Figure 3 Discussion Figure 4 5. Conclusion Acknowledgements References