Rapid Assessment of Insect Fragments in Flour Milled from Wheat Infested with Known Densities of...

SAMPLING AND BIOSTATISTICS

Rapid Assessment of Insect Fragments in Flour Milled from WheatInfested with Known Densities of Immature and AdultSitophilus oryzae (Coleoptera: Curculionidae)

MICHAEL D. TOEWS,1 JOEL PEREZ-MENDOZA,2 JAMES E. THRONE, FLOYD E. DOWELL,ELIZABETH MAGHIRANG, FRANK H. ARTHUR, AND JAMES F. CAMPBELL

USDAÐARS Grain Marketing and Production Research Center, 1515 College Avenue, Manhattan, KS 66502

J. Econ. Entomol. 100(5): 1714Ð1723 (2007)

ABSTRACT Milling wheat, Triticum aestivum L., infested with low densities of internal feedinginsects can result in ßour containing insect fragments. The Food and Drug Administration (FDA)enforces a standard or defect action level stating that a maximum of 75 insect fragments per 50 g ofßour is allowed. However, the relationship between level of infestation and number of resultingfragments is not well documented, and a more rapid method for enumerating insect fragments isneeded. We characterized the number of insect fragments produced from milling small lots of wheatspiked with known densities and life stages of Sitophilus oryzae (L.) (Coleoptera: Curculionidae).Insect fragments were enumerated with near-infrared spectroscopy (NIRS), a quick nondestructiveprocedure, and with the industry standard ßotation method. Results showed that an individual smalllarva, large larva, pupa, or adult produced 0.4, 0.7, 1.5, and 27.0 fragments, respectively. NIRS-predictedcounts of �51 (from small larvae), �53 (from large larvae), �43 (from pupae), or 0 (from adults)indicated that there were �75 actual fragments in that sample, because the upper bound of associated95% inverse prediction conÞdence intervals was less than the standard; NIRS-predicted counts of �98,�117, �108, or �225 fragments (same life stages as above) signaled that these ßour samples contained�75 actual fragments. These data suggest that NIRS could be adopted for rapid assessment of insectfragments resulting from relatively low levels of infestation with immature life states, but that it wasnot accurate enough for enumerating insect fragments, relevant to FDA standards, resulting fromadults.

KEY WORDS rice weevil, stored products, near-infrared spectroscopy (NIRS), internal insectinfestation, insect fragments

Milling of wheat, Triticum aestivum L., infested withstored-product insects that develop and feed insideindividual kernels, also known as internal infestingspecies, produces insect fragments in the Þnishedproduct (Sachdeva 1978, Perez-Mendoza et al. 2005).Internal insect infestations in bulk wheat, unlike ex-ternal insects, are not removed during the grain clean-ing process before milling. Internal infestations can bedetected before milling by holding wheat and waitingfor adults to emerge from kernels (Storey et al. 1982),but this takes too long to be practical. Counts of insect-damaged kernels (IDK) are used to grade the levelof infestation in wheat, but this method is unreliablebecause there is no correlation between insect-

damaged kernels and current insect infestation(WingÞeld and Pedersen 1985, Russell 1988, Perez-Mendoza et al. 2004). More rapid detection methodsinclude traditional techniques such as x-ray exam-ination or crack and ßoat (AACC 1995) and moreadvanced or experimental methods like near-infraredspectroscopy (NIRS), enzyme-linked immunosorbentassay (ELISA), or x-ray imaging (Quinn et al. 1992,Dowell et al. 1998, Karunakaran et al. 2003, Maghiranget al. 2003, Haff and Slaughter 2004, Toews et al. 2006).Most of these detection methods have not been com-mercially adopted due to practical constraints withspeed, small sample size, detection accuracy, or equip-ment expense.

The exact relationships among differing insect spe-cies, life stages, and infestation densities, and theresulting fragments in milled products are largely un-known. Sachdeva (1978) estimated that Sitophilus gra-narius (L.) (Coleoptera: Curculionidae) adults pro-duced 3 times more fragments than wheat infestedwith larvae. Perez-Mendoza et al. (2005) found thatRhyzopertha dominica (F.) (Coleoptera: Bostrichidae)

Mention of trade names or commercial products in this publicationis solely for providing speciÞc information and does not imply rec-ommendation or endorsement by the U.S. Department of Agriculture.

1 Corresponding author: Department of Entomology, Coastal PlainExperiment Station, University of Georgia, 122 S. Entomology Dr.,Tifton, GA 31793-0748 (e-mail: [email protected]).

2 Current address: Schendel Pest Services, 1826 S. Kansas Ave.Topeka, KS 66612.

adults and pupae produced 23 and 2.6� more frag-ments, respectively, than large larvae. Both of thesestudies showed that insect fragment counts increasedproportionally with infested kernel densities. In theUnited States, the Food and Drug Administration reg-ulates insect fragments in wheat ßour with an averagedefect action level (DAL) of 75 fragments per 50 gßour (FDA 1997). Fragment counts are currentlybased on the standard ßotation method (AOAC 1996).This method is labor-intensive, time-consuming, andexpensive because each sample must be carefullyheated and digested with hydrochloric acid beforeharvest of the ßoating insect fragments with suctionÞltration. The fragments must then be counted whileviewed under a dissecting microscope.

Several studies previously described novel tech-niques to rapidly detect and quantify insect fragments.Kitto (1991) developed a sandwich ELISA for detec-tion and quantiÞcation of myosin, an insect muscleprotein. This technology is available in kits that can beused on whole grain or ßour samples. Schatzki et al.(1993) determined that myosin content of individualsincreased from 0.1 to 9.0 �g from Þrst through fourthinstars, decreased to 1.9 �g in pupae, and resolved at6.6 �g in adult S. granarius. Brader et al. (2002) dem-onstrated that the ELISA procedure was more accu-rate than x-ray analyses, cracking, and ßotation, or thestandard ßotation method for detecting S. granariusinfestations under speciÞc conditions. Perez-Men-doza et al. (2003) investigated using NIRS to predictinsect fragments in ßour, but their results showedlimited success classifying fragments below the FDAdefect action level. Recently, a more successful studydemonstrated the utility of NIRS to enumerate R.dominica fragments resulting from milling grain spikedwith large larvae, pupae, or adults (Perez-Mendoza etal. 2005). Perez-Mendoza et al. (2005) showed that aBruker Optics Fourier-transform (FT) near-infrared(NIR) spectrometer (Bruker Optics, Billerica, MA)produced the most accurate fragment estimates inwheat ßour among the four spectrometers tested.Sitophilus oryzae (L.) (Coleoptera: Curculionidae)

is a cosmopolitan internal infesting insect species(Rees 2004). The life cycle averages 35 d at 27�C and69% RH, but larval development is generally completein only 22 d at these conditions (ShariÞ and Mills1971). Oviposition rates of 344Ð384 eggs per femaleare possible under ideal conditions (Birch 1945). Fe-males chew a hole and deposit individual eggs insidegrain kernels where the individuals remain for theduration of the immature life stages. A gelatinous ma-terial is deposited behind the egg, effectively protect-ing the developing immature from exposure to insec-ticides applied directly to the grain. Freshly emergedadults remain inside the kernels for a few days beforechewing out. In addition to insect fragments, internalinfestations also may contaminate grain-derived foodswith metabolic waste and caste skins. Bernton andBrown (1967) presented data showing that �10% ofhumans had positive allergic reactions to dermal ap-plications of S. oryzae-dialyzed extracts.

Development of a rapid, accurate, and inexpensivemethod for quantifying insect fragments in Þnishedßour is highly desirable to the milling industry. Newmethods that meet these simple criteria are likely toreplace existing methods. Although the number offragments is regulated, there is little information avail-able on how many fragments are produced by millingwheat with a known number of infested kernels. Sci-ence-based thresholds that relate known infestationswith number of insect fragments could greatly im-prove a millerÕs ability to produce acceptable ßour.The objectives of this study were to 1) characterizethe number of fragments in Þnished wheat ßour re-sulting from milling known densities of wheat kernelscontaining small larvae, large larvae, pupae, or newlyeclosed (preemergent) adults of S. oryzae; and 2)develop thresholds relative to each life stage and testthe ability to predict fragment densities by using NIRSfor rapid assessment of the number of insect fragmentsin a ßour sample.

Materials and Methods

Insect Cultures. Sitophilus oryzae colonies werereared in the laboratory in 0.9-liter glass jars Þlled with�300 g of hard red winter wheat ÔJaggerÕ (Sears et al.1997) equilibrated to 13.5% moisture content. Jarswere Þtted with Þlter paper and wire mesh lids with420-�m openings to prevent insect escape. The col-ony, originating from eastern Kansas and collectedbefore 1960, was maintained in the laboratory at 27.0 �0.5�C and 60 � 5.0% RH with a photoperiod of 14:10(L:D). Rearing density was �100 starting adults per300 g of wheat.

Kernels containing S. oryzae of the correct life stagewere selected using radiographic analyses. Foundingadults were removed after 24 h and the colonies werereturned to the growth chamber for 14 d (small lar-vae), 25 d (large larvae), �30 d (pupae), or �35 d(preemergent adults). Double-sided adhesive tapewas used to hold the wheat kernels from the insectcolonies to previously exposed x-ray Þlms for easyhandling. The Þlms were then individually x-rayed andprocessed following the methods described in Toewset al. (2006). Kernels containing preemergent adultswere selected based on the presence of adults in ker-nels that had not yet chewed an exit tunnel to theoutside of the kernel. Kernels containing a single in-sect of the targeted life stage were manually culledusing forceps and grouped by life stage in aluminumbaking dishes. These kernels were immediately placedin a forced air laboratory oven for 1 h at 130�C to killthe insects. Kernels were then stored in sealed plasticbags in the freezer at �27�C. The body and tunneldimensions (length and width) of 25 representativesmall larvae, large larvae, and adults were measuredfrom the radiographs by using a stage micrometer.Insect Infestation Densities. Ten levels of infesta-

tion for each life stage were selected based on pre-liminary data, suggesting a range of insect fragmentsbelow and above the defect action level. Uninfestedwheat was obtained immediately after harvest directly

October 2007 TOEWS ET AL.: INSECT FRAGMENTS IN FLOUR 1715

from a grower in Dickinson Co., KS, who used stan-dard agronomic practices. The wheat was cleanedwith multiple passes over a mechanical dockage tester(model XT2, Carter Day International, Minneapolis,MN) to ensure all foreign material and potential ex-ternal infesting insects were removed. Subsamples ofthe clean wheat were x-rayed as described above, butthey showed no signs of insect infestation. In prepa-ration for milling, 100 � 0.1-g samples of clean wheat(�3,150 kernels) were weighed on a microbalanceand stored in heavy plastic bags. Infested kernels werethen added to the clean wheat at the following ex-perimental densities: 0, 10, 20, 40, 60, 80, 100, 120, 140,or 160 kernels (small or large larvae); 0, 5, 10, 15, 20,30, 40, 50, 60, or 70 kernels (pupae); and 0, 1, 2, 3, 4,5, 6, 7, 8, or 9 kernels (adults). For consistency, theequivalent number of sound kernels was removedfrom each 100-g sample before introducing the in-fested kernels. Five replicates of each infestation den-sity were milled.Milling. Wheat samples (100 g) were individually

milled on an experimental Brabender Quadrumat Sr.mill (type 12-10-N87, C. W. Brabender Instruments,Inc., Hackensack, NJ) at the Hard Winter WheatQuality Laboratory located at the USDAÐARS GrainMarketing and Production Research Center, Manhat-tan, KS. This 4-roll milling system is commonly used toproduce production equivalent test ßours, and it hasa milling efÞciency of �60%; therefore, each 100 g ofwhole wheat yielded 60.0 � 5.0 g of ßour. Wheatsamples, initially at 11.4% moisture, were tempered to15% moisture and then milled starting with the neg-ative controls (0 infested kernels) and continuingthrough the increasing infestation densities within aparticular life stage. Five hundred grams of cleanwheat was milled between life stage treatments toremove insect fragments that may have accumulatedinside the mill. Flour was stored in 300-ml polyethyl-ene jars with tight Þtting lids.SpectralDataCollection.The Bruker OpticÕs Matrix

I FT-NIR spectrometer, which uses the diffuse reßec-tance technique, was used to collect spectral data(835Ð2502 nm) of wheat ßour containing varying S.oryzae fragments. This instrument was described indetail by Armstrong et al. (2006). Each ßour samplewas poured in a glass-bottomed cylindrical bowl (85mm in diameter) equipped with a stirring paddle tocontinuously mix the sample while rotating the bowlduring which time multiple reßectance scans werecollected. Two replicates, each consisting of the spec-tra of 100 scans per ßour sample, were averaged intoone Þnal spectrum that was assigned to that replicate.Fragment Counts.A direct count of the insect frag-

ments from each milled sample was obtained to com-pare against the NIRS-predicted results. The standardßotation method (AOAC 1996) was scaled-up to col-lect fragments in the 60 � 5-g ßour replicates. Brießy,the AOAC procedure prescribes digesting the ßourwith hydrochloric acid and then separating the hy-drophobic insect fragments from the remaining hy-drophilic plant material in a mineral oil and watermixture. After settling, the oil was washed with iso-

propanol though lined Þlter paper that retained non-digested solids. After moistening with �2 ml of dis-tilled water, the resulting Þlter papers were examinedunder a dissecting microscope with darkÞeld lightingat 30� magniÞcation. Visual identiÞcation of S. oryzaefragments followed previously published methods(Kurtz 1956, Gentry et al. 1991) and practice on pre-liminary data sets previously scored by an experiencedtechnician.Data Analyses. Standard regression techniques

were used to describe the relationship between NIRS-predictions and the resulting number of S. oryzae frag-ments recovered using the standard ßotation method.Regardless of the two NIRS-predicted response vari-ables modeled (NIRS-predicted infested kernels andNIRS-predicted insect fragments), these estimates arebased on the total mass of the cumulative fragmentsdetected in the sample. Data were Þt using Table-Curve 2D software (Systat Software Inc. 2002) andÞnal prediction equation parameters or coefÞcientswere based on the pattern of residuals, lack-of-Þt tests,and r2 values (Draper and Smith 1981). ConÞdenceintervals (95% CI) were used to estimate the intervalaround the observed mean for insect fragments fromthe same population, whereas prediction intervals(95% PI) were used to relate the current Þndings onfragment numbers to a future observation. SpeciÞ-cally, we used the upper boundary of insect fragmentsfrom the prediction interval to show how many in-fested kernels containing S. oryzae of a known agecould be milled and still produce ßour with acceptablenumber of fragments based on the FDA standard.

NIR spectra were analyzed with partial least squares(PLS) regression analysis by using GRAMS (ThermoElectron Corp. 2006), a spectral analysis software.Independent variables (number of infested kernels ornumber of insect fragments) by life stage (small lar-vae, large larvae, pupae, adults, or mixed life stages)were assigned to corresponding spectral data. Param-eters chosen in the PLSPlus/IQ software were PLS-1calibration type, cross-validation diagnostic type withone Þle out, mean centering data preparation, and nopath length or baseline corrections. This method ofcross-validation attempts to emulate predicting “un-known” samples by using the training data itself; in aone-Þle out cross-validation, a single sample is left outand predicted using the model developed using theremaining data set (Galactic Industries Corporation1996). The wavelength range of 1,000Ð2,500 nm wasused in calibration model development to eliminatethe noise observed in the �1,000-nm region. Param-eters used to evaluate calibration models developedfor each insect life stage included the coefÞcients ofdetermination (R2), standard error of cross validation(SECV), and PLS regression coefÞcients. The SECVwas used as an indicator of the “best” number ofindependent variables or factors needed for the cali-bration model. The PLS regression coefÞcients give anindication of the wavelengths (shown as positive andnegative peaks in the plot) that are more heavilyweighted at wavelengths that contributed most to thecalibration model.

1716 JOURNAL OF ECONOMIC ENTOMOLOGY Vol. 100, no. 5

Regression models were developed to explain therelationship between predicted number of S. oryzaefragments from the NIRS calibration models and ac-tual number of fragments obtained with the standardßotation method. Inverse prediction conÞdence in-tervals on linear regression models (Neter et al. 1996)were calculated to demonstrate how NIRS predictedcounts of insect fragments (a quick and nondestruc-tive approximate measure) could be used to estimatethe number of actual insect fragments (a precise butdestructive and time-consuming measure) with 95%conÞdence.

Results

S. oryzae body dimensions generally changed withlife stage, whereas tunnel dimensions changed little.Tunnels in wheat kernels did not increase from largelarva through preemergent adult life stages, whereasindividual body length and width changed signiÞ-cantly with life stage. Tunnel length and width morethan doubled between the small larvae and large lar-vae life stages (Table 1). Interestingly, pupae weresigniÞcantly longer and wider than adults.

Older insects and infestation density increased thenumber of insect fragments observed in the Þnishedßour. Samples produced by milling more infested ker-nels yielded a proportional increase in insect frag-ments (Table 2). Small larvae contributed fewer frag-ments than similar numbers of large larvae. A singlesmall larvacontributed0.4�0.03 fragments(95%CI�0Ð1.8; 95% PI � 0Ð15.1), whereas an individual largelarva contributed 0.7 � 0.04 fragments (95% CI �0Ð8.1; 95% PI � 0Ð26.7). In comparison, a single pupa

contributed 1.5 � 0.1 fragments (95% CI � 0Ð10.8; 95%PI � 0Ð31.4) and a single adult contributed 27.0 � 1.8fragments (95% CI � 11.2Ð46.3; 95% PI � 0Ð109.2). Alinear relationship explained much of the variabilitybetween the number of infested kernels and the num-ber of resulting insect fragments in the ßour (Fig.1AÐD). Appendages of the head and thorax, sclero-tized internal organs, and general body cuticle wereobserved in the adult fragments, but immature frag-mentswereprimarilycomprisedofheadcapsule,man-dibles, and tarsal segments. The 95% prediction inter-vals suggest that the maximum number of infestedwheat kernels that could be milled in a 100-g wheatsample and still produce ßour below the FDA defectaction level of 75 fragments in 50 g ßour (upper 95%CL would be below 90 fragments in 60 g of ßour) was155 kernels containing small larvae (95% PI � 54.5Ð89.5), 90 containing large larvae (95% PI � 43.7Ð89.7),39 containing pupae (95% PI � 36.5Ð89.7), and 0containing adults (95% PI � 0Ð80.0).

NIRS calibration models developed with these datasatisfactorily predicted both the number of infestedkernels in that ßour sample and number of insectfragments in the ßour sample. For estimation of insectfragments, PLS regression coefÞcients obtained byNIRS showed that important wavelengths used to dif-ferentiate insect fragments from the ßour matrix oc-curred at �1,180, 1,360, 1,440, 1,620, 1,725, 1,765, 1,900,1,920, and 1,950 nm (Fig. 2). Similar PLS regressioncoefÞcients were used to develop calibration modelsto estimate the number of infested kernels that weremilled to produce that sample. The coefÞcients ofdetermination (R2) suggested that the calibrationmodels described comparatively more of the variation

Table 1. Mean � SEM (n � 25) S. oryzae body and tunnel dimensions (millimeters) by life stage

Life stage Body length Body width Tunnel length Tunnel width

Small larvae 0.98 � 0.05d 0.41 � 0.01c 1.84 � 0.09b 0.66 � 0.02bLarge larvae 2.54 � 0.06c 1.42 � 0.03a 4.13 � 0.06a 1.61 � 0.02aPupae 3.64 � 0.07a 1.40 � 0.02a 4.17 � 0.06a 1.62 � 0.03aPreemergent adults 3.46 � 0.04b 1.04 � 0.02b 4.07 � 0.04a 1.59 � 0.02a

Means within columns followed by the same letter are not signiÞcantly different (F range � 286.4Ð475.8; df � 3, 96; P � 0.01; one-wayANOVA). Means separations follow the lsmeans test (P � 0.05) (SAS Institute 1999).

Table 2. Mean � SEM (n � 5) number of insect fragments recovered in flour milled from 100 g of wheat containing kernels infestedwith S. oryzae small larvae, large larvae, pupae, or adults

Small larvae Large larvae Pupae Preemergent adults

No. infestedkernels

No.fragments

No. infestedkernels

No.fragments

No. infestedkernels

No.fragments

No. infestedkernels

No.fragments

0 0.0 � 0.0 0 0.0 � 0.0 0 0.0 � 0.0 0 0.0 � 0.010 2.8 � 0.9 10 5.8 � 1.1 5 12.5 � 1.1 1 37.8 � 7.420 5.6 � 1.0 20 13.8 � 2.0 10 20.6 � 2.1 2 50.0 � 8.040 15.0 � 2.5 40 30.8 � 2.1 15 23.0 � 2.5 3 84.6 � 12.360 27.8 � 5.1 60 45.6 � 4.5 20 26.2 � 4.2 4 116.6 � 6.580 35.8 � 4.5 80 66.8 � 3.3 30 56.2 � 5.0 5 136.2 � 18.5100 41.8 � 4.8 100 85.0 � 5.3 40 59.2 � 4.5 6 190.4 � 28.9120 54.0 � 2.4 120 90.4 � 8.0 50 80.3 � 9.5 7 229.2 � 24.1140 57.4 � 3.1 140 101.6 � 6.9 60 97.8 � 11.9 8 227.2 � 13.0160 83.4 � 7.1 160 105.4 � 6.0 70 na 9 265.8 � 34.1

na, no data available.


for the number of infested kernels containing smalland large larvae than for kernels containing pupae oradults (Table 3). Similarly, the calibration models forfragments from small and large larvae explained morevariation in the data than the calibration models forfragments from pupae or adults (Table 4). Directmodel comparisons between predictions of infestedkernels versus fragments are difÞcult because modelÞtting implied the need for different numbers of fac-tors for each life stage; differing numbers of factorsshould be interpreted similar to regression modelswith multiple independent variables.

Modeling the relationship between NIRS-predictedinfested kernels (or fragments) and the known num-ber of infested kernels (or observed fragments) ne-cessitated different types of curves. Lack-of-Þt testsand patterns of residuals suggested that only a linearrelationship existed between the NIRS-predictednumber of infested kernels and the actual density ofinfested kernels (Fig. 3AÐE). Interpretation of lack-of-Þt tests implied that quadratic models best ex-plained the relationship between the NIRS-predictednumber of insect fragments and the observed frag-ments (data not shown); however, these data werealso Þt to linear models (Fig. 4AÐE), because wewanted to calculate inverse prediction conÞdence in-tervals, which are based on a linear curve. The 95%inverse prediction conÞdence intervals showed that

predictions of 51, 53, 43, 0, and 21 fragments (for smalllarvae, large larvae, pupae, adults, or mixed stages,respectively) would fall below the DAL of 75 frag-ments in 50 g of ßour (upper 95% CI would be below90 fragments in 60 g of ßour). Similarly, NIRS-pre-dicted fragment counts of 98, 117, 108, 225, and 152fragments (from small larvae, large larvae, pupae,adults, or mixed stages, respectively) would be abovethe FDA defect action level (lower 95% CI would beabove 90 fragments in 60 g of ßour).

There were obvious differences in the time andequipment requirements to determine the number ofinsect fragments in ßour using the two methods. Forthe NIRS process, the spectrometer required �1 minper sample to complete the scans. The technicianrequired an additional minute to log the data on thecomputer, transfer the ßour back to the storage jar,thoroughly clean the sampling cup with compressedair, manually enter a new sample number, and load anew sample into the sample cup. In comparison, thestandard ßotation method required �2.5 h per batchof six samples (�75 samples could be processed duringthis time by using NIRS). Although we processed abatch of six samples simultaneously with the ßotation,this required six sets of glassware and other laboratoryequipment. Equipment-wise, the NIRS method re-quired only the NIRS scanner and a can of compressedair. By comparison, each sample processed with the

Fig. 1. Relationship between number of infested kernels/100 g of wheat and S. oryzae insect fragments resulting fromsmall larvae (A), large larvae (B), pupae (C), and preemergent adults (D). Numbers in parentheses after equation parametersindicate standard errors of parameter estimates.


ßotation method required a sieve, reßux apparatus,Wildman trap ßask, several beakers, hot-plate withstirring bar, watch glass, and ßask in addition to hot tapwater, hydrochloric acid solution, isopropanol solu-tions, mineral oil, detergent, and tetrasodium ethyl-enediaminetetraacetate.

Discussion

Various parameters contribute to the number offragments produced by a single insect. Sachdeva(1978) estimated that milling wheat containing deadS. oryzae produced �2.6 times more fragments thanlive individuals. Perez-Mendoza et al. (2005) reportedthat a single R. dominca larva or pupa produced vir-

tually identical numbers of fragments, and a similarpattern was shown here with S. oryzae. Rhyzoperthadominica (F.) adults produced only one half the num-ber of fragments (Perez-Mendoza et al., 2005) ob-served here with S. oryzae adults. The procedure fol-lowed by Perez-Mendoza et al. (2005) involved dryingthe specimens in a laboratory oven, but they did notfreeze the insect-infested kernels before milling. It isunknown whether freezing may cause the sclerotizedadult cuticle to shatter into more fragments duringmilling. WingÞeld and Pedersen (1985) reported thatinsect life stage, species, body size, and moisture con-tent affected fragmentation, whereas Harris et al.(1952) showed that mill size, ßow, and equipmentwere also important. We hypothesize that substan-tially more fragments were obtained with S. oryzaecompared with R. dominica adults, because S. oryzaeadults are up to 1.5 mm longer (Rees 2004) and havelonger appendages (legs, antennae, and snout) com-pared with R. dominica.

Analyses of spectra generated with NIRS highlightdifferences in chemical constituents among biologicalmaterials (Ridgway and Chambers 1996). The chitinof insect procuticle is comprised of repeatingn-acetyl-glucosamine sugars (Chapman 1982); CH3 and CH2

are commonly occurring moieties in insect cuticularlipids (Lockey 1988). The CH, CH2, CH3, and C�Ochemical constituents present in chitin were impor-tant parameters that may have allowed for predictionof insect fragments in ßour as shown in the PLS re-gression coefÞcients plots. Some wavelengths (�1,620and 1,765 nm) that were important for identifying S.oryzae insect fragments were identical to those usedby Perez-Mendoza et al. (2005) to quantify R. do-minica insect fragments in ßour. These regions gen-erally correspond with the Þrst and second CH ab-sorption overtones of CH, CH2, and CH3 groups(Murray and Williams 1990, Shenk et al. 1992, Os-borne et al. 1993). Likewise, Maghirang et al. (2003)also found NIR wavelengths at �1,370 and 1,610 nmas important for identifying whole S. oryzae insidewheat kernels. Each independently developed cal-ibration model identiÞed similar PLS regression co-efÞcients across studies and types of samplingequipment, thereby suggesting robust utility ofthese wavelengths.

Several factors complicate the development of amore accurate NIRS calibration model. For example,there is some error associated with direct enumeration

Table 4. Performance of NIRS calibration models devel-oped for enumerating S. oryzae fragments in wheat flour (60 g)by life stage

ModelNo.

factorsSECV na R2

Small larvae 3 15.25 43 0.71Large larvae 5 16.89 48 0.81Pupae 3 20.99 39 0.60Preemergent adults 6 69.36 48 0.50Mixed stages 5 54.19 178 0.42

a n is number of observations.

Fig. 2. PLS regression coefÞcients for NIRS calibrationmodels developed to enumerate S. oryzae insect fragmentsresulting from small and large larvae (A) or pupae and adults(B) in wheat ßour.

Table 3. Performance of NIRS calibration models developedfor enumerating S. oryzae infested kernels in wheat flour (60 g) bylife stage

ModelNo.

factorsSECV na R2

Small larvae 7 18.50 43 0.88Large larvae 6 15.10 48 0.92Pupae 5 9.31 39 0.83Preemergent adults 4 1.93 48 0.56Mixed stages 5 31.96 178 0.58

a n is number of observations.


of insect fragments by using the ßotation method.WingÞeld and Pedersen (1985) milled insect-infestedwheat and then sent Þve representative ßour samplesto two separate reputable laboratories; on average,one laboratory reported 200 more insect fragments inthe samples compared with the other. Similarly,Brader et al. (2002) documented wide variation infragment counts among three laboratories that re-ceived subsamples of ßour produced by milling wheatspiked with known numbers of infested kernels. Afterextensive study and practice on preliminary samples,a single person counted the fragments in our study toreduce variability due to processing the samples.Therefore, any processing error should be reducedand be consistent across the samples. The Þtting of alinear curve without signiÞcant lack of Þt to the plots

of infested kernels versus number of fragments at leastsuggests good precision in counting the fragments. Asecond factor complicating calibration is differentsized fragments. Although NIRS-predicted counts arebased on the total mass of the cumulative fragments,numbers of fragments quantiÞed with the standardßotation method are irrespective of the total mass.Thus, it is difÞcult to calibrate the novel detectiontechnique to the approved standard when they are notmeasuring exactly the same parameter. This argumentis supported by the fact that linear models were abetter Þt for predicting the number of infested kernels(a relatively constant mass), but quadratic modelswere a better Þt for predicting the number of frag-ments (a variable number of fragments could com-prise the same mass).

Fig. 3. Relationship between NIRS-predicted and actualnumber of S. oryzae infested kernels in ßour samples (60 g)produced from wheat infested with small larvae (A), largelarvae (B), pupae (C), preemergent adults (D), and mixedlife stages (E). Numbers in parentheses after equation pa-rameters indicate standard errors of parameter estimates.


Adoption of NIRS for insect fragment analyses hasseveral advantages over the standard ßotation methodand ELISA procedure. NIR spectrometers are widelyavailable, relatively inexpensive, and presently used ingrain processing facilities to measure quality attributesof wheat ßour, such as ash, moisture, and proteincontent, and quality of animal feeds such as crudeprotein, fat, and starch. NIRS is much faster (�2 minto process a sample) compared with the standardßotation method that requires �2.5 h to process asample and the ELISA procedure that requires justover an hour to process a sample (Brader 1997). Evenwhen adjusting for the ability to process more thanone sample at a time with the non-NIRS methods,NIRS was more efÞcient. Furthermore, NIRS estima-tion of insect fragments requires no special sample

preparation and produces no disposal hazards. Thestandard ßotation method requires a continuous sup-ply of laboratory chemicals (i.e., hydrochloric acid,mineral oil, and isopropanol), whereas the ELISA pro-cedure also requires purchase of disposable kits andreagents for each sample. The ELISA method to detectmyosin also has several technical issues that impact itsaccuracy: there are variable amounts of myosin indifferent life stages (Schatzki et al. 1993) and myosindegrades during the Þrst 2 wk after insect death, butit remains relatively stable after that initial decay pe-riod (Atui et al. 2007). After the calibration model isdeveloped, NIRS determination does not requirehighly skilled labor for operation. By comparison, thestandard ßotation method requires detailed training inmicroanalytical entomology to recognize the insect

Fig. 4. Relationship between NIRS-predicted and ob-served S. oryzae fragments in ßour samples (60 g) producedfrom wheat infested with small larvae (A), large larvae (B),pupae (C), preemergent adults (D), and mixed life stages(E). Numbers in parentheses after equation parameters in-dicate standard errors of parameter estimates.


fragments (Kurtz and McCormack 1965), and the my-osin ELISA procedure requires training in immuno-logical detection technique. Finally, the standard ßo-tation and ELISA procedures are both destructivesampling methods, whereas NIRS is nondestructive.

Disadvantages to using NIRS for detection and es-timation of insect fragments in wheat ßour generallyrelate to the imprecise nature of the estimates, com-plex development of robust calibration models, ongo-ing spectrometer calibration, and consistency acrossseveral individual instruments. Inverse prediction in-tervals developed here and in Perez-Mendoza et al.(2005) suggest that NIRS-predicted immature frag-ment counts below �50 fragments forR. dominica andS. oryzae would not exceed the FDA threshold of 75fragments in a 50-g sample. Likewise, NIRS-predictedfragment counts above 105 fragments would likelyhave �75 actual fragments in a 50-g sample. These datasuggest that one would not be able to determinewhether the number of fragments in a sample wasabove or below the threshold if NIRS-predicted frag-ment counts were between �50 and 105 fragments;similarly, NIRS-predicted fragments resulting frommilling wheat infested with adult S. oryzaewere highlyvariable. Because NIRS data are quick and inexpen-sive, this situation could be improved by analyzing afew more samples from that lot of ßour to try todetermine the actual fragment level. There are cur-rently no data on processing ßour samples with morethan one insect species present, but this may not be aserious issue because NIRS can differentiate amongwhole adults (Dowell et al. 1999). Development of acalibration model is critical for NIRS to be useful forestimating insect fragments in ßour samples. Severalcommercial laboratories have the expertise to developthese models. To address the need for continuousNIRS calibration, at least one manufacturer of NIRspectrometers offers an internet-enabled analyticalsystem where an expert remotely monitors and con-tinuously calibrates the system (Dominy and Buckley2003). A second beneÞt of this service is that theexpert could interpret each sample on a per fee basis.

Although incoming wheat will likely contain mixedlife stages, we hypothesize that methods of quantify-ing insect fragments should be focused on fragmentsproduced by small- and medium-sized larvae. Incom-ing wheat in most commercial facilities is cleaned withentoleters (impact machines), scalpers, and aspirationbefore milling (Mills and Pedersen 1992). Wheatthrown by centrifugal force inside an entoleter duringthe cleaning process would likely break hollow ker-nels, such as those chewed by large larvae. Data pre-sented here show that the cavity size inside a wheatkernel remained static from the late larval throughpreemergent adult life stages, suggesting that estima-tion of fragments from these life stages may not beimportant. Broken kernels and newly exposed insectsresulting from entoleters would be easily separatedfrom the sound kernels; therefore, no insect fragmentsfrom these sources would be evident in the Þnal mill-stream. Unfortunately,notallmillsuse this typeofclean-ing equipment and there are no hard data on entoleter

use during the grain cleaning process at domestic mills.Rigorous scientiÞc study on the performance of entol-eters is needed to test this hypothesis.

Acknowledgments

We thank Laura McLaughlin at the Grain Quality andStructure Research Unit (Grain Marketing and ProductionResearch Center) for milling the wheat samples, and AnnRedmond for conducting the acid hydrolysis procedure oneach ßour sample. Jaclyn Rowan, Orrin Holle, Caleb Doll,Matt Kennedy, Luis Sainz, and Susan Romero provided ex-cellent technical support. The FT-NIRS spectrometer wasprovided by the QTA group of the Cognis Corp. (Cincinnati,OH). We greatly appreciate Terry Arbogast and ThomasPearson who provided helpful comments to an earlier manu-script draft. This work was funded with partial support fromUSDAÐCSREES (RAMP) under agreement 00-511-01-9674.

References Cited

[AACC] American Association of Cereal Chemists. 1995.Approved methods of the American Association of CerealChemists, 9th ed. American Association of Cereal Chem-ists, St. Paul, MN.

[AOAC] Association of Official Analytical Chemists. 1996.16.5.11 AOAC OfÞcial method 972.32, Light Þlth (pre- andpost-milling) in ßour (white), p. 18. In OfÞcial methods ofanalysis of Association of OfÞcial Analytical Chemists In-ternational, 16th ed. Association of OfÞcial AnalyticalChemists International, Gaithersburg, MD.

Armstrong, P. R., E. B. Maghirang, F. Xie, and F. E. Dowell.2006. Comparison of dispersive and Fourier-transformNIR instruments for measuring grain and ßour attributes.Appl. Eng. Agric. 22: 453Ð457.

Atui, M. B., P. W. Flinn, S.M.N. Lazzari, and F. A. Lazzari.2007. Detection ofRhyzoperthadominica larvae in storedwheat using ELISA: the impact of myosin degradationfollowing fumigation. J. Stored Prod. Res. 43: 156Ð159.

Bernton, H. S., and H. Brown. 1967. Insects as potentialsources of ingestant allergens. Ann. Allergy 25: 381Ð387.

Birch, L. C. 1945. The inßuence of temperature, humidity,and density on the oviposition of the small strain ofCalandra oryzae L. and Rhizopertha dominica Fab.(Coleoptera). Aust. J. Exp. Biol. Med. Sci. 23: 197Ð203.

Brader, B. 1997. Are food sanitation assays meaningful?Cereal Foods World 42: 759Ð760.

Brader, B., R. C. Lee, R. Plarre, W. Burkholder, G. B. Kitto,C. Kao, L. Polston, E. Dorneanu, I. Szabo, B. Mead, B.Rouse, D. Sullins, and R. Denning. 2002. A comparisonof screening methods for insect contamination in wheat.J. Stored Prod. Res. 38: 75Ð86.

Chapman, R. F. 1982. The insects: structure and function,3rd ed. Harvard University Press, Cambridge, MA.

Dominy, S. F., and E. Buckley. 2003. Alternative analysis.World Grain, June: 66Ð69.

Dowell,F.E., J.E.Throne, andJ.E.Baker. 1998. Automatednondestructive detection of internal insect infestation ofwheat kernels by using near-infrared spectroscopy. J.Econ. Entomol. 91: 899Ð904.

Dowell, F. E., J. E. Throne, D. Wang, and J. E. Baker. 1999.Identifying stored grain insects using near-infrared spec-troscopy. J. Econ. Entomol. 92: 165Ð169.

Draper, N. R., and H. Smith. 1981. Applied regression anal-ysis, 2nd ed. Wiley, New York.

[FDA] Food and Drug Administration. 1997. 578.450Wheat ßourÐadulteration with insect fragments and


rodent hairs (CPG 7104.06). Compliance Policy GuidesManual, Chapter 5, Foods, color and cosmetics, Subchapter 578Ðprocessed grain. (http://www.fda.gov/ora/compliance_ref/cpg/default.htm#1998).

Galactic Industries Corporation. 1996. PLSplus/IQ forGrams/32 and Grams/386. Galactic Industries Corpora-tion, Salem, NH.

Gentry, J. W., K. L. Harris, and J. W. Gentry, Jr. 1991.Microanalytical entomology for food sanitation con-trol. Association of OfÞcial Analytical Chemists, Mel-bourne, FL.

Haff, R. P., and D. C. Slaughter. 2004. Real-time x-ray in-spection of wheat for infestation by the granary weevil,Sitophilus granarius (L.). Trans. ASAE 47: 531Ð537.

Harris, K. L., J. F. Nicholson, L. K. Randolph, and J. L.Trawick. 1952. An investigation of insect and rodentcontamination of wheat and wheat ßour. J. Am. Off. Anal.Chem. 35: 115Ð157.

Karunakaran, C., D. S. Jayas, and N.D.G. White. 2003. Softx-ray inspection of wheat kernels infested by Sitophilusoryzae. Trans. ASAE 46: 739Ð745.

Kitto, G. B. 1991. A new rapid biochemical technique forquantitating insect contamination in grain. Bull. Assoc.Oper. Millers, March: 5835Ð5838.

Kurtz, O. L. 1956. IdentiÞcation of stored products insectsby the micromorphology of the exoskeleton. III. Identi-Þcation of the larval fragments and their signiÞcance insanitation analyses. J. Assoc. Off. Anal. Chem. 39: 990Ð1014.

Kurtz, O. L., and T. H. McCormack. 1965. Comparison ofAACC and AOAC methods for extraneous materials inßour. J. Assoc. Off. Anal. Chem. 48: 554Ð558.

Lockey, K. H. 1988. Lipids of the insect cuticle: origin, com-position and function. Comp. Biochem. Physiol. B 89:595Ð645.

Maghirang, E. B., F. E. Dowell, J. E. Baker, and J. E. Throne.2003. Automated detection of single wheat kernels con-taining live or dead insects using near-infrared reßec-tance spectroscopy. Trans. ASAE 46: 1277Ð1282.

Mills, R., and J. Pedersen. 1992. A ßour mill sanitation man-ual. Eagan Press, St. Paul, MN.

Murray, I., and P. C.Williams. 1990. Chemical principles ofnear-infrared technology, pp. 17Ð34. InP. C. Williams andK. H. Norris [eds.], Near-infrared technology in the ag-ricultural and foods industries. American Association ofCereal Chemists, St. Paul, MN.

Neter, J.,M.H.Kutner,C. J.Nachtsheim, andW.Wasserman.1996. Applied linear regression models, 3rd ed. Irwin,Chicago, IL.

Osborne, B. G., T. Fearn, and P. H. Hindle. 1993. PracticalNIR spectroscopy with applications in food and beverageanalysis. Longman ScientiÞc & Technical, Harlow,United Kingdom.

Perez-Mendoza, J., J. E.Throne,F.E.Dowell, and J.E.Baker.2003. Detection of insect fragments in wheat ßour bynear-infrared spectroscopy. J. Stored. Prod. Res. 39: 305Ð312.

Perez-Mendoza, J., P. W. Flinn, J. F. Campbell, D. W.Hagstrum, and J. E. Throne. 2004. Detection of stored-

grain insect infestation in wheat transported in railroadhopper-cars. J. Econ. Entomol. 97: 1474Ð1483.

Perez-Mendoza, J., J. E. Throne, E. B. Maghirang, F. E.Dowell, and J. E. Baker. 2005. Insect fragments in ßour:relationship to lesser grain borer (Coleoptera: Bostrichi-dae) infestation level in wheat and rapid detection usingnear-infrared spectroscopy. J. Econ. Entomol. 98: 2282Ð2291.

Quinn, F. A., W. Burkholder, and G. B. Kitto. 1992. Immu-nological technique for measuring insect contaminationof grain. J. Econ. Entomol. 85: 1463Ð1470.

Rees, D. 2004. Insects of Stored Products. CSIRO Publish-ing, Collingwood, VIC, Australia.

Ridgway, C., and J. Chambers. 1996. Detection of externaland internal insect infestation in wheat by near-infraredreßectance spectroscopy. J. Sci. Food Agric. 71: 251Ð264.

Russell, G. E. 1988. Evaluation of four analytical methods todetect weevils in wheat: granary weevil, Sitophilus gra-narius (L.), in soft white wheat. J. Food Prot. 51: 547Ð553.

Sachdeva, A. S. 1978. Effect of infestation stage, form andtreatment on fragment count in ßour. M.S. thesis, KansasState University, Manhattan, KS.

SAS Institute. 1999. SAS/STAT userÕs guide, version 8. SASInstitute, Cary, NC.

Schatzki, T. F., E. K. Wilson, G. B. Kitto, P. Behrens, and I.Heller. 1993. Determination of hidden Sitophilus grana-rius (Coleoptera: Curculionidae) in wheat by myosinELISA. J. Econ. Entomol. 86: 1584Ð1589.

Sears, R. G., J. M. Moffatt, T. J. Martin, T. S. Cox, R. K.Bequette, S. P. Curran, O. K. Chung, W. F. Herr, J. H.Long, and M. D. Witt. 1997. Registration of ÔJaggerÕwheat. Crop Sci. 37: 1010.

Sharifi, S., and R. B. Mills. 1971. Developmental activitiesand behavior of the rice weevil inside wheat kernels. J.Econ. Entomol. 64: 1114Ð1118.

Shenk, J. S., J. J. Workman, Jr., andM. O.Westerhaus. 1992.Application of NIR spectroscopy to agricultural products,pp. 383Ð431. In D. A. Burns and E. W. Ciurczak [eds.],Handbook of near-infrared analysis. Marcel Decker, Inc.,New York.

Storey, C. L., D. B. Sauer, O. Ecker, and D. W. Fulk. 1982.Insect infestation in wheat and corn exported from theUnited States. J. Econ. Entomol. 75: 827Ð832.

Systat Software Inc. 2002. TableCurve 2D version 5.01.Systat Software Inc., Point Richmond, CA.

Thermo Electron Corp. 2006. Grams32/AI version 8.0,PLSPlus/IQ chemometric application. Thermo ElectronCorp., Salem, NH.

Toews, M. D., T. C. Pearson, and J. F. Campbell. 2006. Im-aging and automated detection of Sitophilus oryzae(Coleoptera: Curculionide) in pupae in hard red winterwheat. J. Econ. Entomol. 99: 583Ð592.

Wingfield, J., and J. Pedersen. 1985. Insect fragment levelsin ßour milled from selected commercial wheat blends.Bull. Assoc. Oper. Millers April 1985.

Received 28 February 2007; accepted 17 July 2007.


Rapid Assessment of Insect Fragments in Flour Milled from Wheat Infested with Known Densities of...

Documents

Transcript of Rapid Assessment of Insect Fragments in Flour Milled from Wheat Infested with Known Densities of...