HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Free Via Open Access Abstract Full Text (PDF) Interpretive Summary Alert me when this article is cited Alert me if a correction is posted Services Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Google Scholar Articles by Kaylegian, K. E. Articles by Barbano, D. M. PubMed PubMed Citation Articles by Kaylegian, K. E. Articles by Barbano, D. M.

Impact of fatty acid composition on the accuracy of mid-infrared fat analysis of farm milks¹

K. E. Kaylegian^*, D. A. Dwyer, J. M. Lynch^*, D. E. Bauman, J. R. Fleming and D. M. Barbano^*,2

^* Northeast Dairy Foods Research Center, Department of Food Science, and
Department of Animal Science, Cornell University, Ithaca, NY 14853
USDA, Agricultural Marketing Service Southwest Milk Marketing Area, PO Box 110939, Carrollton, TX 75011

² Corresponding author: dmb37{at}cornell.edu

	ABSTRACT

TOP FOOTNOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION CONCLUSIONS ACKNOWLEDGMENTS REFERENCES

 
Our objective was to determine whether data from a previous study using model milk emulsions to characterize the influence of variation in fatty acid chain length and unsaturation on mid-infrared (MIR) fat predictions could be used to identify a strategy to improve the accuracy of MIR fat predictions on a population of farm milks with a wide variation in fatty acid chain length and unsaturation. The mean fatty acid chain length for 45 farm milks was 14.417 carbons, and the mean unsaturation was 0.337 double bonds per fatty acid. The range of fatty acid chain lengths across the 45 farm milks was 1.23 carbons, and the range in unsaturation was 0.167 double bonds per fatty acid. Fat B (absorbance by the carbon-hydrogen stretch) MIR predictions increased and fat A MIR (absorbance by the ester carbonyl stretch) predictions decreased relative to reference chemistry with increasing fatty acid chain length. When the fat B MIR fat predictions were corrected for sample-to-sample variation in unsaturation, the positive correlation between fat B and fatty acid chain length increased from a coefficient of determination of 0.42 to 0.89. A 45:55 ratio of fat B corrected for unsaturation and fat A gave a smaller standard deviation of the difference between MIR prediction and reference chemistry than any ratio of the fat B (without correction for unsaturation) and fat A or either fat B or fat A alone. This demonstrates the technical feasibility of this approach to improve MIR testing accuracy for fat, if a simple procedure could be developed to determine the unsaturation of fat in milk rapidly and to correct the fat B reading for the effect of unsaturation before being combined with fat A.

Key Words: infrared milk analysis • fatty acid composition

	INTRODUCTION

TOP FOOTNOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION CONCLUSIONS ACKNOWLEDGMENTS REFERENCES

 
In mid-infrared (MIR) milk analysis, fat has historically been measured using the absorbance of infrared light at the fat B wavelength (3.48 µm; C-H stretch) and the fat A wavelength (5.73 µm; C=O stretch). The fat A measurement cannot compensate for variation among milks in fatty acid chain length and assumes that fat in all milks has the same chain length (Biggs and McKenna, 1989). The fat B measurement compensates for differences in fatty acid chain length (Biggs and McKenna, 1989), but assumes that the fat in all milks has the same amount of unsaturation. Variation among milks in unsaturation of fat will cause errors in the fat B prediction of milk fat. Variation in fatty acid composition of milk fat from farm to farm is created by differences in feeding practices. In general, feeding strategies that increase the fat content of the diet of the cow will cause changes in fatty acid composition, particularly if the fat is able to bypass hydrogenation in the rumen.

It is well documented that when the fatty acid chain length in milk fat increased, values for fat A are lower than with reference chemistry because the fat A measurement responds to the number of ester linkages (Dunkley et al., 1977; Franke et al., 1977; Stegeman et al., 1991). Kaylegian et al. (2009) reported that increasing the chain length increased the difference (i.e., MIR minus reference chemistry) between the MIR fat B prediction and reference chemistry by 0.0429% and the fat A prediction by –0.0566% fat per unit increase in carbon number per 1% fat. Increasing unsaturation decreased the difference (i.e., MIR minus reference chemistry) between the MIR prediction of fat and reference chemistry for fat B by –0.4021% and increased fat A by 0.0291% fat per unit increase in double bonds per 1% fat (Kaylegian et al., 2009). Clearly, both chain length and unsaturation have a large influence on fat B, but separately, changes in unsaturation and chain length cause the fat B and fat A MIR fat predictions to deviate from reference chemistry in opposite directions.

Milks from individual farms can be sorted by the magnitude of the difference (both positive and negative) between the MIR fat B and fat A readings. Divergence of fat B and fat A readings on an unknown sample indicates that the unknown milk being tested contains fat that differs from the mean chain length and unsaturation of the fat in the calibration milks used to adjust the secondary slope and intercept. Therefore, milks that have more divergence between MIR fat B and fat A predictions are likely to have a fatty acid composition that differs significantly from the mean fatty acid composition of the milk used to calibrate that instrument. Our practical experience in routine testing of large numbers of samples in multiple laboratories within the USDA Federal Milk Markets is that every instrument calibrated with the same milks that has intercorrection factors controlled by the same system of samples will have the same differences between fat B and fat A readings on the same unknown milks (data not shown).

Biggs and McKenna (1989) concluded that a 73:27 ratio of fat B and fat A gave better agreement with reference chemistry than either fat B or fat A. Biggs and McKenna (1989) justified this ratio based on the explanation that triglycerides in milk are approximately 27% by weight glycerol and 73% by weight fatty acids. This explanation is not correct. Glycerol as a percentage of the molecular weight of a milk fat trigylceride with a mean fatty acid chain length of 14.5 carbons and a mean unsaturation of 0.33 double bonds per fatty acid is approximately 11%, not 27%. However, it remains that Biggs and McKenna (1989) found that the 73:27 ratio gave a lower standard deviation of the difference (SDD) between reference chemistry and MIR-predicted chemistry than either fat B or fat A alone. Our objective was to determine whether data from a previous study using model milk emulsions to characterize the influence of variation in fatty acid chain length and unsaturation on MIR fat predictions could be used to identify a strategy to improve the accuracy of MIR fat predictions on a population of farm milks with a wide variation in fatty acid chain length and unsaturation.

	MATERIALS AND METHODS

TOP FOOTNOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION CONCLUSIONS ACKNOWLEDGMENTS REFERENCES

 
Experimental Design
Milks were obtained from individual farms from 3 different areas of the United States (New York, Texas, and Vermont) during the summer of 2005. Our goal was to select a group of farms with milks having a wide range of fatty acid chain length and unsaturation. Milks were tested for fat, true protein, and anhydrous lactose with a Fourier transform (FT) MIR milk analyzer with the fat A, fat B, protein, and lactose using the optimized virtual filter wavelengths reported previously (Kaylegian et al., 2009). The differences in MIR fat predictions (i.e., fat B minus fat A) were used as the basis for selection of farms. Milks that have a fatty acid chain length and unsaturation similar to the milks used to calibrate the instrument will have fat B and fat A predictions that agree. The more the fatty acid chain length and unsaturation of milk differs from that of the calibration milks, the larger will be the difference between the MIR fat B and fat A predictions. The observed differences between fat B and fat A can be either positive or negative. The USDA Federal Milk Market Laboratories selecting the milks were asked to provide milks that had as wide a range of fat B minus fat A values as possible.

The farm milks were analyzed by chemical methods for fat, true protein, lactose, FFA content, and fatty acid composition. The mean fatty acid chain length and unsaturation for each milk were calculated from the fatty acid composition. The variation in MIR fat B and fat A predictions for the farm milks that was due to fatty acid chain length and unsaturation was compared with data reported for a model system (pure monoacid triglycerides and pasteurized skim milk) in which the chain length and unsaturation were varied systematically (Kaylegian et al., 2009). Approaches to improve the accuracy of MIR fat prediction using calculations to compensate for variation in fatty acid chain length and unsaturation were evaluated based on the data from 45 farms.

Source of Farm Milk
Milks were chosen from farms (by 3 USDA Federal Milk Market Laboratories) that were known to have a large difference between the MIR fat A and fat B readings. Milks obtained from farms in New York (n = 8), Texas (n = 27), and Vermont (n = 10) were packed in wet ice and shipped overnight to Cornell University for analysis.

Chemical Analyses
Chemistry values for each milk were determined using AOAC (2000) methods for fat by modified Mojonnier ether extraction (method 989.05; chapter 33.2.26), true protein by Kjeldahl nitrogen analysis (method 991.22; chapter 33.2.13), and lactose by enzymatic analysis (method 2006.06, 33.3.67). The FFA content was determined using a modified copper soap method (expressed as mEq of FFA/kg of milk; Ma et al., 2003).

Fatty Acid Composition
The fatty acid composition of each farm milk was determined in duplicate using the method described by Lynch et al. (2005). An example of the calculation to convert the GLC data from a fatty acid methyl ester (FAME) basis to a fatty acid basis and the calculation of the mean chain length and unsaturation are shown in Table 1. The calculation began with the relative area percentage of GLC output on a FAME basis. A butter oil reference standard (CRM164, Commission of the European Communities, Community Bureau of Reference, Brussels, Belgium) was analyzed under the same conditions as the milk fat from the farm milks to obtain the GLC area percentage for the standard. The ratio of the GLC area percentage divided by the known weight percentage of the standard was used to calculate a recovery factor for each FAME. These recovery factors were multiplied by the GLC FAME area percentage to calculate the recovery corrected percentage FAME for each fatty acid. The individual recovery adjusted FAME were normalized to a sum of 100% for the known FAME in each sample. The adjusted FAME data were converted to a fatty acid basis by multiplying by a correction factor (molecular weight of each fatty acid divided by the molecular weight of its FAME) to remove the molecular weight of the methyl ester group for each fatty acid, and were then normalized to a sum of 100% to obtain the relative fatty acid composition based on the fatty acids not FAME. The relative fatty acid composition was used to calculate the mean fatty acid chain length and unsaturation by converting each fatty acid to a millimolar basis and weighting each fatty acid by multiplying by either the number of carbons or the number of double bonds in the fatty acid chain, respectively. The mean fatty acid chain length was calculated (Table 1) by dividing the total fatty acid concentration weighted for carbon number (i.e., 6,177.071 mmol) by the total fatty acid concentration (i.e., 426.053 mmol). The mean unsaturation was calculated by dividing the total fatty acid concentration weighted for the number of double bonds (i.e., 139.351 mmol) by the total fatty acid concentration (i.e., 426.053 mmol).

Statistical Analyses
All regression analyses were performed using the PROC REG function in SAS (version 8e, SAS Inst. Inc., Cary, NC). A regression analysis was performed to determine whether there was a correlation between fatty acid chain length and unsaturation, between reference chemistry fat, true protein, or lactose of the milks and the fatty acid chain length or unsaturation, between the FFA content of the milk and MIR-predicted fat (B and A) or protein, and between fat B minus ether, fat A minus ether, or fat B minus fat A and fatty acid chain length.

	RESULTS AND DISCUSSION

TOP FOOTNOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION CONCLUSIONS ACKNOWLEDGMENTS REFERENCES

 
Composition of Farm Milk—Chemical Analysis
The fatty acid composition for farm milks from New York, Texas, and Vermont with the minimum and maximum fatty acid chain length and unsaturation are shown in Table 2. In all 3 states, the milk with the shortest fatty acid chain length also had the smallest number of double bonds per fatty acid (Table 2). It is interesting to note that in New York and Vermont, the milk with the longest fatty acid chain had the largest number of double bonds per fatty acid (Table 2). Three out of 3 farms that had milks with the shortest chain length also had milks with the lowest number of double bonds, and 2 out of the 3 farms that had milks with the longest chain length had milks with the highest number of double bonds, indicating that there should be a positive correlation between fatty acid chain length and unsaturation.

View larger version (18K): [in this window] [in a new window]   Figure 1. Change in unsaturation (double bonds per fatty acid) as a function of fatty acid chain length (carbons) for 45 farm milks from New York (n = 8), Texas (n = 27), and Vermont (n = 10).

The difference between MIR fat B prediction and reference chemistry was positively correlated (P < 0.001) with chain length (Figure 2a), and the difference between fat A prediction and reference chemistry was negatively correlated (P < 0.001) with chain length (Figure 2b). These observations are consistent with those reported by Kaylegian et al. (2009), in which the fat B readings increased and the fat A readings decreased relative to reference chemistry as fatty acid chain length increased.

View larger version (39K): [in this window] [in a new window]   Figure 2. Mid-infrared fat prediction minus fat reference chemistry as a function of fatty acid chain length for 45 farm milks from New York (n = 8), Texas (n = 27), and Vermont (n = 10): (a) fat B; (b) fat A.

View larger version (20K): [in this window] [in a new window]   Figure 3. Mid-infrared predicted fat B minus fat A values as a function of fatty acid chain length for 45 farm milks from New York (n = 8), Texas (n = 27), and Vermont (n = 10).

Improving the MIR Prediction of Milk Fat Content by Adjustment of Fat B MIR-Predicted Values for a Change in Unsaturation
Using model emulsions of skim milk and pure monoacid triglycerides, Kaylegian et al. (2009) determined that fat B readings decreased by –0.4021% per unit increase in unsaturation (double bonds per fatty acid) per 1% fat. The fat B readings from the 45 farm milks (Table 3) were corrected for the effect of variation in unsaturation by using the following equation: [y = (fat B) + {0.4021 x [(unsaturation of milk tested) – (mean unsaturation of the calibration set)]} x (fat B)]. The value used for mean unsaturation of the modified milk calibration set was determined by running a regression analysis on fat A minus ether (y) as a function of number of double bonds (x) for the 45 farm milks. The calculated value (using the regression equations) for double bonds per fatty acid for which the difference between fat A and reference chemistry was zero corresponded to 0.337 double bonds per fatty acid. This value was close to the average unsaturation (0.327) of the calibration milks used in this study that was measured independently by GLC and that is reported in Table 1.

The correlation of fat B minus reference chemistry with mean fatty acid chain length was improved from a coefficient of determination of 0.42 (Figure 2a) to 0.89 (Figure 4) by correcting fat B for the variation in total unsaturation. This major improvement in the correlation of fat B data with a change in fatty acid chain length demonstrated that the correction greatly reduced the effect of unsaturation on the fat B results. The factor (i.e., –0.4021%) used for correction, based on the study of model milk systems (Kaylegian et al., 2009), was varied to higher and lower values in the correction of fat B for unsaturation, but it did not improve the coefficient of determination of the relationship between fat B corrected for unsaturation minus reference chemistry and fatty acid chain length (data not shown). This indicated that the correction value for removal of the effect of variation in fatty acid unsaturation at the fat B wavelength determined with model milk emulsions that varied only in unsaturation (Kaylegian et al., 2009) worked well with the farm milks that varied in chain length and unsaturation simultaneously.

View larger version (22K): [in this window] [in a new window]   Figure 4. Mid-infrared predicted fat B corrected for the effect of unsaturation (US) minus fat reference chemistry as a function of fatty acid chain length for 45 farm milks from New York (n = 8), Texas (n = 27), and Vermont (n = 10).

The range of wavelengths used in FT MIR milk analyzers is from 3.3 to 10 µm (3,000 to 1,000 cm^–1). It is possible that data from areas of the MIR spectrum other than the fixed-filter wavelengths could be used with multivariate statistics to achieve better accuracy of MIR fat predictions in milk. To explore the feasibility of this option, fat was extracted from milks that differed greatly in unsaturation, and MIR spectra were collected from 3.3 to 10 µm using an FT MIR spectrometer (Nicolet Instrument Corp., Madison, WI) with a ZnSe attenuated total reflectance crystal accessory (Thermo Spectra-Tech, Madison, WI) with a resolution of 4 cm^–1 and 256 scans per sample using OMNIC software (OMNIC E.S.P., Nicolet Instrument Corp.). The MIR spectra of the pure milk fats differed slightly from each other at the fat B (C-H stretch) and fat A (C=O stretch) wavelengths [3.51 and 5.72 µm (2,851 and 1,748 cm^–1), respectively]. Kaylegian et al. (2009) evaluated milk spectra from 3 to 10 µm for a series of milk emulsions that differed only in unsaturation and reported that sequential differences in the MIR absorbance of these samples were not present at wavelengths other than fat B and fat A wavelengths. This may indicate that multivariate approaches such as classical least squares regression, inverse least squares regression, partial least squares regression, and principal component regression could have a difficult time correctly compensating for variation in fatty acid chain length and unsaturation beyond what has been done with the traditional fixed-filter approach using intercorrection factors. In fact, Luinge et al. (1993) evaluated 55 raw producer milks for calibration and validation performance, and concluded that there were no significant differences between reference chemistry minus FT MIR fat predictions using multivariate approaches and reference chemistry minus MIR fat predictions using a fixed-filter approach. Similar conclusions were made by van de Voort et al. (1992). Thus, although multivariate calibration approaches would seem to have the potential to provide improved accuracy of fat testing by MIR, the reality is that no one has been able to demonstrate a significant performance improvement with multivariate approaches in comparison with a properly run fixed-filter approach.

	CONCLUSIONS

TOP FOOTNOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION CONCLUSIONS ACKNOWLEDGMENTS REFERENCES

 
The mean fatty acid chain length was 14.417 carbons and the mean unsaturation was 0.337 double bonds per fatty acid for 45 farm milks collected from New York, Texas, and Vermont. The range of fatty acid chain lengths was 1.23 carbons and the range in unsaturation was 0.167 double bonds per fatty acid among 45 farms. Fat B MIR fat predictions increased and fat A MIR predictions decreased relative to reference chemistry with increasing fatty acid chain length. When the fat B MIR fat predictions were corrected for sample-to-sample variation in unsaturation (measured by GLC analysis of each milk), the positive correlation between fat B and fatty acid chain lengths increased from a coefficient of determination of 0.42 to 0.89. A 45:55 ratio of fat B corrected for unsaturation and fat A gave a smaller SDD between MIR prediction and reference chemistry fat values (0.0160) than any ratio of the fat B (without correction for unsaturation) and fat A or either fat B or fat A alone. The present study demonstrated the technical feasibility of this approach. If a rapid procedure could be developed to determine the unsaturation of fat in milk and could be used in real-time testing to correct the MIR fat B readings, improved accuracy of MIR milk fat predictions could be achieved.

	ACKNOWLEDGMENTS

TOP FOOTNOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION CONCLUSIONS ACKNOWLEDGMENTS REFERENCES

 
The authors thank Cornell University staff members Maureen Chapman, Jessica Mallozzi, and Karen Wojciechowski for their technical assistance, and Lloyd Metzger (South Dakota State University, Brookings) for FT MIR spectra of milk fats of different unsaturation. The authors thank the Test Procedures Committee of the USDA, Dairy Programs, Federal Milk Markets, for their financial support of this research.

	FOOTNOTES

TOP FOOTNOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION CONCLUSIONS ACKNOWLEDGMENTS REFERENCES

 
¹ Use of names, names of ingredients, and identification of specific models of equipment is for scientific clarity and does not constitute any endorsement of product by authors, Cornell University, or the Northeast Dairy Foods Research Center.

Received for publication November 18, 2008. Accepted for publication January 8, 2009.

	REFERENCES

TOP FOOTNOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION CONCLUSIONS ACKNOWLEDGMENTS REFERENCES

 


AOAC. 2000. Official Methods for Analysis. AOAC Int., Gaithersburg, MD.

Biggs, D. A. and McKenna, D.. 1989. Alternative methods for infrared analysis of fat in milk: Interlaboratory study. J. AOAC 72:724–734.

Dunkley, W. L., Franke, A. A. and Smith, L. M.. 1977. Determining fat, protein, and lactose in milk from cows fed formaldehyde protected sunflower-soybean supplement. J. Dairy Sci. 60:45–52.[Abstract/Free Full Text]

Franke, A. A., Dunkley, W. L. and Smith, N. E.. 1977. Influence of feeding protected tallow on accuracy of analysis by infrared milk analyzer. J. Dairy Sci. 60:1870–1874.[Abstract/Free Full Text]

Franke, A. A., Dunkley, W. L. and Smith, N. E.. 1981. Influence of dairy rations with whole cottonseed or extruded soybeans on accuracy of analysis for milk components. J. Dairy Sci. 64:2185–2191.[Abstract/Free Full Text]

Kaylegian, K. E., Houghton, G. E., Lynch, J. M., Fleming, J. R. and Barbano, D. M.. 2006. Calibration of infrared milk analyzers: Modified milk versus producer milk. J. Dairy Sci. 89:2817–2832.[Abstract/Free Full Text]

Kaylegian, K. E., Lynch, J. M., Fleming, J. R. and Barbano, D. M.. 2009. Influence of fatty acid chain length and unsaturation on mid-infrared milk analysis. J. Dairy Sci. 92:2485–2501.[Abstract/Free Full Text]

Kaylegian, K. E., Lynch, J. M., Fleming, J. R. and Barbano, D. M.. 2007. Lipolysis and proteolysis of modified and producer milk used for calibration of mid-infrared milk analyzers. J. Dairy Sci. 90:602–615.[Abstract/Free Full Text]

Luinge, H. J., Hop, E., Lutz, E. T. G., van Hemert, J. A. and de Jong, E. A. M.. 1993. Determination of the fat, protein and lactose content of milk using Fourier transform infrared spectrometry. Anal. Chim. Acta 284:419–433.[CrossRef]

Lynch, J. M., Barbano, D. M., Bauman, D. E., Hartnell, G. F. and Nemeth, M. A.. 1992. Effect of a prolonged release formulation of N-methionyl bovine somatotropin (sometribove) on production and composition of milk fat. J. Dairy Sci. 75:1794–1809.[Abstract]

Lynch, J. M., Barbano, D. M., Schweisthal, M. and Fleming, J. R.. 2006. Precalibration evaluation procedures for mid-infrared milk analyzers. J. Dairy Sci. 89:2761–2774.[Abstract/Free Full Text]

Lynch, J. M., Lock, A. L., Dwyer, D. A., Noorbakhsh, R., Barbano, D. M. and Bauman, D. E.. 2005. Flavor and stability of pasteurized milk with elevated levels of conjugated linoleic acid and vaccenic acid. J. Dairy Sci. 88:489–498.[Abstract/Free Full Text]

Ma, Y., Barbano, D. M. and Santos, M.. 2003. Effect of CO₂ addition to raw milk on proteolysis and lipolysis at 4°C. J. Dairy Sci. 86:1616–1631.[Abstract/Free Full Text]

Robertson, N. H., Dixon, A., Nowers, J. H. and Brink, D. P. S.. 1981. The influence of lipolysis, pH and homogenization on infra-red readings for fat, protein and lactose. S. Afr. J. Dairy Technol. 13:3–7.

Sjaunja, L.-O. 1984. Studies on milk analyses of individual cow milk samples. II. Factors affecting milk analyses by infrared techniques under laboratory conditions. Acta Agric. Scand. 34:260–272.

Stegeman, G. A., Baer, R. J., Schingoethe, D. J. and Casper, D. P.. 1991. Influence of milk fat higher in unsaturated fatty acids on the accuracy of milk fat analyses by the mid-infrared spectroscopic method. J. Food Prot. 54:890–893.

van de Voort, F. R., Kermasha, S., Smith, J. P., Mills, B. L. and Ng-Kwai-Hang, K. F.. 1987. A study of the stability of record of performance milk samples for infrared milk analysis. J. Dairy Sci. 70:1515–1523.[Abstract/Free Full Text]

van de Voort, F. R., Sedman, J., Emo, G. and Ismail, A. A.. 1992. Assessment of Fourier transform infrared analysis of milk. J. AOAC 75:780–785.

This Article Free Via Open Access Abstract Full Text (PDF) Interpretive Summary Alert me when this article is cited Alert me if a correction is posted Services Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Google Scholar Articles by Kaylegian, K. E. Articles by Barbano, D. M. PubMed PubMed Citation Articles by Kaylegian, K. E. Articles by Barbano, D. M.