Changes in human meibum lipid with meibomian gland dysfunction using principal component analysis

Abstract

Changes in the phase transition temperatures and conformation of human meibum lipid with age and meibomian gland dysfunction have been quantified but with analysis of less than 1% of the infrared spectral range. The remaining 99% of the spectral range was analyzed with principal component analysis (PCA) and confirms our previous studies and reveal further insights into changes that occur in meibum with age and disease. Infrared spectra of meibum from 41 patients diagnosed with meibomian gland dysfunction (Md) and 32 normal donors (Mn) were measured. Principal component analysis (PCA) was used to quantify the variance among the spectra and meibum protein was quantified using the infrared carbonyl and amide I and II bands. A training set of spectra was used to discriminate between Mn and Md with an accuracy of 93%. This shows that certain spectral regions (eigenvectors) contain compositional and structural information about the changes that occur with the principal component (variable), meibomian gland dysfunction. The spectral features of the major eigenvector indicate that Md contains more protein and relatively less CH(3) and cis = CH band intensity compared to Mn. The amount of protein was confirmed from relative infrared band intensities. Our study supports the idea that compositional differences result in meibum that is less fluid and more viscous with meibomian gland dysfunction so that less lipid flows out of the meibomian gland orifice as observed clinically. This study also demonstrates the power of the combination of infrared spectroscopy and PCA as a diagnostic tool that discriminates between Mn and Md.

Figure 1

Mahalanobis groupings of standards for eigenvectors 1 versus 2 calculated by principal component analysis for A) =CH/CH₂, standards in Table 1, B) CH₃/CH₂, standards in Table 1. C) Protein standards composed of lysozyme and wax.

Figure 2

Principal component analysis was used on a training set of standards in Table 1 and lysozyme protein mixtures to predict A) =CH/CH₂, B) CH₃/CH₂, and C) protein amide/wax C=O. The actual and predicted values are graphed.

Figure 3

Average Infrared spectra of human meibum (□ □ , bold) Md. (□ □, gray) Mn.

Figure 4

‘PRESS’ plots showing the optimum factors to use (arrow). A) Age component. B) Meibomian Gland Dysfunction and Normal Score constituent.

Figure 5

The % of total variance for each eigenvector (factor) related to the ‘meibomian gland dysfunction /normal score’ constituent.

Figure 6

A) A training set was used to discriminate between the spectra of meibum from normal donors and the spectra of meibum from donors with meibomian gland dysfunction. This shows that the infrared spectra must contain compositional and structural information about the changes that occur with meibomian gland dysfunction. A score above 50 (vertical line) is considered a sample with meibomian gland dysfunction. B) The same training set used for 4A was used to predict the age of the donors with meibomian gland dysfunction within ± 20 y at a 95 % confidence limits. (•) Mn group from: Borchman et al 2010. (○) Md group. (□□ □) Linear regression fit and (---) 95 % confidence limits of Mn group data from: Borchman et al.

Figure 7

Mahalanobis groupings for select pairs of eigenvectors from the principal component analysis of meibomian dysfunction scores.

Figure 8

Infrared spectra of A) Cholesterol palmitate. B) Oleyloleate C) Eigenvector 1 for the constituent ‘meibomian gland dysfunction /normal score’. D) egg white lysozyme. Numbers are infrared band assignments: 1) Amide A, 3282 cm⁻¹ 2) Amide 1, 1656 cm⁻¹ 3) Amide 2, 1525 cm⁻¹ 4) ,1380 cm⁻¹ 5) C-C twist, 720 cm⁻¹.

Figure 9

Infrared spectra of A) Eigenvector 4, B Eigenvector 3, C) Eigenvector 2.

Figure 10

Infrared spectra of A) Average spectrum of Mn age greater than 50 years. B) Eigenvector 1. C) Difference spectra of the average spectrum of Mn age greater than 50 years minus eigenvector 1. D) Spectrum of the wax oleyloleate.

Figure 11

There was very little difference between the average Md spectrum (top bold) and the average Mn spectrum (top thin). (second from bottom) difference spectrum average Md minus average Mn. The difference spectrum resembled that of a protein. (bottom) Infrared spectrum of anhydrous lysozyme.

Figure 12

Infrared spectra of the carbonyl, amide 1 and bands for average human meibum Md (□ □ , bold, upper) and curve fit (□ □, gray, lower).

Figure 13

Infrared spectra at 24.8 °C on a AgCl window. Top: A) Average of human meibum (□ □ , bold, upper) Md. (□ □, gray, lower) Mn. B) Bovine sphingomyelin. Bottom: Fingerprint region for A) Oleyloleate B) Average of human meibum (□ □ , bold, upper) Md. (□ □ , gray, lower) C) Bovine sphingomyelin D) lysozyme.

Figure 14

Infrared spectra of (thin gray) average Md, (bold black) average Md plus eigenvector 2.

Figure 15

The relationships between the relative amide band intensity and phase transition temperature of human meibum. Lipid phase transition temperatures for normal samples were taken from Borchman et al., 2010. Lipid phase transition temperatures for Md were taken from Foulks et al., 2010. Values are average ± the standard error of the mean. Age ranges were : normal child, 13 years and below; normal adolescent, 14 to 32 years; normal adult 33 years and above. The same samples were used to measure the lipid phase transition temperature and the relative intensity of the amide bands.

Figure 16

The relationships between the relative amide band intensity and lipid order of human meibum. Lipid order for normal samples were taken from Borchman et al., 2010. Lipid order for Md were taken from Foulks et al., 2010. Values are average ± the standard error of the mean. Age ranges were : normal child, 13 years and below; normal adolescent, 14 to 32 years; normal adult 33 years and above. The same samples were used to measure lipid order and the relative intensity of the amide bands.