Identification of tear lipid biomarkers in women with dry eye disease and the impact of intense pulsed light therapy: a case-control study

Abstract

Background

The tear-film lipid layer (TFLL) constitutes the outermost barrier of the ocular surface, reducing evaporation and stabilising the tear film. In aqueous-deficient dry eye (ADDE) and Meibomian-gland dysfunction (MGD), compositional changes in the TFLL compromise this protective role. The present study was designed to characterise the tear-lipid fingerprints associated with ADDE and MGD, to compare them with those of healthy subjects, and to assess the impact of intense pulsed-light (IPL) therapy on the tear lipidome in MGD.

Methods

In a multicentre, prospective, observational–interventional case-control pilot study, 52 participants were enrolled in two phases: a discovery cohort (9 ADDE, 15 MGD, 13 controls) and an independent validation cohort (15 additional subjects). Tear lipids were profiled by ultra-high-performance liquid chromatography–mass spectrometry (UHPLC-MS). Unsupervised principal-component analysis (PCA) explored global variance; supervised partial least-squares discriminant analysis (PLS-DA) and orthogonal PLS-DA (OPLS-DA) defined group differences and yielded candidate biomarkers, with model robustness confirmed by permutation testing and CV-ANOVA. MGD participants received IPL at baseline, day 15, and day 45; clinical metrics and tear samples were obtained before and after therapy.

Results

A total of 176 lipid species were identified and quantified in positive- and negative‐ion modes (ESI + and ESI–). Supervised PLS-DA clearly separated ADDE, MGD and control samples, while OPLS-DA highlighted 48 lipids that differed significantly among groups (p < 0.05). Both dry-eye subtypes were characterised by a pronounced depletion of lysophospholipids (LPE, LPC, LPG, LPI; fold change < 0.5) and an enrichment of (O-acyl)-ω-hydroxy fatty acids (OAHFA; fold change > 2) relative to controls. Cholesteryl esters (ChE) showed a subtype-specific elevation only in the MGD-versus-control comparison (fold change > 2). Permutation testing and CV-ANOVA confirmed the robustness of the ADDE-versus-MGD discrimination model. Although IPL therapy significantly improved clinical metrics such as tear break-up time and lissamine-green staining, the changes observed in the tear lipid profile were not statistically significant.

Conclusions

Dry-eye subtypes appear to possess discrete lipidomic signatures; consequently, the lipid panel identified here could serve as a set of potential therapeutic targets. The dissociation between clinical improvement and lipidomic stability after IPL indicates that the therapy may benefit the ocular surface through mechanisms other than large-scale remodelling of the tear-film lipid layer, highlighting the need to explore complementary therapeutic pathways.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12886-025-04315-1.

Background

Dry eye disease (DED) is a multifactorial condition that affects the ocular surface and that is characterised by disruption of the tear film (TF). In DED, instability of the TF is provoked by hyperosmolarity, in conjunction with other symptoms that include ocular surface inflammation and damage, as well as neurosensory abnormalities. According to the DEWS report (2017: [1], TF homeostasis and stability plays an important role in the aetiology of DED, and on its effects on vision. The TF is a relatively thin layer (3–5 μm: [2, 3], the structural and functional properties of which are primarily driven by its chemical composition [4], mainly lipids, proteins, mucins and electrolytes. This TF fulfils several vital functions, which include protection against pathogens and lubrication of the ocular surface, as well as forming a smooth refractive surface to ensure optimal vision as it constitutes the first refractory layer of the eye [5].

The tear film lipid layer (TFLL) covers the surface of the cornea [5–8], helping to prevent evaporation and TF breakdown. Indeed, it is possible that TF stability is compromised in DED due to undesirable alterations to the TFLL [9]. Shortening of the TF break-up time (TBUT) is directly related to TF thickness [10, 11], which might reflect changes in the TF lipid composition. Meibomian glands (MGs) are fundamental in maintaining TFLL homeostasis, holocrine glands that are embedded into the tarsal plates and that secrete the lipid-rich waxy meibum through their orifices in the lid margin. This meibum combines with the aqueous tear secretion and glycocalyx to form the semi-viscous TF at the ocular surface [12]. Consequently, Meibomian gland dysfunction (MGD) involves terminal duct obstruction, or conditions in which there are qualitative or quantitative changes in MG secretion [13].

The TFLL contains a complex mixture of polar (e.g., phospholipids, sphingolipids and fatty acids) and non-polar lipids (including wax esters -WEs, cholesterol esters –ChEs- and diesters), the latter forming a lipophilic layer at the outermost air-lipid interface, while the former separate the TFLL from the underlying polar aqueous interface [14]. Non-polar lipids in the TFLL primarily combat the evaporation of aqueous tears, although they also establish a protective barrier against microbes and organic matter [15]. Conversely, the polar lipids in this layer probably enhance the spread of the hydrophobic/non-polar lipids through the aqueous tears by reducing the surface tension, or through protein-protein interactions (e.g., with lipocalin, lysozyme and surfactant proteins).

DED is one of the most common ophthalmological diseases, generally contemplated as two distinct entities: aqueous-deficient dry eye (ADDE) that is characterized by reduced lacrimal gland activity; and evaporative MGD. DED is thought to affect between 5% and 50% of individuals, and in people over 50 years of age it is more prevalent in women than in men [16]. Currently, the main treatment for DED is tear supplementation (e.g., artificial tear use), while warm compresses can be used to relieve MG obstruction by liquefying any solidified oil in plugged oil glands, while Meibomian gland expression (MGX) or consumption of omega-3 fatty acid supplements may also be contemplated, in conjunction with prescription medications like cyclosporine, steroid drops or oral (e.g. tetracycline) [16]. Intense pulsed light (IPL) is a relatively new therapy thought to improve DED symptoms [17], which has historically been used in dermatology and is often used to treat acne rosacea, acne vulgaris, hyperpigmentation, essential telangiectasias, unwanted hair and photodamaged skin. IPL is a form of photodynamic therapy that utilizes non-coherent, broad-spectrum light and it appears to be beneficial in DED provoked by MGD [18–21].

Here, the aim was to define the tear lipid profile in the two main types of dry eye, ADDE and MGD, comparing it to that of healthy subjects in order to detect alterations in lipid composition. Moreover, changes to the tear lipid profile induced by IPL in patients with MGD were evaluated, which may be relevant when considering using this approach to treat DED.

Methods

Subjects

Following approval from the Euskadi Ethics Committee (PI2019080), we conducted a multicentre, prospective, observational, interventional case-control pilot study in which 52 participants were consecutively enrolled by qualified clinicians. The study comprised two sequential phases: a discovery phase, used for biomarker identification, that included 9 patients with ADDE, 15 patients with chronic MGD, and 13 healthy subjects (Ctrl); and a validation phase, for which an independent cohort of ten additional dry-eye patients and five healthy controls was recruited. All procedures conformed to the tenets of the Declaration of Helsinki and the ARVO Statement for research involving human subjects. Written informed consent was obtained from every participant after the study’s purpose and potential consequences had been fully explained. Patients and control volunteers were recruited at Basurto University Hospital (Bilbao, Spain), Miguel Servet University Hospital (Zaragoza, Spain), and the Instituto Clínico Quirúrgico de Oftalmología (Bilbao, Spain).

Suitability for inclusion on the study was assessed by clinical examination that included a slit-lamp examination of the lid margin and DED symptoms were confirmed in all participants based on their Ocular Surface Disease Index (OSDI) score. Initially, the tear osmolarity (OSM) was assessed using the TearLab Osmolarity system (OcuSense Inc., San Diego, CA, USA), following the manufacturer’s instructions. Before conducting any diagnostic tests or administering eye drops, the OSM was measured in each eye using the disposable test cards provided with the osmometer. The TBUT was determined by biomicroscopy, with the participants initially instructed to blink naturally in order to evenly distribute the instilled fluorescein across the corneal surface. This distribution was confirmed on a wide beam biomicroscope with a cobalt blue filter, with the time elapsed between eye opening and the appearance of the first dark or black spot/region on the cornea recorded (in seconds) to establish the TBUT. The same procedure was then carried out on the contralateral eye. Subsequently, Lissamine green (LG) staining was evaluated according to the Oxford scale and a type I Schirmer’s (SCH) test was conducted on all the participants following the instillation of a drop of 0.5% proparacaine ophthalmic solution. The same exploratory protocol was carried out in the month preceding IPL therapy and one month after the three sessions of IPL. Loss of MG function in MGD patients was quantified in different ways and staged from 0 to 3 according to the scale developed by Arita et al. [22]. The quantity and quality of the meibum was also assessed and both values were scored from 0 to 3: Quantity scored as ‘0’ if all the glands functioned, ‘1’ if only some of them functioned, ‘2’ if only a few glands functioned and ‘3’ if none functioned; Quality scored as ‘0’ if the meibum was clear and transparent, ‘1’ if it was a thick or cloudy discharge, ‘2’ if it was cloudy and granular or with detritus and ‘3’ if it looked like toothpaste.

The following exclusion criteria were applied: ocular surgery in the three months prior to the study, a systemic condition (e.g. active allergies), or medication use that might interfere with the interpretation of the results (e.g., anti-inflammatory agents), or the concurrent use of topical medications other than artificial tears.

The clinical variables were explored following a sequential approach, moving from the less invasive to more invasive tests to avoid altering the measurements. The participants were instructed not to use eye drops for two hours before any study visit.

Tear fluid collection

Tear samples were collected from the lower conjunctival sac using 10 µl glass capillary tubes, as described previously [23]. Tear samples were sequentially collected from both eyes of each participant, following a one-week washout period in the case of using artificial tears. For MGD patients, tear samples were collected before (MGD T0) and at the end of IPL treatment (MGD T1). All samples were stored at − 80 °C until lipid extraction.

Intense pulsed light (IPL) procedure

IPL therapy was performed on each participant at baseline, and then 15 and 45 days later. A modular laser multi-application platform (E-Eye, Houdan, France) was used to administer this therapy to the periorbital area. During the IPL procedure, patients were required to wear opaque goggles and a coolant was applied around the eyelids of both eyes, with any make-up and contact lenses removed before treatment. Depending on the patient’s tolerance, each subject received 10 light pulses, using a 590 nm filter at an intensity of 12–16 J/cm² depending on the Fitzpatrick skin type.

Chemicals and standards

Optima^® LC/MS-grade water, methanol, acetonitrile (ACN), 2-propanol, formic acid and the Pierce LTQ Velos ESI Positive/Negative Ion Calibration Solutions were obtained from Fisher Scientific (Fair Lawn, NJ). Ammonium formate and toluene were purchased from Sigma-Aldrich (Sigma Chemical Co., St Louis, MO), while the Splash™ LipidoMix™ and the ceramide/sphingoid internal standards (ISs) were purchased from Avanti Polar Lipids Inc. (Alabaster, AL). In-house synthesis of the 16-(oleoyloxy) hexadecanoic acid (OAHFA) standard was carried out as described previously [24]. The Wax Ester (WE) 15,15,16,16,17,17,18,18,18-d9 (WE 18:1(d9)/26:0) standard was obtained by treating deuterated oleic acid (OA) with SOCl₂ [25]. Both compounds were characterized by ¹H and ¹³C nuclear magnetic resonance (NMR) spectroscopy and high-resolution mass spectrometry (MS), and their purity was established as > 95%.

Lipid extraction and UHPLC-MS analysis

Lipid extraction from the individual tear samples (10 µl) was performed in Eppendorf tubes using the isopropanol protocol described elsewhere [24]. UHPLC-MS/MS analyses were conducted following the protocol previously detailed by Acera A [26]. The experiments utilized a Vanquish UHPLC system (Thermo Fisher Scientific, USA) equipped with a reversed-phase column (Acquity UPLC C18 CSH™ 2.1 × 100 mm, 1.7 μm) and a pre-column (Acquity UPLC C18 CSH™ 2.1 × 5 mm, 1.7 μm; VanGuard). The UHPLC system was coupled to a Q Exactive HF-X hybrid quadrupole-Orbitrap mass spectrometer (Thermo Fisher Scientific, USA) operating in data-dependent acquisition mode. A 10 µl injection volume was used, and all samples were resuspended in 70 µl of methanol: toluene (9:1, v/v) and maintained at 10 °C prior to injection into the UHPLC-MS/MS system.

Design of the analysis

A blank and a diluted IS 9:1 (v/v) MeOH: Toluene solution was included at the beginning and end of the run to test for any possible contamination or carry-over effect, and to check the chromatography resolution, sensitivity and exact mass. A quality control (QC) sample, prepared by combining 10 µl of each sample and each lipid extract was reconstituted with MeOH: Toluene (9:1, v/v), injected regularly every ten injections throughout the run to monitor the sensitivity and stability of the UHPLC–MS platform. This QC sample was also used to condition the system at the beginning of the analysis. In a previous test, at least 15 injections of a sample containing the matrix studied were seen to be necessary to stabilize the system. The injection order of the study samples was randomized to minimize the effect of any instrumental drift arising from column degradation or contamination of the MS source on the evaluation of repeatability within each pathology (Fig. 1).

Fig. 1.

Outline of the injection sequence. [1] Three blanks (MeOH:Toluene 9:1, v/v); [2] diluted internal standard solution (IS); [3] 15 quality controls for system conditioning (QCsys); [4] one QC sample run after every ten injections to check for sensitivity and stability; [5] 10 randomized experimental samples; [6] a QC sample followed by an IS; and finally, [7] a blank

MS data processing

All the MS data were acquired and processed using the Xcalibur 4.1 software package (version 4.1, Thermo Fisher Scientific, Waltham, MA, USA), while the LipidSearch software version 4.2.27 (Mitsui Knowledge Industry, Tokyo, Japan) was used to identify and quantify the lipid species in these complex biological samples. One of the key processing parameters was searching the individual data files for the product ion MS/MS spectra of lipid precursor ions. MS/MS fragment ions were predicted for all the precursor adduct ions measured within 5 ppm. The product ions that match the predicted fragmentations within a 5 ppm mass tolerance were used to calculate a match score and the candidates providing the highest-quality match were identified. Next, the search results from the individual positive and negative-ion files from each sample group were aligned within a retention window (0.1 min), and the data were merged for each annotated lipid. The annotated lipids were then filtered to reduce false positives according to the established criteria. (I) Main ion: In ESI+, the main [M + H]⁺ ion was observed for lysophosphatidylcholines (LPCs), phosphatidylcholines (PCs), lysophosphatidylethanolamines (LPEs), phosphatidylethanolamines (PEs), phosphatidylserines (PSs), sphingomyelins (SMs), ceramides (Cers), hexosylceramides (Hex1Cers), dihexosylceramides (Hex2Cers); the main [M + NH4]⁺ ion was observed for lysophosphatidylinositols (LPIs), phosphatidylinositols (PIs), lysophosphatidylglycerols (LPGs), phosphatidylglycerols (PGs), triacylglycerols (TGs), cholesterol esters (ChEs) and WEs; In ESI-, the main [M-H]⁻ ion was observed for LPEs, LPGs, LPIs, LPSs, PIs, PSs and OAHFAs. (II) Identification grade (ID quality filter): (A) lipid class and fatty acids (FAs) were fully identified; (B) lipid class and some FAs were identified; (C) lipid class or FAs were identified; and (D) lipid ID obtained via other fragment ions (H₂O loss, etc.). (III) Correlation of carbon number with the retention time or degree of saturation with retention time in each lipid class. Finally, (IV) the ratio of the areas of the chromatographic peaks of biological as opposed to blank samples should be > 1.5.

Quantification was carried out by normalizing the extracted monoisotopic ion peak area of each native lipid species to the intensity of the extracted monoisotopic ion peak area of the IS (Eq. 1). The ISs used in this study were chosen to avoid those present in the native tear samples (Supplementary Table S1).

1

Statistical analysis

The descriptive analysis was performed using absolute and relative frequencies in the case of categorical variables, and the mean and standard deviation (SD) in the case of continuous variables, using the R-Statistics software 4.3.0 (Lucent Technologies, New Jersey, USA). The samples’ normal distribution was assessed with the Shapiro-Wilk test and the statistical significance of the mean differences was measured using Student’s t-test. The calculated p-values determine the probability of an association with the lipids in the dataset, with a p-value < 0.05 based on a Fisher’s exact test considered significant. Sample size was determined to compare mean lipid concentrations across three independent groups. A two-tailed Student’s t test for independent samples was used as a conservative proxy for the post-hoc contrasts of a one-way ANOVA. Setting the type-I error (α) at 0.05 and the desired power (1 − β) at 0.80, and assuming equal variances between groups, the minimum detectable difference was defined as one pooled standard deviation of the target lipids.

Differences in lipid composition among the patient groups were examined with both univariate and multivariate approaches. In the univariate analysis, volcano plots were constructed using Student’s t-test (α = 0.05) and a minimum two-fold change; multiple comparisons were adjusted with a 5% Benjamini–Hochberg false-discovery rate (FDR). Pairwise lipid abundances were expressed as group-specific lipid ratios, and fold changes between 0.5 and 2 were considered biologically non-significant. Global variation in lipid fingerprints among ocular-disease groups was first explored with unsupervised principal-component analysis (PCA). Supervised partial least-squares discriminant analysis (PLS-DA) was then applied to build and validate predictive models, while orthogonal PLS-DA (OPLS-DA) was interrogated to pinpoint candidate biomarkers for dry-eye disease. All multivariate analyses were performed in SIMCA v17.0.2 (Umetrics, Umeå, Sweden).

The resulting p-values for the goodness of fit of the model (R2) and the cross-validated prediction ability (Q2) were confirmed using permutation tests, and p-values were obtained from a CV-ANOVA (Analysis of Variance testing of Cross-Validated predictive residuals) that compares the fit of two models to the same data. This function tests the residuals of the linear regression between cross-validated scores of the predictive OPLS component and the response Y, with the variation of Y around its mean. The p value assumes the hypothesis of equal residuals of the two models [27].

Validation phase

After generating predictive models that describe lipid profiles specific to each dry-eye subtype, a pilot validation was conducted in an independent cohort consisting of dry-eye patients (ADDE and MGD) and control subjects.

Results

Patients

Participants were divided into two cohorts—a discovery cohort and a validation cohort. The discovery cohort comprised 37 volunteers, 15 patients with chronic MGD, 9 patients with ADDE, and 13 healthy Ctrl. The mean age of the patients with MGD was 62.26 (± 11.23) years of age and that of the patients with ADDE was 54.70 (± 19.60) years old, while the Ctrl subjects had a mean age of 48.54 (± 13.74) years.

The pilot validation study recruited fifteen participants, including five patients with chronic MGD (60.4 ± 5.3 years old), five patients with ADDE (58.0 ± 8.4 years old) and five healthy controls (55.6 ± 12.7 years old).

All the participants enrolled onto the study completed the examinations and treatment in full, and no pain or discomfort was experienced during the testing.

At baseline—before any treatment was initiated—the groups were compared across all clinical variables to establish the ocular-surface status of ADDE and MGD patients relative to female controls in both cohorts analysed (Table 1).

Table 1.

Comparison of the clinical variables between the ctrl, ADDE and MGD women

Variable	Ctrl	ADDE	p-value	MGD	p-value
LG	0.00 ± 0.00	3.67 ± 0.9	< 0.001*	1.69 ± 1.10	< 0.02*
OSDI	4.06 ± 7.01	61.80 ± 16.05	< 0.001*	42.88 ± 12.65	< 0.001*
SCH (mm)	20.78 ± 2.34	2.89 ± 6.50	< 0.001*	16.27 ± 10.51	0.0992
TBUT (s)	12.15 ± 2.01	9.00 ± 0.7	0.009*	5.61 ± 0.99	< 0.01*

The values correspond to the mean and SD, and the asterisks indicate significant differences relative to the controls, p-value < 0.05

Abbreviations: ADDE Aqueous deficient dry eye, Ctrl Control, LG Lissamine green, MGD Meibomian gland dysfunction, OSDI Ocular Surface Disease Index, SCH Schirmer’s test, TBUT Tear film break-up time

All clinical variables differed significantly between women with ADDE and healthy controls. In the MGD group, every parameter except tear volume measured with the SCH test also showed significant differences from controls. When the two dry-eye subtypes were compared with each other, all clinical variables except the OSDI score diverged significantly. Both patient groups had markedly higher OSDI scores than controls, indicating discomfort and impaired visual quality regardless of aetiology.

The SCH test revealed a pronounced reduction in tear production in ADDE patients (2.89 ± 6.50 mm) relative to MGD patients (16.27 ± 10.51 mm; p < 0.001). Conversely, TBUT was longer in ADDE patients (9.00 ± 0.70 s) than in MGD patients (5.61 ± 0.99 s; p = 0.01), reflecting greater tear-film instability in the latter. LG staining followed the same pattern, with more extensive staining in ADDE (3.67 ± 0.90) than in MGD (1.69 ± 1.10; p = 0.02), suggesting more severe epitheliopathy in the aqueous-deficient group.

The MGD cohort from the discovery phase subsequently underwent IPL therapy. Clinical parameters recorded before and after treatment (Table 2) demonstrated an overall improvement in ocular-surface status following IPL.

Table 2.

Comparison of the clinical variables in MGD patients pre- and post-IPL treatment

Variable	MGD T0	MGD T1	p-value
LG	1.88 ± 1.17	0.93 ± 0.97	0.02*
OSDI	42.08 ± 19.84	34.70 ± 24.01	0.36
OSM (mOsm/l)	301.57 ± 8.90	298.17 ± 10.66	0.35
SCH (mm)	19.27 ± 11.47	18.37 ± 9.95	0.82
TBUT (s)	3.13 ± 1.56	6.45 ± 1.50	< 0.001*
MEIBS	1.95 ± 0.55	1.65 ± 0.43	0.003*
MEIBI	1.87 ± 0.66	1.72 ± 0.59	0.014*
EGTS	1.75 ± 0.55	1.40 ± 0.59	0.008*
EGTI	1.75 ± 0.91	1.22 ± 0.41	0.017*
EGLS	1.9 ± 0.71	0.92 ± 0.63	0.001*
EGLI	2 ± 1.02	1.15 ± 0.69	0.002*

The asterisks indicate significant differences relative to the controls: p-value < 0.05

Abbreviations: LG Lissamine green, MGD T0 Meibomian gland dysfunction before treatment, MGD T1 Meibomian gland dysfunction after IPL treatment, OSDI Ocular Surface Disease Index, OSM, Tear osmolarity, SCH Schirmer’s test, TBUT Tear film break-up time, MEIBS Meiboscore upper eyelid, MEIBI Meiboscore lower eyelid, EGTS Upper eyelid meibum quantity, EGTI Lower eyelid meibum quantity, EGLS Upper eyelid meibum quality, EGLI Lower eyelid meibum quality

Lipidomic analysis of ocular pathologies in the discovery phase

In the samples analysed, 176 lipid species were identified and quantified as individual lipid species in the ESI+ and ESI- modes (Supplementary Table S2, Figure S1). These lipids belonged to 17 lipid sub-classes from six different lipid classes: glycerophospholipids (LPC, LPE, LPI, LPG, LPS, PC, PI and PS); sphingolipids (Cer, Hex1Cer, Hex2Cer and SM); glycerolipids (TGs); sterol lipids (ChE); OAHFAs; and WEs. The lipid composition of the ADDE and MGD tear samples (before treatment, MGD T0) was compared with the tear lipid profile of the controls.

We applied a multivariate statistical analysis to the data to investigate the differences in the baseline lipid profiles in the tear samples from ADDE and MGD patients. An unsupervised PCA was used assess the dispersion of the data according to the patient's DED type and to assess the possibility of classifying the experimental groups based on the presence of distinct lipids. The first three principal components explained 63% of the total variance in the data, and the R2 (goodness of fit, 0.704) and Q2 (the ability of prediction, 0.556) highlighted the quality of the parameters from the PCA score plot (Figure 2A). Hence, the expression of specific lipids appeared to be sufficient to distinguish the ADDE samples from the Ctrl and MGD samples. Interestingly, and as seen in the space defined by the two first principal components in the bi-dimensional plot of the sample scores, ADDE samples grouped in the same cluster, and MGD and Ctrl samples were found in different clusters. Moreover, greater dispersion was observed in the MGD samples, with lipid profiles more similar to those of women in the Ctrl group (Figure 2A). In this case, the algorithm did not discriminate between Ctrl and MGD groups, and an overlap between MGD T1 and MGD T0 was evident.

Fig. 2.

Multivariate statistical analysis of tears from healthy and evaporative or aqueous deficient dry eye (ADDE) patients. A The principal component analysis (PCA) of lipidomes obtained from the independent tear samples by UHPLC-MS are illustrated. The symbol shape and colour indicate the ocular pathology: ADDE (), MGD T0 (), MGD T1 (), Ctrl () and QC (). Binary comparisons constructed through a partial least square-discriminant analysis (PLS-DA): (B) MGD T0 vs. Ctrl, (C) ADDE vs. Ctrl, and (D) ADDE vs. MGD T0

A supervised PLS-DA statistical model was constructed to maximise the differences between the two classes and to test the ability of lipid biomarkers to classify samples according to their pathology. All binary comparisons produced a net group separation (MGD T0 vs. Ctrl, ADDE vs. Ctrl, and ADDE vs. MGD T0: Figure 2B-D), indicating that each type of DED has a characteristic lipid profile. The differences between the lipid patterns were striking in some cases, such as between ADDE and Ctrl or ADDE and MGD T0 (Figure 2C, D), while they were subtle between MGD T0 and Ctrl (Figure 2B). Permutation tests were carried out to validate the PLS-DA prediction models, recording the cumulative values of R2 and Q2 for 999 random rearrangements of the Y variables, and reconstructing the models (Table 3).

Table 3.

Validation of the PLS-DA models in the discovery phase

Model	Principal ComponentsMinimum1		R2	Q2	Q2 Intercept	p-Value2
MGD T0 vs. Ctrl	2	0.550		0.350	0.002	0.035
ADDE vs. Ctrl	3	0.876		0.557	0.032	0.024
ADDE vs. MGD T0	1	0.521		0.318	−0.107	0.010

Statistically significant differences: p-value < 0.05

Abbreviations: ADDE Aqueous deficient dry eye, MGD T0 Meibomiam gland dysfunction before treatment, Ctrl Healthy subjects

¹Minimum number of principal components required to explain the maximum variance

²The validity of the model according to that minimum number of principal components

The zero or negative intersection of the Q2Y line of tendency obtained for all the models demonstrated that the PLS-DA data were well adjusted, and the clustering was not by chance. An additional test using p-values obtained from CV-ANOVA also confirmed the validity of the models. Hence, the R2, Q2, permutation Q2 and p-values (Table 3) indicated that the models established reliably discriminated the type of DED based on the lipid fingerprint, indicating that the tear composition is altered in MGD and ADDE relative to healthy individuals, and between them. To extract putative tear lipid biomarkers for patients with MGD and ADDE, we performed an OPLS-DA study of the lipid fingerprints, employing the following threshold parameters: VIP >1 p [1] ≥ 0.06 or p [1] ≤-0.06, p(corr) [1] ≥ 0.2 or p(corr) [1] ≤ -0.2. This approach selected the top discriminant lipids (Table 4 and Fig. 3).

Table 4.

Discriminant lipid metabolites in women with evaporative (MGD T0) and aqueous deficient (ADDE) dry eye extracted by OPLS-DA in discovery phase

Lipid	Lipid class	Adduct Characterized	m/z	RT (min)	MGD T0/Ctrl		ADDE/Ctrl		ADDE/MGD T0
					Regulation	p-value	Regulation	p-value	Regulation	p-value
OAHFA18:1/24:1	OAHFA	-H	6.455.820	13.26	up	0.022
OAHFA18:1/28:1	OAHFA	-H	7.016.440	14.36	up	0.007
OAHFA18:1/30:1	OAHFA	-H	7.296.750	14.70	up	0.016	up	0.085
OAHFA18:1/31:0	OAHFA	-H	7.457.080	15.25	up	0.008	up	0.027
OAHFA18:1/31:1	OAHFA	-H	7.436.910	14.93	up	0.008	up	0.118
OAHFA18:1/32:2	OAHFA	-H	7.556.900	14.75	up	0.013	up	0.055
OAHFA18:1/32:3	OAHFA	-H	7.536.750	14.38	up	0.008	up	0.068
OAHFA18:1/33:0	OAHFA	-H	7.737.390	15.61	up	0.042	up	0.040
OAHFA18:1/33:1	OAHFA	-H	7.717.240	15.30	up	0.007	up	0.134
OAHFA18:1/34:1	OAHFA	-H	7.857.390	15.50	up	0.031	up	0.024
OAHFA18:1/34:2	OAHFA	-H	7.837.250	15.17	up	0.012	up	0.034
OAHFA18:1/34:3	OAHFA	-H	7.817.070	14.78	up	0.012	up	0.058
OAHFA18:1/36:1	OAHFA	-H	8.137.710	15.84	up	0.024	up	0.037
ChE18_1	ChE	0	6.686.340	15.83	up	0.302
ChE20_0	ChE	0	6.986.810	16.59	up	0.029
ChE20_1	ChE	0	6.966.650	16.24	up	0.056
ChE20_2	ChE	0	6.946.500	15.87	up	0.067
ChE22_0	ChE	0	7.267.120	16.94	up	0.013
ChE22_1	ChE	0	7.246.970	16.62	up	0.027
ChE22_2	ChE	0	7.226.810	16.27	up	0.044
ChE22_3	ChE	0	7.206.600	15.92	up	0.069
ChE23_0	ChE	0	7.407.280	17.11	up	0.012
TG(18:0_16:0_18:1)		0	8.607.830	16.14	up	0.035
LPE(15:0)	LPE	-H	4.392.700	0.85			down	0.001
LPE(18:1)	LPE	-H	4.793.010	0.99	down	0.017	down	0.007
LPE(16:0)	LPE	-H	4.532.850	0.96	down	0.003	down	0.002
LPE(18:0)	LPE	-H	4.813.170	1.30	down	0.008	down	0.004
LPC(15:0)	LPC	+H	4.813.170	0.84			down	0.019	down	0.028
LPC(16:0)	LPC	+H	4.953.320	0.93			down	0.002	down	0.117
LPC(17:0)	LPC	+H	5.093.480	0.99			down	0.000	down	0.061
LPC(18:0)	LPC	+H	5.233.640	1.25			down	0.002	down	0.004
LPC(18:1)	LPC	+H	5.213.480	0.98			down	0.001	down	0.092
LPC(20:1)	LPC	+H	5.493.790	1.28			down	0.000	down	0.004
LPG(14:1)	LPG	-H	4.542.330	0.70			down	0.022	down	0.005
LPG(15:0)	LPG	-H	4.702.640	0.69			down	0.060	down	0.039
LPG(17:1)	LPG	-H	4.962.800	1.04			down	0.153	down	0.084
LPG(18:1)	LPG	-H	5.102.960	0.86			down	0.008	down	0.003
LPG(18:2)	LPG	-H	5.082.800	0.69			down	0.024	down	0.007
LPI(15:0)	LPI	-H	5.582.810	0.66			down	0.015	down	0.048
LPI(16:0)	LPI	-H	5.722.960	0.70			down	0.000	down	0.001
LPI(18:0)	LPI	-H	6.003.280	0.97			down	0.017	down	0.008
LPI(18:1)	LPI	-H	5.983.120	0.80			down	0.013	down	0.004
LPI(18:2)	LPI	-H	5.962.960	0.67			down	0.008	down	0.017
LPI(20:4)	LPI	-H	6.202.960	0.67			down	0.020	down	0.020
LPS(15:0)	LPS	-H	4.832.600	0.69			down	0.005
Cer(d15:1_17:0)	Cer	+H	5.094.810	5.08			down	0.002
Cer(d15:1_17:1)	Cer	+H	5.074.650	4.39			down	0.000
Cer(d16:1_16:0)	Cer	+H	5.094.810	4.86			down	0.000

Statistically significant differences: p-value < 0.05

Fig. 3.

Representative selection of discriminant lipids through an OPLS-DA. The lipidome of aqueous deficient dry eye (ADDE) patients, Meibomian Gland Dysfunction (MGD) patients and healthy subjects (Ctrl) were evaluated by OPLS-DA to select candidate biomarkers. In the score plots of the two main components (A), discriminant lipids were visualized as S-plots (B), VIP plots (C) and loading plots calculated with the jack-knife algorithm D. Red and blue denote the variables that are more and less abundant in ADDE or MGD patients than the Ctrls, respectively

The OAHFA, ChE, LPE, LPC, LPG and LPI lipid classes showed the most substantial discriminatory power across groups (see Table 4 for the identities, the detection parameters and the typical changes of these species). Intergroup differences were confirmed by a univariate statistical analysis and in particular, through the p-values from the Student’s t-test using the Benjamini-Hochberg FDR of 5% as the correction for the multiple tests.

Volcano plots were also obtained, employing a fold change >2 and p-value ≤0.05 (Student's t-test: Figure 4) to highlight the increase or decrease in lipid concentrations in the different groups studied. As this approach had a less stringent discriminant power than the OPLS-DA threshold we assigned, the univariate analyses expanded the list of potential biomarkers. Indeed, new molecular species were typically added within the same classes defined by the OPLS-DA (Supplementary Table S3).

Fig. 4.

Volcano plots comparing the lipids in MGD T0 vs Ctrl, ADDE vs Ctrl and ADDE vs MGD T0

IPL treatment and changes in tear lipid signatures

When analysing the lipid profile in patients subjected to IPL treatment, the relative abundances in tears did not change significantly, although their appeared to be a relative enhancement of most lipid families after IPL treatment (Table 5).

Table 5.

The change in concentration of the different lipid classes in the tear of MGD patients as a result of IPL treatment

MGD T1/MGD T0
Class	Fold	p-value
Cer	1.01	0.724
ChE	0.96	0.915
Hex1Cer	0.88	0.153
Hex2Cer	1.57	0.282
LPC	1.23	0.152
LPE	1.44	0.072
LPG	1.05	0.915
LPI	1.22	0.648
LPS	1.11	0.603
OAHFA	1.08	0.805
PC	0.86	0.448
PE	1.14	0.695
PI	1.10	0.704
PS	0.94	0.865
SM	1.15	0.198
TG	1.20	0.437
WE	0.86	0.596

Statistically significant differences: p-value <0.05

The three lipid sub-classes that increased in concentration most significantly after IPL treatment were Hex2Cer (1.57-fold), LPE (1.44-fold) and LPC (1.23-fold), although in no case were these differences statistically significant. The PC (0.86-fold) and WE (0.86-fold) families were lipids that experienced a drop in concentration after treatment. The mild effect of IPL therapy on tear lipid composition was studied in more detail, considering the main components. Utilizing the PCA, a larger dispersion in the lipid profile was observed after treatment than prior to IPL, although the analysis revealed no trend in compositional dispersion between the MGD T0 and MGD T1 groups (Fig. 5).

Fig. 5.

PCA diagrams of the MGD T0 and MGD T1 groups. There was no significant distinction in the trends of the lipid clusters before and after treatment: MGD T0, Meibomian Gland Dysfunction before treatment; MGD T1, Meibomian Gland Dysfunction after treatment

Correlation between the clinical variables and the lipid profile in discovery-phase patients

When analysing the correlation between the different classes of lipids and the clinical variables in the distinct study groups, a negative correlation was observed between the LPC and LPE lipids in the ADDE patient's tear and LG staining. Regarding the DED symptomatology measured through the OSDI questionnaire, a positive correlation was observed between the symptoms and the relative amount of the LPC, LPG and LPI lipids in the MGD patient's tear (Supplementary Table S4)(Fig. 6).

Fig. 6.

The lipid volcano plot comparing MGD T1 to MGD T0

Validation Phase

Based on the analysis of the samples collected during the validation phase, the classification performance of the different models (MGD T0 vs CTRL, ADDE vs CTRL, and MGD T0 vs ADDE) was assessed. While all models displayed high sensitivity, their specificity was low, indicating limited accuracy in correctly classifying the groups (Table 6).

Table 6.

Validation of the PLS-DA models in the validation phase

Model	Spec.	Sens.	Accuracy
MDG T0/CTRL	0.538	0.800	0.697
ADDE/CTRL	0.846	0.778	0.818
MGD T0/ADDE	0.111	0.850	0.621

Discussion

Due to the development of increasingly sensitive lipidomic techniques, it is now possible to analyse tear samples quantitatively on an individual basis. Indeed, UHPLC-MS analysis has proven to be a very sensitive method to quantify TF lipids in both positive and negative ion modes. Here, we analysed the relative abundance of polar and non-polar lipids in the tears of healthy women, and women with MGD or ADDE, in order to assess the TF alterations related to specific DED pathologies, and to evaluate changes in their TF lipid profiles. In addition, we evaluated how the tear lipid profile in MGD patients might be affected by IPL therapy.

In this pilot study, lipid profiles were only characterized in women to eliminate hormonal bias. The higher prevalence of DED in women consistently shown suggests there may be sexual dimorphism in the human lacrimal gland. Sex steroid hormones play a crucial role in the development of sexual dimorphism, generating differences between males and females unrelated to reproduction that are ubiquitous in real-world situations [28]. A total of 85.6% of subjects with DED were seen to present signs of MGD or a combination of MGD and ADDE, indicating a predominance of MG dysfunction in DED [29]. With women having a 50% to 70% higher risk of developing DED [30], particularly pronounced after menopause, female sex and increasing age are significant risk factors for the development of evaporative dry eye [31]. As age advances and the levels of sex steroid hormones decrease, the roles of androgens and oestrogens become critical in the pathophysiology of lacrimal gland dysfunction. Oestrogens and androgens seem to exert opposite effects on DED pathogenesis. Although the exact role of oestrogens is not fully understood, several in vitro, animal and human studies suggest that oestrogens decrease sebaceous gland secretion, inhibit lipogenesis and promote ocular surface inflammation. This could be attributed to the competitive interaction that oestrogen has with androgen receptors, thereby inhibiting androgen-induced MG activity [32].

We demonstrate here that a specific lipid profile in tears is associated with each type of DED. In the ADDE cohort, reduced tear volume (as indicated by the SCH test) does not imply a uniform decrease in all the lipids present in tears relative to the Ctrl group. In fact, a significant increase in the concentration of long-chain hydroxylated FAs (OAHFAs) and a marked decrease in lysophospholipids like LPC, LPE, LPG, LPI and LPS, were observed. In the MGD cohort, an increase in OAHFA concentrations was evident relative to the Ctrl group, as well as in ChE and TGs, the change in the latter two being specific to the MGD cohort. The findings regarding OAHFA concentrations in the tears of patients with both types of DED are at odds with previous reports [33–35]. Here, an increase in these species was identified in the tears of both DED cohorts relative to healthy tears. The relative amount of OAHFAs in tear samples, measured in negative ion mode, may be associated with TF stability [36]. Indeed, most OAHFAs were inversely associated with the rate of pre-corneal TF thinning, suggesting the abundance of these OAHFAs decreases in association with an increased rate of tear thinning or evaporation. However, certain OAHFA species (18:1) are positively correlated with the rate of pre-corneal TF thinning, in accordance with the data presented here. Given the complex lipid composition of the TFLL, different OAHFAs may interact distinctly with the mucoaqueous phase and non-polar lipids, potentially fulfilling different roles in relation to tear evaporation. It is also possible that the TFLL’s capacity to reduce evaporation depends on a delicate homeostatic balance among various OAHFAs in the TFLL. An imbalance in the concentration of specific OAHFA species may alter the surface tension of the TF and tear homeostasis. Our results also differ from some earlier data as no significant changes in OAHFA concentration were observed in MGD tear samples following IPL treatment [35]. These findings indicate the need for further studies on larger sample sizes to confirm the reproducibility of these data.

Alternatively, a significant decrease in some lysophospholipids was observed in the ADDE cohort, in contrast to the OAHFA class of lipids. This profile demonstrates a specific pattern for the ADDE group, as not all lipid classes were reduced in conjunction with the reduction in tear volume characteristic of this type of DED. Lipid secretion from the lacrimal gland is regulated by neural, hormonal and vascular factors. Reduced lipid secretion due to lacrimal gland dysfunction leads to TF destabilization. Enzyme activity, such as that of acyltransferase, may deplete the lysophospholipids in tears through reacylation, reconverting them into phospholipids [37]. Additionally, LPC and LPE levels were negatively correlated with LG staining, indicating that higher LG scores that reflect a stronger conjunctival epitheliopathy were associated with less LPC and LPE in tears. These ADDE-specific lipid profiles may pave the way for new approaches in the treatment of this ocular condition.

There is also evidence linking the differences in lipids with sex and age. In women, TGs increase with age [38] and here, high levels of TGs were seen in the tear of women with MGD, especially TG (18:0_16:0_18:1), although this type of lipid was not found in the tear of women with ADDE. An analysis of the TG composition in the tear of women with MGD revealed a significant difference between these patients and healthy women. A link between dyslipidaemia and MGD has already been shown, including elevations in TGs, and this is more likely to occur in patients with MGD than in age- and sex-matched Ctrl adults [39]. By contrast, patients with MGD have a higher incidence of hypertriglyceridemia than the general population. Further studies will be needed to establish whether tear TGs are related to MGD in humans [40]. Another interesting observation was the increase in ChE in MGD. Although ChEs represent a small proportion of the total lipids in most tissues, they plays a crucial role in various age-related inflammatory disorders [41]. Elevated serum cholesterol levels have also been reported in moderate to severe cases of MGD [40, 42], which supports our findings. IPL therapy aims to achieve clinical improvements by heating the MGs to liquefy the meibomian, activating fibroblasts, improving the synthesis of new collagen fibres, eradicating Demodex and decreasing the bacterial load on the eyelids [43]. Moreover, IPL can reduce telangiectasia to help the MGs restore their hypoxic environment and regular activity [44]. This effect is produced by photothermolysis of haemoglobin in the superficial vessels of the skin (achieved with yellow light around a wavelength of 578 nm) and by inducing thrombosis of telangiectasias [45].

The clinical data obtained here demonstrate that IPL produced regeneration of the corneal and conjunctival epithelium, as witnessed by the significant decrease in LG staining. A considerable increase in TBUT is consistent with previous findings, demonstrating that routine application of heat to the inner surface of the eyelids provokes a consistent increase in both TBUT and the number of MGs producing fluid secretion during the treatment period [35, 46]. By contrast, no significant changes in SCH test values were observed after IPL, which is not surprising since patients with MGD had no discernible tear dysfunction, even at baseline (average SCH >10 mm), nor were there significant changes in their tear OSM. Accordingly, we postulate that tear lipid abnormalities associated with MGD could result from altered lipid function rather than an absolute lipid deficiency per se, which is consistent with our observation that eyelid warming and the relief of disease symptoms were not associated with significant changes in lipid composition. An increase in lipids would facilitate the release of fresh tears and meibomian secretions to the ocular surface, also eliminating toxic contaminants and other products metabolized into reactive lipid species. No significant changes in tears were observed in women with MGD for any of the lipid species studied, although there was an increase in some lipids after IPL treatment (Hex2Cer, LPE, LPC, LPI and TG). This increase in polar lipids may favour the anchoring of the TF to the aqueous part of the ocular surface and maintain good surface tension. The lack of statistically significant changes in the tear lipid profile may be due to the protocol used. It is currently recommended to use more shots in each treatment and for longer periods than those used here. IPL equipment is continually being updated and improved, moving from 500 nm to 400 nm wavelengths, providing enhanced biocidal efficacy. Furthermore, recent meta-analyses report that IPL therapy increases TBUT, reflecting improved tear-film stability, yet this benefit does not always translate into symptomatic relief, as OSDI scores often remain unchanged [47, 48]. Likewise, IPL alone has not been shown to increase TFLL thickness, suggesting that the lipid composition may not change appreciably. Collectively, these reviews underscore the need for standardized treatment protocols and further studies that correlate tear lipidomics with harmonized IPL regimens and adjunctive therapies. Taken together, the lack of statistically significant global lipid-profile changes in our cohort does not undermine the clinical efficacy of IPL; rather, it highlights the need for larger, standardised trials using next-generation devices and more sensitive analytic endpoints.

Lastly, but not least, there are lipid species that were not identified here, such as diesters. The chemical nature of diesters in meibum was first proposed based on the structures and the quantities of the moieties within these lipids [49]. These diesters were separated by column chromatography and thin-layer chromatography at the class level, then hydrolysed into their components and derivatised into esters for identification and quantification by gas chromatography [49]. However, no intact diester molecules were detected to confirm these structures due to technological limitations until 2010, when diester molecules were detected for the first time by MS and the corresponding molecules confirmed by MS/MS [50]. The fragmentation patterns of these molecules reported confirmed the corresponding chemical structures for two of the four types of diesters proposed previously, namely ωType I-ST diesters (cholesteryl esters of OAHFA) and α,ω Type II diesters (diacylated α,ω-diols), while the structures of the other two types of diesters proposed were not confirmed [49]. This demonstrates the importance of detecting lipids as intact molecules. Molecules of these two types of diesters were also subsequently detected in independent studies but named as cholesteryl esters of OAHFA [51] and diacylated α,ω-diols [52].

Conclusions

In conclusion, the lipid profile in the tear film of ADDE patients differs from that of MGD patients, highlighting the intricate lipid interactions unique to each type of dry eye disease. OPDLS analysis identified distinctive lipid profiles, uncovering specific imbalances between ADDE and MGD. Notably, the increase in certain OAHFA species observed in both types of dry eye represents a novel finding, counter to previous reports. This imbalance in OAHFAs, ChEs, and reduced lysophospholipids contributes to tear film instability, reflected in shorter breakup times and intensified vital staining, both indicators of an altered ocular surface.

This study, which focuses on a female cohort and compares two types of DED, underscores the need for further investigation into lipid components and lipid-lipid interactions to deepen our understanding of lipid dynamics in the ocular surface of DED patients. Expanding future studies to include a broader range of sexes, ages, and larger cohort sizes will strengthen these findings. Although IPL treatment demonstrates clinical improvements in these women, increasing the number of shots and light intensity may be necessary to determine if these modifications produce significant changes in tear lipid profiles and enhance tear film stability.

Abbreviations

ACN

Acetonitrile

ADDE

Aqueous deficient dry eye

Cer

Ceramide

ChEs

Cholesterol esters

Ctrl

Controls

DED

Dry eye disease

EGLI

Lower eyelid meibum quality

EGLS

Upper eyelid meibum quality

EGTI

Lower eyelid meibum quantity

EGTS

Upper eyelid meibum quantity

FAs

Fatty acids

FDR

False Discovery Rate

Hex1Cer

Hexosyl ceramide

Hex2Cer

Dihexosyl ceramide

IPL

Intense pulsed light

IRON

Iterative rank-order normalization

ISs

Internal standards

LG

Lissamine green

LPA

Lysophosphatidic acid

LPC

Lysophosphatidylcholine

LPE

Lysophosphatidylethanolamine

LPG

Lysophosphatidylglycerol

LPI

Lysophosphatidylinositol

LPS

Lysophosphatidylserine

MEIBI

Meiboscore lower eyelid

MEIBS

Meiboscore upper eyelid

MGD

Meibomian gland dysfunction

MGs

Meibomian glands

MGX

Meibomian gland expression

MS

Mass spectrometry

NMR

Nuclear magnetic resonance

OAHFAs

(O-acyl)-w-hydroxy fatty acids; OA, oleic acid

OPLS-DA

Orthogonal partial least square-discriminant analysis

OSDI

Ocular surface disease index

OSM

Osmolarity

PA

Phosphatidic acid

PC

Phosphatidylcholine

PCA

Principal component analysis

PE

Phosphatidylethanolamine

PG

Phosphatidylglycerol

PI

Phosphatidylinositol

PLSDA

Partial least square-discriminant analysis

PS

Phosphatidylserine

QC

Quality control

SCH

Schirmer's

SD

Standard deviation

SM

Sphingomyelin

TBUT

Tear break-up time

TF

Tear film

TFLL

TF lipid layer

TG

Triacylglycerol

UHPLC

Ultra high-performance liquid chromatography

VIP

Variable importance in projection

WEs

Wax esters

Authors’ contributions

AA and EV supervised and wrote the article. AJM, SLP, JD, have included the patients in the study, conducted clinical tests, and collected tear samples. AJM, MR and AB have performed the IPL treatment. BA and OI have conducted lipidomic and statistical analyses. AA, EV, AJM, SLP, JD, MR, AB, DR, XP, NR, BA, OI supervised the work and revised the draft versions of the manuscript. All authors revised and approved the final manuscript. The authors declare that they have no conflicts of interest.

Funding

The authors wish to acknowledge the financial support received to carry out this work from: MINECO-Retos Fondos Feder (RTC-2016- 48231), Gobierno Vasco (IT 1510 22), PIBA 2020-1-0026, MINECO-Retos (PID2019-111139RB-I00), ELKARTEK (KK-2019-00086) to EV and FISS-21-RD21/0002/0041 to AA.

Data availability

The datasets used and analyzed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

This study was carried out by medically qualified personnel after receiving approval from the Euskadi Ethics Committee (code PI2019080). The analysis followed the tenets of the Helsinki Declaration on Biomedical Research in Humans and adhered to the ARVO statement involving human subjects. Before tear collection, written informed consent was obtained from all the participants, having fully explained the nature and possible consequences of the study.

Competing interests

The authors declare no competing interests.