Lipidomics Reveals Circulating Lipid Biomarkers for Retinopathy of Prematurity in Preterm Infants

Abstract

Objective

Retinopathy of prematurity (ROP) is a multifactorial eye disease affecting children born premature and is a leading cause of blindness in preterm infants worldwide. Although it has primarily been associated with high oxygen supplementation from respiratory support, there are indications that additional metabolic factors, like circulating lipids, may play a role in the disease’s pathophysiology.

Design

An exploratory study on the development of ROP in preterm infants was conducted in Denmark during 2018 and 2019. Infants who developed a maximum of stage 1 ROP were classified as having mild retinopathy, whereas those who developed stage 2 or 3 were classified as having severe retinopathy.

Participants

The study involved 110 preterm infants born before 32 weeks of gestational age.

Methods

During hospitalization in the neonatal wards, the infants were screened for ROP, and blood samples were collected every 2 weeks. A total of 485 lipid species were analyzed using lipidomics methodology, and mixed linear models were applied.

Main Outcome Measures

The association of lipids in early life (postnatal weeks 3–4) and their change throughout the study period was investigated.

Results

All lipid classes, involving 310 lipid species, changed significantly during the neonatal period. In early postnatal life, the lipid profiles of some classes (especially phosphatidylcholines and ether-linked phosphatidylcholines) were associated with the severity of ROP. In infants with stage 2 or 3 ROP, glycerophospholipids and sphingolipids changed more slowly compared with infants with no ROP. Similarly, glycerophospholipid pathways were enriched in infants with ROP.

Conclusions

The lipidomic plasma profile in preterm infants shows significant change across the neonatal period, involving all lipid classes. The association with ROP suggests that lipid metabolism may also play a role in ROP pathogenesis. Dyslipidemia associated with ROP should be addressed in further studies.

Financial Disclosure(s)

The authors have no proprietary or commercial interest in any materials discussed in this article.

Preterm infants born before 32 weeks of gestation face significant challenges due to their early transition from the intrauterine environment. Physical factors, exposure to oxygen and bacterial colonization, as well as the lack of maternal nutrients limit normal growth and increase the risk of both short- and long-term complications. Over the past decade, advances in neonatal care have improved the survival rates of extremely premature infants, but adverse outcomes remain common. As one part of the attempts to prevent negative health outcomes, supplemental nutrition, particularly lipids and amino acids, has become a focus area.1, 2, 3, 4, 5, 6, 7, 8, 9, 10 This has driven increasing interest in understanding the molecular mechanisms related to preterm growth and development.

Retinopathy of prematurity (ROP) is a neurovascular retinal disease and a leading cause of childhood blindness worldwide.¹¹ Known risk factors for ROP include low gestational age (GA), low birth weight (BW), small for gestational age (SGA), increased oxygen supplementation, and amount of days on parenteral nutrition.12, 13, 14, 15, 16 Previous literature further suggests that poor postnatal growth is a risk factor. All these factors possibly contribute to the metabolic disturbances involved in the development of ROP.¹⁷ Gestational age at birth and BW are the basis of most ROP screening program criteria.¹⁸ Researchers propose that growth retardation at birth might also be a useful indicator for prediction of ROP.¹⁹ Ongoing research is focused on screening methods based on algorithms combining the birth-related parameters (GA, BW, and SGA) with postnatal growth to predict ROP development.²⁰ Currently, severe ROP is treated with laser or anti-VEGF drugs, but despite treatment, some infants develop unfavorable outcomes.²¹ To improve long-term visual outcomes, it is desirable to determine biomarkers and therapeutic targets for ROP development at an early stage, allowing prevention or modulation of the disease occurrence.

Recent research has increasingly focused on the role of lipids in neonatal development and their association with ROP.22, 23, 24, 25, 26, 27, 28 The human retina is rich in lipids that are involved in retinal vascular diseases.²⁹ A recent study found that fatty acids (FAs) are the main energy source for photoreceptor mitochondria.³⁰ The Mega Donna Mega trial, a recent randomized multicenter trial, has shown that enteral supplementation of long-chain polyunsaturated FAs (LCPUFAs, such as docosahexaenoic acid and arachidonic acid) from birth until term may positively affect ROP development and other inflammation-related conditions in extremely preterm infants.^9,28 The same authors found significant differences of serum phospho- and sphingolipids during the first couple of postnatal weeks, with sphingosine-1-phosphate having the strongest association with severe ROP.²²

Although current research mostly describes only a few selected lipids, there is a need to characterize a broader spectrum, which can be accomplished using novel analytical techniques. Using mass spectrometry, lipidomics aims to analyze hundreds of lipid species that may play various roles in the human body.31, 32, 33 Lipidomics supplements classical analyses of clinical lipids (such as low-density lipoprotein [LDL] or high-density lipoprotein cholesterol) by providing higher specificity and identifying potential lipids involved in signaling between cells and organs and in immune responses. In this study, we present a lipidomic analysis of plasma samples from a prospective preterm cohort, aiming to determine pathophysiologic alterations of lipids across the neonatal period and explore associations between lipidome changes and ROP development.

Methods

Ethics Statement

This study was conducted in accordance with the tenets of the Declaration of Helsinki and was approved by the Danish Scientific Ethics Committee (H-23000653, De Videnskabsetiske Komitéer, Region Hovedstaden, Borgervænget 3, st. 2100 København) and the Data Protection Agency. Informed consent was obtained from parents or legal guardians before participation.

Study Population

Four neonatal departments in the Capital Region of Denmark participated in this exploratory prospective cohort study. Infants born in 2018 and 2019 were identified through the regional medical record system and local neonatal department lists. Handling physicians obtained informed consent from the caregivers. The inclusion criterion was GA <32 weeks, whereas exclusion criteria included death before term, parental withdrawal, or transfer to another hospital outside the region. We recruited 110 infants but excluded 4 due to death, high fever, and technical error during chemical analysis. The final study population consisted of 106 preterm infants. Thirty-six plasma samples were removed due to insufficient plasma amounts, repeated sampling for the time period, or missing metadata, leaving 286 samples for investigation. Blood samples were taken from the 14th postnatal day and every 2 to 3 weeks until discharge home without ROP or reaching term. Sampling distribution can be found in Figure S1 (available at www.ophthalmologyscience.org).

Clinical Patient Data

As per a previously published study, clinical data were extracted from the hospital system and cross-checked with the hospital medical record system for inconsistencies.¹⁷ The birth characteristics data included GA (in days), BW (in grams), and sex as assigned at birth. Small for gestational age at birth was defined as BW >2 standard deviations below the mean BW expected for the given GA.³⁴ Weight gain was calculated as the average weight gain from birth until the postnatal age at sampling (expressed as grams per week).³⁵ The routinely reported clinical data gathered longitudinally during the neonatal period included head circumference (in centimeters), length (in centimeters), weight (in grams), blood glucose level, and the amount of parenteral nutrition given. Several neonatal comorbidities were retrieved from the medical record system: ductus arteriosus persistence (defined as clinically significant by the attending neonatologist), intraventricular hemorrhage (IVH; stage 1-4), necrotizing enterocolitis (NEC; any stage), bronchopulmonary dysplasia (BPD), sepsis (i.e., at least one positive blood culture), major congenital malformation, and ROP (i.e., stage). For each of these parameters, we documented the infants’ postnatal age.

Neonate Nutrition

In our study population, infants received either their mother’s milk or donor human milk from the first day of life, supplemented with parenteral nutrition. A nutritional calculation sheet was used to ensure that each infant received adequate calories, protein, and fat in accordance with the European Society for Paediatric Gastroenterology, Hepatology and Nutrition guidelines (https://espghan.info/published-guidelines/). By the end of the first week of life, most infants had transitioned to full enteral feeding with human milk. Nearly all infants received human milk fortifiers (bovine protein supplements) mixed into the milk to support weight gain.

ROP Screening and Classification

The preterm infants received ROP screening by an experienced examiner that screened for ROP from the fifth postnatal week and until they saw mature retinal conditions.²⁰ In case of ROP, screening intervals were adjusted by the attending ophthalmologist according to disease severity. In case of early discharge to home, the infants continued ROP screening as planned, and also participated in the study, as an outpatient. Retinopathy of prematurity was described according to the international classification of ROP, 3rd edition.^36,37 The infants were separated according to maximum ROP stage reached: 81 no-ROP, 4 mild ROP (stage 1), and 21 moderate to severe ROP (stage 2–3), according to the worst eye.

Sample Preparation

Blood was collected in lithium heparin tubes and centrifuged to obtain plasma, which was subsequently used for lipid extraction. The sampling, plasma preparation, and storing procedure have been previously described by Bjerager et al.¹⁷ Lipids were extracted using a method based on the Folch protocol.³⁸ Instrumental analysis was performed using reversed-phase chromatography for lipid separation, coupled with high-resolution mass spectrometry detection. The standardized procedure was previously reported by Kronborg et al.^39,40

Samples were pseudo-anonymized and randomized into a new sequence before extraction. Ten microliters of plasma were aliquoted into 1.5 mL microcentrifuge tubes, and 10 μL 0.9% NaCl in water was added, followed by 28 μL internal standard working solution (Equisplash mix; Avanti). The concentration of stable isotope-labeled internal standard lipids was 10 mg/L, dissolved in extraction solvent (CHCl₃: MeOH = 2:1, v/v). An additional 92 μL extraction solvent was then added, and the suspension was vortexed. Proteins were left to precipitate for 30 minutes on ice. After precipitation, the suspension was centrifuged at 10 000 rpm for 3 minutes at 4 °C. Sixty microliters of extraction solvent were mixed with 60 μL of the lower phase from samples into liquid chromatography vials. After analysis in negative mode, samples were evaporated under a stream of nitrogen, redissolved in 240 μL extraction solvent, and analyzed in positive mode.

Chromatographic sequences included solvent runs, and 20% of control samples were prepared in the same way as experimental samples. Control samples included National Institute of Standards and Technology plasma 1950 (NIST SRM 1950; Merck), Sigma serum H4522 (H4522; Merck), pooled experimental samples, and blank samples. In blank samples, plasma was replaced with 0.9% NaCl.

Liquid Chromatography-Mass Spectrometry Analysis

Liquid chromatography separation was achieved with a high-performance liquid chromatograph Infinity II (Agilent), whereas mass detection was performed with a TimsTof Pro2 spectrometer (Bruker).

Lipids were separated using a Waters BEH C18-column (1.7 μm × 2.1 mm × 100 mm) with a matching precolumn, using a gradient of mobile phases. Mobile phase A was water containing 0.1% formic acid and 10 mM ammonium acetate. Mobile phase B was composed of isopropanol and acetonitrile (1:1, v/v) with the same additions of formic acid and ammonium acetate. The flow rate was 0.4 mL/min, and the temperature was maintained at 50 °C. At the beginning of the chromatographic run, the content of mobile phase B was 35%, which increased to 80% in 2 minutes. After an additional 5 minutes, it increased to 100%, where it was kept stable for 4 minutes. The system was then re-equilibrated to the initial conditions. The injection volume was 1 μL in positive mode and 2 μL in negative ionization mode.

Mass detection was performed as described by Kronborg et al³⁹ in profiling mode, ranging from 100 to 1350 m/z in both positive and negative modes. The acquisition speed was 2 Hz. Additional acquisition experiments were conducted to supplement the in-house lipid library using data-dependent acquisition mode and parallel accumulation serial fragmentation acquisition mode on control samples. Identification of compounds for lipidomics was performed in a targeted manner based on a premade library of target lipids in plasma (in-house library). The target list for lipidomics analysis is based on the consensus lipidomics paper by Bowden et al⁴¹ and the Metaboscape 2023b (Bruker) lipid group annotation procedure. Lipid names were adapted from the LIPID MAPS classification.⁴²

Data Processing

Raw data acquired by the Bruker TimsTof Pro2 were converted to MzML format using Proteowizard. Peak extraction was performed in a targeted manner with MZmine 2 software.⁴³ The MZmine procedure included data import, mass detection, target metabolite extraction, sample alignment, minimum intensity filter, feature raw filter, gap filling, and data export. During peak extraction, we retained only lipid features detected in >80% of the samples to ensure data robustness. Remaining missing values were imputed using the limit of detection. Peak tables created in this way were normalized using an in-house R normalization script with a median normalization procedure and reported as normalized areas.

Statistical Analysis

Premature infants were divided into 3 groups for statistical analyses: infants developing mild retinopathy (stage 1 in the worst eye), severe retinopathy (stage 2 or 3 in the worst eye) and infants not developing ROP. Continuous variables are presented as median [interquartile range], and categorical variables are reported as frequency (n) and percentage. Differences between groups at baseline were tested using the Kruskal-Wallis test for continuous variables and the chi-square test for categorical variables. Lipidomics data were log2-transformed before analysis. A linear mixed model was used to explore the difference in lipid levels between groups. Sampling periods were defined as consecutive 2-week intervals of postnatal age. When an infant had multiple samples within the same period, the sample closest to the midpoint of that interval was retained, ensuring at most 1 sample per infant per period. The model was first applied to the entire data set with all periods to explore the ROP-affected lipid profiles compared with controls:

Model I:

∼ ROP∗Period + GA + SGA + Sex + total parenteral nutrition + Sepsis + IVH + BPD + NEC + ductus arteriosus persistence + 1│subject_id

Then the model was applied to data from period 2 (postnatal age 14–28 days), which were mostly sampled before the onset of ROP, to discover prediction markers for ROP:

Model II (early biomarker detection):

∼ ROP+ GA + SGA + Sex + total parenteral nutrition + Sepsis + IVH + BPD + NEC + ductus arteriosus persistence

In the associated statistical tables, model coefficients are reported, including false discovery rate (FDR) corrected (q) and noncorrected P values. To identify the association of clinical parameters with lipids, significant lipids from previous sections were selected for Spearman correlation analysis. A heatmap of the absolute value of the change rate of lipids was plotted to visualize the difference in trends between groups. Lipidome metabolic pathways were explored with the tool BioPAN on LIPID MAPS Lipidomics Gateway (https://lipidmaps.org/biopan/). A two-tailed P value <0.05 was considered statistically significant. P values were adjusted using the Benjamini-Hochberg procedure to control for FDR. All statistical analyses were performed in R-4.3.1.

Results

Based on the statistical analyses conducted, the Results and consequently the Discussion are structured into distinct sections. The first section addresses phenotypic parameters, comorbidities, and ROP classification. This is followed by an examination of lipid alterations and changes in selected clinical parameters during the postnatal period, including their interrelationships. Subsequent sections explore associations between ROP and lipid profiles, presented separately for the early postnatal phase and the entire observation period. Finally, we conclude with a pathway analysis that contextualizes these findings within human metabolic processes.

Clinical Characteristics

The study investigated 106 infants, of which 24 were SGA (23%) and 61 were male (55%) (Table 1). Among all infants, 47 had at least one neonatal comorbidity. These infants had a lower GA (GA 28 weeks [26 weeks and 1 day; 29 weeks and 2 days]; and a lower BW 950 g [814 g; 1203 g]) than those without comorbidities. Retinopathy of prematurity was the most frequently observed morbidity, with a prevalence of 24%. Infants with ROP were more often born immature, SGA, and had more comorbidities (e.g., BPD, sepsis). They were also more frequently treated with total parenteral nutrition than infants without ROP.

Table 1.

Baseline Characteristics

	Overall	No ROP	Stage 1 ROP	Stage 2-3 ROP	P
N	106	81	4	21
Median GA, wks; days [IQR]	29;1 [27;4, 30;5]	29;5 [28;1, 31;0]	28;0 [27;1, 29:2]	26;5 [25;3, 29;0]	<0.001
Median birth weight, g [IQR]	1230 [950, 1518]	1311 [1013, 1609]	800 [614, 1066]	910 [712, 1120]	<0.001
Female, n (%)	45 (42.5)	35 (43.2)	0 (0.0)	10 (47.6)	0.202
SGA, n (%)	24 (22.6)	13 (16.0)	4 (100.0)	7 (33.3)	<0.001
TPN, n (%)	7 (6.6)	2 (2.5)	0 (0.0)	5 (23.8)	0.002
BPD, n (%)	16 (15.1)	8 (9.9)	0 (0.0)	8 (38.1)	0.004
Sepsis, n (%)	16 (15.1)	8 (9.9)	1 (25.0)	7 (33.3)	0.024
PDA, n (%)	22 (20.8)	14 (17.3)	0 (0.0)	8 (38.1)	0.065
NEC, n (%)	6 (5.7)	3 (3.7)	0 (0.0)	3 (14.3)	0.154
IVH, n (%)	9 (8.5)	9 (11.1)	0 (0.0)	0 (0.0)	0.219

BPD = bronchopulmonary dysplasia; GA = gestational age at birth; IQR = interquartile range; IVH = intraventricular hemorrhage (stage 1-4); NEC = necrotizing enterocolitis (any stage); PDA = patent ductus arteriosus; ROP = retinopathy of prematurity; SGA = small for gestational age; TPN = total parenteral nutrition.

Stage 1 is mild ROP; Stage 2-3 is moderate to severe ROP.

Throughout the postnatal period, minor differences in growth rates were observed among the 3 study groups. Infants with moderate to severe ROP (stage 2 and 3) exhibited the slowest growth rate (–20.08 g/week compared with the no-ROP group) as shown in Figure S2 (available at www.ophthalmologyscience.org).

Changes of Plasma Lipid Profile across the Neonatal Period

To investigate the influence of postnatal age on lipid profiles, we used a linear mixed model adjusting for potential confounders (model I; Fig 1). All lipid classes involving 310 lipid species out of 485 changed significantly during the neonatal period (Table S1, available at www.ophthalmologyscience.org). Among these, 77 lipid species increased in concentration, belonging to the following lipid classes: ether-linked phosphatidylcholines (PCOs), phosphatidylinositols (PIs), ether-linked phosphatidylethanolamines (PEOs), diacylglycerols (DGs), and sphingolipids (e.g., hexosylceramides).

Figure 1.

Mean change rate of 485 lipids during the postnatal period. Change of lipids were plotted by lipid head group and species. Significance: empty dot - nonsignificant; full dot - P < 0.05; full dot with a line - P < 0.01. The mean change rate of lipids was calculated from linear mixed model coefficients representing the average change of lipid level per period on log2 scale. Benjamini-Hochberg procedure was used for multiple testing correction. CE = cholesterol ester; Cer = ceramide; DG = diacylglycerol; dMePE = dimethylphosphatidylethanolamine; FA = fatty acid; HexCer = hexosylceramide; Hex2Cer = dihexosyl ceramide; LdMePE = lysodimethylphosphatidylethanolamine; LPC = lysophosphatidylcholine; LPCO = ether-linked lysophosphatidylcholine; LPE = lysophosphatidylethanolamine; LPEO = ether-linked lysophosphatidylethanolamine; PA = phosphatidic acid; PC = phosphatidylcholine; PCO = ether-linked phosphatidylcholine; PE = phosphatidylethanolamine; PEO = ether-linked phosphatidylethanolamine; PG = phosphatidylglycerol; PI = phosphatidylinositol; PS = phosphatidylserine; SHexCer = sulfatide; SM = sphingomyelin; TG = triglyceride.

Conversely, 35 lipid species decreased in concentration, belonging to the following lipid classes: phosphatidic acid, lysophosphatidylethanolamine, ether-linked lysophosphatidylcholines, FAs, cholesterol ester (CE), and sphingolipids (e.g., sulfatides and ceramides [Cer]).

The remaining lipid species that changed over time were distributed across various lipid classes: dihexosyl Cer (Hex2Cer), sphingomyelins, phosphatidylcholines (PCs), lysophosphatidylinositols, lysophosphatidylcholines, ether-linked lysophosphatidylethanolamines, phosphatidylglycerols (PGs), phosphatidylserines (PSs), and triglycerides (TGs).

Changes of Selected Clinical Parameters

Similar to lipids, some clinical parameters also change in early life (Table S2, available at www.ophthalmologyscience.org). The most significant changes were observed in weight, postnatal weight gain, and hemoglobin levels. Changes in oxyhemoglobin, carboxyhemoglobin, and blood urea nitrogen were less evident. No statistically significant differences were observed in blood glucose levels.

Correlation between Lipid Species and Clinical Parameters

Certain lipid species were associated with clinical parameters from routine blood samples (Fig S3, available at www.ophthalmologyscience.org). After FDR correction, 46 lipid species changed significantly, spanning 8 glycerophospholipid classes (PC, PCO, phosphatidylethanolamine [PE], PEO, PG, and PI), 2 sphingolipid classes (sulfatide and Cer), 1 cholesterol lipid (CE), and TGs. Among these, PI exhibited a strong positive correlation with growth (i.e., weight and postnatal weight gain). In contrast, PC was negatively correlated with growth.

Changes in Selected Clinical Parameters According to ROP Severity

Statistical models I and II were applied to assess significant differences between groups in selected clinical parameters. Based on model II, early postnatal albumin levels measured during weeks 3 and 4 were lower in infants who developed ROP stage 2 or 3 compared with those in the no-ROP group (Fig S4, available at www.ophthalmologyscience.org). When model I was applied, additional differences were observed: infants with ROP stages 2 and 3 exhibited reduced growth and slower weight gain compared with no-ROP infants. They also showed lower rates of change in hemoglobin levels. Other parameters did not differ significantly between groups (Fig S5, available at www.ophthalmologyscience.org).

Early Postnatal Profiles of Lipid Species According to ROP Severity

To assess early postnatal lipid changes associated with the development of ROP, we applied a linear mixed model (model II; Fig 2). For this analysis, the sample taken closest to the midpoint of the third and fourth postnatal weeks was chosen, with adjustments for potential confounders. We found 19 glycerophospholipid and 2 sphingolipid species that differed significantly (Fig S6, Table S3, available at www.ophthalmologyscience.org). No significant lipid species were regulated in infants with mild ROP compared with the no-ROP group. However, in infants who developed stage 2 or 3 ROP, 10 lipid species from the glycerophospholipid class (PC, PE, and PS) and FAs were lower, whereas sphingolipids (Hex2Cer) and glycerophospholipids (PCO and PEO) were higher. The lipid profiles also exhibited different direction of changes in certain lipid classes (especially TGs) depending on the severity of ROP.

Figure 2.

Volcano plot of early lipid profile in moderate-severe ROP groups (stage 2, 3) compared with the no-ROP group (stage 0). The lipid classes that were most altered were glycerophospholipids and sphingolipids. Statistical significance is assigned with a line and set to P = 0.05. Red dot is assigned to lipids that were increased and blue dot to lipids that were decreased in moderate-severe ROP group. The estimate indicating the log2 fold change between the 2 groups was calculated from linear mixed model coefficients, and Benjamini-Hochberg procedure was used for multiple testing correction. dMePE = dimethylphosphatidylethanolamine; FA = fatty acid; Hex2Cer = dihexosyl ceramide; PC = phosphatidylcholine; PCO = ether-linked phosphatidylcholine; PE = phosphatidylethanolamine; PEO = ether-linked phosphatidylethanolamine; ROP = retinopathy of prematurity.

Change of Lipid Species across the Neonatal Period According to ROP Severity

To investigate postnatal lipid changes associated with the development of ROP across the neonatal period, we used a linear mixed model (model I; Fig 3). The mean lipid levels across the postnatal period were similar in the group with stage 1 ROP and the control group. Thirty lipid species were differentially altered after FDR correction, showing diverse postnatal change rates across the 3 study groups (Table S4, available at www.ophthalmologyscience.org). Infants with stage 2 or 3 ROP exhibited distinct change rate patterns compared with the other 2 groups, characterized by a generally lower rate of lipid abundance change. However, a few lipid species, such as TGs, had higher change rates in the stage 2 or 3 ROP group compared with the stage 1 ROP group.

Figure 3.

Heatmap of change rates of the 30 top significant lipids across the entire neonatal period in stage 1 and stage 2/3 ROP groups compared with no-ROP group. The mean change rate of lipids was calculated from linear mixed model coefficients, and Benjamini-Hochberg procedure was used for multiple testing correction. CE = cholesterol ester; dMePE = dimethylphosphatidylethanolamine; LPC = lysophosphatidylcholine; LPE = lysophosphatidylethanolamine; PC = phosphatidylcholine; PCO = ether-linked phosphatidylcholine; PEO = ether-linked phosphatidylethanolamine; PG = phosphatidylglycerol; PI = phosphatidylinositol; ROP = retinopathy of prematurity; SHexCer = sulfatide; TG = triglyceride.

Lipid Metabolism Pathways According to ROP Severity

Changes in lipid networks at different stages of ROP were analyzed using BioPAN methodology for pathway analysis. Results from an early period (period 2) are presented in Figure 4 and results from the entire observed infancy in Figure S7 (available at www.ophthalmologyscience.org). In Figure 4, infants with stage 1 ROP already showed a changed pathway involving PS and PE. The group with stage 2 or 3 ROP exhibited an activated pathway converting DG → PC → PS → PE compared with controls, indicating the process of PE biosynthesis was affected. Additionally, the TG biosynthesis pathway was enriched in the more severe ROP group, with an upregulated DG → TG conversion.

Figure 4.

Lipid networks in period 2 for (A) mild ROP and (B) moderate-severe ROP, compared with no-ROP infants, generated by BioPAN. Active lipids are represented by green nodes, and active pathways are colored with green shadow. Green and purple arrows indicate active and suppressed pathways with Z scores (|Z| > 1.645 at P < 0.05), respectively. DG = diacylglycerol; PA = phosphatidic acid; PC = phosphatidylcholine; PE = phosphatidylethanolamine; PS = phosphatidylserine; ROP = retinopathy of prematurity; TG = triglyceride.

Discussion

Preterm infants are prone to multiple comorbidities that appear immediately after birth or later in life. Our results focus on ROP and its link to metabolism. Most reports¹⁴ have associated ROP with GA, BW, and oxygen saturation. In recent years, the metabolism connection has also been researched.^44,45 Clinical evidence that interventions targeting lipid metabolism can affect ROP occurrence has been provided by Hellström et al,^27,28 in which the authors predominantly focused on LCPUFAs. Our results complement these findings and suggest that ROP occurrence and its severity are also associated with other lipid species.

Preterm Neonates Are Affected by Multiple Morbidities

Being born SGA is a strong predictor of poor postnatal growth and neurodevelopment,⁴⁶ and it can also lead to the development of cardiometabolic disorders later in life.⁴⁷ Both immaturity and SGA are also known as major risk factors for the development of ROP, and these factors have previously been used in a risk-based prediction model for ROP treatment.48, 49, 50 Besides ROP, premature infants have severe comorbidites (e.g., sepsis, IVH), patent ductus arteriosus, NEC, and BPD. Notably, metabolic disturbances in glucose and lipid metabolism often precede these adverse events.^17,32,51

Postnatal Growth Rate Is Associated with ROP Development

Consistent with previous research,^20,52 our findings indicate that slower postnatal growth is associated with the development of stage 2 or 3 ROP. VanderVeen et al⁵³ proposed that optimizing nutritional intake, particularly lipids and total caloric consumption, could improve ROP outcomes. Beyond nutrition, impaired postnatal growth may also stem from physiological limitations, such as the gastrointestinal tract’s capacity to absorb and process nutrients.⁵⁴ Additionally, underdeveloped lungs and BPD have been linked to growth delays,⁵⁵ as have systemic stressors like sepsis or infections, which can further hinder growth.⁵⁶

Plasma Lipid Composition Is Extensively Modified in the Postnatal Period

Recently, Burugupalli et al³² described changes in lipid profiles in the Australian Barwon Infant Study, which is based on normal term born infants. Consistent with our findings, the study reported that the changes of lipid profiles were mostly pronounced during early postnatal life. However, although they observed a predominantly increasing trend in lipid features in the first 6 months, our study, which includes only preterm infants, suggested a mixed trend of both upregulation and downregulation. Differences in lipid profiles between preterm and term babies were also observed in cord blood lipid measurements, in which preterm babies had higher levels of total cholesterol, LDL, and very LDL cholesterol, and lower values of TGs.⁵⁷ The authors also concluded that gestational week 34 is crucial for lipid metabolism development. Moreover, dyslipidemia may persist further in life, and children that were born SGA in particular show increased values of total cholesterol, LDL cholesterol, and TGs and lower values of high-density lipoprotein cholesterol, which can contribute to development of cardiovascular diseases.⁵⁸ Understanding lipid metabolism coupled with an appropriate nutritional protocol can improve a catch-up growth phase and decrease risks associated with prematurity and low BW.⁵⁹

Lipids Are Correlated with Clinical Measurements

Although correlations do not establish causality in pathologic development, they can reveal connections between seemingly unrelated aspects of metabolism and prompt formation of new hypotheses for future studies. Our results show that a range of clinical measurements correlated with plasma lipids. As expected, there is a positive correlation between weight and CEs.⁶⁰ Cholesterol esters are also positively correlated with levels of albumin, bilirubin, hemoglobin, and oxyhemoglobin. The 2 CEs most strongly associated with BW and albumin, CE 22:6 and CE 20:4, contain LCPUFAs. Another interesting group of lipids associated with clinical parameters are glycerophosphoinositols (PIs). They play a role in the innate immune system and are abundant in lung mucosa.⁶¹ Our study showed that they are positively associated with infant growth rate and negatively associated with albumin, creatinine, and hemoglobin levels. These findings suggest that infants with better growth are better protected against airway infections due to higher values of PIs. It should also be noted that inositol supplements have been proven inefficient in preventing respiratory distress syndrome.⁶²

Plasma Composition of Nonconjugated FAs Changes during the Neonatal Period and ROP Development

Arachidonic acid (C20:4) and docosahexaenoic acid (C22:6) are LCPUFAs and are important for neonatal development.⁶³ During prenatal growth, they are selectively extracted from the maternal circulation and are enriched in fetal metabolism.⁶⁴ Although supplementation results in higher plasma concentrations, there is still a need to standardize FA sources and dosages for the prevention of prematurity-associated diseases.^65,66 Several studies have suggested that lipid deficiency negatively impacts the development of the retina as well.^26,44 In a mouse model, supplementation with LCPUFAs reduced pathologic retinal angiogenesis,²⁵ and recently, Hellström et al²⁸ reported a 50% reduction in severe retinopathy in extremely premature infants after LCPUFA supplementation. Fu et al³⁰ further proposed that dietary LCPUFA increases circulating adiponectin levels and the expression of the retinal adiponectin receptor, thereby preventing the development of ROP.^30,67 The levels of nonconjugated C22:6 and C20:4 FAs in our study showed only a weak trend with ROP, whereas some of the PC-conjugated lipids had higher statistical significance. In the severe ROP stages 2 and 3, we also recorded a deficiency of C18:3 FA, which is most likely α-linolenic acid and another LCPUFA.

Plasma Composition of Sphingolipids Changes during the Neonatal Period and ROP Development

The class of sphingolipids is characterized by a sphingoid base backbone. It includes diverse subgroups, among which sphingomyelin and Cer are the most prevalent in human plasma. Ceramides can be further subdivided into hexosylceramide, Hex2Cer, and sulfatide.⁶⁸ The most commonly attached hexose sugar is glucose, but other sugars cannot be excluded. Sphingolipids are involved in multiple pathways and cellular processes in the organism, including growth regulation, cell trafficking, apoptosis, and inflammatory responses.⁶⁹ Because of this, they play a significant role in major pathologies such as cancer, neurodegenerative disorders, autoimmune diseases, metabolic syndrome, and skin integrity disorders. Multiple researchers have also reported their importance in preterm birth and its comorbidities. For example, Rusconi et al⁷⁰ hypothesized a link to NEC in a case-control study in which cases had decreased abundances of Cer and increased content of sphingomyelin, speculating a link to inflammatory disorders. Thomas et al⁷¹ reviewed the role of sphingosine-1-phosphate signaling in connection to oxidative lung injury and BPD. Sphingolipids are also involved in multiple retinopathy-related pathological processes, and sphingolipid-targeted drugs are good candidates for retinopathic disease therapeutic options.⁷² The complexity of the pathway⁷³ offers vast space for biomarker discovery. Similarly to our study, Nilsson et al²² concluded that the sphingolipid pathway changes drastically during the neonatal period and that ROP is associated with certain sphingolipid species, which could be used for prediction of severe ROP. In line with these observations, our study also suggests Hex2Cer as a potential marker for severe ROP prediction.

Plasma Composition of Glycerophospholipids Changes during the Neonatal Period and ROP Development

Glycerophospholipids are a diverse class of lipid molecules with 2 FAs attached to the glycerol backbone via an ester bond (PC and PE) or via an ether bond (PCO and PEO). Other major groups are PIs, PSs, phosphatidic acids, and PGs. Besides their extensive presence in all organs as part of cell membranes, glycerophospholipids also perform important tasks, such as transporting FAs around the body. In neonatology, they have gained interest because of the placenta’s selective ability to transfer LCPUFAs to the fetus.^74,75 According to these studies, docosahexaenoic acid and arachidonic acid play a crucial role in neonatal nutrition due to their role in the autoimmune response and antioxidant activity. This was also confirmed by Hellström et al²⁸ in the case of ROP. Additionally, the observed downregulated PC biomarkers in our study suggest that other polyunsaturated phospholipids and FAs are also involved in the process.^22,32 These findings could help further optimize polyunsaturated FA supplements for preterm infants. Besides downregulated polyunsaturated ester-linked PCs, we also identified upregulated PCOs as early biomarkers of ROP. Although PCOs are reported in cases of neurological disorders, cancer, and metabolic disorders,⁷⁶ their roles are mostly unknown in the field of neonatology, which raises the need for future research.

Albumin and Hemoglobin Are Associated with ROP Occurrence and Development

Both clinical parameters, albumin and hemoglobin, have already been extensively researched, and our data confirms previous findings.^77,78 Preterm infants are known to have lower levels of hemoglobin. This is associated with early neurologic functioning in neonates, and they often require blood transfusions to prevent adverse effects.⁷⁹ Our results suggest that hemoglobin levels decrease during the postnatal period, and this decrease is even more evident in infants developing ROP.

In contrast, an early clinical biomarker that showed significance was albumin. Recent studies^17,80 demonstrated an association between low BW, GA, and low plasma albumin levels. Our data show that low plasma albumin levels in weeks 3 and 4 after birth are associated with stage 2 or 3 ROP, which could indicate a lack of amino acid load in the body contributing to reduced growth gain.

Network Analysis Suggests a Shift in Lipid Pathways due to ROP Development

Pathway analysis revealed that ROP induces a shift in lipid conversion even before its clinical onset, and it identified genes that may play a role in these molecular pathways. For example, the enzyme diacylglycerol acyltransferase 2 converts diglycerols into triglycerides.⁸¹ In our data, this shift from DGs to TGs is evident in early ROP development and may also be involved in retinoid metabolism and potentially connected to metabolic dysfunction-associated steatotic liver disease pathology.⁸²

The pathway associated with severe ROP involves the conversion of diacylglycerophosphates (phosphatidic acids) into glycerophosphoserines (PSs). The predicted genes responsible for this conversion encode CDP-diacylglycerol synthase 1 and phosphatidylserine synthase 1. The role of this part of the pathway in ROP is not well understood, but Frostegård et al⁸³ reported PSs to be a novel lipid target acting upstream of VEGF. Suppression of PS signaling using the candidate drug Annexin A5 could improve or supplement treatments with anti-VEGF drugs in retinal vascular diseases.

It should be noted that infants with severe ROP often receive higher amounts of parenteral lipid intake. Although we can adjust for intake at the individual-sample level, the overall nutritional exposure of each infant cannot be fully accounted for due to methodological limitations. Consequently, the observed lipid profile shifts may reflect both alterations in lipid metabolism and differences in parenteral nutrition. These findings should therefore be interpreted with caution.

Strengths and Limitations

The prospective design of this study is a strength. The study cohort is recruited from a geographic area representing preterm infants born in the Capital Region of Denmark. The 4 hospitals followed the same protocol for enteral and parenteral nutrition, which further strengthens the consistency of the study findings. A limitation is that the pathophysiology of ROP is very complex, and as such, this cohort study cannot make conclusions on disease causality. Biomarkers presented in results are associated with ROP. Their role in pathophysiologic changes should be confirmed in further treatment trials to prove causality. Normal values of lipid concentrations for developing children are not known, so severity of dyslipidemia is difficult to characterize. The small number of subjects with ROP stage 1 diagnosis severely limits the statistical power for that stage. Finally, lipids are identified by their head group, number of carbons, and number of double bonds. The exact FA composition, including double bond position, is lacking, and therefore, there are still possibilities for improvement in methodology.

Conclusion and Perspectives

The plasma lipidomic profile in preterm infants exhibits change over the neonatal period involving all major lipid classes. In infants with stage 2 or 3 ROP, significant alterations were observed in glycerophospholipids and sphingolipids, with most biomarkers showing a general trend toward decreasing concentrations. Additionally, there was an enrichment in glycerophospholipid metabolic pathways, with increasing levels of reactants correlating with greater disease severity. Based on the information acquired, further trials should be designed with a focus on correcting dyslipidemia in preterm infants.

Manuscript no. XOPS-D-25-00665.