Serum Olink proteomics-based exploration of novel predictive biomarkers for necrotizing enterocolitis

Summary

Necrotizing enterocolitis (NEC) is a life-threatening gastrointestinal emergency in preterm infants, yet reliable serum candidates for presymptomatic risk stratification remain lacking. In this exploratory study, serial pre-diagnostic serum samples from a prospective discovery cohort were profiled using the Olink proteomics platform. NEC cases and controls were individually matched for gestational age, birth weight, and sampling time. Top-ranked proteins were assessed via enzyme-linked immunosorbent assay in an independent validation cohort and examined in surgical intestinal tissues. Fibroblast growth factor 5 (FGF5) showed the highest discriminative performance with an area under the receiver operating characteristic curve of 0.848. Lower circulating FGF5 levels were reproduced in the validation cohort, although the apparently perfect discrimination was imprecise because of the small sample size. Intestinal FGF5 expression was also reduced in NEC specimens. These findings support FGF5 as a promising exploratory biomarker for early NEC risk assessment, warranting larger multicenter validation and mechanistic investigation.

Graphical abstract

Highlights

• •Four downregulated serum proteins are novel ultra-early biomarkers for preterm NEC
• •FGF5 shows the strongest temporal stability, validated by ELISA and tissue staining
• •Recombinant FGF5 attenuates TNF-α in IEC-6 cells, suggesting a protective role

Health sciences

Introduction

Necrotizing enterocolitis (NEC) is one of the most severe gastrointestinal emergencies during the neonatal period. It affects 7%–9% of extremely low birth weight infants¹ and remains a leading cause of death in infants born at a gestational age (GA) < 28 weeks between postnatal day 15 and postnatal day 60.² Current clinical management is largely supportive and often fails to halt rapid disease progression.³ Approximately 20%–50% of NEC patients ultimately require surgery due to intestinal perforation or progressive necrosis,⁴ and postoperative survivors may develop structural complications and long-term functional impairment. At present, NEC is diagnosed based on a composite assessment of clinical manifestations, laboratory examinations, and imaging findings, which means many cases are not recognized until they have progressed to an advanced stage.⁵ Blood-based assays offer practical advantages-minimal invasiveness, feasibility in small infants, and rapid turnaround-highlighting their potential for clinically actionable pre-symptomatic risk stratification.

A multitude of candidate biomarkers have been proposed, yet few show robust pre-diagnostic performance across centers and sampling times. Blood proteomics has garnered increasing interest because it can capture downstream functional alterations closer to phenotype than genomics or transcriptomics.⁶ However, the abundance of blood proteins spans several orders of magnitude,⁷ posing major analytical challenges for existing detection platforms, particularly when neonatal sample volumes are limited.⁸ The Olink platform, based on proximity extension assay (PEA) technology, helps address these constraints and enables multiplex screening of serological biomarkers from small sample volumes.⁹ PEA-based proteomics has been used to build integrated diagnostic models in complex diseases¹⁰ and to profile early postnatal blood proteomic trajectories in extremely preterm infants, providing insights into neonatal development.¹¹

In the clinical diagnosis of NEC, the diagnostic efficacy of hematological indicators is often limited by disease severity and the timing of sample collection.¹² Although studies have used serum proteomics and metabolomics to screen for NEC-associated biomarkers,¹³ most are restricted to single-time-point sampling or specimens collected at the time of diagnosis. Systematic characterization of dynamic changes in the serum proteome across defined pre-diagnostic time windows remains limited, and few studies have integrated omics findings with tissue-level expression data and functional assays. This knowledge gap impedes translational progress, particularly the transition from discovering predictive biomarkers to understanding disease-relevant biology.

Fibroblast growth factor 5 (FGF5), a member of the fibroblast growth factor family, has been implicated in tissue repair and regeneration and in the modulation of inflammatory injury in several experimental contexts.¹⁴ However, its role in the pathogenesis of NEC in preterm infants remains poorly defined. In our preliminary, strictly exploratory Olink-based serum proteomics screening, we observed a consistent decrease in serum FGF5 levels prior to NEC diagnosis, suggesting a potential association with subsequent disease development. Based on these exploratory findings, we hypothesized that down-regulation of serum and intestinal FGF5 precedes the clinical diagnosis of NEC and may reflect impaired mucosal repair capacity. Accordingly, in this prospective cohort, we aimed to (1) delineate time-resolved serum proteomic signatures in the two weeks preceding NEC diagnosis, (2) evaluate the predictive performance of candidate proteins for early NEC identification, and (3) validate and functionally characterize FGF5 as a potential protective mediator of intestinal injury.

Results

Study population and sample collection

From May 3, 2022, to October 16, 2023, 385 preterm infants were screened, of whom 26 developed NEC during hospitalization. Pre-diagnostic serum samples collected within 14 days before NEC diagnosis were available for 15 NEC infants. After Olink quality control (QC) (Figure S1), one NEC sample (N6) failed QC and was excluded from downstream analyses. The final proteomics dataset therefore included 29 infants (14 NEC and 15 matched controls; Figure 1). The median postnatal age at NEC diagnosis was 13 (interquartile range [IQR]: 8–22) days in the NEC group. Baseline characteristics were comparable between groups (Table 1): mean GA at birth was 29.07 ± 2.28 weeks in controls and 29.10 ± 2.43 weeks in NEC infants; median birth weight (BW) was 1,130 (1,015–1,440) g vs. 1,080 (1,005–1,365) g, respectively. No significant between-group differences were observed in perinatal characteristics or early clinical course (all p > 0.05), supporting adequate matching for proteomic comparisons. For external validation, an independent cohort of 489 preterm infants was prospectively recruited between February 25, 2024, and March 18, 2026, with identical inclusion criteria, sample collection protocols, and 1:1 matching criterion applied. Of the 22 infants diagnosed with NEC in this validation cohort, pre-diagnostic serum samples were successfully obtained from 11 infants, who were 1:1 matched with 11 non-NEC control infants.

Figure 1.

Study design and flowchart of cohort selection

From May 3, 2022, to October 16, 2023, a total of 385 preterm infants with gestational age <32 weeks and/or birth weight <1,500 g was enrolled. Among them, 26 developed NEC and 359 remained free of NEC. For the proteomic study, 15 NEC infants with available pre-diagnostic serum samples were matched 1:1 to 15 controls. After Olink quality control (QC) procedures, 15 control samples and 14 NEC samples were retained for analysis. For external validation, an independent cohort of 489 preterm infants was prospectively recruited from February 25, 2024, to March 18, 2026, using identical inclusion and 1:1 matching criteria. GO, Gene Ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes; PPI, protein-protein interaction; IHC, immunohistochemistry; IF, immunofluorescence.

Table 1.

Basic characteristics of the study population

Characteristic	Total (n = 29)	Control (n = 15)	NEC (n = 14)	p
Neonatal factors
Malea	12 (41.4)	7 (46.7)	5 (35.7)	0.825
Gestational age (week)b	29.08 ± 2.32	29.07 ± 2.28	29.10 ± 2.43	0.968
Birth weight (g)c	1110 (1010, 1430)	1130 (1015, 1440)	1080 (1005, 1365)	0.760
1-min Apgar scorea				0.090
Severe asphyxia	2 (6.9)	2 (13.3)	0 (0.0)
Mild asphyxia	8 (27.6)	2 (13.3)	6 (42.9)
Normal	19 (65.5)	11 (73.3)	8 (57.1)
5-min Apgar scorea				1.000
Mild asphyxia	3 (10.3)	2 (13.3)	1 (7.1)
Normal	26 (89.7)	13 (86.7)	13 (92.9)
CRP (μg/mL)c	0.16 (0, 1.4)	0.1 (0, 0.23)	0.9 (0.01, 6.56)	0.058
WBC (×109/L)b	8.94 ± 4.19	9.11 ± 4.74	8.77 ± 3.68	0.828
NEUT %c	2.98 (1.77, 4.92)	2.55 (1.73, 4.48)	3.58 (1.86, 5.57)	0.477
Perinatal factors
Mother’s ageb	32.10 ± 4.95	31.80 ± 4.44	32.43 ± 5.58	0.741
Twinsa	8 (27.6)	6 (40.0)	2 (14.3)	0.215
Antenatal corticosteroid exposurea				1.000
Received	28 (97.0)	14 (93.0)	14 (100.0)
Not received	1 (3.0)	1 (7.0)	0 (0.0)
Cesareana	16 (55.2)	9 (60.0)	7 (50.0)	0.867
PROMa	11 (38.0)	4 (27.0)	7 (50.0)	0.362
Fetal distressa	5 (17.2)	2 (13.0)	3 (21.0)	0.651
Exclusive breastfeedinga	15 (51.7)	9 (60.0)	6 (42.9)	0.581

PROM, premature rupture of membranes; CRP, C-reactive protein; WBC, white blood cell count; NEUT%, neutrophil percentage.

^aNumber of cases (percentage).

^bMean ± SD.

^cMedian (first quartile, third quartile).

Time-resolved serum proteomic signatures preceding NEC

Global serum proteomic profile in the 0–14-day window

A heatmap of normalized protein expression (NPX) values for the 92 proteins was generated to visualize the global expression profiles across the 29 infants in the discovery cohort (Figure S2). Principal component analysis (PCA) based on the NPX matrix showed that the first two principal components (PC1 and PC2) explained 36.02% and 11.78% of the variance, respectively (cumulative 47.80%; Figure 2A). No clear group separation was observed in the PCA space, suggesting that NEC-associated differences were not captured by the major axes of global variance.

Figure 2.

Screening of DEPs between the control and NEC groups in the discovery cohort

(A) Principal component analysis (PCA) of protein expression profiles showing partial separation between the two groups.

(B) Venn diagram of DEPs identified in the 0–7, 8–14, and 0–14 days pre-NEC diagnosis windows.

(C–E) Volcano plots displaying DEPs between the control (Ctrl) and NEC groups during the 0–14, 0–7, and 8–14 days pre-diagnostic windows, respectively. Red dots indicate upregulated proteins, and blue dots indicate downregulated proteins in NEC (|estimate| above the vertical dashed line and p value below the horizontal dashed line).

Differentially expressed proteins and multiple-testing control

Using paired comparisons between NEC cases and their matched controls, we identified four proteins with nominal differences (unadjusted p < 0.05): FGF5, glial cell line-derived neurotrophic factor (GDNF), tumor necrosis factor ligand superfamily member 11 (TRANCE), and urokinase-type plasminogen activator (uPA) (Figure 2C). All four proteins were significantly reduced in NEC infants compared with controls. After correcting for multiple testing across 92 proteins using the Benjamini-Hochberg false discovery rate (BH-FDR), no proteins remained significant at q < 0.05.

Time-window-specific patterns (0–7 days vs. 8–14 days before NEC diagnosis)

To evaluate the temporal dynamics of these disease-associated proteins, we next performed time-resolved analyses in two pre-diagnostic windows, 0–7 and 8–14 days before NEC diagnosis. Using the same criteria as in the overall analysis, we compared protein expression between NEC and control infants within each time window. Among the four differentially expressed proteins (DEPs) identified in the 0–14-day window, FGF5 remained significantly down-regulated in NEC infants in both the 0–7-day (Figure 2D) and 8–14-day (Figure 2E) strata. In contrast, the other three proteins showed less consistent patterns across time windows. Venn diagram analysis visualized the overlap of DEPs across the 0–7-, 8–14-, and combined 0–14-day windows (Figure 2B), again highlighting FGF5 as the only protein persistently differentially expressed in NEC infants across all pre-diagnostic intervals examined. These results support FGF5 as the most temporally stable component of the pre-diagnostic serum signature.

Exploratory predictive performance of serum proteins for NEC

We then evaluated the exploratory predictive performance of the identified DEPs for discriminating NEC from controls. Receiver operating characteristic (ROC) analyses were conducted separately for each of the four DEPs. All four proteins achieved area under the ROC curve (AUC) values >0.70, indicating at least acceptable discrimination. In the 0–14 days pre-NEC window, FGF5 demonstrated the highest AUC (0.848), followed by TRANCE (0.762), GDNF (0.731), and uPA (0.710) (Figure 3A). During the 0–7 days pre-NEC window, FGF5 achieved an AUC of 0.911 (Figure 3B), while in the 8–14 days pre-NEC window, it maintained an AUC of 0.781 (Figure 3C). These findings support the potential utility of this small serum protein panel, and particularly FGF5, for early risk stratification in very preterm infants. Although these analyses were performed in a small cohort and should be interpreted as exploratory, the consistently high AUC of FGF5 and its central contribution to discriminating NEC from controls support its candidacy as a promising early serum marker.

Figure 3.

Diagnostic performance of serum DEPs across pre-diagnostic windows and validation of the FGF5 predictive model

(A–C) The area under the receiver operating characteristic curve (AUC) values of selected DEPs for distinguishing NEC from Ctrl across the 0–14, 0–7, and 8–14 days pre-diagnostic windows, respectively, in the discovery cohort.

(D) ROC curve for the full dataset (n = 29).

(E) Serum FGF5 levels in the validation cohort using enzyme-linked immunosorbent assay (ELISA). ∗∗∗∗p < 0.0001.

(F) Basic characteristics of the validation cohort. ^aNumber of cases (percentage); ^bmean ± standard deviation (SD).

To validate the proteomic findings derived from the discovery cohort, we quantified serum FGF5 concentrations in an independent validation cohort via enzyme-linked immunosorbent assay (ELISA). Consistent with the Olink NPX data, ELISA verification revealed significantly lower serum FGF5 levels in infants before NEC diagnosis compared with controls (Figure 3E); baseline characteristics of this independent cohort are presented in Figure 3F. The direction and magnitude of between-group differences closely mirrored those from the Olink analysis, strengthening the evidence for FGF5 downregulation as a pre-diagnostic serum feature in very preterm infants. However, the AUC of 1.0 with 100% sensitivity and specificity in this small validation cohort likely reflects overfitting due to limited sample size and low between-subject variability, precluding definitive conclusions. Independent validation in larger cohorts is required before clinical translation can be considered.

Internal validation was performed on the discovery cohort using the Firth logistic regression model, which exhibited favorable apparent discriminative performance (Figure 3D) with an AUC of 0.848 (95% confidence interval [CI]: 0.686–0.967). At the optimal cutoff value of 0.402, the model achieved a sensitivity of 0.929, a specificity of 0.667, and an accuracy of 0.793. Stratified 3-fold cross-validation showed consistent performance (Figure S3), with a median AUC of 0.840 (IQR: 0.045), median sensitivity of 0.750 (IQR: 0.100), and median specificity of 0.600 (IQR: 0.100). Standard bootstrap validation further supported these findings, with a median AUC of 0.857 (95% CI: 0.686–0.967) and most bootstrap AUC values falling between 0.70 and 0.95. The 0.632+ bootstrap-adjusted AUC was 0.846, closely aligning with the original estimate. Collectively, AUC values derived from the four internal validation approaches were highly consistent (all > 0.840), indicating that pre-NEC serum FGF5 has promising discriminatory capacity for early NEC prediction. Nevertheless, given the wide confidence intervals and limited sample size, these findings remain exploratory, and further validation in larger independent cohorts is needed to confirm our results.

Characterization of FGF5 as a candidate NEC marker

FGF5 expression in terminal ileum tissue

Because circulating protein levels may not fully reflect local expression at the site of injury, we next assessed FGF5 expression in terminal ileum tissue. Immunohistochemical staining revealed markedly reduced FGF5 protein signals in the terminal ileum of NEC infants compared with controls (Figure 4A). Immunofluorescence analyses confirmed this finding, demonstrating lower FGF5 expression in the intestinal mucosa of NEC cases (Figure 4B). The concordant reduction of FGF5 in both serum and intestinal tissue suggests that FGF5 dysregulation in NEC is not confined to the circulation but also occurs at the primary site of pathology. Together, these serum and tissue data support FGF5 as a biologically relevant marker linked to NEC-associated intestinal injury and provide a rationale for further functional studies on its potential protective role in intestinal barrier homeostasis.

Figure 4.

Comparative analysis of FGF5 expression in NEC vs. control groups across tissue and cellular models

(A) Immunohistochemical staining and semi-quantitative scoring of FGF5 in terminal ileum tissues from control (Ctrl) (n = 3) and NEC (n = 3). Scale bars, 20 μm.

(B) Immunofluorescence staining of FGF5 in terminal ileum tissues from Ctrl (n = 4) and NEC (n = 3), with quantification of fluorescence intensity. Scale bars, 200 μm.

(C) Western blot analysis of FGF5 protein expression in IEC-6 cells from NEC (n = 3) and Ctrl (n = 3) groups.

(D) Reverse transcription quantitative polymerase chain reaction (RT-qPCR) analysis of FGF5 mRNA expression in IEC-6 cells (Ctrl: n = 5; NEC: n = 5).

(E) RT-qPCR assessment of the effect of FGF5 on inflammation in IEC-6 cells: Ctrl (n = 5), FGF5 treatment (n = 5), and NEC (n = 4). Data are presented as mean ± standard deviation (SD). ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001.

To further explore the protective role of FGF5 in intestinal epithelial cells, we used the rat small intestinal crypt epithelial cell line IEC-6 as an in vitro model. Western blotting and reverse transcription quantitative polymerase chain reaction (RT-qPCR) were performed to detect FGF5 protein and mRNA expression in control and NEC cellular models, respectively. Results showed that both protein (Figure 4C; uncropped blots are provided in Figure S4) and mRNA (Figure 4D) expression of FGF5 in the NEC group were significantly lower than those in the control group, consistent with our previous in vivo findings, suggesting that endogenous FGF5 may be involved in maintaining intestinal homeostasis.

Functional effects of FGF5 in experimental models of intestinal injury

Subsequently, we treated the NEC cellular model with 50 ng/mL recombinant FGF5 protein. RT-qPCR showed that TNF-α mRNA expression was significantly decreased in the FGF5 intervention group compared with the NEC group (Figure 4E). In summary, FGF5 is not only an early serum marker for NEC but may also exert intestinal protective effects by downregulating inflammatory factors, thereby identifying FGF5 as a novel candidate target for mechanistic investigations and therapeutic interventions in NEC.

Gene Ontology (GO) functional enrichment analysis was performed on the identified DEPs, and the enrichment profiles are visualized in Figures 5A and 5B. The molecular functions (MF) mediated by DEPs were predominantly enriched in signaling receptor activity (p = 2.5 × 10⁻⁴) and binding (p = 0.019), with all DEPs engaging in receptor-mediated signal transduction processes. For cellular components (CC), enriched annotations included extracellular space (p = 0.0019) and extracellular region (p = 0.0056), and all DEPs were localized in the extracellular environment. Furthermore, DEPs were significantly enriched in biological processes (BP) such as cell migration (p = 5.01 × 10⁻⁵), cell population proliferation (p = 0.00014), and signaling (p = 0.031), with all DEPs participating in the positive regulation of these processes. These findings indicate that extracellular matrix-associated signaling proteins (e.g., FGF5 and GDNF) and cell migration regulators (e.g., uPA and TRANCE) exhibit significant differential expression during NEC development.

Figure 5.

Functional enrichment and PPI network analyses of DEPs

(A) Distribution of DEPs across functional categories in Gene Ontology (GO) classification.

(B) Top 30 significantly enriched GO terms ranked by p values.

(C) Statistics of DEPs across pathway categories in Kyoto Encyclopedia of Genes and Genomes (KEGG) classification.

(D) Top 30 significantly enriched KEGG pathways ranked by p values.

(E) Global PPI network analyzed by the STRING database.

(F) Specific interaction network of DEPs. The q value reflects statistical significance after Benjamini-Hochberg multiple testing correction.

To further investigate the underlying biological processes and disease mechanisms, we performed Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis for DEPs using Fisher’s exact test. Figure 5C shows the hierarchical structure of enriched KEGG pathways and the distribution of DEPs among them. KEGG enrichment analysis revealed that DEPs were significantly enriched in the NF-κB signaling pathway (hsa04064, p = 9.19 × 10⁻⁵, q = 0.0019), breast cancer pathway (hsa05224, p = 0.00025, q = 0.0027), and calcium signaling pathway (hsa04020, p = 0.00104, q = 0.0073) (Figure 5D). These findings suggest that inflammatory response, abnormal cell proliferation, and calcium signaling disruption are key pathological features in NEC progression, further validating the close association between NEC and dysregulated inflammatory responses as well as impaired tissue repair.

To further elucidate functional relationships among proteins, a STRING-based protein-protein interaction (PPI) network was constructed (confidence score ≥700) and visualized in Cytoscape (Figures 5E and 5F). A total of 92 proteins formed 2,535 interaction pairs, significantly exceeding the expected number of 472 pairs (p < 0.001). Within the PPI network, FGF5, a protein upregulated in the control group, exhibited the highest connectivity, with key interactions including FGF5-GDNF (score: 755, strong association), FGF5-TRANCE (score: 448, moderate association), FGF5-uPA (score: 402, moderate association), and uPA-TRANCE (score: 289, weak association). These findings suggest that FGF5 may serve a crucial regulatory role in the early stage of NEC pathogenesis. As a network hub, FGF5 likely exerts bifunctional effects: it may promote intestinal epithelial repair via GDNF-mediated regenerative signaling, while also modulating inflammatory responses through interactions with TRANCE and uPA. Integrating these network predictions with our clinical proteomic and functional data, FGF5 appears to act as a protective mediator that balances mucosal repair and inflammatory damage in NEC.

Discussion

NEC is a devastating gastrointestinal emergency in preterm infants, characterized by intestinal mucosal injury, barrier failure, microbial translocation, and a dysregulated inflammatory response that can progress to systemic inflammation, multi-organ dysfunction, fulminant sepsis, and death.¹⁵ Multiple risk factors have been implicated, including prematurity, low birth weight, feeding practices, gut microbiota dysbiosis, maternal and environmental exposures, and antecedent intestinal ischemia. Early antibiotic exposure, intercurrent infections, and acid-suppressive therapies may further perturb microbial homeostasis and compromise mucosal defenses, thereby increasing NEC susceptibility.¹⁶ Mechanistically, NEC is associated with reduced microbial diversity, excessive activation of mucosal immunity, and host-environment interactions within permissive genetic contexts. Among the proposed pathways, epithelial Toll-like receptor 4 (TLR4) activation has been consistently implicated in experimental and human NEC, linking microbial signals to exaggerated inflammation and impaired epithelial repair.¹⁷ Clinically, the lack of specific early symptoms often delays diagnosis until substantial intestinal injury has occurred, and a proportion of infants require surgical intervention, with subsequent risks of short bowel syndrome, intestinal failure, parenteral nutrition-associated liver disease, and adverse neurodevelopmental outcomes.¹⁸

In this prospective cohort with pre-diagnostic sampling, we profiled serum proteins within two weeks before NEC diagnosis using multiplex proteomics and identified four candidates of DEPs (FGF5, GDNF, TRANCE, and uPA), all reduced in infants who later developed NEC. Time-resolved analyses across 0–7 and 8–14 days pre-diagnosis highlighted FGF5 as the most temporally consistent signal. Importantly, we strengthened the discovery findings via ELISA in an independent external cohort, alongside immunohistochemical and immunofluorescence confirmation of suppressed FGF5 expression in terminal ileal surgical specimens. In an IEC-6 inflammatory injury model, recombinant FGF5 attenuated TNF-α transcription, providing preliminary functional support for a potential protective association. Given the pivotal role of the FGF family in tissue repair,¹⁹ taken together, these clinical, tissue, and experimental data support FGF5 as a promising candidate biomarker linked to impending NEC and justify further mechanistic investigation.

A major unmet need in NEC is reliable early risk stratification that is feasible in very preterm infants. Many previously reported biomarkers show variability across cohorts and are often influenced by sampling time and disease stage, which limits their clinical implementation.^20,21 Our study extends prior efforts by focusing specifically on pre-diagnostic windows and by integrating multiplex serum profiling with tissue-level expression and functional assays—an approach intended to bridge discovery with translational plausibility. Notably, unsupervised PCA did not clearly separate cases from controls, which is not unexpected in heterogeneous neonatal datasets; therefore, we relied on matched comparisons, time-window stratification, and orthogonal validation rather than global variance structure alone to identify robust candidates.

From a biological perspective, FGF5 belongs to the FGF4 subfamily, which is broadly linked to tissue repair and regeneration across multiple organs. Evidence from other FGF4 subfamily members indicates their roles in cell proliferation, migration, angiogenic responses, and survival programs relevant to wound repair.²² For instance, recombinant human FGF4 (rFGF4) upregulates matrix metalloproteinase-9 (MMP-9) and vascular endothelial growth factor receptor 1 (VEGFR1) expression in db/db mouse skin fibroblasts, enhances fibroblast migration, and accelerates wound healing; combined rFGF4 and vascular endothelial growth factor A (VEGF-A) therapy is more effective for diabetic wound healing than either agent alone.²³ In parallel, disruption of angiogenesis-related signaling (e.g., VEGF pathways) may increase vulnerability to hypoxic/ischemic intestinal injury and predispose to NEC.^24,25 In human and mouse wound models, rFGF4 was shown to accelerate wound re-epithelialization by inhibiting Slug protein ubiquitination and degradation, thereby activating the p38 MAPK-GSK3β signaling axis to stabilize Slug protein.²⁶ Notably, these FGF4 subfamily findings do not establish NEC-specific mechanisms for FGF5. The proposed roles of pathways such as Wnt/β-catenin and NF-κB in FGF5-related NEC mechanisms are literature-based speculation, as these pathways were not directly tested. In our cohort, the observed reduction of FGF5 in both serum and intestinal tissue, together with our GO/KEGG and PPI analyses, suggesting that the identified proteins cluster in extracellular signaling and inflammation-related pathways, is consistent with the concept that NEC reflects an imbalance between inflammatory injury and insufficient mucosal repair. However, these network analyses should be interpreted as hypothesis-generating and do not establish causality; rather, they prioritize FGF5-centered interactions (e.g., with GDNF/TRANCE/uPA) for targeted experimental validation.

FGF5 has also been reported to have region-specific expression and regulatory roles in the nervous system, including effects on glial phenotypes in experimental settings.²⁷ It exhibits a region-specific distribution in the brain, with higher expression levels in the frontal cortex, hippocampus, and hypothalamus compared with other regions. Functionally, FGF5 modulates neuronal differentiation and survival through paracrine signaling and significantly influences astrocytic characteristics in vitro. For instance, during astrocyte differentiation, endogenous FGF2 and FGF5 region-specifically modulate the expression of glial fibrillary acidic protein (GFAP), an intermediate filament component, thereby influencing the mechanical stability of the blood-brain barrier²⁸ and participating in the pathological progression of neurological disorders.²⁹ Notably, the relevance of these nervous system-related observations to NEC remains uncertain. Future studies integrating intestinal, immune, and neural readouts in NEC-relevant models may help clarify whether such cross-organ signaling contributes to disease susceptibility or recovery.

In conclusion, this prospective exploratory study identified reduced pre-diagnostic serum FGF5 as a candidate biomarker associated with subsequent NEC in very preterm infants. Tissue-level evidence and preliminary in vitro findings further suggest that FGF5 may be involved in NEC-related epithelial injury responses.

Limitations of the study

The primary limitation of our work is the limited sample size, which diminishes statistical power for screening 92 analytes and elevates the risk of type II errors during multiple testing. As a single-center study, our predictive models (Firth logistic regression, cross-validation, bootstrap resampling) retain a high risk of overfitting; this artifact was also observed in our small external validation cohort, where FGF5 yielded an AUC of 1.0 with perfect sensitivity and specificity driven by minimal inter-subject variation. All findings must therefore be interpreted as strictly exploratory, and clinical translation cannot be pursued without large-scale independent multicenter validation. Second, Olink proteomics only outputs relative protein abundance, hindering the derivation of standardized clinical cutoffs without dedicated clinical assay development. Finally, the IEC-6 cellular model simplifies NEC-related inflammatory injury and fails to capture the full complexity of prematurity, gut microbiota, perfusion abnormalities, and immune-epithelial interactions involved in NEC pathogenesis. Notwithstanding these constraints, our pre-diagnostic serial sampling cohort remains clinically valuable, given practical barriers to serial blood collection in fragile preterm neonates. Future work will prioritize expanded enrollment, multicenter external validation, and physiologically sophisticated preclinical models to dissect FGF5’s mechanistic role in NEC.

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Shushu Li (lishushu@njmu.edu.cn).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• •Raw data were deposited on Mendeley at https://doi.org/10.17632/wtcgxgdxn9.1. For more information, data will be shared by the lead contact upon request.
• •This paper does not report original code.
• •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Acknowledgments

This work was supported by grants from the National Natural Science Foundation of China (grant nos. 82271744, 81971427, and 82101814), the Natural Science Foundation of Jiangsu Province (BK20221182), and the Nanjing Outstanding Youth Fund for Health Science and Technology Development (grant no. JQX23008).

Author contributions

Conceptualization, S.L. and Y.D.; methodology, S.L. and Y.D.; investigation, X.C. and Y.D.; formal analysis, X.C., M.Z., and Y.C.; writing – original draft, S.L. and Y.D.; writing – review and editing, X.C. and X.Y.; supervision, X.C. and X.Y.; writing – manuscript revision, S.H.; funding acquisition, S.H.; project administration, S.H.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Anti-FGF5 primary antibody (rabbit monoclonal, clone EPR14332; used for IHC, IF, WB)	Abcam, United Kingdom (discontinued)	Cat# ab88118; RRID: AB_10860495; Lot: 1048075-1; Dilution: IHC/IF 1:200, WB 1:1000
Recombinant Anti-beta Actin primary antibody (rabbit monoclonal; loading control for WB)	Servicebio, China	Cat# GB15003-100; RRID: AB_2893157; Clone: N/A; Lot: N/A; Dilution: WB 1:8000
HRP-conjugated goat anti-rabbit IgG (H+L) secondary antibody (used for WB)	Servicebio, China	Cat# GB23303; RRID: N/A; Host: Goat; Lot: N/A; Dilution: 1:8000
Fluorophore-conjugated goat anti-rabbit IgG (H+L) secondary antibody (used for IF)	Bioss, China	Cat# N/A; RRID: N/A; Host: Goat; Lot: N/A; Dilution: 1:200
Biological samples
Human serum samples and intestinal tissues	Nanjing Women and Children’s Healthcare Hospital, China	Inclusion criteria: Infants with GA <32 weeks and/or BW < 1500 g. This study (Approval Number: 2023KY-033) was nested within the ongoing “Hospital-Based Cohort Study of Very Preterm/Extremely Preterm Infants” initiated in January 2020 at the Neonatal Department, Nanjing Women and Children’s Healthcare Hospital (Project No.: ZKX19045; Ethics approval: Ning Fu Lun Zi [2018] No. 54).
Chemicals, peptides, and recombinant proteins
Recombinant Human FGF-5 protein (Animal-Free, C-His tag)	MedChemExpress (MCE)	Cat# HY-P700067AF; Lot: 268797; Species: Human; Expressed in E. coli; MW: 28.46 kDa; Purity ≥95%
Critical commercial assays
Olink Target 96 Inflammation Panel (proteomics detection service)	Shanghai Biotree Biomedical Technology Co., Ltd., China	Project No.: AQ-YXY20231018-olink-R; Technology: Proximity Extension Assay (PEA); Detection platform: Fluidigm RT-qPCR platform; Analytes: 92 inflammation-related proteins
Deposited data
Raw experimental data from this study	Mendeley Data	https://doi.org/10.17632/wtcgxgdxn9.1
Experimental models: Cell lines
Cell line: IEC-6 (rat small intestinal epithelial crypt cells)	American Type Culture Collection (ATCC), USA	Cat# CRL-1592; RRID: CVCL_0394; Species: Rat
Software and algorithms
R statistical software (v4.4.1)	R Foundation for Statistical Computing, Vienna, Austria	RRID: SCR_001905; URL: https://www.r-project.org/; Used for statistical analysis
ImageJ software (v1.54d)	National Institutes of Health (NIH), Bethesda, MD, USA	RRID: SCR_003070; URL: https://imagej.nih.gov/ij/; Used for IHC/IF quantification
Other
Human FGF5 ELISA Kit	Yuanju Biological (Shanghai, China)	Cat# YJ26163
10–180 kDa Prestained Protein Marker	Vazyme (Nanjing, China)	Cat# MP102-02

Experimental model and study participant details

Human participants

Study design and cohorts

A prospective observational cohort design was employed in this study, which was conducted at the neonatal intensive care unit (NICU) of Nanjing Women and Children’s Healthcare Hospital. Written informed consent was obtained from the parents or legal guardians of all infants prior to enrollment. Participants recruited between May 3, 2022 and October 16, 2023 constituted the discovery cohort, while those enrolled from February 25, 2024 to March 18, 2026 formed the independent validation cohort. All infants meeting the predefined inclusion and exclusion criteria were included in data collection and statistical analysis. Blinding was not implemented in this preterm infant cohort, as all outcome measurements were objective and free from subjective assessment bias.

Inclusion and exclusion criteria

The inclusion criteria covered preterm infants admitted to the study center NICU with GA < 32 weeks and/or BW < 1500 g. Exclusion criteria were as follows: (i) refusal of informed consent by parents or guardians; (ii) severe congenital anomalies, including chromosomal abnormalities, inborn errors of metabolism, or major central nervous system malformations; (iii) critical illness or severe dysfunction of major organs.

Clinical data collection and outcome definition

GA was determined via the last menstrual period method and calibrated by early pregnancy ultrasound. Apgar scores were classified as normal (8–10), mild asphyxia (4–7), and severe asphyxia (0–3). Maternal factors included maternal age, twin pregnancy, cesarean section, preterm premature rupture of membranes (PROM), fetal distress, antenatal corticosteroid exposure, and exclusive breastfeeding. Routine laboratory indices measured from the same blood draw as the serum sample: C-reactive protein (CRP), white blood cell count (WBC), and neutrophil percentage (NEUT%). NEC was diagnosed and staged according to the modified Bell criteria. Control infants were defined as those without NEC throughout hospitalization and follow-up.

Clinical sample collection and matching strategy

Venous serum samples (residual specimens after clinical laboratory testing) were collected from the preterm infant cohort on postnatal days 1, 7, 14, and 28, and subsequently stored at −80°C. For infants who subsequently developed NEC, serum samples collected within 14 days before clinical NEC diagnosis were considered eligible pre-diagnostic samples. If more than one eligible pre-diagnostic sample was available from the same infant, the sample collected closest to the date of clinical diagnosis was selected to ensure one observation per infant.

Controls were matched 1:1 to cases using a “twin-sibling priority” principle: if an NEC infant had a twin sibling without NEC, the unaffected twin was prioritized as the matched control. If no eligible twin sibling existed, a non-twin control was matched by (i) postnatal age at blood sampling, (ii) BW, and (iii) GA. For non-twin matching, the control serum sample was selected from the same postnatal sampling day as the matched case (e.g., day 7 vs day 7) whenever available.

Separately, intestinal tissues were obtained from surgical resections performed at the Department of Pediatric Surgery. NEC resection tissues constituted the NEC tissue group. Control tissues were obtained from infants with congenital gastrointestinal malformations without overt inflammation and were processed identically.

Cell lines

Cell line information and culture

The rat small intestinal epithelial crypt cell line IEC-6 was purchased from the American Type Culture Collection (ATCC, USA; Cat# CRL-1592; RRID: CVCL_0394). Cells were cultured in Dulbecco’s Modified Eagle Medium (DMEM) supplemented with 10% fetal bovine serum (FBS) and maintained in a humidified incubator at 37 °C with 5% CO₂. STR authentication was not independently performed as cells were sourced directly from the certified ATCC biorepository. With short culture duration and low passage, mycoplasma contamination was excluded via routine microscopic observation.

Cell treatment

To mimic an inflammatory microenvironment relevant to NEC, cells were treated with lipopolysaccharide (LPS, 50 μg/mL) in DMEM containing 3% FBS. For intervention, NEC-model cells were treated with 50 ng/mL recombinant FGF5 protein. The optimal concentration of FGF5 was determined using the Cell Counting Kit-8 (CCK-8) assay.

Method details

Olink proteomics analysis

Proteomics analysis was performed using the Olink Target 96 Inflammation Panel (92 proteins, Table S1). The assay is based on PEA technology and was performed on the Fluidigm RT-qPCR platform for high-throughput detection. The Olink platform normalizes cycle threshold (Ct) values and subsequently applies a log₂ transformation to generate NPX values. One unit of NPX corresponds to a twofold difference in protein concentration; thus, relative NPX values across samples directly reflect relative protein abundance. Missing NPX values were imputed with the per-protein median across all analyzed samples, an approach robust to the non-Gaussian distributions and outliers commonly seen in proteomic datasets. NPX values are directly compatible with a wide range of statistical analyses, including t-tests and ANOVA. Quality control was performed in accordance with Olink guidelines, using internal and external controls to monitor assay performance (Table S2). Samples that failed predefined QC thresholds were excluded prior to downstream analyses.

Immunofluorescence (IF) and immunohistochemistry (IHC)

Paraffin-embedded intestinal tissue sections were deparaffinized, rehydrated, and subjected to antigen retrieval. Sections were blocked with 3% bovine serum albumin (Servicebio, China) for 1 h at room temperature and incubated with primary antibody (Abcam, UK) at 4°C overnight. For IF, sections were incubated with fluorophore-conjugated secondary antibody (Bioss, China) for 1 h at room temperature in the dark, counterstained with DAPI (Sigma-Aldrich, USA), and mounted with anti-fade medium (Servicebio, China). Images were captured using a fluorescence microscope (Carl Zeiss, Germany). For IHC, after incubation with secondary antibody, sections were developed with DAB, counterstained with hematoxylin, dehydrated, cleared, and imaged using a light microscope (Carl Zeiss, Germany). Staining intensity/positive area was quantified using ImageJ (version 1.54d, USA) with identical thresholds across groups by an investigator blinded to group labels.

ELISA

ELISA was performed strictly according to the manufacturer’s instructions for the ELISA kit (Yuanju, China). Absorbance values were read at 450 nm on a multimode microplate reader (BioTek, USA). A standard curve was constructed by plotting standard concentrations against corresponding absorbance values, with a required coefficient of determination (R²) ≥ 0.99. Sample concentrations were calculated based on triplicate wells.

Western blotting

Total proteins were extracted with RIPA lysis buffer (working solution prepared per standard formulation). Separating and stacking gels were prepared per the instructions of the SDS-PAGE Gel Rapid Preparation Kit (Servicebio, China). Protein molecular weights were calibrated using a 10–180 kDa prestained protein ladder (Vazyme, China). After sample loading, electrophoresis was run at a constant 70 V for 45 min, then switched to 140 V for an additional 45 min. Resolved proteins were subsequently transferred to PVDF membranes (Millipore, USA) at 350 mA constant current for 90 min. Following transfer, membranes were blocked in 5% non-fat milk for 2 h at room temperature, then incubated with primary antibodies overnight at 4°C. After repeated washes, membranes were incubated with secondary antibodies for 2 h at room temperature. Following addition of chemiluminescent developing solution, membranes were imaged using the Bio-Rad Protein Electrophoresis Immunoblotting Imaging System (Bio-Rad, USA). Protein band intensities were quantified using ImageJ software.

RT-qPCR

Total RNA was extracted using TRIzol reagent (Thermo Fisher Scientific, USA) per the manufacturer’s protocol. cDNA was synthesized from 1 μg total RNA using R323 Reverse Transcription Premix (Vazyme, China) per instructions. qPCR was performed with SYBR Green qPCR Premix (Servicebio, China) on a real-time PCR system. Reaction conditions: 50°C for 2 min; initial denaturation at 95°C for 10 min; 40 cycles of 94°C for 15 sec and 60°C for 1 min; followed by melting curve analysis for amplification specificity. Primer design for target genes and housekeeping genes was shown in Table S3. Relative gene expression was calculated using the 2^−ΔΔCt method with β-actin as an internal control.

Quantification and statistical analysis

All analyses were performed in R (version 4.4.1). For all analyses, n represents the number of independent samples in each group; detailed statistical results and exact sample sizes for each comparison are provided in the results section. Baseline characteristics were summarized as mean ± standard deviation (SD) for normally distributed variables, median (IQR) for non-normally distributed continuous variables, and n (%) for categorical variables. The 95% CI was used as the precision measure for model performance.

Between-group comparisons used t-test or Mann–Whitney U test for continuous variables and χ² or Fisher’s exact test for categorical variables, as appropriate. In bar graphs, statistical significance is denoted by asterisks: ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001. For Olink proteomics, NPX differences between NEC cases and matched controls were assessed using paired t-tests (or Wilcoxon signed-rank tests where appropriate). To control for multiple testing across the 92 proteins, p values were adjusted using the Benjamini–Hochberg false discovery rate (FDR), and proteins with FDR q < 0.05 were considered significant candidates.

To develop and internally validate the NEC prediction model, we used Firth-penalized logistic regression to predict NEC while mitigating small-sample estimation bias. Model performance was evaluated by plotting the ROC curve and calculating AUC. The optimal NPX cutoff for FGF5 was determined using Youden’s index, with sensitivity, specificity, and accuracy calculated. Three internal validation procedures were applied: (1) Stratified 3-fold cross-validation: data were stratified by NEC status and randomly partitioned into three folds; in turn, each fold served as the test set for a model trained on the remaining two folds, and performance metrics (AUC, sensitivity, specificity) were summarized as median and IQR. (2) bootstrap validation with 1000 resamples, and an AUC was computed for each fitted model; the median AUC and its 95% confidence interval are reported. (3) 0.632+ bootstrap validation: this adjusted bootstrap approach was used to correct for optimism in the standard bootstrap and to obtain a less biased estimate of model performance.

PCA was conducted using singular value decomposition, and scatter plots of the first two principal components were generated to assess sample clustering and detect potential outliers. GO and KEGG pathway enrichment analyses were performed on DEPs using over-representation analysis with hypergeometric testing and FDR correction. PPI networks were constructed using the STRING database (confidence score ≥ 700) and visualized in Cytoscape (v3.9.1). Key subnetworks were identified using MCODE. A two-sided p < 0.05 was considered statistically significant unless otherwise specified.

Additional resources

This study was nested within the ongoing “Hospital-Based Cohort Study of Very Preterm/Extremely Preterm Infants” initiated in January 2020 at the Neonatal Department, Nanjing Women and Children’s Healthcare Hospital (Project No.: ZKX19045; Ethics approval: Ning Fu Lun Zi [2018] No. 54). The cohort established a standardized follow-up protocol and a biospecimen repository. The current sub-study was approved by the Medical Ethics Committee of Nanjing Women and Children’s Healthcare Hospital (Approval Number: 2023KY-033). No public clinical trial registry link is applicable to this observational cohort study.