Causal Associations of Heparin‐Binding Growth and Differentiation Factors With Thyroid Cancer: A Two‐Sample Mendelian Randomization Study

Article type: Research Article

Keywords: heparin-binding EGF-like growth factor, heparin-binding growth/differentiation factors, IGF2, Mendelian randomization, midkine, thyroid cancer

Authors: Xiwei Zhang, Dongqiang Yang, Yan Liu, Yanzhao Wu, Ping Shi ⚬

Affiliations: Department of Head and Neck Surgical Oncology, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, 100034, China, cacms.ac.cn; Department of Radiological Intervention, The Fourth Hospital of Hebei Medical University and Hebei Tumor Hospital, Shijiazhuang, 050000, China; Department of Otolaryngology Head and Neck Surgery, The Fourth Hospital of Hebei Medical University and Hebei Tumor Hospital, Shijiazhuang, 050000, China

License: Copyright © 2026 Xiwei Zhang et al. International Journal of Endocrinology published by John Wiley & Sons Ltd. CC BY 4.0 This is an open access article under the terms of the https://creativecommons.org/licenses/by/4.0/ License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.

Article links: DOI: 10.1155/ije/6165618  |  PubMed: 42466099  |  PMC: PMC13373531

Relevance: Moderate: mentioned 3+ times in text

Full text: PDF (4.5 MB)

1. Introduction

Thyroid cancer, the most common endocrine malignancy, includes various histological subtypes such as papillary, follicular, medullary, and anaplastic thyroid carcinomas (THCAs) [ref. 1]. Global prevalence of thyroid cancer was estimated to be 586,000 cases in 2020, which ranked as the 10th most prevalent cancer [ref. 2]. Notably, anaplastic THCA, though rare, is highly aggressive, with a median survival of just 3–6 months and a 1‐year survival rate of less than 20% [ref. 3, ref. 4]. While differentiated thyroid cancers like papillary and follicular types have relatively high survival rates, the disease often imposes significant physical and psychological burdens due to the need for lifelong monitoring, anxiety over recurrence, and treatments that can reduce quality of life [ref. 4, ref. 5]. This underscores the importance of understanding the etiological factors of thyroid cancer, which remains a critical area of ongoing research. These issues emphasize the critical need for a deeper understanding of the etiological factors of thyroid cancer, which continues to be a central area of research.

Recent research has highlighted the role of heparin‐binding growth and differentiation factors (GDFs), a group of proteins involved in cellular processes including proliferation, differentiation, and angiogenesis [ref. 6]. Heparin‐binding GDFs can bind to heparin and heparan sulfate proteoglycan and trigger downstream signaling pathways of cellular homeostasis and healing processes [ref. 7, ref. 8]. For instance, fibroblast growth factor 1 (FGF1) and FGF2 are critical in wound healing and angiogenesis [ref. 9], whereas insulin‐like growth factor I (IGF1) and IGF2 are crucial for growth and development [ref. 10]. Midkine and pleiotrophin are known for their roles in neurogenesis and cancer progression [ref. 11], while thyroglobulin is a key marker in thyroid function and thyroid cancer monitoring [ref. 12]. Population‐based studies have provided evidence supporting the associations between GDFs and thyroid cancer. For instance, serum midkine and thyroglobulin can serve as effective biomarkers for screening thyroid nodules for differentiated thyroid cancer preoperatively [ref. 13]. However, the presence of confounding factors and reverse causation in observational studies has made it difficult to establish a clear causal link between heparin‐binding GDFs and thyroid cancer. This challenge highlights the need for more rigorous approaches to better clarify the potential causal relationships.

Mendelian randomization (MR) analysis utilizes genetic variants as proxies for modifiable exposures to infer causal relationships between exposures and outcomes in observational studies [ref. 14]. This approach is grounded in the principle that genetic variants are randomly assorted during gamete formation and thus are generally independent of confounding factors in observational studies. MR analysis assumes that the genetic variants used as instruments are associated with the exposure, are not associated with any confounders of the exposure‐outcome relationship, and influence the outcome solely through the exposure [ref. 15].

To investigate the potential causal links between heparin‐binding GDFs with thyroid cancer, a two‐sample MR analysis was conducted in this study. Elucidating the biological mechanisms underlying the causal association can enhance our understanding of thyroid cancer pathogenesis, potentially leading to novel prevention strategies and therapeutic interventions.

2. Materials and Methods

2.1. Study Design

A two‐sample MR study was conducted to investigate the causal links between heparin‐binding GDFs and thyroid cancer. Utilizing data from previously published studies, genetic variants related to heparin‐binding GDFs were collected. Instrumental variables (IVs) were selected according to three core assumptions [ref. 15]: (A) the single nucleotide polymorphism (SNP) is strongly associated with heparin‐binding GDFs, (B) the SNP is independent of known confounders, and (C) the SNP affects thyroid cancer exclusively through its impact on heparin‐binding GDFs. The study methodology is illustrated in Figure 1.

2.2. Data Sources

Summary data for thyroid cancer were obtained from the IEU (https://gwas.mrcieu.ac.uk/datasets/) and FinnGen (https://storage.googleapis.com/finngen-public-data-r10/summary_stats/) database, including subtypes such as malignant and benign neoplasms of the thyroid gland (Table S1). Summary statistics for proneuregulin‐2, thyroglobulin, IGF‐1, heparin‐binding EGF‐like growth factor, midkine, pleiotrophin, IGF2, neuregulin‐1, FGF1, and FGF2 levels were sourced from the genome‐wide association studies (GWASs) catalog (Table S1). To minimize population stratification bias, all SNPs and summary data were derived exclusively from European ancestry populations. The exposure GWASs were conducted in independent European cohorts, while outcome data for malignant thyroid neoplasms came from the FinnGen project (Finnish population) and those for overall thyroid cancer and benign neoplasms from the IEU database (primarily UK Biobank). There is no substantial sample overlap between exposure and outcome datasets. Using open‐source data from ethically compliant sources, such as those adhering to the Helsinki Declaration, does not require additional ethical approval.

2.3. Selection of IVs

SNPs strongly associated with heparin‐binding GDFs were screened based on a p value threshold of < 5 × 10⁻⁶ [ref. 16]. This threshold, rather than the conventional genome‐wide significance level of p < 5 × 10⁻⁸, was adopted because for several exposures the number of independent SNPs meeting the stricter cutoff was fewer than three. An insufficient number of IVs would preclude essential sensitivity analyses, including MR‐Egger regression and the MR‐PRESSO global test. To mitigate potential weak instrument bias associated with the relaxed threshold, F‐statistics were calculated for all included SNPs, and only those exceeding the empirical benchmark of 10 were retained. SNPs with a minor allele frequency (MAF) ≤ 0.01 were additionally excluded [ref. 17]. Linkage disequilibrium (LD) analysis removed SNPs with R ² < 0.001 (window size = 10,000 kb) [ref. 18]. For outcome summary data, missing SNPs were replaced with proxy SNPs showing high LD (R ² > 0.8). Proxy SNPs were identified using the LDlink web–based tool (https://ldlink.nih.gov/?tab=ldproxy) and were selected based on an R ² > 0.8 threshold in the European 1000 Genomes Project population. This high degree of LD ensures that the proxy SNP serves as a reliable surrogate for the original variant, introducing minimal bias to the causal estimates. If no suitable proxy meeting this criterion was identified, the SNP was excluded from the analysis. Palindromic SNPs were eliminated to avoid discrepancies in strand orientation or allele coding. The F‐statistic is calculated using the following formula:

where N denotes the sample size of the GWAS, and R ² represents the proportion of variance in the exposure explained by the IVs in the regression equation [ref. 19].

2.4. MR Analysis

The primary method for assessing causal relationships between heparin‐binding GDFs and thyroid cancer was the inverse‐variance weighted (IVW) approach. IVW calculates a weighted average effect size, with the inverse variance of each SNP serving as the weight [ref. 20]. To validate the findings, additional analyses were conducted using MR‐Egger, weighted median, and weighted mode methods. MR‐Egger adjusts for pleiotropy by accounting for intercept terms, providing precise causal effect estimates [ref. 20]. The weighted median method assumes that at least half of the IVs are valid [ref. 21], while the weighted mode method determines the mode of effect estimates from individual IVs and assigns appropriate weights [ref. 22]. All analyses were performed using the “TwoSampleMR” package in R software (Version 4.0.5).

2.5. Sensitivity Analysis

Cochran’s Q test was used to assess heterogeneity based on IVW estimates, with a significance threshold of p < 0.05 [ref. 23]. A “leave‐one‐out” sensitivity analysis was conducted to identify potentially influential SNPs by iteratively excluding each SNP [ref. 24]. MR‐Egger regression was used to evaluate potential horizontal pleiotropy from genetic variation, with a near‐zero or statistically insignificant intercept term indicating the absence of pleiotropy [ref. 25]. The MR pleiotropy residual sum and outlier (MR‐PRESSO) method identified outliers (i.e., SNPs with p < 0.05), and their removal allowed for re‐estimation of causal associations, thus correcting for horizontal pleiotropy [ref. 26]. Asymmetry in the funnel plot can indicate heterogeneity among IVs.

2.6. Transcriptomic Validation Using the Cancer Genome Atlas (TCGA) Data

To validate our MR findings at the tissue expression level, we performed a differential expression analysis using data from the TCGA–THCA project. RNA‐sequencing data and corresponding clinical information for 519 thyroid cancer patients and 59 normal control samples were downloaded from the UCSC Xena database (https://xenabrowser.net/datapages/). The limma package (v3.62.2) in R (v4.4.2) was used to identify differentially expressed genes between tumor and normal tissues. p values were adjusted for multiple testing using the Benjamini–Hochberg method, with an adjusted p value (p.adj) < 0.05 considered statistically significant.

3. Results

3.1. Selection of IVs

For heparin‐binding GDFs, 163 relevant IVs were identified, with F‐statistics ranging from 20.93 to 79.47 (mean = 24.2). Two IVs were excluded due to the absence of proxy SNPs in the genetic summary for thyroid cancer. Additionally, 9 IVs were missing from the genetic summaries for malignant and benign neoplasms of the thyroid gland. Among these, rs60229479 was substituted by proxy SNP rs6079365, and rs35660775 was replaced with rs10413312. Detailed information on the IVs used is provided in Table S2.

3.2. Causal Effects of Heparin‐Binding GDFs on Thyroid Cancer

The IVW method revealed significant causal relationships between midkine levels (OR = 1.2099, 95% CI: 1.013–1.445, p = 0.0355) and IGF2 (OR = 0.7496, 95% CI: 0.6056–0.9279, p = 0.0081) with thyroid cancer and between heparin‐binding EGF‐like growth factor and malignant thyroid neoplasms (OR = 0.8821, 95% CI: 0.7873–0.9882, p = 0.0304) (Table 1). Although the alternative MR methods (including MR‐Egger, weighted median, and weighted mode) did not yield statistically significant p values, they demonstrated consistent directional trends with the primary IVW estimates, with all odds ratios (ORs) pointing in the same direction (Table S3). Specifically, for midkine levels, the risk estimates were consistently greater than 1.0 across all methods, while for IGF2 and heparin‐binding EGF‐like growth factor, the estimates remained consistently below 1.0. Scatter plots illustrating SNP effect sizes for heparin‐binding GDFs on thyroid cancer, malignant thyroid neoplasms, and benign thyroid neoplasms are shown in Figures 2. Consistent results are presented in the forest plots (Figures 3).

3.3. Sensitivity Analysis

Cochran’s Q statistics indicated significant heterogeneity in the association between FGF1 and malignant thyroid neoplasms (Q = 20.725, p = 0.036) (Table 2). Based on the comprehensive analysis results of this study, we deemed this bias tolerable. This is because the IVW method leverages the effects of multiple genetic variants on the target variable, which can mitigate biases potentially induced by genetic variation. MR‐Egger analysis found no evidence of horizontal pleiotropy in the association between heparin‐binding GDFs and thyroid cancer (all p > 0.05) (Table 2). Similarly, the MR‐PRESSO global test confirmed the absence of horizontal pleiotropy in these associations (all global p > 0.05) (Table 3). Leave‐one‐out analysis showed that no single SNP significantly influenced the causal relationship between heparin‐binding GDFs and thyroid cancer (Figures 4). Additionally, the symmetry of the funnel plots suggested no substantial heterogeneity in these associations (Figures 5).

3.4. Differential Expression Analysis in Thyroid Tumor Tissues

To explore the expression patterns of the identified genes at the tissue level, we analyzed the TCGA–THCA cohort. The analysis revealed that MDK (the gene encoding midkine) was significantly upregulated in thyroid tumor tissues compared to adjacent normal tissues (logFC > 0, p.adj < 0.05). However, no significant differential expression was observed for IGF2 or HBEGF in tumor tissues (Figure 6).

4. Discussion

The IVW analysis revealed significant causal associations of midkine levels and IGF2 with thyroid cancer and between heparin‐binding EGF‐like growth factor and malignant thyroid neoplasms. No horizontal pleiotropy, outlier SNPs, and evident heterogeneity were reported, substantiating the robustness of our findings. Further research is needed to validate these results and explore the underlying mechanisms of thyroid cancer. It is worth noting that while the primary IVW method yielded statistically significant causal estimates, the alternative MR estimators (MR‐Egger, weighted median, and weighted mode) did not reach formal statistical significance (p ≥ 0.05). This discrepancy is a recognized phenomenon in MR studies and can be primarily attributed to differences in statistical power. The IVW method is the most statistically efficient estimator when horizontal pleiotropy is absent. Conversely, robust methods like MR‐Egger and weighted median relax the assumption of valid instrumentation to guard against pleiotropy, but they do so at the cost of statistical power, which typically leads to wider confidence intervals and larger p values, especially in analyses with a modest number of IVs. Importantly, our study demonstrated a clear directional consistency and showed no evidence of horizontal pleiotropy (as indicated by the nonsignificant MR‐Egger intercepts and MR‐PRESSO global test results). Under these conditions, the IVW estimates serve as the most reliable indicators of the potential causal effects, while the directional alignment of the alternative methods supports the qualitative stability of our findings.

The study revealed positive causal associations between midkine levels and thyroid cancer, suggesting that midkine may contribute to thyroid carcinogenesis. Midkine, known for its role in promoting cell proliferation, migration, and angiogenesis, has been implicated in various cancers [ref. 27, ref. 28]. Its overexpression in thyroid cancer can be driving tumorigenic processes through similar mechanisms observed in other malignancies, such as enhancing cellular invasion and resistance to apoptosis [ref. 29]. Consistent with our findings, an Egyptian study revealed serum midkine levels significantly elevated as the thyroid tumor stage progressed [ref. 30]. To further investigate this connection at the tumor source, we examined its expression within the tissue itself. Our analysis of the TCGA–THCA dataset strongly supported this hypothesis, revealing that the mRNA expression of MDK is significantly upregulated in thyroid tumor tissues compared to adjacent normal tissues. This convergence of evidence—from a genetically predicted causal role for circulating levels to confirmed high expression at the tumor site—provides a compelling argument for the protumorigenic function of midkine in thyroid cancer, reinforcing its potential as a biomarker for further investigation.

However, the inverse association between IGF2 levels and thyroid cancer risk contrasted with the typically oncogenic role of IGF. IGF2 is a key regulator of growth and development, and its dysregulation has been linked to various cancers, including colorectal and breast cancer [ref. 31, ref. 32]. A systematic review of the IGF bioregulation system in thyroid nodular disease found that IGF1 expression was consistently elevated in thyroid cancer, whereas IGF2 expression was generally lower in malignant tissue compared to normal thyroid tissue [ref. 33]. To explore whether this protective function is reflected at the tumor site, we analyzed its tissue‐level expression. Interestingly, our transcriptomic analysis of the TCGA–THCA cohort did not show significant differential expression of IGF2 within established tumor tissues. This apparent divergence between the effect of genetically predicted systemic levels and local tumor expression warrants careful interpretation. It suggests the protective effect of higher circulating IGF2 may occur during the early stages of carcinogenesis, rather than influencing the gene expression profile of a fully developed tumor. Alternatively, the causal effect could be mediated through systemic pathways that do not necessitate a change in IGF2 expression within the thyroid tumor itself [ref. 34–ref. 36]. This result challenges the conventional understanding of IGF2’s role in cancer and highlights the need for further investigation into its highly tissue‐specific functions in thyroid carcinogenesis.

A similar layer of complexity was observed for heparin‐binding EGF‐like growth factor, suggesting a protective role for this factor in malignant thyroid neoplasms. Heparin‐binding EGF‐like growth factor is involved in activating the EGFR signaling pathway, which is crucial for cell survival, proliferation, and differentiation [ref. 37]. Our findings contrasted with studies linking high levels of EGF‐like growth factors to poor prognosis in cancers such as glioblastoma and lung adenocarcinoma [ref. 38, ref. 39]. Consistent with our observations for IGF2, our analysis of the TCGA dataset also found no significant changes in HBEGF expression in malignant thyroid neoplasms. This recurring pattern suggests that the protective causal effects inferred from our MR analysis for both these factors might not be directly linked to their transcriptional regulation within the established tumor microenvironment. Instead, their protective mechanisms could be systemic or primarily related to preventing the initial malignant transformation, processes not fully captured by a static gene expression snapshot of an established cancer. Further research is needed to elucidate the mechanisms underlying this protective association.

This study has several important limitations. The primary limitation is the reliance on GWAS data from populations of European descent. This constrains the external validity of our findings, a crucial point of caution given the well‐documented ethnic disparities in thyroid cancer incidence, particularly its higher prevalence in Asian populations. The genetic architecture of complex diseases can vary significantly across ancestries, including differences in allele frequencies, effect sizes, and LD patterns. Consequently, the causal estimates for midkine, IGF2, and HB‐EGF may not be generalizable to other ethnic groups. It is imperative that our findings be replicated in large‐scale MR studies conducted in diverse populations, especially in Asian cohorts, to ascertain whether these causal pathways are universal or population specific. Furthermore, our study utilized summary‐level data, which precluded subgroup analyses by age, sex, or specific cancer histotypes that might have uncovered heterogeneous effects. Finally, to ensure an adequate number of IVs for sensitivity analyses (such as MR‐Egger and MR‐PRESSO), we employed a relaxed p value threshold of < 5 × 10 ⁻⁶ for instrument selection. While this threshold is less stringent than the conventional genome‐wide significance level (p < 5 × 10⁻⁸) and could theoretically introduce weaker genetic instruments, we mitigated this risk by ensuring all included SNPs possessed an F‐statistic substantially greater than 10 (mean F‐statistic = 24.2). The lack of statistical significance in robust estimators like MR‐Egger might be partially influenced by the noise introduced under this relaxed threshold; however, the consistent directional estimates across all models and the absence of detectable pleiotropy suggest that the primary causal findings remain robust. Nonetheless, future studies with larger sample sizes and stronger genetic instruments meeting genome‐wide significance are warranted to confirm these relationships.

5. Conclusions

In summary, this MR study provides evidence for causal associations of genetically predicted midkine levels and IGF2 with thyroid cancer and of heparin‐binding EGF‐like growth factor with malignant thyroid neoplasms. These findings contribute to the understanding of thyroid cancer etiology and highlight directions for future mechanistic investigation.

Nomenclature

GDFs
Growth and differentiation factors

MR
Mendelian randomization

GWAS
Genome‐wide association studies

IVW
Inverse‐variance weighted

FGF1
Fibroblast growth factor 1

IGF1
Insulin‐like growth factor I

IVs
Instrumental variables

SNP
Single nucleotide polymorphism

MAF
Minor allele frequency

LD
Linkage disequilibrium

Author Contributions

Conception and design: Xiwei Zhang and Ping Shi. Administrative support: Ping Shi and Yanzhao Wu. Provision of study materials or patients: Xiwei Zhang and Dongqiang Yang. Collection and assembly of data: Xiwei Zhang and Dongqiang Yang. Data analysis and interpretation: Xiwei Zhang and Ping Shi. Manuscript writing: Xiwei Zhang, Dongqiang Yang, Yan Liu, Yanzhao Wu, and Ping Shi.

Funding

This study was supported by the Medical Science Research Project of Hebei (no. 20240378) and Hebei Provincial Department of Finance Government funded outstanding Medical talents project (ZF2025197).

Disclosure

All authors gave final approval for the manuscript.

Ethics Statement

The authors have nothing to report.

Consent

The authors have nothing to report.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting Information

Additional supporting information can be found online in the Supporting Information section.

References

1. Thyroid Cancer,. International Journal of Otolaryngology and Head & Neck Surgery. (, 2019
2. Special Types of Thyroid Carcinoma,. Histopathology. (, 2018. [DOI | PubMed]
3. Thyroid Cancer: Global Burden and Trends,. Indian J Surg Oncol. (, 2022. [DOI | PubMed]
4. Follow-up of Differentiated Thyroid Cancer–What Should (And what Should Not) Be Done,. Nature Reviews Endocrinology. (, 2018. [DOI]
5. Thyroid Cancer, Recent Advances in Diagnosis and Therapy,. International Journal of Cancer. (, 2021. [DOI | PubMed]
6. Heparin, Heparan Sulphate and the TGF-β Cytokine Superfamily,. Molecules. (, 2017. [DOI]
7. Control of Growth Factor Networks by Heparan Sulfate Proteoglycans,. Annals of Biomedical Engineering. (, 2008. [DOI | PubMed]
8. Heparan Sulfate Proteoglycans: Key Mediators of Stem Cell Function,. Frontiers in Cell and Developmental Biology. (, 2020. [DOI]
9. The Role of Fibroblast Growth Factor (FGF) Signaling in Tissue Repair and Regeneration,. Cells. (, 2021. [DOI]
10. Role of Insulin-Like Growth Factors in Growth, Development and Feeding,. Nutrition and Growth. (, 2013
11. Midkine and Pleiotrophin in Neural Development and Cancer,. Cancer Letters. (, 2004. [DOI | PubMed]
12. Thyroglobulin as a Tumor Marker in Differentiated Thyroid Cancer–Clinical Considerations,. Acta Clinica Croatica. (, 2018. [DOI]
13. Evaluation of Serum Midkine as a Biomarker in Differentiated Thyroid Cancer,. Life Sciences. (, 2015. [DOI | PubMed]
14. Methodological Approaches, Challenges, and Opportunities in the Application of Mendelian Randomisation to Lifecourse Epidemiology: A Systematic Literature Review,. European Journal of Epidemiology. (, 2023. [DOI]
15. Mendelian Randomization,. JAMA. (, 2017. [DOI | PubMed]
16. Physical Activity and Risks of Cardiovascular Diseases: A Mendelian Randomization Study,. Frontiers in cardiovascular medicine. (, 2021. [DOI]
17. Causal Associations Between Serum Amino Acid Levels and Osteoarthritis: A Mendelian Randomization Study,. Osteoarthritis and Cartilage. (, 2023. [DOI | PubMed]
18. Ldlink: A Web-based Application for Exploring Population-specific Haplotype Structure and Linking Correlated Alleles of Possible Functional Variants,. Bioinformatics. (, 2015. [DOI | PubMed]
19. A Review of Instrumental Variable Estimators for Mendelian Randomization,. Statistical Methods in Medical Research. (, 2017. [DOI | PubMed]
20. A Causal Relationship Between Childhood Obesity and Risk of Osteoarthritis: Results From a Two-Sample Mendelian Randomization Analysis,. Annals of Medicine. (, 2022. [DOI | PubMed]
21. Causal Relationship Between Physical Activity, Leisure Sedentary Behaviors and COVID-19 Risk: A Mendelian Randomization Study,. Journal of Translational Medicine. (, 2022. [DOI]
22. Mendelian Randomization Study of Inflammatory Bowel Disease and Bone Mineral Density,. BMC Medicine. (, 2020. [DOI]
23. Causal Effects of Glycemic Traits and Endometriosis: A Bidirectional and Multivariate Mendelian Randomization Study,. Diabetology & Metabolic Syndrome. (, 2024. [DOI]
24. A Causal Relationship Between Appendicular Lean Mass and Atrial Fibrillation: A Two Sample Mendelian Randomization Study,. Nutrition, Metabolism, and Cardiovascular Diseases. (, 2024. [DOI]
25. Interpreting Findings from Mendelian Randomization Using the MR-Egger Method,. European Journal of Epidemiology. (, 2017. [DOI | PubMed]
26. No Genetic Causal Association Between Periodontitis and Arthritis: A Bidirectional Two-Sample Mendelian Randomization Analysis,. Frontiers in Immunology. (, 2022. [DOI]
27. Midkine (MDK) Growth Factor: A Key Player in Cancer Progression and a Promising Therapeutic Target,. Oncogene. (, 2020. [DOI | PubMed]
28. Midkine is a Potential Therapeutic Target of Tumorigenesis, Angiogenesis, and Metastasis in Non-small Cell Lung Cancer,. Cancers. (, 2020. [DOI]
29. Midkine: A Cancer Biomarker Candidate and Innovative Therapeutic Approaches,. European Journal of Breast Health. (, 2024
30. Long Noncoding RNA HOTAIR and Midkine as Biomarkers in Thyroid Cancer,. Egyptian journal of immunology. (, 2023. [DOI]
31. Insulin-Like Growth Factor 2 (IGF2) Signaling in Colorectal Cancer—From Basic Research to Potential Clinical Applications,. International Journal of Molecular Sciences. (, 2019. [DOI]
32. Risk Factors for Breast Cancer and Expression of Insulin-Like Growth factor-2 (IGF-2) in Women With Breast Cancer in Wuhan City, China,. PLoS One. (, 2012. [DOI]
33. IGF Bioregulation System in Benign and Malignant Thyroid Nodular Disease: A Systematic Review,. Vivo. (, 2020. [DOI]
34. TGF-β-Mediated Epithelial-Mesenchymal Transition and Cancer Metastasis,. International Journal of Molecular Sciences. (, 2019. [DOI]
35. Comprehensive Analysis of Immune Infiltration, Gene Correlations, and Traditional Chinese Medicine in Lung Adenocarcinoma,. International Journal of Biological Macromolecules. (, 2025. [DOI]
36. Single-Nucleus RNA Sequencing and Spatial Transcriptomics Reveal an Immunosuppressive Tumor Microenvironment Related to Metastatic Dissemination During Pancreatic Cancer Liver Metastasis,. Theranostics. (, 2025. [DOI | PubMed]
37. Heparin-Binding Epidermal Growth Factor–like Growth Factor as a Critical Mediator of Tissue Repair and Regeneration,. The American Journal of Pathology. (, 2018. [DOI | PubMed]
38. DSE Promotes Aggressive Glioma Cell Phenotypes by Enhancing HB-EGF/ErbB Signaling,. PLoS One. (, 2018. [DOI]
39. Heparin Binding Epidermal Growth Factor-Like Growth Factor is a Prognostic Marker Correlated with Levels of Macrophages Infiltrated in Lung Adenocarcinoma,. Frontiers Oncology. (, 2022. [DOI]