Inverse co-variation of circulating PIWIL2 and PIWIL4 in gastric adenocarcinoma: an exploratory ELISA-based study

Abstract

Background

Gastric cancer is characterized by substantial molecular heterogeneity, and RNA-mediated regulatory mechanisms may contribute to its biological complexity. PIWI-family proteins are central components of the PIWI-interacting RNA pathway, but their circulating patterns in gastric adenocarcinoma remain insufficiently characterized. This study aimed to describe circulating PIWIL1, PIWIL2, and PIWIL4 protein concentrations and explore intra-cohort co-variation patterns in patients with gastric adenocarcinoma without evaluating disease specificity, diagnostic performance, or clinical utility.

Methods

In this cross-sectional, targeted ELISA-based exploratory study, circulating PIWIL1, PIWIL2, and PIWIL4 concentrations were quantified in 93 de-identified gastric adenocarcinoma samples. PIWIL1 and PIWIL2 measurements were available for 88 samples, and PIWIL4 measurements were available for 72 samples. Commercial ELISA kits were used according to the manufacturer’s instructions, but independent in-house serum validation, including dilution linearity, serum parallelism, matrix-interference testing, intra- and inter-assay precision, and orthogonal protein confirmation, was not performed. Spearman’s rank correlation analysis with false discovery rate correction was used to assess exploratory intra-cohort relationships among PIWI-family proteins.

Results

PIWIL1 and PIWIL2 showed relatively stable circulating distributions, whereas PIWIL4 demonstrated pronounced heterogeneity and right-skewness. The most prominent rank-based relationship was an inverse correlation between PIWIL2 and PIWIL4 (Spearman’s ρ ≈ −0.53, q < 0.001), based on 72 complete cases. This correlation remained statistically significant in sensitivity analyses using log-transformed data and exclusion of extreme values. No statistically significant correlations were observed between PIWI-family proteins and CEA or CA 19 − 9 after false discovery rate correction. Exploratory TP53 analyses were not central to the primary objective and are reported descriptively; PIWIL2 showed a modest positive correlation with TP53 (Spearman’s ρ = 0.33), while PIWIL4 showed a weak positive correlation (ρ = 0.19).

Conclusions

This study identified an inverse rank-based co-variation between circulating PIWIL2 and PIWIL4 concentrations within a de-identified gastric adenocarcinoma cohort. However, this observation should be interpreted strictly as a pre-validation, exploratory ELISA-derived intra-cohort signal. Because the study lacked comparator groups, clinical annotation, tissue-level confirmation, functional assays, orthogonal protein quantification, and independent serum validation of the ELISA measurements, the observed correlation cannot be considered an analytically validated proteomic association. Further studies with confirmed assay linearity, serum parallelism, matrix-interference assessment, between-plate reproducibility, re-assay of high-value samples at appropriate dilutions, and independent analytical confirmation are required before any biological or clinical significance can be attributed to this finding.

Introduction

Gastric cancer remains one of the most serious malignancies worldwide and continues to impose a substantial clinical and public health burden. According to GLOBOCAN 2022 estimates, stomach cancer accounted for approximately 968,000 new cases and 660,000 deaths globally, ranking among the leading causes of cancer incidence and cancer-related mortality worldwide [1]. The magnitude of this burden is particularly important because gastric cancer is frequently diagnosed at an advanced stage, when curative treatment options are limited and survival outcomes remain poor. In addition to its high mortality, gastric cancer is characterized by marked molecular and clinical heterogeneity, which complicates early detection, risk stratification, and disease monitoring. While extensive genomic profiling, such as the foundational classifications established by The Cancer Genome Atlas [2], has significantly advanced our understanding of genetic subtypes, translating these structural blueprints into actionable clinical tools remains challenging. Consequently, increasing attention has been directed toward RNA-mediated regulatory systems and epigenetic alterations as additional layers of gastric tumor biology [3]. Non-coding RNAs and their associated binding proteins represent a critical, yet incompletely understood, level of molecular regulation that may contribute to tumor progression beyond the genome [4].

Among these regulatory mechanisms, PIWI-family proteins, most notably PIWIL1, PIWIL2, PIWIL3, and PIWIL4, are central components of the PIWI-interacting RNA pathway. Originally characterized for their essential role in maintaining genomic integrity in germline cells via transposon silencing [5, 6], PIWI proteins are involved in gene regulation through epigenetic and post-transcriptional mechanisms. Increasing evidence suggests that aberrant PIWI-family expression may be involved in somatic tumorigenesis [7]. In different malignancies, PIWI-family proteins have been linked to cancer-related processes such as cellular proliferation, apoptosis regulation, stemness, and metastatic dissemination [8, 9]. However, their functional roles are highly context-dependent, with studies reporting different effects depending on the specific PIWI paralog, tissue of origin, tumor type, and cellular microenvironment [10].

In gastric cancer, the available evidence regarding PIWI proteins remains limited and predominantly tissue-based [11]. While tissue biopsies remain the gold standard for initial diagnosis, they are invasive, provide only a limited snapshot of the disease, and may fail to capture the full spatial and temporal heterogeneity of the tumor [12, 13]. Recent advances in spatial omics have further demonstrated that gastric cancer cannot be fully understood as a uniform molecular entity. Spatial transcriptomics, spatial proteomics, and related spatially resolved approaches allow researchers to map gene expression, protein distribution, immune-cell localization, stromal organization, and tumor–microenvironment interactions directly within tissue architecture. In gastric cancer, these technologies are increasingly used to reveal intratumoral heterogeneity, regional tumor niches, immune-excluded phenotypes, and spatially distinct patterns of tumor–stroma communication. This has important implications for biomarker discovery and patient stratification, as clinically relevant molecular signals may be localized to specific tumor regions and missed by conventional bulk profiling or limited biopsy sampling. However, spatial omics approaches remain dependent on tissue acquisition and therefore share some practical limitations of biopsy-based investigation, particularly invasiveness, sampling bias, and limited feasibility for repeated longitudinal monitoring. In this context, circulating biomarker profiling may be considered complementary rather than competitive: while spatial omics can define where molecular alterations occur within the tumor ecosystem, liquid biopsy approaches may help capture systemic molecular signals that can be evaluated through minimally invasive sampling [14, 15]. Liquid biopsy has gained increasing attention in gastric cancer because circulating tumor-derived signals may provide information on molecular heterogeneity, treatment response, minimal residual disease, and recurrence risk without repeated invasive tissue sampling [16, 17]. Among liquid biopsy components, extracellular vesicles, particularly exosomes, are of special interest because they are actively released by tumor cells and cells of the tumor microenvironment and carry biologically informative molecular cargo, including proteins, DNA fragments, messenger RNAs, non-coding RNAs, and regulatory RNA-binding proteins [18, 19]. Their lipid bilayer protects this cargo from degradation in the circulation, making exosomes attractive candidates for stable blood-based molecular analysis. In gastric cancer, exosome-associated biomarkers may help bridge the gap between tissue-level molecular alterations and systemic disease monitoring. Importantly, exosomes are not merely passive carriers of tumor-derived material; they may also participate in tumor–microenvironment communication, immune modulation, epithelial–mesenchymal transition, metastatic niche formation, and therapy resistance. For this reason, exosome-related liquid biopsy signals may reflect both tumor burden and active biological processes involved in disease progression. Within this broader context, circulating PIWI-family proteins may represent a relevant area for exploratory investigation, particularly because RNA-binding proteins and RNA-mediated pathways remain incompletely characterized in gastric cancer.

Although PIWIL3 is also a member of the PIWI-family and has been implicated in tumor-related regulatory processes in several cancer types, it was not included in the present analysis. The current study focused on PIWIL1, PIWIL2, and PIWIL4 as predefined candidate circulating proteins based on assay feasibility, availability of standardized ELISA-based quantification within the approved protocol, and previous institutional experience with these markers. Therefore, the present analysis should not be interpreted as a comprehensive evaluation of the entire PIWI-family.

In this study, we aimed to describe circulating PIWIL1, PIWIL2, and PIWIL4 protein concentrations in an independent cohort of patients with gastric adenocarcinoma and to explore intra-cohort co-variation patterns among these predefined candidate proteins. The study was designed as a targeted ELISA-based, descriptive analysis and not as a clinical proteomics validation study, diagnostic accuracy study, comparator-controlled biomarker investigation, or mechanistic study. Circulating PIWI-family protein profiles were evaluated alongside established serum tumor markers, including CEA and CA 19 − 9, not to determine discriminatory capacity, but to explore whether PIWI-family proteins showed measurable rank-based relationships with conventional serum markers within the available dataset [20, 21]. Because no healthy control group, benign gastric disease comparator, non-gastric cancer disease control cohort, clinical annotation, tissue-level PIWI expression data, or orthogonal protein quantification was available, the present study cannot establish gastric cancer specificity, diagnostic utility, clinical relevance, or biological mechanism. Its contribution is therefore limited to transparent reporting of an exploratory ELISA-derived intra-cohort correlation that requires validation before any biological or clinical interpretation can be made.

Methods

Study design and cohort

This study was conducted from March 2025 to December 2025 as a cross-sectional biomarker profiling analysis of patients with histologically confirmed gastric adenocarcinoma. The analytical cohort consisted of 93 de-identified gastric adenocarcinoma samples, representing all available samples collected during the defined study period that met the eligibility criteria for biomarker analysis. No additional clinical selection criteria were applied after de-identification.

Because the samples were strictly de-identified before analysis, individual-level demographic, clinicopathological, molecular, treatment-related, and outcome variables could not be linked to the circulating biomarker measurements. Therefore, age, sex, TNM stage, histological subtype, Lauren classification, lymph-node status, metastatic status, Helicobacter pylori status, treatment history, molecular subtype, and outcome data were not available for subgroup or confounder analysis. No aggregate clinical subgroup data could be generated for the present cohort without re-linkage to clinical records, which was outside the scope of the approved de-identified analytical protocol.

Circulating biomarker measurements were available for most cases, with PIWIL1 and PIWIL2 quantified in 88 samples and PIWIL4 in 72 samples, reflecting missing or invalid assay outputs in a subset of cases. The study was designed as a descriptive, hypothesis-generating investigation aimed at characterizing circulating PIWI-family protein concentrations and their intra-cohort co-variation patterns. No longitudinal follow-up or clinical outcome data were included. Accordingly, the study was not powered or designed to evaluate differential protein abundance between disease and non-disease states, diagnostic performance, prognostic relevance, or clinical biomarker utility.

The present gastric adenocarcinoma cohort is entirely independent from the cohort reported in our previous publication by Kldiashvili et al. [22]. The prior study was conducted from 1 March 2024 to 1 November 2024 and included colorectal cancer, breast cancer, prostate cancer, and healthy control participants, but did not include gastric cancer cases. In contrast, the present study included only de-identified samples from patients with histologically confirmed gastric adenocarcinoma collected during the 2025 study period. Therefore, none of the 93 gastric adenocarcinoma samples analysed in the present study were included in the previous publication. Reference [22] is cited only to contextualize previous institutional work on circulating PIWI-related biomarkers and does not represent the source dataset for the current analysis.

Ethical approval was obtained from the Bioethics International Committee of Petre Shotadze Tbilisi Medical Academy, Tbilisi, Georgia, approval protocol No. 53-014–202504161130. The study was conducted in accordance with the Declaration of Helsinki. Participants received information about the study purpose, procedures, potential risks, and benefits, and provided oral informed consent prior to inclusion. The study did not include minors. All samples were anonymized before analysis, and no personally identifiable information was accessible to the researchers.

Sample processing and biomarker quantification

Circulating levels of PIWIL1, PIWIL2, and PIWIL4 were quantified using target-specific, commercially available enzyme-linked immunosorbent assay kits from FineTest: PIWIL1, catalogue number EH5515; PIWIL2, catalogue number EH5516; and PIWIL4, catalogue number EH4482. The study was designed as a targeted ELISA-based circulating protein profiling analysis rather than as untargeted, mass spectrometry-based, or multiplexed clinical proteomics. ELISA was selected because the analysis focused on three predefined candidate PIWI-family proteins and because this platform allowed standardized quantification of circulating concentrations across the available sample set.

No orthogonal protein quantification method, such as multiple reaction monitoring mass spectrometry, parallel reaction monitoring mass spectrometry, proximity extension assay, aptamer-based proteomics, or another independent proteomic platform, was performed in the present study. The FineTest ELISA kits were used according to the manufacturer’s instructions, and each assay plate included calibration standards and blank controls. Independent in-house analytical validation of these assays in human serum was not performed. Specifically, intra-assay precision, inter-assay precision, spike recovery, dilution linearity, freeze–thaw stability, serum parallelism, matrix-interference testing, formal between-plate variability assessment, and orthogonal protein quantification were not evaluated. Therefore, PIWIL1, PIWIL2, and PIWIL4 concentration estimates should be interpreted as exploratory ELISA-derived measurements rather than independently validated proteomics-grade quantitative measurements.

Briefly, collected serum samples were processed and subjected to a 1:4 dilution using the sample diluent buffer provided by the manufacturer. Diluted samples and calibration standards were added to pre-coated microplate wells and incubated with target-specific antibodies. Following standardized washing steps to remove unbound material, enzyme-conjugated secondary antibodies and chromogenic substrates were applied. The reaction was terminated using stop solution, and optical density was measured immediately at 450 nm using a microplate reader. Biomarker concentrations were calculated by interpolation from standard curves and expressed in pg/mL after adjustment for the dilution factor.

Throughout the manuscript, PIWI-family proteins are reported using HGNC-style PIWIL nomenclature: PIWIL1, PIWIL2, and PIWIL4. Because historical literature uses overlapping HIWI, HIWI2, HIWI4, and MIWI-related terms inconsistently, citations were reviewed to avoid attributing paralog-specific findings from one PIWI-family member to another.

Because independent serum validation was not performed, the ELISA-derived concentrations should not be considered analytically validated quantitative protein measurements. Dilution linearity was not confirmed for samples with high PIWIL4 values, and high concentration samples were not re-assayed at higher dilution factors to verify that they fell within a confirmed quantitative range. Serum parallelism and matrix interference testing were also not performed; therefore, matrix-dependent signal effects cannot be excluded. Formal intra-assay precision, inter-assay precision, and between-plate variability were not quantified in the present study. These limitations directly constrain the analytical interpretation of the reported biomarker concentrations and the correlations derived from then.

Standard curve modelling and data preprocessing

Biomarker concentrations were derived from optical density measurements using a four-parameter logistic regression curve-fitting model, consistent with quantitative immunoassay practice [23]. For each assay plate, calibration standards supplied with the commercial ELISA kits and blank controls were included according to the manufacturer’s instructions. Standard curves were generated separately for each analyte and plate using the corresponding calibrator optical density values. Concentrations were interpolated from the fitted four-parameter logistic curve and expressed in pg/mL after adjustment for the 1:4 serum dilution factor.

Curve performance was reviewed before inclusion of plate-level data, including assessment of the expected monotonic signal response across the calibration range and consistency of blank and calibrator readings with the manufacturer’s protocol. A prospectively predefined R² acceptance threshold for standard-curve inclusion was not specified for public reporting, and calibrator replicate coefficient-of-variation acceptance criteria were not prospectively defined. Therefore, plate-level standard curves were not excluded according to a predefined numerical R² cut-off. Curve review was based on qualitative assessment of the expected monotonic dose–response pattern across calibrators and consistency of blank and calibrator readings with the manufacturer’s protocol. This limits the analytical interpretation of the absolute ELISA-derived concentration estimates and should be considered when interpreting correlations derived from these values.

Values falling below the assay’s established limit of detection or producing optical densities comparable to the limit of blank were not interpreted as biological measurements and were recorded as invalid or missing values. High-value observations, including PIWIL4 values exceeding 35,000 pg/mL, were retained in the primary exploratory analysis to preserve the observed data structure. However, because these samples were not re-assayed at higher dilution factors and dilution linearity was not independently confirmed, these values cannot be assumed to represent fully validated quantitative concentrations. Their influence was evaluated statistically through sensitivity analyses, but such analyses do not resolve uncertainty regarding assauy linearity, saturation, hook effect, or matrix-related signal interference.

Data preprocessing and quality control

A structured preprocessing workflow was applied before statistical analysis. Rows corresponding to calibration standards and blank controls were excluded from the analytical dataset. Negative concentration values were treated as invalid assay outputs and recoded as missing values rather than interpreted as true biological measurements. Missing optical density values resulted in corresponding missing concentration values. Extreme values were not removed arbitrarily; instead, their influence was evaluated through sensitivity analyses. The distribution of PIWIL4 demonstrated substantial right-skewness and the presence of high-value observations, which were retained in the primary analysis to preserve the underlying data structure.

Statistical analysis

All statistical analyses were performed using the R software environment (R Foundation for Statistical Computing, Vienna, Austria) [24]. Continuous variables were summarized using median and interquartile range because biomarker concentrations were non-normally distributed. Relationships between PIWI-family protein concentrations were assessed using Spearman’s rank correlation coefficients. To control for multiple testing, p-values were adjusted using the Benjamini–Hochberg false discovery rate procedure [25], and adjusted q-values below 0.05 were considered statistically significant.

The primary correlation analysis between PIWIL2 and PIWIL4 was conducted using complete cases with valid measurements for both proteins. A post-hoc power calculation was performed for the observed PIWIL2–PIWIL4 correlation using Fisher’s z transformation, with a two-sided α = 0.05. Missing data were summarized descriptively for each biomarker. Because no demographic, clinicopathological, molecular, or treatment-related variables were available in the de-identified dataset, formal missing-data mechanism testing, including Little’s MCAR test, could not be performed.

Given the distributional characteristics of the data, including skewness and the presence of extreme values, correlation analyses were complemented by sensitivity analyses using log-transformed values for strictly positive observations. Sensitivity analyses included exclusion of the top 5% of PIWIL4 values, exclusion of the single highest PIWIL4 observation highlighted in Fig. 3, and analysis of log10-transformed positive PIWIL2 and PIWIL4 concentrations. For each sensitivity analysis, Spearman’s ρ, raw p-values, Benjamini–Hochberg adjusted q-values, and complete-case sample size were reported. Additional analyses were performed excluding extreme values to assess their influence on correlation estimates. Exploratory analyses examined rank-based relationships between PIWI-family proteins and selected biomarkers from related pathways, including TP53. These analyses were interpreted as descriptive and hypothesis-generating. Because the analysis was based on continuous ELISA-derived biomarker concentrations and did not involve observer-dependent categorical matching or manual case classification, no kappa-based match validation procedure or kappa cut-off threshold was applicable.

Fig. 3.

Relationship between circulating PIWIL2 and PIWIL4 concentrations. Scatter plot of PIWIL2 and PIWIL4 concentrations in complete cases (n = 72), shown on logarithmic axes. The figure displays the distribution of observed values and the overall rank-based relationship between the two biomarkers. Because the primary analysis was based on Spearman’s rank correlation, no local smoothing curve is shown, to avoid over-interpretation of local non-linear fluctuations in this heterogenous dataset. One visually prominent high-value PIWIL4 observation is highlighted in the figure because higher-dilution re-assay was not performed for analytical confirmation. The observed association was inverse and statistically significant (Spearman’s ρ = −0.53, q < 0.001)

Reporting standards

The study design, statistical analysis, and reporting were conducted in accordance with the STROBE Statement to support methodological transparency and reproducibility [26]. The study was not designed as a MIAPE-compliant mass spectrometry-based proteomics experiment, but as a targeted ELISA-based exploratory circulating protein profiling study.

Data reporting and access

Raw assay-level data were not deposited in a public repository because the dataset derives from human clinical samples and is subject to institutional data protection and ethical access restrictions. De-identified analytical data may be made available from the corresponding author upon validated reasonable request, subject to institutional approval and applicable ethical requirements.

Results

Cohort and data completeness

A total of 93 de-identified gastric adenocarcinoma samples were included in the analysis. PIWIL1 and PIWIL2 measurements were available for 88 samples, while PIWIL4 measurements were available for 72 samples. Therefore, the primary PIWIL2–PIWIL4 correlation analysis was based on 72 complete cases with valid measurements for both proteins. PIWIL4 values were unavailable for 16 of the 88 samples with PIWIL1/PIWIL2 data, corresponding to 18.2% missingness among otherwise evaluable PIWI samples. Missing PIWIL4 values reflected missing or invalid assay outputs rather than deliberate sample exclusion. Because the dataset did not include linked demographic, clinical, molecular, or treatment-related variables, it was not possible to assess whether PIWIL4 missingness was completely at random or associated with patient- or disease-level characteristics.

For the observed PIWIL2–PIWIL4 correlation of approximately ρ = −0.53 with n = 72 complete cases, the post-hoc power was estimated to be > 99% at a two-sided α = 0.05. This indicates that the complete-case sample size was sufficient to detect a correlation of the observed magnitude, although this does not address potential bias arising from missing PIWIL4 data or the absence of clinical annotation.

Distribution of PIWI-family proteins

PIWI-family proteins demonstrated distinct distributional characteristics across the cohort (Fig. 1). PIWIL1 and PIWIL2 were detectable in all evaluable samples, with relatively stable distributions and moderate variability. Median concentrations were approximately 108 pg/mL for PIWIL1 and 1,177 pg/mL for PIWIL2.

Fig. 1.

Distribution of circulating PIWI-family protein concentrations across the cohort. Boxplots show log10-transformed concentrations in pg/mL for PIWIL1 (n = 88), PIWIL2 (n = 88), and PIWIL4 (n = 72). Individual data points are overlaid to show the distribution of observed values. PIWIL4 demonstrated greater inter-individual variability and right-skewness compared with PIWIL1 and PIWIL2. Concentrations were log10-transformed for visualization

In contrast, PIWIL4 exhibited substantial heterogeneity, characterized by a pronounced right-skewed distribution and a wide dynamic range. Concentrations ranged from low detectable levels to values exceeding 35,000 pg/mL. One of the high-value PIWIL4 observations is visually highlighted in Fig. 3 because of its prominence in the bivariate distribution and the absence of higher-dilution analytical confirmation. Although dilution linearity was not independently confirmed and high PIWIL4 samples were not re-assayed at higher dilution factors, the highest PIWIL4 values cannot be considered analytically confirmed quantitative measurements. Therefore, the wide PIWIL4 range should be interpreted cautiously, as it may reflect true biological heterogeneity, assay behavior at the limits of the standard curve, matrix-dependent signal, or a combination of these factors. Although sensitivity analyses assessed the statistical influence of extreme values, they cannot establish the analytical accuracy of the underlying measurements.

Interrelationships within the PIWI protein family

Spearman’s rank correlation analysis was used to examine rank-based relationships among circulating PIWI-family protein concentrations (Fig. 2). The most prominent finding was an inverse correlation between PIWIL2 and PIWIL4 (Spearman’s ρ ≈ −0.53, q < 0.001), representing the strongest rank-based relationship observed in the dataset. One visually prominent high-value PIWIL4 observation is highlighted in Fig. 3 for transparency. Because this sample was not re-assayed at higher dilution factors, its quantitative value should be interpreted cautiously.

Fig. 2.

Spearman correlation heatmap of PIWI-family proteins. Heatmap cells display pairwise Spearman’s rank correlation coefficients (ρ) together with FDR-corrected q-values and statistical significance annotations. The strongest inverse correlation was observed between PIWIL2 and PIWIL4 (ρ = −0.53, q < 0.001). FDR correction was performed using the Benjamini–Hochberg procedure. Significance levels: ns, not significant; * q ≤ 0.05; ** q ≤ 0.01; *** q ≤ 0.001

This relationship is illustrated in Fig. 3, as a scatter plot on logarithmic axes. The point distribution demonstrates substantial heterogeneity, particularly in the lower-to-mid PIWIL2 concentration range and should not be interpreted as showing a uniformity decreasing local pattern across the entire concentration range. However, the overall rank-based association across complete cases was inverse and statistically significant (Spearman’s p = −0.53, q < 0.001). The PIWIL2-PIWIL4 correlation remained statistically significant across sensitivity analyses, including exclusion of extreme values and application of log-transformed data. These findings suggest that the observed inverse rank-based association was not solely driven by extreme values or by the right-skewed distribution between PIWIL1 and either PIWIL2 or PIWIL4 after correction for multiple testing. This result should be interpreted as a statistical co-variation within the available cohort only; the present dataset does not allow determination of whether the pattern reflects gastric cancer biology, cohort composition, assay-platform behavior, matrix effects, or the distributional properties of PIWIL4. However, because the correlation is derived entirely from ELISA measurements that were not independently validated in serum, it should be interpreted as an exploratory statistical signal rather than as an analytically confirmed proteomic association.

Sensitivity analyses

Sensitivity analyses supported the statistical stability of the primary PIWIL2–PIWIL4 correlation within the available dataset. In the primary complete-case analysis, PIWIL2 and PIWIL4 were inversely correlated (n = 72; Spearman’s ρ = −0.533; p = 1.44 × 10⁻⁶; q = 1.92 × 10⁻⁶). After exclusion of the top 5% of PIWIL4 values, corresponding to the four highest PIWIL4 observations, the inverse association remained statistically significant (n = 68; Spearman’s ρ = −0.554; p = 9.42 × 10⁻⁷; q = 1.92 × 10⁻⁶). In the log10-transformed analysis, the rank-based correlation was unchanged (n = 72; Spearman’s ρ = −0.533; p = 1.44 × 10⁻⁶; q = 1.92 × 10⁻⁶). After exclusion of the single highest PIWIL4 observation highlighted in Fig. 3, the inverse association also remained statistically significant (n = 71; Spearman’s ρ = −0.530; p = 2.02 × 10⁻⁶; q = 2.02 × 10⁻⁶). These results (Table 1) indicate that the observed inverse rank-based association was not solely dependent on the highest PIWIL4 observations. However, these sensitivity analyses assess the statistical stability of the correlation conditional on the observed ELISA-derived data and do not establish the analytical accuracy of the measurements themselves. Therefore, the PIWIL2–PIWIL4 association remains an exploratory ELISA-derived signal requiring independent analytical validation. The analyses assess robustness of the correlation conditional on the observed ELISA-derived data; they do not establish analytical accuracy of the measurements themselves. Specifically, sensitivity analyses cannot resolve uncertainty related to dilution linearity, hook effect, saturation, antibody specificity, serum matrix interference, serum parallelism, or between-plate variability. Therefore, the PIWIL2-PIWIL4 correlation should be interpreted as a preliminary ELISA-derived signal requiring independent analytical validation.

Table 1.

PIWIL2-PIWIL4 sensitivity analysis

Analysis	n	Spearman’s p	p-value	BH-adjusted q-value
Primary complete-case analysis	72	−0.533	1.44 × 10− 6	1.92 × 10− 6
Top 5% PIWIL4 values excluded	68	−0.554	9.42 × 10− 7	1.92 × 10− 6
log10-transformed analysis	72	−0.533	1.44 × 10− 6	1.92 × 10− 6
Highest PIWIL4 observation excluded	71	−0.530	2.02 × 10− 6	2.02 × 10− 6

BH-adjusted q-values were calculated across the four PIWIL2–PIWIL4 analyses shown in this table using the Benjamini–Hochberg procedure. The top 5% exclusion removed the four highest PIWIL4 values from n = 72 complete cases

Integration with clinical tumor markers and TP53

To explore whether circulating PIWI-family protein concentrations showed measurable rank-based relationships with conventional clinical tumor markers, PIWIL1, PIWIL2, and PIWIL4 were evaluated alongside carcinoembryonic antigen (CEA) and carbohydrate antigen 19 − 9 (CA 19 − 9). Spearman’s rank correlation analysis did not show statistically significant correlations between PIWI-family protein concentrations and either CEA or CA 19 − 9 after false discovery rate correction (all q > 0.05). Within the limits of this cross-sectional dataset, these findings indicate that circulating PIWI-family protein concentrations did not demonstrate detectable monotonic relationships with the evaluated conventional serum tumor markers. Consistent with prior observations regarding the limited sensitivity of CEA and CA 19 − 9 in gastric cancer [20, 21], this result suggests that PIWI-family proteins may provide biomarker information not reflected by these traditional markers; however, this interpretation remains descriptive and requires validation in clinically annotated cohorts.

An exploratory Spearman correlation analysis was also conducted to examine whether PIWI-family protein concentrations showed rank-based relationships with circulating TP53 [27]. No statistically meaningful correlation was observed between PIWIL1 and TP53. PIWIL2 showed a modest positive correlation with TP53 (Spearman’s ρ = 0.33), while PIWIL4 showed a weak positive correlation (ρ = 0.19). Given the exploratory nature of this analysis, the modest effect sizes, and the cross-sectional study design, these findings should be interpreted cautiously as hypothesis-generating observations rather than evidence of functional interaction, tumor suppressor pathway involvement, or directional biological influence.

Discussion

In this study, we observed an inverse rank-based co-variation between circulating PIWIL2 and PIWIL4 concentrations within a de-identified cohort of patients with gastric adenocarcinoma. This association was statistically detectable in the available dataset and remained present in sensitivity analyses. However, the finding should be interpreted with substantial caution because the ELISA-derived concentrations were not independently validated in serum. Therefore, the observed correlation cannot be considered an analytically confirmed proteomic association, a clinically meaningful biomarker pattern, or evidence of a biologically established PIWIL2-PIWIL4 regulatory relationship. It is more appropriately interpreted as a preliminary ELISA-derived intra-cohort signal whose analytical, biological, and clinical significance remain uncertain.

Moreover, study lacked comparator cohorts, clinical annotation, tissue-level PIWI expression data, orthogonal protein quantification, and functional validation, the observed association cannot be considered evidence of a gastric cancer-specific biomarker pattern, a clinically meaningful proteomic signature, or a biologically established PIWIL2–PIWIL4 regulatory axis. It is more appropriately interpreted as an exploratory ELISA-derived intra-cohort correlation whose biological and clinical relevance remain unknown. The absence of a control cohort is therefore not only a limitation for diagnostic interpretation, but a fundamental constraint on the scientific interpretation of the observed co-variation. Without health controls, benign gastric disease comparators, or non-gastric controls, it is not possible to determine whether a similar PIWIL2-PIWIL4 correlation exists in non-cancer populations or in other clinical conditions. Similarly, without clinical annotation, the observed pattern may reflect unmeasured differences in disease stage, molecular subtype, treatment exposure, demographic composition, inflammatory status, or sample-related factors. Without orthogonal analytical confirmation, assay-platform behavior, antibody specificity, matrix interference, or dilution non-linearity cannot be excluded. These alternative explanations remain unresolved in the present study and preclude any claim of biological specificity or clinical relevance.

Importantly, the observed PIWIL2-PIWIL4 association was based on a global rank-based correlation measure and should not be interpreted as evidence of a uniformity decreasing functional relationship across all concentration ranges. The scatter distribution showed substantial heterogeneity, and local smoothing approaches may produce non-monotonic visual patterns in the presence of skewness, sparse observations, and high-value PIWIL4 measurements. For this reason, the Fig. 3 presents individual observations, aligning the graphical display more directly with the Spearman’s correlation analysis.

The inverse relationship between PIWIL2 and PIWIL4 should therefore be interpreted cautiously as an observational, rank-based co-variation pattern. PIWI-family proteins are central components of the PIWI–piRNA pathway, but individual paralogs may have different and context-dependent roles across tissues, tumour types, and biological compartments; therefore, PIWI/piRNA-related interpretations should be based on paralog-specific evidence rather than extrapolated from broader extracellular vesicle or general biomarker literature [7–10, 28–31]. In the present study, higher circulating PIWIL2 concentrations were statistically associated with lower circulating PIWIL4 concentrations within the available cohort; however, this does not establish that PIWIL2 is upregulated as a driver of gastric cancer progression, that PIWIL4 is suppressed, or that one protein regulates the other. Several alternative explanations may account for the observed inverse correlation, including coordinated transcriptional or post-transcriptional regulation, differential protein release, secretion, degradation, or shedding into the circulation, assay- or matrix-related effects, disease-stage composition, molecular subtype distribution, treatment exposure, demographic confounding, or the pronounced right-skewness of PIWIL4 values. Because the study did not include tissue-level PIWI expression, piRNA profiling, functional assays, longitudinal sampling, clinical annotation, or comparator cohorts, these possibilities cannot be distinguished in the present analysis.

The exploratory integration with conventional serum tumour markers and TP53 should also be interpreted descriptively. The absence of statistically significant rank-based correlations between PIWI-family proteins and CEA or CA 19 − 9 indicates only that no measurable monotonic relationship was detected between these variables in the present dataset. It should not be interpreted as evidence of biological independence, functional separation, or superiority over conventional serum tumour markers. Similarly, the modest positive correlation observed between PIWIL2 and TP53 and the weak positive correlation observed between PIWIL4 and TP53 should not be interpreted as evidence of functional interaction with tumour suppressor pathways. These findings are exploratory and require validation in clinically annotated datasets with tissue-based and functional analyses. Given the known complexity and inconsistency of PIWI-family nomenclature in the literature, paralog-specific mechanistic conclusions were avoided, and findings from one PIWI-family member were not extrapolated to another.

Overall, the present findings should be regarded as a pre-validation analytical observation rather than as a validated clinical proteomics result. The study documents an inverse PIWIL2-PIWIL4 rank-based correlation under clearly defined methodological constraints, but it cannot determine whether this pattern reflects true biological co-variation, cohort composition, assay-platform behavior, matrix effects, plate effects, or distributional properties of PIWIL4. Its relevance depends on future studies that include comparator cohorts, clinical annotation, tissue-level assessment, PIWIL3 measurement, longitudinal sampling, full serum ELISA validation, and orthogonal protein quantification.

Despite these observations, several important limitations must be acknowledged. The most important design limitation is the absence of a comparator cohort. The analytical cohort consisted exclusively of patients with gastric adenocarcinoma and did not include healthy controls, benign gastric disease comparators, or non-gastric cancer disease controls. As a result, the study cannot determine whether the observed inverse PIWIL2–PIWIL4 correlation is specific to gastric cancer or whether similar circulating PIWI co-variation may occur in other physiological or pathological contexts. The absence of a control cohort also precludes assessment of whether absolute PIWIL1, PIWIL2, or PIWIL4 concentrations are elevated, reduced, or otherwise altered relative to non-cancer reference distributions. Consequently, no conclusions can be drawn regarding diagnostic accuracy, discriminatory capacity, gastric cancer specificity, or clinical biomarker utility. The findings should therefore be interpreted strictly as descriptive and hypothesis-generating observations within a gastric cancer cohort.

A further major limitation is the complete absence of clinical annotation. Strict de-identification prevented integration of demographic, clinicopathological, molecular, and treatment-related variables, including age, sex, TNM stage, histological subtype, Lauren classification, Helicobacter pylori status, treatment history, and TCGA-related molecular subtype information. This substantially limits the clinical and biological interpretability of the findings. Without staging data, it is not possible to determine whether the PIWIL2–PIWIL4 inverse correlation reflects disease burden, late-stage enrichment, or other stage-dependent processes. Without histological and molecular subtype information, including EBV-associated, microsatellite instability-high, genomically stable, and chromosomally unstable subtypes, the observed pattern cannot be evaluated in relation to known gastric cancer heterogeneity. Similarly, age and sex could not be assessed as potential confounders of circulating protein concentrations, and treatment status could not be controlled despite the possibility that chemotherapy, immunotherapy, surgery, or other interventions may influence circulating protein dynamics. Therefore, the observed PIWIL2–PIWIL4 co-variation cannot be interpreted as stage-independent, subtype-independent, treatment-independent, demographically adjusted, or universally representative of gastric adenocarcinoma biology. It should instead be considered an unadjusted descriptive intra-cohort correlation requiring validation in clinically annotated datasets.

The cross-sectional design also limits the interpretation of the correlation analyses. Spearman’s rank correlation can describe monotonic relationships between biomarkers at a single time point, but it cannot establish causality, directionality, temporal dynamics, functional dependency, or biological independence. Therefore, the absence of statistically significant correlations between PIWI-family proteins and CEA or CA 19 − 9 should not be interpreted as proof that these biomarkers operate independently. Rather, it indicates that no measurable rank-based correlation was detected in this dataset. In addition, circulating protein levels may not directly reflect tissue expression or functional activity within the tumour microenvironment, as blood-based biomarker shedding dynamics can differ substantially from localized tissue patterns [13, 32–34].

Several analytical limitations directly affect interpretation of the primary finding. Circulating PIWI-family proteins were quantified using a single commercial ELISA platform, an independent in-house validation in human serum was not performed. The study did not assess intra-assay precision, inter-assay precision, spike recovery, dilution linearity, freeze–thaw stability, serum parallelism, matrix interference, or formal between-plate variability. In addition, prospectively predefined standard-curve R² acceptance thresholds and calibrator replicate CV criteria were not specified, meaning that plate-level curve acceptance was based on qualitative curve review rather than predefined numerical performance criteria. This is particularly important for PIWIL4, which showed a wide dynamic range, including values exceeding 35,000 pg/mL. A visually prominent high-value PIWIL4 observation was retained in the analysis and highlighted in the scatter plot for transparency. Because high PIWIL4 samples were not re-assayed at higher dilution factors, it cannot be confirmed that these values fell within a validated quantitative range or were free from saturation, hook effect, or other standard-curve-related artifacts. In addition, without serum parallelism and matrix-interference testing, apparent PIWIL4 concentrations may reflect matrix-dependent signal rather than true circulating protein concentration. Without formal between-plate reproducibility assessment, plate-to-plate technical variation cannot be excluded as a contributor to observed concentration differences across samples. These limitations are not resolved by statistical sensitivity analyses, which evaluate the stability of the correlation given the observer data but do not validate the analytical accuracy of the data themselves. Accordingly, the PIWIL2–PIWIL4 co-variation should be regarded as a preliminary ELISA-derived observation requiring independent technical confirmation.

A further limitation relates to data deposition and proteomics-style reporting. The present study was not originally designed as a MIAPE-compliant mass spectrometry-based proteomics experiment, and raw assay-level data, including optical density readings, complete standard curve outputs, and plate-level metadata, were not deposited in a public repository. Public deposition was restricted because the data derive from human clinical samples and are subject to institutional data protection and ethical access requirements. Although de-identified analytical data may be made available upon validated reasonable request, the absence of unrestricted public deposition limits independent re-analysis and external auditability of the ELISA workflow. This limitation further supports interpreting the findings as exploratory targeted ELISA-based observations rather than proteomics-grade quantitative validation.

Missing data represent an additional limitation. PIWIL4 measurements were available for 72 samples, compared with 88 samples for PIWIL1 and PIWIL2, resulting in approximately 18.2% PIWIL4 missingness among otherwise evaluable PIWI samples. Although the observed PIWIL2–PIWIL4 correlation had high post-hoc statistical power, the missing-data mechanism could not be formally evaluated because no linked demographic, clinical, molecular, or treatment-related variables were available. Consequently, Little’s MCAR test or equivalent assessment of missingness in relation to clinical characteristics could not be performed. The possibility that PIWIL4 missingness was non-random or related to unmeasured sample or patient characteristics cannot be excluded. High post-hoc power for the observed correlation does not compensate for the absence of clinical annotation, comparator cohorts, or formal assessment of the PIWIL4 missing-data mechanism.

Finally, although the present study was performed at the single institution and used ELISA-based quantification of PIWI-family proteins, it is not based on the cohort reported in our previous publication [22]. The prior publication included colorectal, breast, and prostate cancer cases and healthy controls recruited in 2024, whereas the present analysis uses an independent gastric adenocarcinoma cohort collected in 2025. The current study therefore addresses a distinct disease context and a separate analytical dataset. The study also did not include PIWIL3, although this PIWI-family member has been implicated in several malignancies. As a result, the present findings cannot be considered a complete assessment of PIWI-family protein dynamics in gastric adenocarcinoma, and future studies should include PIWIL3 alongside PIWIL1, PIWIL2, and PIWIL4 to determine whether broader PIWI-family co-variation patterns exist.

Conclusions

In conclusion, this study observed an inverse rank-based co-variation between circulating PIWIL2 and PIWIL4 concentrations within a de-identified cohort of patients with gastric adenocarcinoma. This finding should be interpreted strictly as a pre-validation, exploratory ELISA-derived intra-cohort signal. Because the study lacked comparator groups, clinical annotation, tissue-level confirmation, functional assays, PIWIL3 assessment, orthogonal protein quantification, and independent serum validation of the ELISA measurements, the observed correlation cannot be considered as an analytically validated proteomic association. The possibility that the pattern reflects cohort composition, assay-platform behavior, matrix interference, between-plate variation, or the distributional properties of PIWIL4 cannot be excluded. Future studies should first establish analytical validity through dilution linearity, spike recovery, serum parallelism, matrix-interference testing, intra- and inter-assay precision, between-plate reproducibility, re-assay of high-value samples at appropriate dilutions, and orthogonal protein confirmation. Only after such validation, together with clinically annotated comparator-controlled cohorts, can the biological or clinical significance of the PIWIL2-PIWIL4 pattern be assessed.

Author contributions

Ian Alexander Cree contributed to the conception and design of the study and critically revised the manuscript for important intellectual content. Ekaterina Kldiashvili contributed to methodological development, statistical analysis, interpretation of findings, and drafting of the manuscript. Elene Kekelia and Eter Dumbadze contributed to study implementation, laboratory activities, and data preparation. Mariam Abuladze contributed to interpretation of findings, and drafting of the manuscript. All authors approved the final version of the manuscript and agree to be accountable for all aspects of the work, ensuring that questions related to the accuracy or integrity of any part of the study are appropriately investigated and resolved.

Funding

This study was supported by the Shota Rustaveli National Science Foundation of Georgia (SRNSFG), the project FR-24-18776, MB4GAP: Molecular Biomarkers for Gastric Adenocarcinoma Personalization. The SRNSFG was not involved in the study design, data collection, data analysis, data interpretation, writing of the manuscript, or the decision to submit the article for publication.

Data availability

The de-identified analytical dataset used and/or analysed during the current study is available from the corresponding author upon validated reasonable request, subject to institutional approval, applicable ethical restrictions, and data protection requirements. Because the study used human-derived clinical biomarker data, unrestricted public deposition of raw assay-level data was not permitted under the applicable institutional data protection framework. No personally identifiable information is available to the researchers or will be shared.

Declarations

Ethics approval and consent to participate

This study was conducted in accordance with the ethical standards of the Declaration of Helsinki. Ethical approval was obtained from the Bioethics International Committee of Petre Shotadze Tbilisi Medical Academy (Tbilisi, Georgia, Approval No.: 53-01-4-202504161130). All samples were de-identified prior to analysis, and no personally identifiable information was accessible to the researchers. Oral informed consent was obtained from all participants or their legal representatives prior to sample collection.

Consent for publication

All participants provided oral informed consent for the use of their biological samples and associated de-identified data for research purposes, including publication of anonymized and aggregate study findings. No personally identifiable information, individual participant images, or case-level identifiable data are included in this manuscript.

Declaration of generative AI and AI-assisted technologies

During the preparation of this manuscript, the authors used ChatGPT (OpenAI, https://chat.openai.com/) only for language editing, grammar checking, stylistic refinement, and improvement of manuscript readability. ChatGPT was not used for raw data processing, data analysis, statistical computation, figure generation, interpretation of results, formulation of scientific conclusions, or development of the study design or methodology. All AI-assisted edits were critically reviewed and revised by the authors. The authors take full responsibility for the accuracy, integrity, and scientific content of the manuscript.

Competing interests

The authors declare no competing interests.