Clinical relevance of tissue copper, selenium, and cadmium alterations in colorectal cancer

Abstract

Imbalances in essential elements and altered levels of toxic elements are thought to play a role in colorectal cancer (CRC) development. This study assessed the concentrations of selected elements in CRC tissues and explored their associations with clinical parameters. Paired samples of tumor tissue and adjacent noncancerous tissue were collected from 62 patients with confirmed CRC. After microwave-assisted acid digestion, ten elements (Al, Cr, Mn, Fe, Ni, Cu, Zn, Se, Sr, Cd) were quantified using ICP-MS. Statistical analyses and machine-learning methods were applied to identify features discriminating clinical and demographic subgroups. Significant differences between tumor and healthy tissues were observed for Mn, Cu, Zn, Se, Sr, and Cd, with further variation linked to Tumor Node Metastasis (TNM) classification, clinical stage, and place of residence. Cadmium levels were higher in the healthy tissues of women (p_adj = 0.021), while selenium concentrations showed a decreasing trend with advancing disease (Stage I vs. III; p = 0.017, p_adj = 0.051). Boruta analysis identified lymph node status, TNM stage, and elevated tumor copper levels as key features associated with CRC. Elemental profiles in CRC tissues differ markedly from those in adjacent healthy tissues and exhibit meaningful associations with disease stage and sex. The distinctive copper pattern observed in tumor samples, together with clinical parameters, suggests that elemental profiling may complement standard diagnostic approaches and contribute to a deeper understanding of CRC biology.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-37256-w.

Introduction

Colorectal cancer (CRC) is a leading global cause of cancer morbidity and mortality, and its burden continues to rise despite screening/treatment advancements. Globally in 2022, CRC was the third most diagnosed cancer (9.6% of new cases) and the second leading cause of cancer deaths (9.3%), per IARC’s Global Cancer Observatory¹. A substantial proportion of CRC cases occur in high/very high Human Development Index (HDI) countries, which exhibit considerably elevated incidence rates¹. CRC etiology, though only partially understood, is widely accepted as multifactorial, involving genetic predispositions, lifestyle, and environmental exposures. Environmental factors (including diet and toxic exposures) are estimated to contribute up to 80% of all cancer cases², with trace elements (TEs) representing a significant component of these environmental influences^3–5.

Recent epidemiological studies⁶ highlighted associations between increased environmental concentrations of cadmium (Cd), arsenic (As), hexavalent chromium (Cr), and nickel (Ni) with higher regional CRC incidence, independent of individual-level risk factors. TEs are known to play essential roles in maintaining cellular stability, enzymatic activity, and hormonal function, while also modulating the toxicity of other elements, such as aluminum (Al) and lead (Pb)^7,8. Disruptions in elemental homeostasis can lead to cellular damage, oxidative stress, DNA damage, and ultimately malignant transformation^5,8,9. Oxidative stress and angiogenesis are proposed central processes in CRC pathogenesis, where elemental imbalances contribute by dysregulating antioxidant enzymes (e.g., superoxide dismutase (SOD), glutathione reductase, glutathione peroxidase) and promoting lipid peroxidation^10,11.

Within the specific context of CRC, the homeostasis of several trace elements is mechanistically linked to tumorigenesis. Copper (Cu) is a critical factor in promoting angiogenesis, a process vital for tumor vascularization and growth, and acts as a cofactor for enzymes involved in invasion and metastasis¹². Manganese (Mn) is the central component of mitochondrial manganese superoxide dismutase (Mn-SOD), and its dysregulation in cancer contributes to the accumulation of reactive oxygen species and mitochondrial dysfunction^4,13. Similarly, zinc (Zn) is essential for the function of over 300 enzymes and transcription factors, including the tumor suppressor p53, and its imbalance can affect cell proliferation and apoptosis^14–16. In contrast, selenium (Se) plays a largely protective role, as it is an integral component of antioxidant selenoproteins that defend against oxidative damage, and its depletion has been correlated with an increased risk of CRC^17,18. Furthermore, exposure to toxic elements such as cadmium (Cd), a recognized human carcinogen, is suspected to drive colorectal carcinogenesis by inducing oxidative stress, disrupting DNA repair pathways, and promoting a pro-inflammatory state^19,20.

Although the interaction between trace elements and colorectal tumorigenesis is not yet fully elucidated, existing studies suggest that alterations in essential TEs such as zinc (Zn), iron (Fe), and copper (Cu) may influence cancer development and progression^21,22. Zn, for example, may participate in telomerase activity and inhibit NADPH oxidase/Cu/Zn-SOD, while Cu is implicated in pro-angiogenic signaling that supports carcinogenesis. Sohrabi et al.⁵ reported levels of several TEs, including thallium (Tl), Zn, Pb, Cr, and Cu, as significantly elevated in CRC tissues compared to healthy controls.

Despite these insights, systematic assessment of elemental concentrations in paired colorectal tumor and adjacent healthy tissues is rare, and existing findings are inconsistent^23–25. Therefore, more investigation is needed to clarify the roles of essential and toxic elements in CRC biology and their potential as clinical biomarkers. Direct analysis of CRC tissue is of particular significance as it provides an unfiltered snapshot of the biochemical alterations within the tumor microenvironment itself. As these tissues are routinely collected during standard clinical procedures, they represent a highly relevant and accessible resource for research. Their analysis is crucial for elucidating the fundamental pathophysiology of the disease and serves as the foundational step for discovering robust biomarkers that could later be translated to less invasive diagnostic methods^26,27. Despite increased understanding of CRC molecular mechanisms, current diagnostic/prognostic tools face significant limitations. Colonoscopy, for example, though effective, is invasive and unsuitable for frequent screening²⁸. Moreover, there remains a lack of sensitive and specific tissue- or fluid-based biomarkers that could aid in early detection, stratification of disease progression, and personalized treatment planning^29,30. Investigating elemental imbalances in colorectal tissues thus offers a novel approach. Elemental profiles reliably associated with disease stage or tumor characteristics could serve as biomarkers, either through direct tissue analysis or via extrapolation to circulating biofluids. Thus, the clinical relevance of these elemental signatures surpasses pathophysiological insight, offering a path to better CRC diagnostics and precision oncology.

In this study, we comprehensively assessed the concentrations of ten elements: Al, Cr, Fe, Ni, Cu, Zn, Cd, manganese (Mn), selenium (Se), and strontium (Sr) in matched tumor and adjacent non-tumorous colorectal tissue samples resected from CRC patients. Using inductively coupled plasma mass spectrometry (ICP-MS) and a combination of traditional statistical approaches with machine learning feature selection (Boruta algorithm)³¹, we examined relationships between tissue element profiles and key clinical/demographic factors (e.g., disease stage, TNM (Tumor Node Metastasis) classification, lymph node status, sex, residence). Environmental exposure was specifically considered via subgroup analysis by urbanization level, reflecting potential environmental trace metal exposure gradients. Our results aim to improve understanding of CRC trace element dynamics and explore their use in disease stratification and biomarker development.

Results

Patient characteristics

Patient demographic and baseline clinical characteristics for the study cohort are detailed in Table 1. The cohort consisted of 34 males and 28 females (n = 62). With respect to clinical staging, 12.9% (n = 8) of patients were Stage 1, 38.7% (n = 24) were Stage 2, 40.3% (n = 25) were Stage 3, and 8.1% (n = 5) were Stage 4, with Stage 3 being the most frequently observed. The presence of distant metastases (M1 status) was recorded in 8.1% (n = 5) of the patient population.

Table 1.

Patients summary.

No	Variable	Stats / Values	Freqs (% of Valid) / Distinct Values	Valid	Missing
1	Sex; (categorical);	M; W;	34 (54.8%); 28 (45.2%);	62; (100.0%);	0; (0.0%);
2	Age; (quantitative);	Mean (sd) : 70.0 (8.48); min - max: 35–86;	24 distinct values	62; (100.0%);	0; (0.0%);
3	Clinical_stage; (categorical);	1; 2; 3; 4;	8 (12.9%); 24 (38.7%); 25 (40.3%); 5 ( 8.1%);	62; (100.0%);	0; (0.0%);
4	TNM_classification; (categorical);	1; 2; 3; 4;	2 ( 3.2%); 9 (14.5%); 42 (67.7%); 9 (14.5%);	62; (100.0%);	0; (0.0%);
5	Nodules; (categorical);	0; 1; 2;	33 (53.2%); 16 (25.8%); 13 (21.0%)	62; (100.0%);	0; (0.0%);
6	Metastases; (categorical);	0; 1;	57 (91.9%); 5 ( 8.1%);	62; (100.0%);	0; (0.0%);
7	Neoadjuvant; (categorical);	0; 1;	50 (80.6%); 12 (19.4%)	62; (100.0%);	0; (0.0%);
8	Residence; (categorical);	City_Over100k; Town_10kTo50k; Town_50kTo100k; Town_UpTo10k; Village;	17 (27.4%); 8 (12.9%); 5 ( 8.1%); 8 (12.9%); 24 (38.7%)	62; (100.0%);	0; (0.0%);
9	Residence_Code; (categorical);	0 (~ Village); 1 (~ Town_UpTo10k); 2 (~ Town_10kTo50k); 3 (~ Town_50kTo100k); 4 (~ City_Over100k);	24 (38.7%); 8 (12.9%); 8 (12.9%); 5 ( 8.1%); 17 (27.4%);	62; (100.0%);	0; (0.0%);

sd: standard deviation, min-max: minimal and maxial values.

Trace element concentration differences between tumor and adjacent healthy colorectal tissue

A paired Wilcoxon signed-rank test was performed to compare the concentrations of selected TEs between tumor tissue and adjacent healthy tissue. Statistically significant differences were observed for several elements. Initial analysis, before statistical correction, indicated higher levels of Al, Mn, and Cu in tumor tissue than in healthy tissue. Conversely, Se and Cd levels were found to be significantly lower in tumor tissue. Cr, Fe, Ni, Zn, and Sr did not show statistically significant differences in their levels between the two tissue types at this stage. Following multiple comparison adjustment, Mn and Cu levels remained significantly higher in tumor tissue, while Se and Cd levels remained significantly lower. The difference initially noted for Al was no longer statistically significant following this correction. All detailed elemental concentrations and statistics are available in Supplementary Table 2.

Stratified comparison of elemental concentrations in paired tumor and healthy tissues

Paired Wilcoxon comparison of metal concentrations in tumor and healthy tissues revealed significant elemental differences when stratifying by clinical variables. Stratification of the paired tissue analysis by clinical stage revealed significant, stage-dependent differences in the concentrations of several elements (Fig. 1). In earlier stages of the disease, a pattern of elemental depletion was predominant in tumor tissue. Specifically, selenium (Se) concentrations were significantly lower in tumors compared to adjacent healthy tissue in both Stage 1 (median 10.77 vs. 22.02 µg/g, respectively; p = 0.0078) and Stage 2 (median 3.32 vs. 11.11 µg/g, respectively; p = 0.012). This trend of depletion in tumors was also observed for strontium (Sr) in Stage 1 (p = 0.039) and for zinc (Zn) and cadmium (Cd) in Stage 2 (p = 0.029 and p = 0.0017, respectively). In contrast, a different pattern emerged in Stage 3, where manganese (Mn) and copper (Cu) concentrations were significantly elevated in tumor tissue. The median concentration of Mn was markedly higher in tumors than in healthy tissue (4.15 vs. 1.95 µg/g, respectively; p = 0.00042), and a similar, highly significant elevation was observed for Cu (median 61.68 vs. 38.43 µg/g, respectively; p = 3.81e-5). No significant differences between paired tissues were observed for these elements in other stages.

Fig. 1.

Comparative elemental concentrations in paired healthy and tumor tissues across Clinical stages. The figure presents boxplot comparisons of elemental concentrations between paired healthy and tumor tissue samples, with data further subgrouped by patient Clinical stage. Each panel corresponds to a specific element: (A) Mn, (B) Cu, (C) Zn, (D) Se, (E) Sr, and (F) Cd. Within each panel, comparisons are shown for distinct Clinical stage categories. P-values from paired Wilcoxon signed-rank paired tests comparing healthy versus tumor tissue concentrations within each clinical stage subgroup are included. The y-axis denotes elemental concentrations (units as appropriate, See Methods); the x-axis within each subgroup differentiates between healthy and tumor tissue.

When the data was stratified by TNM classification, significant differences between paired healthy and tumor tissues were exclusively observed in the large subgroup of patients classified as TNM 3 (N = 42), as shown in Fig. 2. Within this specific subgroup, both manganese (Mn) and copper (Cu) were found at significantly higher concentrations in tumor tissue. The median concentration of Mn increased from 2.19 µg/g in healthy tissue to 3.99 µg/g in tumors (p = 0.0032), while a similar elevation was seen for Cu (median 50.73 vs. 37.17 µg/g, respectively; p = 0.00019). Conversely, selenium (Se) levels were significantly lower in the tumor tissue of this same patient group (median 7.01 vs. 8.78 µg/g, respectively; p = 0.049). No significant differences were found within other TNM classification subgroups.

Fig. 2.

Comparative elemental concentrations in paired healthy and tumor tissues according to TNM classification. This figure displays boxplot comparisons of elemental concentrations between paired healthy and tumor tissue samples, with data further organized by patient TNM_classification categories. Each panel is dedicated to a specific element: (A) Mn, (B) Cu, (C) Zn, (D) Se, (E) Sr, and (F) Cd. Within each panel, comparisons of paired tissue concentrations are presented for distinct TNM_classification groups. P-values from paired Wilcoxon signed-rank tests assessing differences between healthy and tumor tissue within each TNM_classification subgroup are provided on the plots. The y-axis represents elemental concentrations (units as appropriate, See Methods); the x-axis within each subgroup distinguishes between healthy and tumor tissue.

Analysis of the data stratified by Residence Code, used as a proxy for environmental exposure, revealed significant differences in elemental concentrations between paired tissues in specific subgroups (Fig. 3). In the subgroup of patients residing in villages (Code 0), copper (Cu) concentrations were found to be significantly higher in tumor tissue compared to adjacent healthy tissue (median 47.14 vs. 35.14 µg/g, respectively; p = 0.0012). Conversely, in the subgroup of patients from large cities (Code 4), both selenium (Se) and cadmium (Cd) levels were significantly lower in tumor tissue. The median concentration of selenium decreased from 17.91 µg/g in healthy tissue to 8.02 µg/g in tumors (p = 0.0011), while cadmium levels declined from a median of 2.28 µg/g to 0 µg/g (p = 0.0080). Additional metrics are available in Supplementary Table 3.

Fig. 3.

Comparative elemental concentrations in paired healthy and tumor tissues by residence code as an environmental factor. This figure presents boxplot analyses comparing elemental concentrations in paired healthy versus tumor tissues, with data further stratified by Residence_Code, reflecting different levels of urbanization and potential environmental influences. The panels correspond to individual elements: (A) Mn, (B) Cu, (C) Zn, (D) Se, (E) Sr, and (F) Cd. For each element, paired tissue concentrations are visualized across different Residence_Code categories (0: ~Village, 1: ~Town_UpTo10k, 2: ~Town_10kTo50k, 3: ~Town_50kTo100k, 4: ~City_Over100k). P-values derived from paired Wilcoxon signed-rank tests for healthy versus tumor tissue comparisons within each residence category are indicated. The y-axis denotes elemental concentrations (units as appropriate, See Methods), while the x-axis within each displayed subgroup differentiates between healthy and tumor tissue.

Association of elemental concentrations with clinical and demographic characteristics

Stratified Wilcoxon tests by patient characteristics revealed distinct associations. Patient sex was significantly associated with differences in healthy tissue Cd (Cd_Healthy): females (N = 28, median 2.687) had higher levels than males (N = 34, median 0.000; p_adj = 0.021). Clinical stage correlated with elemental variations: healthy Se (Se_Healthy) tended to decrease with stage (e.g., Stage 1 [N = 8, med 22.02] vs. Stage 3 [N = 25, med 8.307] and Stage 1 vs. Stage 2 [N = 24, med 11.105] both yielded p_adj = 0.051). Healthy Ni (Ni_Healthy) suggested distributional differences (Stage 3 vs. Stage 1, p_adj = 0.064) despite identical medians. Other nominal (p < 0.05) stage-related variations (Cd, Al, Cu, Cr in healthy tissue) and all TNM-based differences were non-significant following correction (p_adj > 0.05 and > 0.11 respectively). Lymph node status (Nodules) showed only a borderline link for tumor Cd (Cd_Tumor) (N2 vs. N0/N1, p_adj = 0.053).

Stratification by Residence_Code (indicator of urbanization from villages [Code 0] to large cities [Code 4]) revealed several nominal elemental differences (p < 0.05), such as higher healthy Se (Se_Healthy) in large city residents (Code 4, N = 17, med 17.907) versus village dwellers (Code 0, N = 24, med 8.578; p = 0.025). Other nominal findings for Ni_Healthy, Cr_Tumor, and Cd_Healthy when stratified by residence also emerged. However, no environmental associations remained statistically significant after multiple comparison adjustment (p_adj > 0.23). Full results are in Supplementary Table 4.

Comparison of elemental ratios in paired tumor and healthy tissues

Analysis of paired tissue samples revealed significant alterations in several key elemental ratios. A marked elevation was observed in both the copper-to-zinc (Cu/Zn) and manganese-to-zinc (Mn/Zn) ratios within tumor tissue compared to adjacent healthy tissue. The median Cu/Zn ratio increased from 0.063 to 0.093 (p = 4.5e-7), while the median Mn/Zn ratio increased from 0.0032 to 0.0051 (p = 0.00047); both findings remained highly significant after correction for multiple comparisons (adj. p = 1.8e-6 and adj. p = 0.00095, respectively).

In contrast, the iron-to-copper (Fe/Cu) ratio was significantly lower in tumor tissue (median = 12.36) than in the corresponding healthy tissue (median = 16.93), a difference that also remained significant after adjustment (p = 0.018; adj. p = 0.024). Finally, no statistically significant difference was observed for the selenium-to-cadmium (Se/Cd) ratio between the paired tissue groups (p = 0.629). A complete summary of the statistical metrics, including effect sizes and test statistics, is presented in Supplementary Table 5.

Association between elemental ratios and clinical and demographic characteristics

Further analysis was conducted to explore associations between elemental ratios and key clinical and demographic characteristics. After correction for multiple comparisons, two significant associations were identified, both involving the iron-to-copper (Fe/Cu) ratio in tumor tissue.

A significant difference was observed between sexes, with female patients exhibiting a markedly lower tumoral Fe/Cu ratio (median = 10.28) compared to male patients (median = 15.69; adj. p = 0.0087). Additionally, neoadjuvant treatment status was associated with this ratio; patients who received neoadjuvant therapy had a significantly higher tumor tissue Fe/Cu ratio (median = 16.74) than treatment-naïve patients (median = 11.91; adj. p = 0.030).

Although several other comparisons yielded nominal significance (p < 0.05), such as differences in the Cu/Zn ratio by TNM classification or the Fe/Cu ratio by residence code, these associations did not remain statistically significant after applying the Benjamini-Hochberg correction. A complete summary of these analyses is presented in Supplementary Table 6.

Mining complex Clinical-Elemental data: machine learning feature selection for cancer classifications

The validated Boruta algorithm identified key predictive features for Clinical_stage and TNM_classification (Fig. 4). For Clinical_stage determination, primary predictors were Nodules (lymph node status) and TNM_classification achieving complete (100%) selection frequency in full and subsampled datasets. Notably, Cu_Tumor_Increased (elevated copper levels in tumor tissue) was also robustly identified (100% in full and 68% subsampled data). Similarly, for TNM_classification as the target, Clinical_stage (100% in full and 92% subsampled data) and Cu_Tumor_Increased (100% in full and 76% subsampled data) were the most significant, stable features. Supplementary Fig. 1 shows complete feature selection results for both clinical variables.

Fig. 4.

Stable features identified by boruta algorithm for patient stratification by key cancer clinical classifications. The plot displays the stable features identified by 100 iterations of the Boruta algorithm as key for patient stratification, according to the methodology outlined (selection frequency > 50% in both full dataset and 70% random subsample analyses, See Methods). Panel (A) presents the stable features for stratifying patients by Clinical_stage, while panel (B) shows those for TNM_classification. For each identified stable feature (y-axis), the bars (x-axis) illustrate their selection frequency (percentage of iterations confirmed as important) in analyses conducted on the full dataset (e.g., one color) versus the randomly subsampled data (e.g., another color), thereby demonstrating their robustness and consistent importance.

The Boruta algorithm identified Nodules, TNM_classification, and Cu_Tumor_Increased as primary features informative for classifying Clinical_stage. Similarly, for TNM_classification, Clinical_stage and Cu_Tumor_Increased emerged as predominant discriminators. These results suggest that Cu_Tumor_Increased provides robust and consistent discriminatory information across both staging classifications, highlighting its potential to complement established clinical parameters.

Discussion

This study aimed to characterize elemental concentrations in CRC and to investigate potential associations with disease stage using statistical and machine-learning approaches. Analysis of paired tumor and healthy tissues from 62 CRC patients showed significant elemental alterations with potential clinical relevance. Ten tissue metals (Al, Cr, Fe, Ni, Zn, Se, Sr, Cd, Mn, Cu) were evaluated. Among these, tumors showed significantly higher Mn and Cu, and lower Se and Cd, while Al, Cr, Fe, Ni, Zn, and Sr levels did not differ significantly from healthy tissue. These results are consistent with the hypothesis that differential TE profiles are associated with CRC pathogenesis.

Mn, Cu, Zn, Se, Sr, and Cd concentrations varied by clinical stage, TNM classification, and residence area. The observed patterns were highly element- and stage-specific: Mn and Cu concentrations were primarily divergent in Stage 3 tumors, whereas Zn and Cd differences emerged in Stage 2, and Se and Sr trends in Stage 1 (Se variations were also noted in Stage 2). Similarly, TNM classification differences were concentrated in TNM 3 tumors (Mn, Cu, Se). This complex, dynamic dysregulation of TEs during CRC progression suggests a potential mechanism contributing to oxidative stress and malignancy³². Differential Cu expression in patients from rural areas (villages) and Se/Cd depletion in tumors of urban residents support the hypothesis of an environmental exposure-tumor biology interaction, which warrants further investigation.

Trace elements are not uniformly distributed in human tissues, and previous studies have reported inconsistent findings in CRC. For example, Sohrabi et al.⁵ reported higher median levels of Zn, Cr, Cu, Al, and Pb in tumor tissues, but lower Sn and Fe. Lavilla et al.³³ observed elevated Mn, Sn, Cu, Al, Fe, Mg, Ca, K, P, and S in tumor sites, with lower Cd levels. Conversely, Juloski et al.⁴ found lower Cd, Cr, Co, Zn, and Hg in tumors, and higher Cu, Se, Ca, and Mg. Other studies also noted increased Cu, Fe, and Sn or elevated Mg, Cr, Zn, and Si in tumor tissues^24,34. These discrepancies may reflect significant differences in methodologies, such as the use of paired versus non-paired samples, but also environmental or socioeconomic differences, as supported by Tomlinson et al.⁶, who showed that environmental TE exposure correlates with cancer incidence and is significantly shaped by socioeconomic status.

Elevated tumor Mn in our study is plausibly related to overexpressed Mn-dependent enzymes like Mn-SOD (involved in controlling reactive oxygen species, ROS). Mn has also been implicated in activating p53-related pathways in CRC^10,35,36, although its precise role remains to be fully elucidated. Similarly, Cu is known to be involved in carcinogenesis and tumor progression both in vitro and in vivo^37,38, and acts as a cofactor for more than 30 enzymes including ceruloplasmin, ascorbate oxidase, lysine oxidase, dopamine-hydroxylase, cytochrome oxidase, and tyrosinase³⁹. Cu also promotes ROS generation via catalytic redox cycling and binds chromatin DNA bases, potentially inducing strand breaks, which is a key carcinogenic step^40,41. The increased Cu concentrations in tumor tissue observed in our cohort are consistent with Sohrabi et al.⁵.

Selenium is vital for redox balance, immune response, and blocking cancer pathways. The selenoproteins aid in apoptosis, anti-angiogenesis, and cell cycle control. Se is also explored in clinical use, including nanoformulations⁴². While Se may also support DNA repair and cancer suppression⁴³. Our finding of lower Se levels in tumor tissue compared to healthy tissue is consistent with the hypothesis regarding its protective role. However, excessive Se might promote cancer by inducing oxidative stress and abnormal cell proliferation⁴⁴. In line with Zhang et al.⁴⁵, SELENBP1 expression, a selenium-associated marker, has been inversely correlated with poor CRC differentiation, although no clear association with lymph node metastasis has been established⁴⁶. Separately, when assessing patient characteristics, our analysis found that healthy tissue cadmium (Cd) levels were significantly higher in females than in males, a finding consistent with previous reports on blood Cd concentrations⁴⁷.

Beyond the analysis of individual element concentrations, studying critical elemental ratios provides deeper insight into the functional dysregulation of the tumor microenvironment. The significantly elevated copper-to-zinc (Cu/Zn) and manganese-to-zinc (Mn/Zn) ratios in cancerous tissue are consistent with indicators of systemic oxidative stress. An increased Cu/Zn ratio is often interpreted as reflecting impaired cytosolic antioxidant defenses (via the enzyme Cu/Zn-SOD) and a simultaneous shift towards a pro-inflammatory and pro-angiogenic state associated with tumor progression^48–50. Concurrently, the elevated Mn/Zn ratio may suggests a profound mitochondrial dysfunction, reflecting the high level of reactive oxygen species production characteristic of altered cancer cell metabolism⁵¹. The significantly lower iron-to-copper (Fe/Cu) ratio observed in tumors may have important biological implications, reflecting a relative enrichment of copper over iron. This copper predominance could be linked to the promotion of angiogenesis, a critical process for tumor survival, growth, and metastasis. These findings are consistent with the concept that the tumor microenvironment actively supports the development of its own vascular network^52–54. Additionally, the association of this ratio with clinical factors is also noteworthy. The lower Fe/Cu ratio in female patients may reflect underlying physiological differences in iron and copper homeostasis between sexes^55,56. More strikingly, the higher Fe/Cu ratio in patients who received neoadjuvant therapy suggests a treatment-induced metabolic shift, possibly related to an influx of iron from therapy-induced cell death (e.g., ferroptosis) and inflammation, a phenomenon that warrants further investigation.

We identified a relative increase in copper in tumor compared with healthy tissue (Cu_Tumor_Increased) as a robust feature associated with clinical stage and TNM classification, contrasting with two recent meta-analyses that reported no consistent Cu differences between CRC patients and controls in blood or tissue⁵⁷. However, these studies predominantly used absolute Cu values from non-paired samples and showed high heterogeneity, especially in serum studies (I² > 97%). In contrast, our approach used paired tissue analysis and machine learning to account for intra-individual variability, highlighting relative copper dysregulation (vs. patient’s baseline) as a potentially context-sensitive marker. This may explain why increased tumor Cu consistently emerged as a meaningful feature, despite inconsistent findings from aggregate-level studies. The well-established roles of copper in tumor biology, including angiogenesis, proliferation, and redox regulation⁵⁸, together with our observation of patient-specific increases in tumor copper, underscore the biological significance of this element. These findings indicate that copper dysregulation represents a meaningful feature of disease status that merits further investigation. While further validation is needed, this potential is consistent with ongoing research, including clinical trials of copper-targeting strategies like chelation therapy (reviewed in⁵⁸, reflecting copper’s growing recognition in cancer pathophysiology and therapy.

The interpretation of these findings must take several inherent limitations into account. First, the study was based on a relatively small sample size collected from a single center, which limits the generalizability of the results and warrants caution when interpreting the machine-learning–based feature selection. Consequently, these exploratory observations should be regarded as preliminary and require validation in larger, independent cohorts. However, the statistical robustness of the significant findings is supported by internal validation, as demonstrated by consistent feature selection across multiple validation subsets and adequate statistical power achieved in post-hoc analyses. Secondly, the cross-sectional, single-time-point study design allows only for the identification of associations, precluding the establishment of definitive causality in elemental dysregulation. Crucially, the use of a paired design and per-patient relative metrics (ratios and ‘Increased’ features) substantially controls for systemic intra-individual variability, mitigating the impact of several unmeasured confounders. Despite this methodological control, the study still lacked comprehensive data on specific lifestyle and regional confounders, such as detailed exposure history and dietary habits, which were unavailable and warrant future systematic assessment. Thus, findings require confirmation in larger, multicenter studies with more detailed lifestyle/environmental adjustments. Despite these limitations, this study offers novel data on CRC TE profiles and is, to our knowledge, the first in a general Polish population to analyze trace elements in paired tumor and healthy tissues.

Conclusions

In conclusion, this study demonstrates pronounced alterations in tissue trace-element profiles in colorectal cancer, particularly involving copper, selenium, and cadmium, with clear associations related to disease stage, sex, and environmental factors. The paired-tissue design revealed that the relative increase in copper within tumor tissue—rather than its absolute concentration—correlates with clinical stage and TNM classification. This suggests that patient-specific elemental shifts may represent a biological feature of disease progression rather than a strictly clinical marker. Moreover, the observed selenium depletion and cadmium distribution patterns underscore the complex interplay between elemental homeostasis, tumor biology, and individual patient characteristics. Although these tissue-based findings provide valuable insights into CRC pathobiology and highlight its heterogeneity, they should be regarded as preliminary. Future studies integrating comprehensive lifestyle and exposure data will be essential to fully elucidate the role of elemental dysregulation in colorectal cancer.

Materials and methods

Patients and tissue samples

This cross-sectional study was conducted from June 2021 to May 2023 at the University Clinical Hospital No. 1 in Lublin, Poland. The study enrolled 62 consecutive patients with a histopathologically confirmed diagnosis of colorectal cancer (CRC) who were scheduled for colectomy. Inclusion and exclusion criteria were established for patient recruitment. Inclusion criteria were: (i) age 18 years or older, (ii) a histopathological diagnosis of CRC, and (iii) provision of written informed consent. Exclusion criteria were: (i) age younger than 18 years, (ii) chronic kidney disease (stages G3-G5, according to KDIGO guidelines⁵⁹, (iii) diagnosed maldigestion or malabsorption syndromes, (iv) requirement for parenteral nutrition, and (v) refusal to provide informed consent. The tissue samples used for this research were residual material from routine histopathological examinations. The status of each sample as either healthy or CRC tissue was confirmed by a certified histopathologist, and no additional invasive procedures were performed for sample collection. Patient demographics (age, gender, residence) were from questionnaires. Staging used TNM classification³²; records were reviewed for neoadjuvant treatment. Ethical approval (Medical University of Lublin Bioethics Committee, KE-0254/122/2021, Lublin, Poland) and hospital director permission were obtained. The study adhered to Declaration of Helsinki principles, and all participants provided pre-inclusion written informed consent. Moreover, all methods were performed in accordance with the relevant guidelines and regulations.

Tissue Preparation and assessment of trace elements

Tissues (0.03–0.65 g) were collected after colectomy into polypropylene sterile vials and stored in deep freeze (-80 °C). All samples underwent washing in deionized water with a resistivity of 18.0 MΩ cm (Ultrapure Milipore DirectQ 3UV-R (Merck KGaA, Darmstadt, Germany). Samples were dried for 60 min in a laboratory oven at 80 °C. Then, the dried samples were placed in pre-washed digestion vessels. Next, samples were digested with 10 mL of digestion mixture, that is, 7 mL of 69.0% nitric acid (HNO₃) solution + 3 mL of 30% m/m (H₂O₂). Both reagents were of Suprapur^® grade (Merck, Darmstadt, Germany). The digestion was carried out in Multiwave 5000 Digestion System (Anton Paar, Innsbruck, Austria) using Teflon^® vessels. The microwave-assisted sample preparation was conducted in a closed system, as described previously⁶⁰. The temperature was set at 180 °C. After cooling, deionized water was added to dilute the solutions (1 : 10 v/v) after digestion. Solutions were stored at 4 °C until analysis by Inductively Coupled Plasma Mass Spectrometry (ICP-MS).

The concentration of studied elements was determined with an XSeries 2 ICP-MS instrument (Thermo Fisher Scientific, Bremen, Germany) equipped with a collision/reaction cell operated using 7% H2/He gas mixture (Linde Gaz Polska, Poland), an ASX-510 autosampler (CETAC, Omaha, Nebraska, USA) and the PlasmaLab software (Thermo Fisher Scientific). Each sample was analyzed in triplicate. The confirmation of linearity was performed based on the lack-of-fit test and regression model test using ANOVA. The limit of detection (LOD) of the method was calculated based on a 3×standard deviation of 100 analytical blanks (Supplementary Table 1). Analytical accuracy was checked using certified reference material (CRM) with the serum matrix (Seronorm Trace Elements serum L2, Sero, Billingstad, Norway) with certified concentrations of analyzed elements. Additional non-matrix matched CRMs—EP-H-2 (EnviroMAT Drinking Water), and EU-H-3 (EnviroMAT Waste Water) (SCP Science, Quebec, ON, Canada)—were applied for validation of the analytical procedure. The concentrations found were in good agreement with the certified values. Recoveries were in the range of 95–110%. All solutions of multielemental (Merck, Germany) and monoelemental (Merck, Germany) ICP-MS standards were prepared daily by the dissolution of the reference materials in water and used for the calibration. One-point calibration was performed. Standards, blanks, and samples were analyzed with ¹⁰³Rh (10 µg/L; Merck, Germany) serving as the internal standard.

All elemental concentrations reported in this study are expressed in micrograms per gram (µg/g) of tissue.

A detailed description of the ICP-MS parameters is provided in our previous work⁶¹.

Statistical analysis

Metal concentrations were quantified in paired healthy and CRC tissue samples obtained from each patient. Based on their biological relevance, four key elemental ratios were calculated for each tissue sample: copper-to-zinc (Cu/Zn), selenium-to-cadmium (Se/Cd), iron-to-copper (Fe/Cu), and manganese-to-zinc (Mn/Zn). To compare metal concentrations between the paired healthy and tumor tissues from the same subject, the paired Wilcoxon signed-rank test was employed. This test was applied for each metal and also within subgroups defined by clinical variables such as clinical stage and TNM classification. For comparisons of metal concentrations between independent groups (e.g., when assessing differences based on categorical clinical variables, not in a paired manner), the unpaired Wilcoxon rank-sum (Mann-Whitney U) test was utilized. The log2 fold change (logFC) between groups was calculated based on the group means, using the formula log2((meanGroup2 + 1)/(meanGroup1 + 1)), where a pseudo-count of 1 was added to each mean for stability. Effect sizes were assessed using the rank biserial correlation coefficient (rrb) and the Hodges-Lehmann estimator of the median difference between groups. These metrics were employed to quantify the magnitude of observed differences independently of sample size and P-values, thus complementing the primary significance testing. Furthermore, post-hoc statistical power was calculated (Power_shiehpow_norm in Supplementary Tables) to ensure that the study was adequately powered (typically ≤ 0.80) to reliably detect the reported significant effects. P-values from these analyses were adjusted for multiple comparisons using the Benjamini-Hochberg method to control the false discovery rate (FDR). A significance level of p < 0.05 (after FDR adjustment where applicable) was generally considered statistically significant. The Residence_Code variable was derived by categorizing the patient’s place of residence into five groups based on population size: Village (coded as 0), Town up to 10k inhabitants (1), Town from 10k to 50k inhabitants (2), Town from 50k to 100k inhabitants (3), and City over 100k inhabitants (4) (See Table 1).

Binary features for subsequent machine learning analysis were engineered from elemental concentrations (Al, Cr, Mn, Fe, Ni, Cu, Zn, Se, Sr, Cd) in paired tumor and healthy tissue samples. Firstly, ‘_Increased’ features were derived by binarizing the comparison of each element’s tumor concentration against its corresponding healthy tissue concentration on a per-patient basis (evaluating ‘Element_Tumor > Element_Healthy’). Secondly, ‘_High’ features were generated for both tumor and healthy tissues by binarizing each elemental concentration relative to the cohort’s median for that specific element and tissue type (evaluating ‘Element_Tissue > median(Element_Tissue)’). These derived binary metrics served as inputs for the predictive models.

To identify the most important metal features that distinguish between tissue subgroups, the Boruta algorithm for R was implemented (version 8.0.0)⁶². The robustness of the feature selection was validated by performing 100 iterations of the Boruta algorithm on random 70% subsets of the data, followed by 100 iterations on the full dataset. Features consistently confirmed as important (selected in > 50% of iterations in both training and validation sets) were considered stable and significant discriminators. The ‘getImpFerns’ argument was used within the Boruta algorithm. All statistical analyses were performed using R software (version 4.4.1)⁶³. All statistical analyses were conducted using standard functions from publicly available R packages ensuring the full reproducibility of the analysis, as the methods rely solely on established, documented resources.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

AB and AP conceived the idea for the study. MK, JW, AP and AB contributed to the design of the research and methodology. MK was responsible for the software and its validation. MK, AP, GJN, AU, BP, PK, WM, BŚ conducted the formal analysis. The investigation was carried out by JW, ASZ, AP, GJN, AU, BP, PK, WM, BŚ and AB. Data curation was handled by MK, AP, JW. The original draft of the manuscript was prepared by AP, JW and AB. It was revised by MK and AB. Images were created by MK. Supervision was provided by AB. Funding acquisition was coordinated by WM and AB.All authors have read and approved the final version of the manuscript. All authors declare that they have no conflicts of interest.

Funding

This research was supported by the Medical University of Lublin under grant numbers DS 653 and DS 162.

Data availability

Data are available from the first and the corresponding author upon reasonable request.

Declarations

Competing interests

All authors have read and approved the final version of the manuscript. All authors declare that they have no conflicts of interest.The authors declare no competing interests.