Elemental analysis by Metallobalance provides a complementary support layer over existing blood biochemistry panel-based cancer risk assessment

Abstract

Despite the benefit of early cancer screening, Japan has one of the lowest cancer screening rates among developed countries, possibly due to there being a lack of "a good test" that can provide sufficient levels of test sensitivity and accuracy without a large price tag. As a number of essential and trace elements have been intimately connected to the oncogenesis of cancer, Metallobalance, a recent development in elemental analysis utilizing the technique of inductively coupled plasma mass spectrometry has been developed and tested as a robust method for arrayed cancer risk screening. We have conducted case-control epidemiological studies in the prefecture of Chiba, in the Greater Tokyo Area, and sought to determine both Metallobalance screening's effectiveness for predicting pan-cancer outcomes, and whether the method is capable enough to replace the more conventional antigen-based testing methods. Results suggest that MB screening provides some means of classification potential among cancer and non-cancer cases, and may work well as a complementary method to traditional antigen-based tumor marker testing, even in situations where tumor markers alone cannot discernibly identify cancer from non-cancer cases.

Keywords: Cancer biology; Chemical biology; Epidemiology; Mass spectrometry; Preventative medicine.

Figure 1. Metallobalance (MB) as a complementary test for pan-cancer screening.

Conventional serological examinations of tumor markers by methods such as electrochemiluminescence immunoassays (ECLIA), is only limited to a small known panel of markers, e.g., CEA, CA19-9, PSA and CA125; conversely, only a limited number of cancer susceptibility may be reportable (bottom right example; shaded grid indicates reportable type of cancer per marker; GI, gastrointestinal; BR, breast; LN, lung; BD, bladder; PS, prostate; OV, ovarian). Metallobalance (highlighted in red) on the other hand, utilizes induced collision plasma assisted mass spectrometric analysis to ionize serological analytes and produces a profile of 17 elemental levels. When combined cancer risk outcomes can generate logistic regression models to characterize susceptibility to a large number of cancer types simultaneously.

Figure 2. ICP-MS levels of certain elements may provide well differentiable indicators for cancer occurrence.

(A) log odds-ratios (OR) of elements tested in MB screening. Red shaded regions indicate those with log OR > 0 for cancer occurrence; results shown of average ORs from 1000 random SMOTE sampling trials. (B) ROC analysis of MB logistic regression classifier models for P₀ and P₁; black, all elements; red, model constructed only from those with log₁₀OR >0; AUC, area under curve. (C, D) Distributional overlap of study cohorts by UMAP. Labeled colors indicate various conditions, either by cohort population (C) or by elemental levels (D). S (left) and Zn (right) are shown for representative purposes, gradients of light to dark purple indicate low to high elemental levels;P₀, non-cancer control; P₁, cancer cases.

Figure 3. Differences in effect size in MB profiles.

(A) Representative cases of elements with small (lung cancer, left) and large (ovarian cancer, right); (B) heat map of various different cases and corresponding effect sizes of P1, with color gradients of d indicated in the upper left corner; (C) comparison of Cu (left) and Cs (right) levels between P₀ and P₁ cohorts. M, male; F, females. Asterisk, p < 0.05.

Figure 4. Behavior of tumor biomarkers in cancer and non-cancer cases.

Shaded yellow region indicates normal thresholds (upper limit, bottom right) for each respective marker. Vertical and horizontal axes indicate pairings of tumor biomarkers and concentration.

Figure 5. Identifying critical elements as support indicators for tumor biomarkers.

(A) Two-dimensional comparison of generalized linear regression coefficients non-cancer (P₀, horizontal axes) vs. cancer (P₀, vertical axes) of MB elements; solid reference line has a slope of 1. (B) Comparison of mean Mo levels among subjects with normal marker ranges; P₀ red; P₀, blue.

Figure 6. Receiver-operating characteristics of a random forest classifier model for P₀ and P₁.

Representative result from one SMOTE sampling trial. AUC, area under curve.