MS-Based Ionomics: A Validated ... - ACS Publications

Apr 6, 2017 - Nestlé Institute of Health Sciences, 1015 Lausanne, Switzerland. ‡ ... ICP-MS/MS-based ionomics was used to analyze human serum of 12...
0 downloads 0 Views 1MB Size
Article pubs.acs.org/jpr

ICP-MS/MS-Based Ionomics: A Validated Methodology to Investigate the Biological Variability of the Human Ionome Tobias Konz,† Eugenia Migliavacca,† Loïc Dayon,† Gene Bowman,† Aikaterini Oikonomidi,‡ Julius Popp,‡,§ and Serge Rezzi*,† †

Nestlé Institute of Health Sciences, 1015 Lausanne, Switzerland Old Age Psychiatry, Department of Psychiatry, CHUV, 1011 Lausanne, Switzerland § Leenaards Memory Center, Department of Clinical Neurosciences, CHUV, 1011 Lausanne, Switzerland ‡

S Supporting Information *

ABSTRACT: We here describe the development, validation and application of a quantitative methodology for the simultaneous determination of 29 elements in human serum using state-of-the-art inductively coupled plasma triple quadrupole mass spectrometry (ICP-MS/MS). This new methodology offers high-throughput elemental profiling using simple dilution of minimal quantity of serum samples. We report the outcomes of the validation procedure including limits of detection/ quantification, linearity of calibration curves, precision, recovery and measurement uncertainty. ICP-MS/MS-based ionomics was used to analyze human serum of 120 older adults. Following a metabolomic data mining approach, the generated ionome profiles were subjected to principal component analysis revealing gender and agespecific differences. The ionome of female individuals was marked by higher levels of calcium, phosphorus, copper and copper to zinc ratio, while iron concentration was lower with respect to male subjects. Age was associated with lower concentrations of zinc. These findings were complemented with additional readouts to interpret micronutrient status including ceruloplasmin, ferritin and inorganic phosphate. Our data supports a gender-specific compartmentalization of the ionome that may reflect different bone remodelling in female individuals. Our ICP-MS/MS methodology enriches the panel of validated “Omics” approaches to study molecular relationships between the exposome and the ionome in relation with nutrition and health. KEYWORDS: ionomics, ionome, multielemental profiling, triple quadrupole ICP-MS



INTRODUCTION Elements are essential in every structural or functional aspect of life.1 While carbon, hydrogen, oxygen and nitrogen are the four most abundant organic elements in living organisms, many others own fundamental roles in protein, peptide, RNA and DNA structure.2 Elements exhibit also a broad range of biological functions from enzymatic catalysis, cellular signaling, mitochondrial activity, hormone synthesis, to redox and osmotic balance. Two very explicit examples of the many vital roles of elements are the ones of iron in oxygen transport by hemoglobin as well as magnesium in intracellular biologically active adenosine triphosphate. Other elements have detrimental effects on health such as the heavy metals cadmium, mercury and lead, which can cause serious toxicological damages.3 The postgenomic era has accelerated the development of novel methodological approaches to capture biological information related to these elements at gene, protein and metabolite levels with the ultimate goal to acquire in-depth knowledge of the molecular organization and function in living organisms. Many of the so-called “Omics” approaches enable the study of the genome, proteome and metabolome in a comprehensive manner and examples are increasingly reported © 2017 American Chemical Society

on their successful application in epidemiological and clinical research.4−6 One of the greatest premises of these approaches is to depict the molecular cross-talk between genes and environmental factors to understand the determinants from health status to homeostatic loss and disease. The field of biomarker discovery has then become a research priority to characterize disease risk factors and prevalence at population level. Biomarkers can also facilitate personalized healthcare throughout data driven patient stratification.7 An overarching challenge for biomarkers, beyond many possible utilizations for prediction and/or classification purposes, relies in their actual ability to explicit molecular relationships with the exposome. The exposome is defined as the sum of all environmental exposures at every life stage8,9 including physical, chemical, lifestyle and dietary factors. The impact of the exposome on living organisms can be measured using various approaches such as proteomics, lipidomics and metabolomics, the later targeting the real end points of physiological regulatory processes in biological fluids and tissues.7 The challenge is then to decipher the molecular relationships between the Received: February 1, 2017 Published: April 6, 2017 2080

DOI: 10.1021/acs.jproteome.7b00055 J. Proteome Res. 2017, 16, 2080−2090

Article

Journal of Proteome Research

scandium, gallium, indium, tellurium and bismuth at 100 ng mL−1 was mixed online with the samples via a T-connector. Standard clinical routine analytes were determined by using an Architect plus ci4100 clinical chemistry and immunoassay platform (Abbott Laboratories, Lake Bluff, IL, USA).

exposome and the phenome, which is defined as the complete set of phenotypic traits of a cell, tissue, organ, or whole organism. However, whereas numerous methodologies to analyze organic biomolecules have become very popular, the quantitative profiling of elements has not been fully established as a standard “top down” approach to investigate the human exposome-phenome interactions. Yet, elements such as micronutrients which can be found in the diet (calcium, magnesium, potassium, zinc, iron, manganese, iodine, cobalt, selenium), and environmental pollutants (cadmium, mercury, lead) are major determinants of the exposome with many health implications. Ionomics aims at measuring the entire elemental composition of a living organism (i.e., the ionome) and its dynamics relative to genetic, physiological and metabolic variability.10 As such, ionomics opens a complementary window to organic-based “Omics” approaches to investigate the exposome/phenome relationships. Ionomics has been previously developed and applied to study the biology and homeostasis of nutrients in plants.10−12 Other applications were reported in fish,13 unicellular organisms14,15 and human cells.16 Elemental profiling has also been applied in mammalian systems, including humans to identify alterations of saliva redox status in periodontitis,17 serum markers of osteoarthritis,18 schizophrenia,19 familial amyloid polyneuropathy,20 Alzheimer disease (AD),21 metabolic16,22 and occupational health.23 Furthermore, ionomics has been applied to define element reference ranges in the general population.24−28 A series of methodologies mainly based on inductively coupled plasma mass spectrometry (ICP-MS) was published for different biological fluids including saliva,17,29 blood serum23,26,30−34 and plasma,24,35 whole blood,25−27,31,34,35 cerebrospinal fluid36,37 and urine.23,24,27,38 Those methods showed different degrees of performance in terms of validation (recovery, intra- and interday variability and limit of detection (LOD)). To our knowledge, all the reported methods show limitations in at least one of the important performance criteria for ionomics such as full method validation, number of biologically relevant elements measured in a single run, high throughput, and low sample consumption. Moreover, biologically important elements such as sulfur and phosphorus cannot be incorporated into simultaneous profiling of multiple elements due to polyatomic interferences that cannot be resolved by single quadrupole ICP-MS.39 We here report the development and validation of a comprehensive and quantitative method encompassing the measurement of 29 biologically relevant elements (B, Mg, Al, P, S, K, Ca, Ti, V, Cr, Mn, Fe, Co, Ni, Cu, Zn, As, Se, Br, Rb, Sr, Mo, Cd, Sn, I, Cs, Ba, Hg and Pb) in human serum for high throughput ionomic study.



Chemicals and Materials

All solutions were prepared by using 18 MΩ cm−1 deionized water obtained from a Milli-Q system (Millipore, Bedford, MA, USA). All chemical substances used for analysis were of highest grade of purity available. The diluent solution was composed of 5% 1-butanol (99.9%, Sigma-Aldrich, St. Louis, MO, USA), 0.05% EDTA (ethylenediaminetetraacetic acid, 99.995% trace metals basis, Sigma-Aldrich), 0.05% triton X-100 (BioXtra, Sigma-Aldrich) and 0.25% ammonium hydroxide (SigmaAldrich). Solutions for external calibration and online-internal standard were prepared from ICP standards (Merck Millipore, Darmstadt, Germany), while the tuning solution was purchased from Agilent Technologies (Santa Clara, CA, USA). Human Samples and Study Population

A pool of human serum (Type AB Male) was purchased from Biopredic International (Saint Grégoire, France). The human serum reference material Seronorm Trace Elements Serum L-2 was obtained from Sero (Billingstad, Norway). The application of the human serum pool and human reference material was approved by the Ethics Committee Vaud, Switzerland. Serum samples were also obtained from 120 community dwelling older adults with a mean age of 70.4 years (standard deviation (SD) = 7.9), 43 participants were males and 77 were female. Subjects were recruited among outpatients referred to the Memory Clinics, Old age psychiatry, Departments of Psychiatry, and the Leenaards Memory Center, Department of Clinical Neurosciences, University Hospital of Lausanne (Switzerland) or recruited from the community through advertisement or among the spouses of memory clinic patients. The Ethics Committee Vaud, Switzerland approved the clinical protocol and all participants signed written informed consent. Venous punctures were performed between 8:30−9:30 am after overnight fasting using serum S-Monovette from Sarstedt (Nümbrecht, Germany). After 20−30 min the serum samples were centrifuged and aliquoted in polypropylene tubes from Treff AG (Degersheim, Switzerland) and stored at −80 °C. To reduce the risk of batch effects, the specimens were analyzed randomly in terms of gender and age. The preparation of blanks, calibration solutions, human serum samples and quality controls (QCs) was conducted using metal free centrifuge tubes (VWR, Radnor, PA, USA). Sample Preparation Procedures

EXPERIMENTAL SECTION

All sample preparation steps were conducted at room temperature. Before analysis, each serum sample was thawed only once and homogenized for 10 s using a vortex mixer. Subsequently, the samples were diluted (1:10) using the diluent solution. For the preparation of the spiked serum samples, a human serum pool was aliquoted and spiked with ICP standards. Three different concentration levels, adjusted to the concentration range of the respective elements in human serum were prepared. To avoid cross contamination of the analytes, two different sets of spiked samples were prepared: set 1 consisting of level 1−3 (Mg, P, S, K, Ca and Mo) and set 2, level 1−3 (B, Al, Ti, V, Cr, Mn, Fe, Co, Ni, Cu, Zn, As, Se, Br, Rb, Sr, Cd, Sn, I, Cs, Ba Hg and Pb).

Instrumentation

All ICP-MS experiments carried out in this study were performed using an Agilent 8800 triple quadrupole ICP-MS (Agilent Technologies, Tokyo, Japan) operated in low matrix plasma mode. The mass spectrometric device was equipped with an integrated autosampler, a concentric nebulizer and a Scott double-pass spray chamber. The instrument was tuned prior analysis by using a multielement tuning solution to verify the functional conditions of the mass spectrometer. A selection of the optimized parameters of the ICP-MS instrument is described in Table S1 Supporting Information. An internal standard (ISTD) solution containing beryllium (200 ng mL−1), 2081

DOI: 10.1021/acs.jproteome.7b00055 J. Proteome Res. 2017, 16, 2080−2090

Article

Journal of Proteome Research

Table 1. Calculated LOD, LOQ and Required LOQ in ng mL−1 and Coefficient of Determination of the Analytical Method analyte

transition [gas mode]

LOD [ng mL−1]

LOQ [ng mL−1]

required LOQ [ng mL−1]

correlation coefficient (r)

Boron Magnesium Aluminum Phosphorus Sulfur Potassium Calcium Titanium Vanadium Chromium Manganese Iron Cobalt Nickel Copper Zinc Arsenic Selenium Bromine Rubidium Strontium Molybdenum Cadmium Tin Iodine Cesium Barium Mercury Lead

11 B [no gas] 24 Mg [He] 27 Al [He] 31 → 47 P [O2] 32 → 48 S [O2] 39 K [He] 43 Ca [He] 48 → 64 Ti [O2] 51 → 67 V [O2] 52 → 68 Cr [O2] 55 Mn [He] 56 Fe [He] 59 Co [He] 60 Ni [He] 63 Cu [He] 66 Zn [He] 75 → 91 As [O2] 80 → 96 Se [O2] 79 Br [He] 85 Rb [He] 88 Sr [He] 98 Mo [He] 111 Cd [He] 118 Sn [no gas] 127 I [no gas] 133 Cs [He] 138 → 154 Ba [O2] 202 Hg [no gas] 208 Pb [no gas]

0.0147 0.3740 0.0267 3.8915 0.2732 1.4485 12.1088 0.0027 0.0001 0.0038 0.0037 0.0246 0.0002 0.0027 0.0029 0.4082 0.0003 0.0084 0.0851 0.0015 0.0048 0.0020 0.0001 0.0087 0.0202 0.0105 0.0014 0.0008 0.0004

0.0444 1.1334 0.0809 11.7924 0.8280 4.3894 36.6934 0.0082 0.0004 0.0114 0.0111 0.0745 0.0007 0.0082 0.0088 1.2370 0.0008 0.0255 0.2578 0.0044 0.0144 0.0062 0.0002 0.0263 0.0613 0.0318 0.0041 0.0025 0.0013

0.700 1820.0 0.080 6300.0 103320.0 12600.0 8817.2 0.010 0.0015 0.0080 0.0300 55.00 0.0030 0.0130 74.00 42.0 0.0500 7.40 120.0 1.313 1.20 0.027 0.0013 0.03 6.50 0.05 0.040 0.021 0.012

0.9999 0.9998 0.9990 0.9997 0.9997 0.9998 0.9999 0.9992 0.9994 0.9989 0.9998 0.9998 0.9996 0.9996 0.9999 0.9999 0.9999 1.0000 0.9998 0.9999 1.0000 0.9982 0.9988 0.9980 0.9999 0.9998 0.9998 0.9997 0.9995

Analytical Validation

thoroughly homogenized by using a vortex mixer. The solution was immediately transferred into a perfluoroalkoxy alkane (PFA) vial and diluted 1:10 with diluent solution before analysis. The robust recovery was calculated according to eq 1:

Calibration Curves and Residuals. The external calibration method was selected for this method validation. The calibration range of each element was determined according their expected physiological concentrations in human blood serum. Each ICP standard was carefully analyzed for potential contaminations during the method development. As Mg and Mo ICP standards carried significant amounts of Pb, Cd and I, respectively, two calibration curves (a: Mg, P, S, K, Ca, Mo and b: B, Al, Ti, V, Cr, Mn, Fe, Co, Ni, Cu, Zn, As, Se, Br, Rb, Sr, Cd, Sn, I, Cs, Ba Hg, Pb) were implemented to account for the effects of these contaminations on the estimation of their endogenous levels in blood serum. Calibration solutions were spiked with typically 10 concentration levels for each element. Internal standards were applied to each sample types (blank, calibration standard, spiked-serum, certified reference material and authentic samples) in order to account for potential physical matrix effects such as sample transport and plasma fluctuations. The residuals for each element were calculated from the respective calibration curves. LOD/LOQ. For the determination of LODs and limits of quantification (LOQ) blank samples (diluent solution) were spiked with different concentration levels of each element (9 levels). Each sample was analyzed in triplicate. The LODs and LOQs were determined according to the recommendation of the International Conference on Harmonization (ICH) guidelines.40 Selectivity. of the method was evaluated against a certified reference material, Seronorm Trace elements in serum L-2. To each vial, 3 mL of Milli-Q water was added and the sample was

Recoveryrobust =

ConcMedian × 100 Conc ref

(1)

Trueness and Precision Experiments. A human serum pool was aliquoted and spiked at three different concentration levels as discussed previously. One serum aliquot was applied as blank sample. The experiment was conducted by the same analyst on 6 days, covering as many sources of variation as possible. Each sample was analyzed in triplicate. The obtained results for trueness/recovery were calculated according to eq 1 using the median of the determined concentration. Measurement Uncertainty. Measurement uncertainty (u) was calculated using eq 2, with the coefficient of variation of intermediate reproducibility (CV(iR)) and the corrected relative standard deviation of the recovery (RSD(Rec)corrected). This approach is based on existing validation data as proposed by Barwick and Ellison.41,42 In this case, the precision and trueness studies provide the necessary data required to calculate measurement uncertainty. Precision and trueness contributions are combined together to obtain the overall uncertainty. For the calculations, the uncertainty related with the concentration of the reference sample has to be expressed in terms of standard deviation. For spiked samples the uncertainty is generally negligible compared with the method variability and considered as equal to zero. 2082

DOI: 10.1021/acs.jproteome.7b00055 J. Proteome Res. 2017, 16, 2080−2090

Article

Journal of Proteome Research Table 2. Validation Results for Trueness/Recovery, Precision (Repeatability and Intermediate Reproducibility) and Measurement Uncertainty of the Analytical Method

standard uncertainty analyte

ref. value [ng mL−1]

median of results [ng mL−1]

trueness, recovery [%]

rep. CV(r) [%]

inter. rep. CV(iR) [%]

u [ng mL−1]

rel. u [%]

B Mg Al K Ca Ti P S V Cr Mn Fe Co Ni Cu Zn As Se Br Rb Sr Mo Cd Sn I Cs Ba Hg Pb

5.33 1066.7 2.67 15644 4978 0.462 3556 53333 0.021 0.096 0.377 53.33 0.064 0.196 58.67 53.33 0.160 5.333 122.7 9.60 9.07 0.072 0.007 0.114 4.80 0.444 2.87 0.125 0.049

6.15 999.4 2.71 14707 4780 0.512 3555 50120 0.022 0.106 0.375 49.90 0.059 0.190 53.28 50.24 0.170 5.440 135.6 9.14 8.75 0.068 0.007 0.107 4.97 0.410 2.75 0.123 0.045

115.2 93.7 101.8 94.0 96.0 110.9 100.0 94.0 104.9 110.8 99.6 93.6 93.0 97.2 90.8 94.2 106.4 102.0 110.6 95.3 96.5 95.0 95.8 94.1 103.5 92.3 95.6 98.8 92.5

1.3 2.0 10.4 1.8 4.2 9.0 3.3 3.1 3.8 6.5 3.8 4.7 3.4 4.8 2.0 3.2 2.2 3.8 2.5 1.6 1.4 6.7 7.8 9.3 2.4 4.2 3.7 2.9 2.2

1.5 1.9 18.9 4.2 5.2 11.5 8.4 8.0 7.6 6.5 5.7 9.2 5.8 4.4 4.2 3.5 1.9 6.1 11.7 3.7 1.8 10.8 7.8 13.7 4.2 5.2 6.3 6.2 5.2

0.40 39.5 0.55 806 282 0.070 322 4273 0.002 0.008 0.023 4.93 0.004 0.009 3.61 2.31 0.007 0.355 17.2 0.42 0.25 0.008 0.001 0.016 0.22 0.028 0.20 0.008 0.003

6.5 4.0 20.1 5.5 5.9 13.4 9.1 8.6 8.1 7.8 6.0 9.8 7.3 4.5 6.8 4.6 4.0 6.5 12.6 4.5 2.8 11.5 8.0 14.5 4.4 7.0 7.5 6.7 6.8

2 u = Median CV(iR )2 + RSD(Rec)corrected

sulfur, iron, tin and iodine are shown in Figure S1 Supporting Information. This finding and the randomly distributed points of the residuals confirm a linear correlation between the detector response and the concentration of the elements in the calibration solutions. Consequently a linear fitting can be applied for quantification.

(2)

Statistical Analysis. All statistical analysis was conducted using the software R version 3.3.1. Linear models were performed to establish the association between mineral serum concentrations, after logarithmic transformation, and demographic variables including gender and age. Differences between two groups were evaluated using t tests, after logarithmic transformation, and Wilcoxon tests. Principal component analysis (PCA) was performed by conducting a singular value decomposition of the centered and scaled log10 transformed elemental concentrations, using prcomp function.

b. Method LOD/LOQ



RESULTS AND DISCUSSION The developed method was subjected to an in-depth single laboratory validation to evaluate the (a) linearity of the calibration curves, (b) method LODs and LOQs, (c) selectivity, (d) trueness, (e) precision, and (f) measurement uncertainty.

The determined LOQs should be below the lowest concentrations expected in the sample. These values are obtained by dividing the lowest concentration of an element in serum by 10 in order to reflect the 10-fold dilution of the sample. As can be extracted from Table 1, the determined LOQs are below the targeted LOQs for most analytes. However, the values for aluminum and chromium are slightly above the respective threshold. In the case of Al, the achieved LOQ is very close the targeted value (0.081 compared to 0.08 ng mL−1) while for Cr (0.011 to 0.008 ng mL−1) the nominal difference is more pronounced.

a. Linearity of the Calibration Curves

c. Selectivity, Matrix Interferences in Serum Samples

To evaluate the linearity of the calibration curves, the correlation coefficient (r) was determined, revealing values of r ≥ 0.999 for most of the elements under evaluation (r ≥ 0.995 is considered to be linear).43 Table 1 summarizes the observed results. Furthermore, all residual plots were visually evaluated with respect to the corresponding regression lines and showed that points were randomly distributed around the regression line. The calibration curves and respective residual plots of

For this methodology, isotopic transitions in the respective gas modes (no gas, He or O2) were chosen to remove relevant spectral interferences. To investigate if the elements are detected in the absence of relevant matrix and spectral interferences, a certified reference material was analyzed (see Table S2 Supporting Information). The median recovery values for all analytes under evaluation in the certified samples ranged from 82.9 to 108.7%. The recovery results for Sn and Cs were 2083

DOI: 10.1021/acs.jproteome.7b00055 J. Proteome Res. 2017, 16, 2080−2090

Article

Journal of Proteome Research

are increasingly reported. In this vein, some scientists applied this technique to compare serum levels of trace elements between healthy controls and Schizophrenia or AD patients, respectively.19,21 Despite the fact that essential elements such as I, K, S and Br are not included in the panels, no validation results have been presented. Harrington et al. applied two different techniques to determine Ca, K, Mg and Na (ICPoptical emission spectrometry (OES)) and Cr, Fe, Co, Cu, Zn Se and Mo (ICP-MS) in human serum.31 Here, elements measured by ICP-OES showed good interday reproducibility (≤10%), while some elements (Cr, Co, Fe, Mn, and Mo) determined by ICP-MS failed to meet the criteria for interday reproducibility (up to 118%). Zhao et al. reported the measurement of 64 trace elements in the serum of female osteoarthritis patients and age-matched controls.18 It appeared that most elements under evaluation with mass >89 amu (33 elements) showed relatively poor intra- and interday precision results (>20%). Moreover, the evaluation of the recovery was carried out for only 14 elements, revealing significant deviations from the theoretical values (Ca, Ti, Zn < 80%, Al and 65Cu > 120%). The quantitative strategy reported by Verni et al. encompasses 28 trace elements and toxic heavy metals in human serum.33 For most of the elements, notable validation results were obtained. However, the biological relevant elements V, Cr, I and Br were not taken into consideration. In comparison to most of the published methodologies for holistic elemental profiling, the herein presented method was validated in an exhaustive manner. The reported analytical figures-of-merit confirm the fitness-for-purpose. Moreover, this method encompasses a holistic spectrum of biological important elements such as the heteroatoms S and P, the essential trace elements Mn, Fe, Co, Cu, Zn, As, Se, Br, Mo and I, potentially biorelevant elements such as B, Al, Ti, V, Cr and Ni, the electrolytes Mg, K and Ca, elements with similar chemical behavior than electrolytes Rb, Sr, Cs and Ba, and toxic heavy metals like Sn, Cd, Hg and Pb. Such a holistic approach reduces the need for preanalytical literature research in order to define study-specific target elements. In addition, the use of a single method-based strategy also eliminates intermethod bias, allows faster analysis while requiring less sample volume (a single aliquot of 150 μL is required). All these advantages enable the application of the proposed methodology to clinical studies were high number of samples and limited sample volume is available. We then applied our ionomic method to the analysis of serum element profiles from 77 female and 43 male individuals within an age range from 49 to 85 years. Serum specimens were randomly analyzed to prevent from putative gender and agedriven batch effects. Median values of the measured elements and their respective known reference ranges are reported in Table 3. Our element profiling method enabled to quantify S taking advantage of the MS/MS to eliminate the limiting interference between S ions (32S) with polyatomic oxygen species (16O2+). Sulfur was the most abundant element in serum with a median concentration of 1.063 mg mL−1. Sulfur belongs to the ten most abundant elements in the human body and is essential to life under various speciation forms. The hetero element is present in many proteins either incorporated in the amino acids cysteine and methionine or as iron sulfur cluster and can be found in a variety of molecules (i.e., glutathione, homocysteine, taurine, biotin, thiamine, coenzyme A, lipoic acid). Yet knowledge stills missing on the definition of optimal sulfur status in humans. Most of the other measured

not considered since the concentration of these elements in the samples was below the LOQs (0.025 vs 0.0263 ng mL−1 and 0.002 vs 0.0318 ng mL−1, respectively). The obtained results demonstrated the specificity of the method as practically all analytes can be accurately measured in the presence of other elements in the sample matrix that causes matrix and potentially spectral interferences. d. Trueness

The recovery results for spiked serum level 2 are represented in Table 2. The lowest recovery was obtained for Cu (90.8%) while the highest recovery is attributed to B (115.2%). The mean of all recoveries was 98.9%. The values for practically all analytes in spiked serum (see Table 2) were in good agreement with the recovery ranges recommended by the FDA (U.S. Department of Health and Human Services, Food and Drug Administration)44 and are within 15% of the reference values. e. Precision (Repeatability/Intermediate Reproducibility)

Coefficient of variation of reproducibility CV(r) and CV(iR) of the analytes obtained on 6 different days should be under 15% and may not exceed 20%. The obtained precision results using robust statistics are summarized in Table 2. The CV(r) ranged from 1.3 to 10.4% with a mean of 4.1% for all analytes while the CV(iR) varied between 1.5% and 18.9% (mean 6.6%). For Al the CV(iR) was between 15% and 20% which may be due to the low amount of spiked analyte. f. Measurement Uncertainty

During the validation experiment, standard uncertainties between 2.8% and 20.1% were observed (see Table 2) and agree very well with values expected for external calibration and are significantly higher compared with methods applying isotope dilution analysis for quantification.45 Possible errors in the preparation of the calibration solutions and spiked samples, particularly in the case of low concentrations, increase these uncertainties and may be responsible for some of the observed values. Comparison to Other Methodologies. Early methodologies applied to holistic mineral profiling in human serum or plasma were rather based on spectroscopic techniques27,46 or high resolution or sector field (SF) ICP-MS23,28 than on single quadrupole ICP-MS. In 1999 and 2001, Muniz et al. applied a SF-ICP-MS based method to determine a panel of 15 most biologically relevant elements (Al, Ca, Cr, Mn, Fe, Co, Cu, Zn, Se, Rb, Sr, Mo, Cd, Pb and U).28,32 However, no validation data was presented. Since then, only a limited number of SF-ICPMS methodologies were reported.20,24 Judging on the available validation data, these methodologies seem to be well suited for the quantitative analysis of up to 17 trace elements. Nevertheless, the fact that in almost two decades the number of analytes has not substantially increased tends to demonstrate that SF-ICP-MS has reached its technical limits in terms of multielemental analysis. This trend might also be related to the introduction of multipole-collision/reaction cells for quadrupole-based ICP-MS, allowing a more efficient removal of polyatomic interferences which in turn hamper the proper detection of elements with m/z < 80. This is particularly reflected by studies in which collision/reaction cells have not been applied; challenging elements were either not determined or the obtained values indicated the presence of spectral interferences, leading to overestimated concentrations.30 Therefore, it is not surprising that the number of holistic methodologies using collision/reaction cell equipped ICP-MS 2084

DOI: 10.1021/acs.jproteome.7b00055 J. Proteome Res. 2017, 16, 2080−2090

Article

Journal of Proteome Research

that our method measures total serum phosphorus unlike routinely used spectrophotometric methods that are limited to the quantification of inorganic phosphates. We used principal component analysis (PCA) to analyze the global variance of the measured ionomic profiles. The first (PC1) and second (PC2) principal components accounted for 12.1 and 8.0% of the total ionome variance, respectively (Figure 1a). The scores plot of PC1 and PC2 displays interindividual variability of the serum ionome with a partial comapping of the samples according to gender along PC1. The examination of the corresponding loading plot (see Figure 1b) revealed that serum levels of Ca, P, Rb and Cu were the most influential elements in the observed gender ionome differences. These findings were further evaluated by computing univariate statistical analyses on the concentrations of these elements. Statistically significant increases of levels of Ca, P and Cu in serum from female subjects were confirmed (p-values