Molecular Phenotyping of Oxidative Stress in Diabetes Mellitus with Point-of-care NMR system

Diabetes mellitus is one of the fastest growing health burdens globally. Oxidative stress which has been implicated to the pathogenesis of diabetes complication (e.g., cardiovascular event) were, however, poorly understood. We report a novel approach to rapidly manipulate the redox chemistry (in a single drop) of blood using point-of-care NMR system. We exploit the fact that oxidative stress changes the subtle molecular motion of water-proton in the blood, and thus inducing a measurable shift in magnetic resonance relaxation properties. This technique is label-free and the whole assays finish in a few minutes. Various redox states of the hemoglobin were mapped out using our newly proposed two-dimensional map, known as T1-T2 magnetic state diagram. We demonstrated the clinical utilities of this technique to rapidly sub-stratify diabetes subjects based on their oxidative status (in conjunction to the traditional glycemic level), to improve the patient risk stratification and thus the overall outcome of clinical diabetes care and management. (155 words) Key Points for Summaries A novel approach to rapidly manipulate the redox chemistry (in a single drop) of blood using point-of-care NMR system. Assessment of the oxidative status, in conjunction to their glycemic level allows sub-stratification of diabetes subjects which was demonstrated clinically. Visual Abstract

relaxation properties. This technique is label-free and the whole assays finish in a few 23 minutes. Various redox states of the hemoglobin were mapped out using our newly proposed Diabetes mellitus (DM) is one of the fastest growing health burdens that is projected to affect 2 592 million people worldwide by 2035 1 . DM is defined by a persistent elevation of plasma 3 glucose concentration. Under chronic hyperglycemic condition, glucose is non-enzymatically 4 attached to protein (glycation), which has deleterious effects on their structure and function. 5 Hence, glycated hemoglobin A1c (HbA1c), which reflects the overall glycemic burden of an 6 individual over the previous 2─3 months, is increasingly used to diagnose the disease 2 . It is 7 also recommended for monitoring long-term glucose control of DM patients, and for risk 8 stratification 3,4 . 9 However, HbA1c does not adequately reflect all the disease associated risk factors. In 10 particular, restoring HbA1c level to near-normal level does not necessarily translate into a 11 significant reduction of cardiovascular event, a diabetes complication commonly associated 12 with oxidative stress 5 . In addition, subjects with stable chronic hyperglycemia due to 13 glucokinase mutations were found to have unexpectedly lower prevalence of 14 micro/macrovascular complication. A major pathological effect of diabetes mellitus is the 15 chronic oxidative-nitrosative stress and recently reported carbonyl 6 and methylglyoxal 16 stress 7 , which drives many of the secondary complications of diabetes including 17 nephropathy, retinopathy, neuropathy, and cardiovascular diseases 8 . Oxidative-nitrosative 18 stress can damage nucleic acids, lipids and proteins, which severely compromise the cellular 19 health and induce a range of cellular responses leading ultimately to cell death [9][10][11] . Direct 20 measurement of oxidative stress and susceptibility in patients may improve the prediction 21 of disease associated risks related to oxidative stress, and hence improve the long term 22 diabetes care and management program 12,13 . 23 Currently, an individual's oxidative status cannot be easily characterized in detail using 24 routinely available biomarkers 14 in clinical practice and/or at point of care. This has impeded 25 the understanding of the pathological effects of acute and prolonged exposure to oxidative 26 stress. The reactive oxygen species (ROS) and reactive nitrogen species (RNS) are often 27 reactive and short-lived, may disrupt the redox state of biological tissues/cells (e.g., red 28 blood cells (RBCs), plasma). Several methods have been developed to detect the redox 29 properties of the blood using the optical 13,15 or magnetic properties 16,17 of the inorganic iron-1 chelate of hemoglobin (Hb) and plasma albumin.
2 Electron spin resonance is commonly used to detect the ROS/RNS directly 18,19 . However, the 3 approach is hampered by inherent sample stability issues and limited sensitivity 20 . Stable 4 molecular products formed from reactions with ROS/RNS, such as the oxidation targets (e.g., 5 lipid, protein, nucleic acid) are measurable using a range of spectrophometric assays and 6 mass spectrometry (MS) 21 . Nevertheless, fluorescent-staining often causes cell-toxicity 21,22 , 7 and therefore these assays may not provide information that reflects in vivo conditions. 8 Ultraviolet-visible light spectroscopy has poor spectral resolution, and limited sensitivity. technique to reveal detailed chemistry of these species, yet requires substantial sample 12 preparation and therefore difficult to be deployed as a rapid screening tool 24 . 13 We herein report an approach to rapidly quantify the composite redox state of the 14 Hb/plasma by direct measurement of proton relaxation rates of (predominantly) bulk water 15 using a bench-top sized micro magnetic resonance relaxometry (micro MR) system (Figures 16 1A-B) 25,26 . The non-destructive nature of the micro MR analysis allows oxidative stress to be 17 artificially introduced in ex vivo environment using different biochemical compounds (e.g., 18 nitrite, peroxide) in a controlled manner ( Figure 1C). This allows functional assessment of 19 the oxidative susceptibility, tolerance and capacity of a given sample. This yields significantly 20 richer and clinically useful information about the oxidative stress levels of the blood within 21 an individual, as compared to routine biomarkers. To enumerate the various redox states of 22 the Hb (e.g., Fe 2+ , Fe 3+ , Fe 4+ , and globin-associated radical Fe 4+ ) and the plasma, two

26
MR relaxometry is a technique to measure relaxation rate, which can be obtained by 27 acquiring spin-echoes of (predominantly) water content of the cells/tissues using 28 conventional nuclear magnetic resonance (NMR) spectroscopy and magnetic resonance 29 imaging (MRI) system. Recent advances in NMR system miniaturization have raised the 1 prospect of applying these techniques in point-of-care diagnostic setting. These applications 2 include immuno-magnetic labeling based detection (e.g., tumor cells 27-29 , tuberculosis 30 and 3 magneto-DNA detection of bacteria 31 ) and label-free micro MR detection of various diseases 4 (e.g., oxygenation 32 /oxidation 26 level of the blood and malaria screening 25,33 ). 5 We applied micro MR analysis on whole blood samples to stratify diabetic subjects into 6 subgroups based on their oxidative status levels in association with their glycemic control 7 ( Figure 1F). Assessment of oxidative status by measuring the redox state of whole blood was 8 shown to be highly time-and patient specific, revealing information that is potentially critical 9 for clinical diagnostic, monitoring and prognostic purposes. process of oxidative (and nitrosative) stress involve the process of electron transfer, which 18 lead to the eventual formation of oxidized products. The oxidized product is much more 19 stable and measurable using proton NMR relaxometry.

20
Here, we chemically induced ( Figure 1C and Methods Online) and characterized various 21 redox states of the red blood cell and represented them using T1-T2 magnetic resonance 22 relaxation state diagram ( Figure 1D). Each Hb species has specific oxidation states (e.g., Fe 2+ , 23 Fe 3+ , Fe 4+ , globin-associated radical of Fe 4+ or its' corresponding complexes) that are bound 24 to specific neighboring proteins, and dissipate energy via unique relaxations mechanism in 25 both the longitudinal (T1) and transverse (T2) relaxation frames. The T2 and T1 relaxation 26 times measurement were performed using the standard Carr-Purcell-Meiboom-Gill (CPMG) 27 pulse sequence 36 and inversion-recovery observed by CPMG, respectively. The pairing of 28 both relaxation times forms a specific T1─T2 relaxometry coordinate, which is unique to each 1 redox state.

2
These relaxation times reflect predominantly the bulk water, which came into contact with 3 macromolecular proton (e.g., hemoglobin, albumin) 37 . Water is an attractive natural 4 molecular network probing system as it forms hydrogen bonds with practically all others 5 macromolecule (e.g., protein, metabolite) that are present in human circulation 37,38 . 6 Therefore, a subtle change of the molecular environment can induce a measurable change in 7 the proton relaxation rate. Among the early works on relaxation rate dependent on the blood 8 oxygenation level were carried out by Thulborn et. al.,39 Gomori et. al.,40 and eventually used 9 to measure brain activity known as functional MRI 41 .

10
Oxyhemoglobin (oxy-Hb) which has the lowest reduced ferrous (Fe 2+ ) state is the 11 predominant Hb species in circulation. The oxy-Hb can be provisionally assigned to the 12 center of the state diagram, which has four quadrants (i.e., Q1, Q2, Q3 and Q4). Due to the 13 semi-solid structure of RBC and oxidation process which reduces the proton relaxation time, 14 the redox pathways of RBC mapped out predominantly in Q1.

15
Electrons in the d sub-orbital of iron hemoglobin can exist in various paired or unpaired 16 conditions, rendering them into two possible magnetic states, i.e., diamagnetic and 17 paramagnetic states, respectively. Hb with at least one unpaired electron, i.e. deoxygenated 18 hemoglobin (deoxy-Hb), methemoglobin (met-Hb), nitrosyl hemoglobin (nitrosyl-Hb), and 19 oxo-ferryl radical exhibit the effect of paramagnetism with much larger bulk magnetic 20 susceptibility than its' diamagnetic counterparts i.e., oxy-Hb, ferryl-Hb, and hemichrome   Figure 1A). Redox 2 titration profile showed a strong dose-dependent curve, where both T1 and T2 relaxation 3 times reduced gradually as progressively higher proportion of RBCs were oxidized and 4 increased the volume paramagnetic susceptibility, when the nitrite concentrations were 5 increased from 50 nM to 10 mM (Figures 2A-B). As the blood sample transformed to a 6 complete paramagnetic state (T2=92.8 ms, T1=190.0 ms) from the initial diamagnetic states 7 (T2=149.0 ms, T1=620.0 ms), the A-ratio dropped from 4.16 to 2.02 (R 2 >0.95, Figure 2C). As 8 the volume paramagnetic susceptibility increased, this causes the T1-T2 trajectory to move 9 downward in Q1 ( Figure 2D).

10
The dose-dependent reaction was lost when excess of nitrite (>10 mM) was introduced. This  The amplitudes of the oscillation decreased as the nitrite concentration was increased from 26 500 µM to 4 mM ( Figure 2H). At much higher nitrite concentration (>10 mM), the reaction 27 curve decayed rapidly in an exponential manner with an increasingly dampened oscillation.
Interestingly, the corresponding kinetic profiles followed an identical path over time in the 1 T1-T2 trajectories as the nitrite concentration was increased ( Figure 2G). The oxidation 2 process drove all the trajectories toward a common coordinate (T2= 92.8 ms, T1= 190.0 ms), 3 where all the RBCs were converted fully into met-Hb. For low nitrite concentration (e.g., 500 4 M) however, the T1-T2 trajectory circulated around the origin and did not reach the 5 eventual met-Hb coordinates.

Mellitus.
A cross sectional study was carried out to stratify DM subjects based on their 8 oxidative status. DM subjects (n=185) who had HbA1c measured in the outpatient clinic as 9 part of their clinical care (random blood sample) were included in this study. These subjects 10 had HbA1c ranging from 4% to 16% and the subjects were classified into good glycaemic 11 control (<7.0% HbA1c) and poor glycaemic control (>8.0% HbA1c) subgroups 2 . Healthy young 12 male subjects (n=32; age range of 21 to 40 years, fasting glucose below 5.6 mmol/L, average 13 HbA1c of 5.16 (±0.32) %, and body mass index below 23.5 kg/m 2 ) were separately recruited 14 as control subjects. The collected whole blood in EDTA-anticoagulated tubes were 15 centrifuged (14 000 g, 5 min) to separate the RBCs and plasma. The micro MR analysis was 16 performed blindly on freshly collected fasting blood samples or otherwise kept at 4˚C within 17 2 hours. Other clinical laboratory tests (e.g., HbA1c) were performed in parallel. The spread of the baseline was large for the good glycaemic control group, which suggests a 20 large between-subject variability of nitrosative susceptibility, despite having similar 21 glycaemic level ( Figure 4E). Using the nitrosative susceptibility (Anitrosative-ratio), which could 22 be derived hypothetically from this study and the traditional index of glycaemic control 23 (HbA1c), DM subjects could be stratified into four distinct quadrants (i.e., Q1 to Q4). This 24 approach singled out a minority group in Q3 (subgroup III), who had good glycaemic control, 25 and yet had (high) nitrosative stress Anitrosative-ratio that was at 75 th percentile that of typical 26 DM subjects with poor glycaemic control and at 95 th percentile of the healthy control 27 subjects.

29
Baseline Study: Glycation and Glycoxidation of Plasma. Increased blood glucose 1 promotes non-enzymatic glycation of plasma proteins, which include the albumin, alpha-2 crystalline, collagen, and low-density lipoprotein. A large proportion of total serum protein 3 is attributable to serum albumin 44,45 . Glycation and oxidative damage cause protein 4 modification, which affects the protein functionality 46 . The micro MR analyses were 5 performed at room temperature (26°C). Each T1-T2 coordinate represents the composite 6 redox properties of one subject's plasma ( Figure 5A). The baseline readings of the DM 7 subjects have much shorter T1 and T2 relaxation times, and it was well separated from the 8 healthy non DM subjects (blue). Notably, DM subjects with poor glycemic control, in 9 particularly DM subjects of >10% HbA1c subgroup (mean A-ratio of 2.52), seen a strong 10 departure from the healthy controls (mean A-ratio of 2.13) ( Figure 5B).

11
The marked reduction in relaxation states was attributed to an increase in glycation and 12 glycoxidation of the serum albumin, known as glucose toxicity. As a result of increased  The micro MR analyses were performed before (black squares) and after (red circles) the 4 mixing ( Figures 6A-C).

5
The results of this stress test revealed a large spread of T1-T2 coordinates for DM subjects 6 ( Figures 6A-B), indicating marked variation in their peroxidative susceptibility as 7 compared to healthy controls ( Figure 6C). Lower anti-oxidant capacity (or increase in 8 peroxidative stress susceptibility) of plasma is indicated by reduction in T1 and T2 relaxation 9 coordinates (red circles). As more oxidized plasma was formed, the T1 relaxation time 10 reduced much faster than T2 relaxation time, and hence the reduction in Aperoxidative-ratio 11 ( Figure 6D), which was in agreement with in vitro validation (Supplementary Figure 8c).

12
DM subjects had much higher plasma peroxidative susceptibility as compared to non-DM 13 counterparts ( Figure 6E). The normalized plasma peroxidative stress susceptibility can be 14 defined by the difference between the Abaseline-ratio and the Aperoxidative-ratio (Figure 6e). Note 15 that the plasma baseline (black) measurements of this cohort having similar positive 16 correlation with glycaemic levels ( Figure 6D) were in agreement with the previous cohort 17 measured independently 49 ( Figure 5). Exposure to peroxyl compound leads to an increased 18 formation of disulfide bonds in albumin and human non-mercaptalbumin, which was also 19 observed in several others pathological state 50,51 . The proposed peroxidative susceptibility 20 measurement that is independent of HbA1c can be used to stratify the DM subjects into 21 subgroups, which provide insight into the oxidative status (susceptibility and damage) in 22 personalized manner ( Figure 6E).

1
We have developed a highly sensitive approach to accurately detect and quantify the redox 2 (and hence oxidative/nitrosative) state and the subtle molecular motion changes of blood 3 samples, inferred based on the relaxation measurement. This is the first demonstration of 4 the unique magnetic resonance relaxation properties of the various hemoglobin states, 5 which were mapped out using the proposed magnetic state diagram. The measurement of 6 redox properties in plasma/erythrocytes can provide a useful parameter for functional 7 phenotyping of many biological pathways to better understand disease pathophysiology.

8
This technology has vast potential to be applied for clinical disease diagnosis, prognosis and 9 monitoring, given that the specificity of the oxidative stress in association with the disease 10 state can be further improved in near future.

11
The platform presented here has several innovative features and is readily adaptable for laborious technique such as gas-or liquid-chromatography mass spectrometry has to be 1 employed, limiting its' utility as diagnostic tools.

2
Further clinical validation is needed to compare current proposed biomarkers with 3 isoprostanes, and a combined assessment may yield even richer information. A long-term 4 follow-up and large-scale prospective study is currently underway to evaluate the diagnostic 5 performance of this technique. This accurate and rapid technique for quantification of 6 oxidative stress may be included in future risk stratification models where subjects with 7 single or multiple complications can be streamlined based on their oxidative index. This 8 work opens up new opportunities for molecular phenotyping of oxidative stress in a rapidly 9 and systematic manner for various chronic diseases (e.g., cancer) and a range of hematology 10 applications (e.g., sepsis), including the acquired and congenital diseases such as enzymatic 11 deficiency, Hb synthesis defects (e.g., Thalassemia), and Hb molecular defects (e.g., sickle 12 cells anemia, unstable Hb).  The authors declare no competing financial interests. One technology disclosure related to 5 this technology was filed.  Aldrich, was mixed into 180 µL of the prepared blood. Sodium salicylate treated RBCs. 20 µL 1 of the desired concentration (as described in the Text) of sodium salicylic were then mixed 2 into 180 µL of prepared blood. Preparation of Oxo-ferryl Hb. Oxo-ferryl Hb was prepared in 3 two steps. The RBCs were first treated with sodium nitrite (similar to the protocol described 4 above) to convert the RBCs into met-Hb. Hydrogen peroxide were then added into the met-5 Hb using the same protocol as described above. Preparation of Nitrosyl-Hb. The nitrosyl Hb 6 was prepared in two steps. The RBCs were first converted into deoxygenated Hb (similar to 7 the protocol described below) and treated with sodium nitrite using the same protocol as

FIGURE LEGENDS
1 Figure 1: Functional sub-Phenotyping of Oxidative Stress with micro MR analysis approach. 2 (a) Schematic illustration of the micro MR assays performed in this work. Once the patient's 3 blood is collected via venipuncture, necessary biochemical assay is performed in blood 4 aliquots. Chemical reagent (e.g., nitrite, peroxide) is mixed with the fresh blood and 5 incubated for an interval of 10 min (unless mentioned otherwise) in selected concentration.

6
Micro capillary tubes were then used to sample the biological samples i.e., RBCs/plasma. 7 Standard centrifugal force (3000 g, 1 minute) was used to separate and concentrate the 8 packed RBCs from the buffer to avoid possible hematocrit variation in patients. The capillary 9 tubes were then slotted into the rf-probe for micro MR analysis and the read-out completes 10 in less than 5 minutes. Proton NMR of predominantly the bulk water of red blood cells (and 11 plasma) were adjusted to resonance frequency of 21.57 MHz. The portable micro MR system 12 developed in this work consists of a benchtop console, detection circuit coil mounted on a 13 micro stage and a palm-sized 0.5 T permanent magnet, and a temperature controller to 14 stabilize the magnetic field within the chamber and biological sample under measurement.

15
(b) The rf pulse sequences used were standard CPMG pulse sequence and standard inversion 16 recovery experiment (with CPMG detection) for the T2 relaxations, and T1 relaxations 17 measurements, respectively. In order to obtain high signal-to-noise ratio under relatively 18 inhomogeneous magnetic environment, an array of echoes (a few thousands) within a very 19 short echo interval (in the order of s) were used to acquire spin-echoes from less than 4 L 20 sample volume of packed RBCs or plasma.  with poor glycaemic controls (n=62) and good glycaemic control (n=50) subgroup as 26 compared to healthy non DM subjects (n=20). The subjects with poor glycaemic controls 27 were further sub-divided into >8% HbA1c (n=47) and >10% HbA1c (n=15) subgroups. The statistical significance was calculated using the Student's T-Test (two-tailed, unequal 1 variance). coordinates of RBCs taken before (black) and after (red) nitrite treatment for subjects with 5 (a) poor glycaemic control (n=39), and (b) good glycaemic control (n=28). (c) Its' 6 corresponding distribution based on A-ratio index. The statistical significance was 7 calculated using the Student's T-Test (two-tailed, unequal variance). (d) The diagnostic 8 accuracy as calculated using ROC curve for RBCs taken before (black) and after (red) the 9 stress test. The probability diagnostic accuracy is quantified as Area Under the Curve (AUC).

10
(e) A quadrant chart of diabetic subjects stratified into subgroups based on their oxidative 11 status (nitrosative stress) in association with their glycemic levels (e.g., HbA1c) as compared 12 with healthy non DM subjects (n=23). The proposed method segregated effectively the 13 subgroup III subjects (good glycemic control and yet high nitrosative stress) from the rest of 14 the cohorts. Note that the Y-axis (nitrosative stress) were inversely represented as compared 15 to quadrants shown in Figure 1F. DM subjects (blue, n=24), subjects with good glycaemic control (green, n=55) and subjects 20 with poor glycaemic control (red, n=39). (b) The corresponding A-ratio against the subjects 21 with poor glycaemic control (n=39) and good glycaemic control (n=55) subgroups, as 22 compared to healthy non DM subjects (n=24). The subjects with poor glycaemic controls 23 were further subdivided into >8% HbA1c (n=14) and >10% HbA1c (n=25) subgroups. The 24 statistical significance was calculated using the Student's T-Test (two-tailed, unequal 25 variance). (c) The diagnostic accuracy of RBCs (gray) and plasma (red) taken from subjects 26 with good glycaemic control with respect to healthy non DM subjects. The number of 27 subjects (n) were indicated on the parentheses (non-DM, good glycaemic control). Abaseline -Astress ) in association with their glycaemic levels (e.g., HbA1c) as compared with 7 healthy non DM subjects. Note that the Y-axis (peroxidative stress) were inversely 8 represented as compared to Y-axis (nitrosative stress) in quadrant shown in Figure 4E.