Salivary lactate and 8-isoprostaglandin F2α as potential non-invasive biomarkers for monitoring heart failure: a pilot study

Heart failure (HF) is a cardiovascular disease affecting about 26 million people worldwide costing about $100 billons per year. HF activates several compensatory mechanisms and neurohormonal systems, so we hypothesized that the concomitant monitoring of a panel of potential biomarkers related to such conditions might help predicting HF evolution. Saliva analysis by point-of-care devices is expected to become an innovative and powerful monitoring approach since the chemical composition of saliva mirrors that of blood. The aims of this study were (i) to develop an innovative procedure combining MEPS with UHPLC-MS/MS for the simultaneous determination of 8-isoprostaglandin F2α and cortisol in saliva and (ii) to monitor lactate, uric acid, TNF-α, cortisol, α-amylase and 8-isoprostaglandin F2α concentrations in stimulated saliva samples collected from 44 HF patients during their hospitalisation due to acute HF. Limit of detection of 10 pg/mL, satisfactory recovery (95–110%), and good intra- and inter-day precisions (RSD ≤ 10%) were obtained for 8-isoprostaglandin F2α and cortisol. Salivary lactate and 8-isoprostaglandin F2α were strongly correlated with NT-proBNP. Most patients (about 70%) showed a significant decrease (a factor of 3 at least) of both lactate and 8-isoprostaglandin F2α levels at discharge, suggesting a relationship between salivary levels and improved clinical conditions during hospitalization.

In the acute phase of HF, the body tries to maintain an adequate cardiac function by activating a large number of compensatory mechanisms including neurohormonal mechanisms 10 . Among these, hypothalamic-pituitary-adrenal axis, sympathetic and the renin-angiotensin-aldosterone systems are usually activated during HF to maintain circulatory homeostasis and blood pressure 11,12 . Cortisol and α-amylase are well-known chemicals related to these conditions and thus might be useful indicators for monitoring HF evolution 13 . Earlier studies have shown that HF is also associated with hyperlactatemia 14 , oxidative stress 15 , hyperuricemia 16 and inflammation 17 , exacerbating the pathophysiology of the HF. Hyperlactatemia can be viewed as part of the stress response that includes an increased metabolic rate, the activation of the sympathetic nervous system, an accelerated glycolysis and a modified bioenergetic supply 14 . Blood lactate levels increase during ischemia and after ischemia-reperfusion injury due to an altered anaerobic metabolism and an inadequate tissue perfusion 18,19 . Moreover, several studies have highlighted that oxidative stress (OS), defined as an imbalance between antioxidant defenses and production of reactive oxygen species (ROS), is enhanced in HF and may cause cellular dysfunction, DNA damage, protein and lipid peroxidation [20][21][22][23][24] . Isoprostanes are compounds produced in vivo from the oxidative peroxidation of arachidonic acid and thus provides an accurate assessment of OS both in vitro and in vivo 25 . In the setting of HF, hyperuricemia is often associated with reduced exercise capacity, inflammation markers, endothelial dysfunction, oxidative stress and diastolic dysfunction 26 . The increased blood levels of uric acid (UA) depends of both enhanced production resulting from OS and to a decreased excretion due to renal failure 27 . Tumour necrosis factor alpha (TNF-α) is one of the cytokines involved in the pathogenesis of HF 28 , leading to cardiomyocyte, hypertrophy, fibrosis and negative inotropic effects 28,29 .
In the last decades, the unobtrusive monitoring of health conditions and drug therapies by the analysis of fluids that can be collected in a non-invasive way (e.g. breath, saliva, sweat, and wound exudate) has attracted much attention [30][31][32][33][34] . Saliva, whose chemical composition mirrors that of blood, can be collected in a non-invasive way by easy sampling procedures requiring some cautions [35][36][37] . Compared to blood and its derivatives, it is safer to handle and transport 38 , and its simpler chemical composition makes it particularly suitable for human biomonitoring in combination with POC devices [39][40][41] .
The aims of this study were i) to develop an innovative procedure combining micro-extraction by packed sorbent (MEPS) with ultra-high-performance liquid chromatography coupled to electrospray ionization triple-quadrupole mass spectrometry (UHPLC-ESI-MS/MS) for the simultaneous determination of 8-isoprostaglandin F 2α (8-isoPGF 2α ) and cortisol in saliva and ii) to monitor lactate, uric acid, TNF-α, cortisol, α-amylase and 8-isoPGF 2α concentrations in stimulated saliva samples collected from 44 HF patients during their hospital stay due to acute HF. We hypothesize that changes in the chemical composition of patients' saliva during recovery of baseline conditions due to therapies are specular to changes occurring at home when patients drift towards acute conditions. Reliable biomarkers predicting HF flares are needed to develop sensing devices usable at home, from the patient themselves or caregivers, that may provide an early advice of the building up of acute conditions. This paper aims at identifying the target molecules and providing the basic knowledge needed for the development of this kind of devices.

Results
Development of MEPS procedure for the determination of 8-isoprostaglandin F 2α and cortisol in saliva. The optimization of the MEPS procedure maximized the extraction efficiency of 8-isoPGF 2α and cortisol in saliva samples. We investigated dilution ratio of the sample, sampling cycles, composition of the washing solution and volume of the elution solvent as possible parameters affecting MEPS performance.
Proteins, mucins and other interferences in the matrix may cause a premature deterioration of the MEPS sorbent performance and/or a cartridge occlusion 42 . To prevent these issues, saliva sample can be diluted with water and then filtered using a syringe filter prior to MEPS extraction. The influence of the sample to water dilution ratio, i.e. 1:2, 1:5 and 1:8 v/v, on the analyte peak area was investigated. For this purpose, nine aliquots (500 µL each) of pooled saliva samples, spiked with 8-isoPGF 2α (50 pg/mL) and cortisol (500 pg/mL), were diluted with LC-MS water to achieve the desired dilution ratio and then filtered at 0.2 µm before starting the MEPS procedure. The entire volume of each sample, namely 1500, 3000 and 4500 µL, was loaded up and discharged 3, 6 and 9 times, respectively. The target analytes were eluted with 50 µL of methanol, so that the sample aliquot volume (500 µL) to solvent elution volume ratio was fixed at 10. The filtration step and the sample to water dilution ratio did not significantly (p > 0.05) affect the analytes peak area, but the 1:2 v/v dilution ratio was soon excluded because of sorbent occlusion just after ten MEPS extractions. Thus, we decided to use 1:5 sample to water dilution ratio and filter the resulting aqueous solution as the best compromise in terms of duration of the MEPS procedure and cartridge lifetime.
Methanol was added to the aqueous washing solution used during the MEPS procedure 43 , and no significant analyte loss was observed when the washing step was carried out with a 95:5 v/v water:methanol mixture.
A quantitative elution of the analyte was achieved by a single withdraw-eject cycle of 50 µL of methanol, since the content of 8-isoPGF 2α and cortisol in the methanol used in the second withdraw-eject cycle was below the limit of detection.
Analytical figures of merit of MEPS procedure. Five working standard solutions were analysed in triplicate to obtain calibration curves in the range 50-1250 pg/mL and 500-12500 pg/mL for 8-isoPGF 2α and cortisol, respectively. A known amount (20 ng/mL) of internal standard (IS), i.e. 8-iso prostaglandin F 2α -d4 (8-isoPGF 2α -d4), was added to these solutions before to MEPS-UHPLC-ESI-MS/MS analysis. For 8-isoPGF 2α , calibration curves were constructed by plotting the analyte to IS peak area ratios (A/A IS ) versus the corresponding concentration ratios (C/C IS ), whereas for cortisol the analyte peak areas were plotted versus the corresponding concentrations. The best calibration plots (a = m × b) were evaluated using the Deming regression and the slope (average ± standard deviation) for 8-isoPGF 2α and cortisol resulted 1.2 ± 0.1 and (1.3 ± 0.1) × 10 4 , respectively.
The presence of matrix effect was ruled out by comparing the slopes of the calibration curves (i.e. standard solutions and spiked samples) at a confidence level of 95%.
Limit of detection (LOD) was estimated as three times the standard deviation (s.d.) of the "low level spiked blank" sample, i.e. 50 pg/mL for both 8-isoPGF 2α and cortisol. Each saliva sample was prepared five times and then analysed with the MEPS-UHPLC-ESI-MS/MS approach. A LOD equal to 10 pg/mL was obtained for both 8-isoPGF 2α and cortisol.
Standard spiked saliva samples were analysed three times within the same day and on three consecutive days to evaluate the intra-and inter-day recovery, respectively. For both analytes, the recovery was almost quantitative (95-110%) with intra-and inter-day precisions below 10%.
Six aliquots of pooled saliva samples were prepared by spiked a known amount of 8-isoPGF 2α (50 pg/mL) and cortisol (500 pg/mL). Three samples (S1) were analysed according to the procedure described in 4.5.2 whereas the other aliquots were absorbed in three Salivette polyester swabs to evaluate the analyte recovery from the collection device. Sample (S2) was recovered by centrifuging the device at 7000 rpm and 4 °C for 5 min and then analysed with the same method. The recovery was calculated as S2 to S1 8-isoPGF 2α and cortisol concentration ratio. For both analytes the mean recovery and the corresponding relative standard deviation resulted 95% and 10%, respectively.
These results confirmed that MEPS-UHPLC-ESI-MS/MS method allows a reliable quantification of 8-isoPGF 2α and cortisol in saliva samples.

Stability of 8-isoprostaglandin F 2α solutions.
Cortisol is a very stable analyte both in water solution either in saliva samples 44 . The stability of standard working solutions and saliva samples was evaluated at −80 °C, whereas the stability of extracted samples at 4 °C. The initial concentrations of the analyte (t = 0 h) were used as the reference values and stability was evaluated by analysing the variance (ANOVA) at a confidence level of 95%.
Standard working solutions of 8-isoPGF 2α were stable for the entire observational period (one month), whereas extracted samples in methanol were stable throughout the duration of a typical sequence of chromatographic analyses (storage in the autosampler over 24 h at 4 °C). The 8-isoPGF 2α concentration in saliva was stable at a confidence level of 95%, both after one week storage at 4 °C and six months at −80 °C.
Monitoring heart failure during hospitalization: a pilot study. Thirteen (30%) out of the 44 patients enrolled had been diagnosed for HF at least 2 years before their inclusion in the pilot study, and 9 of them had one HF episode within 6 months from their first hospital admission (t 0 ). Over 90% of the HF patients showed comorbidities such as diabetes mellitus (n = 17), hypertension (n = 29), dyslipidaemia (n = 22), chronic obstructive pulmonary disease (n = 9), chronic kidney disease (n = 14), chronic liver disease (n = 2), coronary artery disease (n = 18), atrial fibrillation (n = 15), and oncological diseases (n = 6). Table 1 reports the demographic and clinical data of HF patients, both at the hospital admission and discharge.
The Mann-Whitney test revealed statistically significant gender differences only for weight and body mass index (p < 0.05).
Compared to the hospital admission (acute phase) (Table 1) A stepwise regression analysis (Supplementary Table S1) was applied to identify the variables more closely related to the blood level of NT-proBNT, a sensitive marker of HF.
The model had the coefficient of determination (R 2 ) of 0.7 and p-value < 0.001. The stepwise regression allowed the variables to be reduced to five. Since the literature already describes left ventricular ejection fraction, oxygen saturation and estimated glomerular filtration rate as risk or detrimental factors for HF [45][46][47] , we therefore focused on lactate and 8-isoPGF 2α as new potential predictors for monitoring HF patients.
When patients were grouped in one of four categories of the NYHA functional class (Fig. 1), only NT-proBNP content in blood and salivary lactate were significantly different among the four classes (confidence level of 95%), whereas salivary 8-isoPGF 2α was not correlated to the NYHA classes.
Salivary levels of lactate and 8-isoPGF 2α levels at the admission and at the discharge were not significantly different (confidence level of 95%) when patients were grouped according to the comorbidities. As an example, median (lower and upper quartiles) salivary lactate concentrations calculated for diabetic (n = 17) and non-diabetic subgroup (n = 27) resulted 1610 μM (700-3700 μM) and 2050 μM (1250-4105 μM), respectively. In the same way, these patients showed similar salivary levels of 8-isoPGF 2α [35 pg/mL (25-45 pg/mL) vs. 40 pg/ mL (25-60 pg/mL)]. Figure 2 shows for each patient the changes over time of salivary concentrations of lactate and 8-isoPGF 2α , normalized with respect to the admission level.
The majority of HF patients (about 70%) had a remarkable three-fold decrease of lactate and 8-isoPGF 2α concentration in saliva over time (Fig. 2). Eighteen patients, belonging to NYHA class III and IV and characterized www.nature.com/scientificreports www.nature.com/scientificreports/ by low LVEF (20-30%) and high blood NT-proBNP (1200 pg/mL or higher) values at admission, showed a very marked decrease of lactate (5 to 10-fold). Only ten patients showed a decrease of 8-isoPGF 2α at least of a factor of three, and five patients had stable (within ±30%) lactate and 8-isoPGF 2α levels (about 500 µM and 30 pg/mL, respectively) during their stay at the hospital. Moreover, two more patients had similar results in terms of lactate levels, although they showed high 8-isoPGF 2α concentrations, i.e. from 50 up to 100 pg/mL. Interestingly, the decrease of both lactate and 8-isoPGF 2α levels was not affected by the presence of comorbidities such as diabetes mellitus, dyslipidaemia, and kidney disease.

Characteristics
Hospital admission (t 0 ) Hospital discharge (t d ) p-value

Discussion
Isoprostanes represent a novel class of compounds produced in vivo from the oxidative peroxidation of arachidonic acid, whereas cortisol is a well know marker of chronic stress related to the hypothalamic-pituitary-adrenal axis activity. In particular, 8-isoPGF 2α is considered a representative example of these compounds and a reliable biomarker of free radical mediated lipid peroxidation in vivo 48 . Such biomarkers are commonly determined in biological fluids by enzyme-linked immunosorbent assays (ELISA) 44,49 , although commercially available kits show drawbacks related to antibody cross-reactivity which lead to partially inaccurate results 50 . A solid phase extraction (SPE) procedure coupled to well-established LC-MS or GC-MS methods may overcome these issues 51,52 , but the need of larger volumes of sample represents a clear disadvantage, especially in the case of saliva analysis. Micro-extraction techniques such as MEPS overcome this limitation and improve the analytical performance of the MS methods. Our results showed that MEPS is a good alternative to classical clean-up procedures, with a low consumption of organic solvents (<100 µL/measurement), a reusable sorbent (over 100 extractions with the same cartridge) and reduced the cost of analysis. In addition, the MEPS approach permits to automate the extraction procedure, thus enhancing the reproducibility of the analytical method. Compared to other methods 50,53,54 , our procedure showed a satisfactory recovery (95-110%) and an adequate detection limit (10 pg/mL), which allowed a reliable determination of 8-isoPGF 2α and cortisol in real samples at the low concentration levels expected in saliva samples. www.nature.com/scientificreports www.nature.com/scientificreports/ Validated procedures were used to monitor over time a panel of potential biomarkers such as lactate, uric acid, TNF-α, cortisol, α-amylase and 8-isoPGF 2α in saliva samples collected from 44 HF patients during their hospitalization. We used stepwise regression to fit a linear model on thirty-three variables to predict blood NT-proBNP level. In this way, the variables of each saliva sampling were assumed as an independent input vector, thus leading to a 193-by-33 input data set and an output variable. Our model only consisted of those variables with p-value < 0.05 whereas variables with p-value > 0.1 were removed. We forced the linear model to include only a linear term for each variable and a constant term. The stepwise regression allowed to identify lactate and 8-isoPGF 2α as potential predictors of NT-proBNP, which represent one of the most promising HF biomarker 55,56 . The results are in line with the clinical conditions observed in the acute phase and disease severity, most patients being (about 70%) in NYHA class III (n = 18) and IV (n = 14) at hospital admission. The acute phase of HF is characterized by a reduced ability of the heart to pump blood that leads to a decreased perfusion of organs www.nature.com/scientificreports www.nature.com/scientificreports/ and tissues 57 . In the attempt to restore an adequate perfusion, several neurohormonal mechanisms are activated. Natriuretic peptides are released from the myocardium, increasing the excretion of water and electrolytes and inducing vasodilation 58 . Although beneficial in the early stages of HF, these compensatory mechanisms often lead to a vicious cycle worsening HF 57 . The reduced tissue perfusion increases anaerobic glycolysis, i.e. the conversion of pyruvate to lactate 14 . Plasma lactate levels result from the balance between production (e.g. in skeletal muscle) and clearance, mainly due to liver metabolism (about 70%) and kidney removal 59 . The reduced cardiac output following HF limits the ability of liver and kidney to remove lactate and contributes to increase lactate plasma levels. An additional contribution comes from oxidative stress, which induces subtle changes in intracellular pathways and redox signalling, and causes cellular dysfunction and damage 15 . In addition, the presence of ROS activates a broad variety of pathways with the induction of apoptosis 60 .
The decrease of lactate and 8-isoPGF 2α concentrations during hospitalization are likely related to the improved clinical conditions due to treatment. In fact, all patients received diuretics and β-blockers in order to stabilize their clinical conditions and minimize damages to target organs. The improvement of clinical conditions was confirmed by the concomitant decrease of blood NT-proBNP level. Patients suffering from both HF and other comorbidities showed the same decreasing trend of salivary lactate and 8-isoPGF 2α and, then, they were indistinguishable from the other patients. As a consequence, it is possible to hypothesize that salivary levels of both lactate and 8-isoPGF 2α could be mainly related to cardiac decompensation rather than other conditions. These results were further confirmed when multi ROC curve was built to assess the capability to distinguish admission (acute phase) and discharge point. In fact, a combination of ROC curve analyses of lactate and 8-isoPGF 2α showed better sensitivity and specificity, confirming that the concomitant determination of such compounds has a potential value in HF monitoring.
These observations are consistent with the hypothesis that lactate levels and 8-isoPGF 2α could be useful in guiding the treatment of acute HF. In addition, assuming that the high lactate and 8-isoPGF 2α salivary levels observed in acute conditions build up progressively over time, their monitoring would allow to prevent acute conditions. The avoidance of acute conditions and hospital readmissions would improve the quality of life and possibly prolong survival.
We conclude that this pilot study identified lactate and 8-isoPGF 2α as possible non-invasive biomarkers for HF monitoring, and that saliva analysis has potential applications in clinical practice.
All the liquid solutions and saliva samples were stored in sterile polypropylene containers from Eppendorf (Milan, Italy). Phenex ™ -RC syringe filters (0.2 μm regenerate cellulose, 4 mm of diameter) were from Phenomenex (California, USA).
Salivette roll-shaped polyester swabs were purchased from Sarstedt (Nümbrecht, Germany). A Macherey Nagel pehanon narrow range (6.0 < pH < 8.1) pH paper strips (Düren, Germany) with a resolution of 0.3 pH units was used to estimate the salivary pH.
Preparation of standard solutions and quality control samples. Stock solutions of triethanolamine (265 mM), tetra-n-butylammonium bromide (90 mM), 9-chloromethyl-anthracene (10 mM) were prepared by dissolving the required amounts of the pure compounds in acetonitrile. A stock solution of uric acid (970 μg/ mL) was prepared by dissolving a weighed amount of the pure compound in ultrapure water and adding NaOH (10 M). On the same way, a stock solution of cortisol (1000 μg/mL) was prepared by dissolving the appropriate amount of the pure compound in acetonitrile. A stock solution of 8-isoPGF 2α (1000 µg/mL) was prepared by dissolving 1 mg of pure compound in 1 mL of a mixture of acetonitrile and methanol (50:50 v/v) and stored at −20 °C in amber vials for 12 months. A stock solution of 8-isoPGF 2α -d4 in methyl acetate (100 µg/mL) was stored at −20 °C for 12 months.
Working solutions of lactate, uric acid, and cortisol were prepared gravimetrically by an appropriate dilution of the stock solutions with LC-MS water. All stock and working solutions were stored in amber vials at 4 °C and prepared monthly. Working solutions, containing 8-isoPGF 2α in the range 0.05-1.25 ng/mL, were prepared through sequential dilutions with LC-MS water from an intermediate stock solution of the analyte, monthly prepared, at 5 µg/mL in a 50:50 v/v acetonitrile:methanol mixture. A working standard mixture of 8-isoPGF 2α -d4 (20 ng/mL) was prepared through sequential dilutions with LC-MS water from the stock solution. All working solutions were prepared weekly and stored at 4 °C.

Equipment.
A VELP Scientifica ZX4 Advanced Vortex Mixer (Usmate, Italy) and an Eppendorf Centrifuge 5804 R equipped with an A-4-44 swinging bucket rotor (Milan, Italy) were used to homogenize and centrifuge the samples, respectively. A Julabo SW22 thermostatic water bath (Milan, Italy) was used to control the temperature (resolution of 0.1 °C). A Lambda 25 Perkin Elmer UV-Visible spectrophotometer (Milan, Italy), equipped with two quartz cuvettes with an optical path of 1 cm, was used to determine α-amylase.
The amount of the collected saliva sample was calculated according to weight differences before and after sampling by using a Radwag AS220/X balance (Milan, Italy).
Study population. The  Sample collection and analysis. Supplementary Fig. S1 shows a typical five-points sampling timing for the monitor of HF patients during hospital stay. Stimulated saliva samples were collected by asking the patients to roll a Salivette polyester swab in the mouth for 2 min at the time of hospital admission (t 0 ), after every 2 days, and up to at the discharge (t d ). Blood samples were drawn at the admission and at the discharge to perform routine clinical analysis, as suggested by the ESC guidelines 1 . Circadian cycle effects and interference from fasting were minimized by collecting samples always in the morning (10.00 to 12.00 AM), 2 h after breakfast. The number of the hospitalizations days was not the same for all patients, on average, each stay lasted approximately 5 days.
After sample collection, salivary pH was measured using a narrow range pH paper. The salivary flow rate (millilitres per minute) was calculated from the ratio between the weight difference (g) of the sampling device before and after sampling and the collection time (min), considering the density of sample equal of 1 g/mL 63 . The sample was recovered by centrifuging the swab at 7000 rpm and 4 °C for 5 min. Saliva was then stored in a polypropylene tube at −80 °C until use, with a maximum storage time of 3 month. Prior to analysis, samples were thawed at room temperature (25 ± 2 °C) and then vortex-mixed for ten seconds to homogenize the sample. All the details of the analytical methods are reported in the supplementary information.

Statistical analysis.
The distribution of variables was tested for normality by the Shapiro-Wilk test.
Continuous variables with a normal distribution were reported as mean ± standard deviation, whereas variables with skewed distribution were described by median with lower (25 th percentile) and upper (75 th percentile) quartiles. The potential differences between groups were evaluated using the t-test and Mann-Whitney test. The statistical relationship between demographic and clinical variables were examined by Pearson's correlation. A two-tailed p-value of <0.05 was considered statistically significant.