Diurnal variation in expired breath volatiles in malaria-infected and healthy volunteers

We previously showed that thioether levels in the exhaled breath volatiles of volunteers undergoing controlled human malaria infection (CHMI) with P. falciparum increase as infection progresses. In this study, we show that thioethers have diurnal cyclical increasing patterns and their levels are significantly higher in P. falciparum CHMI volunteers compared to those of healthy volunteers. The synchronized cycle and elevation of thioethers were not present in P. vivax-infection, therefore it is likely that the thioethers are associated with unique factors in the pathology of P. falciparum. Moreover, we found that time-of-day of breath collection is important to accurately predict (98%) P. falciparum-infection. Critically, this was achieved when the disease was asymptomatic and parasitemia was below the level detectable by microscopy. Although these findings are encouraging, they show limitations because of the limited and logistically difficult diagnostic window and its utility to P. falciparum malaria only. We looked for new biomarkers in the breath of P. vivax CHMI volunteers and found that a set of terpenes increase significantly over the course of the malaria infection. The accuracy of predicting P. vivax using breath terpenes was up to 91%. Moreover, some of the terpenes were also found in the breath of P. falciparum CHMI volunteers (accuracy up to 93.5%). The results suggest that terpenes might represent better biomarkers than thioethers to predict malaria as they were not subject to malaria pathogens diurnal changes.


SUPPLEMENTARY MATERIAL
Section S1. Additional Figures   Supplementary Figure 1: Time course of mean levels of thioethers for P. falciparum CHMI trials (N=7), healthy controls (N=8) and parasitaemia. P. falciparum infection started immediately after the breath sample at time zero. The vertical dotted line represents the time when treatment was administered in the P. falciparum trial. The healthy control data shows the means of three consecutive days of measurements. The diurnal cycle is repeated over the duration of the CHMI trial and the collection time matched as closely as possible to CHMI collections. Figure shows mean(sem). Sem is used as a measure of precision for the estimated mean. Samples with asterisks in blue denote sampling time between 19:00 and 21:00. MTP= X=Days of breath sample collection Day 0=Malaria parasite inoculum ¥ =indicates that two breath samples were collected per volunteer: one sample in the morning (around 7:00) and a second sample 12 hours later ɛ = indicates that four breath samples were collected per volunteer: around 7:00, 11:00, 15:00 and 19:00. ß = indicates day of admission §=number of breath samples missed due to inability of volunteer to provide breath samples Confinement phase=Day 8, 9 and 10 for P. falciparum and Day 10, 11 and 12 for P. vivax

Supplementary
The accuracy of the classification to predict the sample group was considered significant when the accuracy result was above the significance level at 95% confidence interval, i.e. 71.43% (10/14). Classification accuracy rates below the significance level value were considered nonsignificant. Numbers in bold means classification was above statistical significance levels (>71.43%).

Preparation of inocula.
The production of the parasite inoculum has been described previously (2). In brief, laboratory-reared Anopheles stephensi mosquitoes were infected by membrane feeding on a blood meal containing gametocytes with the chloroquine-sensitive P. falciparum clone 3D7, derived from an isolate originally collected from an airport worker in Amsterdam. Ten days after their last blood meal, the mosquitoes were allowed to feed on a healthy male who had no evidence of any blood borne virus infection. A 500 mL unit of blood was taken from the volunteer six hours after he became ill with the development of high fever, 13 days after the mosquito bites. The blood was leucocyte depleted, mixed with the cryopreservation agent Glycerolyte 57, aliquoted into 1-mL cryovials and stored in liquid nitrogen at QIMR.
To prepare inocula for experimental infection studies, aliquots of the P. falciparum cell bank were thawed, washed, and resuspended in injectable saline solution according to a method described elsewhere (3). Parasites were synchronous. It was planned that each injected inoculum would contain ~120,000 erythrocytes, of which ~5,400 were parasite-infected. Each challenge inoculum was dispensed into syringes and stored in sealed plastic bags on ice until administered. The time between thawing and injection was no more than 60 minutes.
For P. vivax, aliquots of the P. vivax cell bank were thawed, washed, and resuspended in injectable saline solution according to a method described elsewhere (4), with the modification that the final red cell pellet was resuspended in normal saline solution rather than Roswell Park Memorial Institute medium. One mL of the resuspended cells was set aside for quantification of parasitaemia by polymerase chain reaction (PCR), and the remainder of the dose dispensed aseptically into a 2-mL syringe and stored on ice until administration. The number of parasite genome equivalents injected into subjects was determined using the quantitative PCR method described below.
Cohorts. Each participant in the P. falciparum and P. vivax was inoculated intravenously on Day 0 with ~1,800 viable Plasmodium falciparum-infected human erythrocytes and ~1,800 viable Plasmodium vivax-infected human erythrocytes. From Day 4 onwards, participants were monitored daily for adverse events and for the unexpected early onset of symptoms, signs or parasitological evidence of malaria.
On the day designated for commencement of treatment, as determined by qPCR results, participants were admitted to the study unit and confined for safety monitoring and antimalarial drug administration, with parasite load and drug levels being monitored. The threshold for commencement of treatment was when PCR quantification was confirmed to be ≥ 1,000 parasites mL -1 . If clinical or parasitological evidence of malaria (the onset of clinical features of malaria) occurred, or if ≥1,000 parasites mL -1 was detected by PCR before the morning of Day 7, allocated treatment was started immediately.
Following treatment with an antimalarial, participants were followed up as inpatients for at least 48 hours, to ensure they tolerated the therapy and to confirm a clinical response then, if clinically well, on an outpatient basis for safety and to monitor for persistence of malaria parasites using PCR. Early intervention with Riamet® was planned if either poor responses or fast responses were seen following initial treatment with experimental drugs, to ensure patient safety.

PCR quantification of Plasmodium parasitaemia.
A consensus Plasmodium species RT-PCR method described elsewhere (5) was modified to make use of TaqMan hydrolysis probe chemistry. The assay amplifies a conserved 199-base pair (bp) target in the multicopy, highly conserved, 18S ribosomal RNA gene. Parasites were quantified from 500 μL of packed red cells. Each sample was tested in duplicate during the study. After completion of the study, all samples were retested in triplicate. When coefficients of variation varied by >20%, samples were re-tested.

S2.2 Healthy volunteer's cohort
The objective of this study was to determine the levels of four specific thioethers in the breath of healthy individuals over a period of three consecutive days. We compared these results with those found in CHMI trials.
The entire study was approved by the CSIRO Health and Medical Research, Human Research Ethics Committee (proposal number: LR 4/2017). Written informed consent was received from participants prior to inclusion in the study.
The primary outcome was to measure levels in counts of allyl methyl sulphide, 1-methylthiopropane, (E)-1-methylthio-1-propene, (Z)-1-methylthio-1-propene and the ratio: (Z/E) 1methylthio-1-propene. Inclusion criteria were: CSIRO staff based at Black Mountain laboratories in Canberra Australia, non-smoking adults aged between 18 and 45 years and must be non-smokers and in good health. Exclusion criteria were: presence of current or suspected serious chronic diseases such as cardiac or autoimmune disease (HIV or other immune deficiencies) and recent acute infectious disease or fever (e.g., sub-lingual temperature ≥ 38.5°C) within the two days prior to the start of breath collection).
There were eight participants. Each participant provided: four breath samples containing at least 1 liter of breath each day at the following times, in the morning before breakfast (around 7:00), 11:00, 15:00 and 19:00. Samples were collected from all participants for three consecutive days. The 7:00 and 19:00 samples were collected by the volunteers in their homes. A total of 96 samples was collected. Breath samples were collected as described in Methods section in the main manuscript.

S2.3 Ambient air collection, breath transfer and analysis
Ambient air sample collection. For P. vivax, every time a breath sample set (N= 8) was collected was collected, we collected three ambient air samples. For healthy control study, we followed the same procedure as per P. vivax.
For P. vivax CHMI trial, 1 litre of ambient air was collected in a 3-L Sample Pro FlexFilm bag (SKC Inc, Pennsylvania), the ambient air was then transferred to a sorbent tube using an electric pump as described below. For the healthy control trials, 1 L of ambient air was collected directly into sorbent tubes using an electric pump (flow: at 200 mL min -1 ).
Transfer of sample into sorbent tubes and storage. Breath and ambient air from the bags were transferred to sorbent tubes using an electric pump. 1 L of the sample at 200 mL min -1 from the bag to the sorbent tube, so all tubes had consistently the volume of sample. The tubes had two layer sorbents comprising 200 mg of Tenax TA and 200 mg of Unicarb (Markes International Limited, UK). After capture of breath volatiles onto sorbent tube they were kept at 4°C for storage and transport.
For the P. falciparum cohort, breath samples were stored between 7-29 days. For the P. vivax cohort, breath samples were stored between 7-26 days and ambient air samples were stored between 23-71 days. For healthy controls, breath samples and ambient air samples were stored between 1-12 days.
Original concentration of the thioethers on the day of collection, were calculated by correcting for decay during storage time using the method described in Section S2.4. A gas chromatograph (7890B Series GC, Agilent Technologies, USA) equipped with a HP-5MS UI GC capillary column (Agilent J&W GC Column) 30 m in length, 0.25mm ID and 0.25 m film thickness was used with the following temperature program. Initial temperature was 35°C, held for 5min, ramped to 250°C at 5°C min -1 and final temperature of 250°C held for 2min. The total run time for the analysis was 50 min. The helium carrier gas flowed at a rate of 1.9 mL min -1 .

Standards run with breath
The QTOF (Agilent Technologies, USA) used an electron ionization source set at a temperature of 230°C. The quadrupole was set to a temperature of 150°C and the collision cell had a nitrogen flow of 1.5 mL min -1 . The emission current was fixed at 35 A and the electron energy at 70eV. The mass range was scanned from 35 to 350 amu at an acquisition rate of 5 spectra sec -1 and a scan time of 200 ms spectrum -1 . Deconvolution was used to identify the compounds while ion extraction was used to calculate the area under peak for each thioether. The spectra was analyzed using Mass Hunter Qualitative analysis, Version B.07.00. In addition, chemical standards were used to confirm the identities of the compounds.  Table S7). Fold change used in this paper was calculated as the ratio between the final peak area and the initial peak area.

S2.4 Thioether decay correction due to storage
Thioethers are relatively unstable compared to many other breath volatiles. Thioethers have low boiling points (88-90°C) and chemically reactive double bonds. CHMI breath samples had to be transported from Brisbane to Canberra (1,100 km), kept at 4°C and stored for a variable number of days prior to analysis. We performed an experiment to determine the stability of thioethers on sorbent tubes when stored at cold temperatures in order to correct estimates of thioether concentrations for decay during storage.
Briefly tubes spiked with known amounts of thioethers were stored at four different temperatures from 6-60°C. Using Arrhenius equation, we established the decay rates for each of the thioethers. The information on the stability of thioethers allowed us to estimate the original concentration of the thioethers at the time of collection by correcting for decay. Full results of the stability studies will be published elsewhere.

S2.5 Mass spectral pre-processing and feature selection for untargeted search of novel biomarkers in P.vivax trial.
Given that the thioethers did not show to have predictive value for P. vivax, we initiated an untargeted search for volatiles associated with P. vivax infection. We took an automated, machine learning approach. In order to identify features of the data that can distinguish breath healthy controls and P. vivax infected individuals and to identify volatiles that are capable of supporting such a classification, we calculated the mutual information (MI) between each "feature", which is the unique combination of a specific value of mass to charge ratio (m/z) and GC retention time (Rt), and the time course (i.e. stage of the infection).
This involved the following steps: Normalization. Samples analyzed by the GC-MS-QTOF instrument on different days were normalized using a spiked mixture of standards as described in Section S2.3. Only one standard (i.e. 2-hexanone) analyzed at the beginning of the run on each day of analysis was used for normalization. The normalization was as follows: x 10 7 where is the original sample reading, is the relevant day's 2-hexanone standard reading and ′ is the normalized reading. The factor 10 7 is used to correct for compound levels.
Peak detection and area under the curve (AUC). Next, we used an automated process (6) to identify chromatographic peaks, calculate the area under each peak and thus attribute the total counts for the particular ion in the sample. This process allowed us to compare the total amount of individual ions across samples. Automated peak detection algorithm was based on (6), which identifies a peak when the signal's second derivative is negative.
The peak detection algorithm yielded a derivative spectrum with the same number of time steps as the original data, with area under the curve at the time step of each detected peak's maximum value, and zero for all other time steps.
Peak Alignment. It is an unavoidable feature of GC that the elution time of a chemical peak can shift slightly between runs even under identical running conditions due to environmental and sample variability. To compare GC-MS data automatically, chemically identical peaks must be aligned to the same time index. This was done using a combination of piecewise alignment (7) and dynamic time warping (8). Alignment accuracy was checked using the Pearson correlation between samples (7).
Feature Selection. Feature selection identifies a subset of features, ∈ 1, 2, 3, … , , where is the total number of features, such that the resulting subset of features gives the best classification performance for the given size constraint on the number of features in . We selected a subset of features by maximizing the mutual information between the selected features , … , and class (9): , , | , This approach minimizes the uncertainty about the class, given the features. The challenges in evaluating the equation above, for a given size constraint, include: (i) estimating the multivalent joint and conditional density function with only a small data set; and (ii) selecting when there is a large number of feature sets to choose from (there are , where is the total number of features combinations for each . We took a simple approach to these issues. To select features, we took the ones with the highest individual mutual information for the class. This means that only mutual information calculations are made for a single feature and avoids the combinatorial explosion of possible feature sets as increases (10,11). This approach is also more appropriate for the current problem of choosing the features, which are the VOC scores after peak detection and calculation of the area under the curve, because it may allow us to identify a subset of compounds that can differentiate the classes.