Engineering genetically-encoded synthetic biomarkers for breath-based cancer detection

Breath analysis holds great promise for rapid, noninvasive early cancer detection; however, clinical implementation is impeded by limited signal from nascent tumors and high background expression by non-malignant tissues. To address this issue, we developed a novel breath-based reporter system for early cancer detection using D-limonene, a volatile organic compound (VOC) from citrus fruit that is not produced in humans, in order to minimize background signal and maximize sensitivity and specificity for cancer detection. We metabolically engineered HeLa human cervical cancer cells to express limonene at levels detectable by mass spectrometry by introducing a single plant gene encoding limonene synthase. To improve limonene production and detection sensitivity twofold, we genetically co-expressed a modified form of a key enzyme in the cholesterol biosynthesis pathway. In a HeLa xenograft tumor mouse model, limonene is a sensitive and specific volatile reporter of tumor presence and growth, permitting detection of tumors as small as 5 mm. Moreover, tumor detection in mice improves proportionally with breath sampling time. By continuously collecting VOCs for 10 hours, we improve sensitivity for cancer detection 100-fold over static headspace sampling methods. Whole-body physiologically-based pharmacokinetic (PBPK) modeling and simulation of tumor-derived limonene predicts detection of tumors as small as 7 mm in humans, equivalent to the detection limit of clinical imaging modalities, such as PET, yet far more economical. Significance Statement We developed a breath-based reporter system using the plant terpene, D-limonene – a volatile secondary metabolite that gives citrus fruit its characteristic scent but is not produced in human tissues – as a biomarker for early cancer detection. Results from this study could pave the way for in vivo gene delivery and tumor-specific expression of exogenous volatile cancer reporters with broad applicability to the early diagnosis of a wide variety of cancers.


Introduction
Breath analysis provides rapid and non-invasive biomolecule detection, with great promise for early cancer detection and surveillance 1,2 . The human body emits hundreds of volatile organic compounds (VOCs) -organic molecules that readily vaporize at room temperature -in the breath 3 .
Breath, a less complex matrix than blood and other bodily fluids, can be sampled easily, painlessly, and inexpensively 1 . Moreover, breath can be directly analyzed using real-time mass spectrometry, reducing the need for sample storage and processing 1 . While no single VOC can reliably signal cancer presence on its own, VOC signatures or "breathprints" have been reported that can distinguish a number of cancers -including lung, colon, breast, and prostate cancers -from benign disease and healthy controls in relatively small study populations [4][5][6] . However, as with liquid biopsies [7][8][9][10][11][12][13][14] , clinical implementation of breath VOCs for early cancer detection is limited by low or variable signal from cancer cells and high background signal from nonmalignant tissues 3,5 .
Furthermore, identification of cancer-specific VOC signatures has been impeded by statistical overfitting -a common pitfall in early stage 'omics approaches due to typically small study populations relative to the numerous endogenous parameters analyzed -limiting their generalizability. 15 Engineered synthetic reporters provide an innovative solution to overcome the detection limitations of endogenous biomarkers. [16][17][18][19][20][21] By effecting diseased cells to express an exogenous biomarker that is not naturally produced in human tissues, background signal from non-diseased tissues is minimized, thereby maximizing sensitivity and specificity. Moreover, exogenous reporters from biochemical classes that are orthogonal to the human metabolome can easily be distinguished from the complex milieu of endogenous molecules by mass spectrometry.
Furthermore, detection of a single exogenous biomarker that uniquely signals disease presence avoids the statistical challenges associated with endogenous VOC analysis. Recent synthetic strategies include exogenous protein biomarkers encoded on in vivo-delivered DNA vectors and selectively secreted into the blood by cancer cells, as well as nanoparticles that release a volatile compound in the breath to signal lung infection or inflammation 17,21 . Genetically-encoded synthetic biomarkers have practical and theoretical advantages, including: 1) integration with clinically established in vivo gene delivery methods, including those used in vaccine development (viral vectors, liposomes, and minicircles) [22][23][24] ; 2) selective expression in many cancer types using tumor-specific promoters and tumoritropic or tumor-targeted vectors 17,25,26 ; and 3) continuous expression throughout the lifetime of the cancer, which can enable repeat monitoring after a single administration. However, there have been no reports thus far of strategies that genetically encode synthetic biomarkers for breath-based detection of cancer.
Here, we combine the high specificity and sensitivity of an exogenous cancer biomarker with the speed, simplicity, and non-invasive nature of breath VOC detection. To genetically encode a VOC biomarker in cancer cells that is distinct from endogenous VOCs, we looked to plants. Humans and plants share a common cholesterol biosynthesis pathway, but in plants this pathway also generates terpenes, the volatile compounds that attract pollinators and protect from herbivorous insects and pathogens [27][28][29] . We hypothesized that the mammalian cell's cholesterol biosynthetic machinery could be exploited to produce plant volatiles by genetically introducing the appropriate exogenous enzymes (Fig. 1). While many plant volatiles require multiple biosynthetic steps, only a single enzyme, limonene synthase (LS) 30 , bridges the cholesterol biosynthesis pathway with production of limonene 27 , the monoterpene that gives citrus fruits their characteristic scent.
Limonene is already used clinically (for example, to treat gallstones and heartburn), has chemopreventive and chemotherapeutic effects in many types of cancers, and is safe at oral doses as high as 100 mg/kg (~7 g for an average 70 kg adult) 31,32 . Due to its wide industrial use, metabolic engineering approaches for increasing limonene biosynthesis have been extensively studied in microbial systems and plants 29,[33][34][35][36] , and have the potential to be adapted to human cancer cells for breath-based diagnosis and eventually -at high expression levels -for therapy. We demonstrate that limonene can be genetically expressed in human cancer cells and can report on early tumor presence and growth in a xenograft mouse model. We also extrapolate our VOC-based detection to humans using a whole-body physiologically-based pharmacokinetic (PBPK) model of VOC biodistribution, metabolism, and exhalation.

Limonene expression and detection in cultured tumor cells.
HeLa cells were transfected with a vector containing LS and eGFP genes under the control of a single CAG promoter ( Figs. 2A, 2B).
Antibiotic selection and FACS sorting for high eGFP expressers yielded a stable cell line containing limonene synthase (HeLa-LS) (Fig. 2C). To maximize limonene production in cultured HeLa-LS cells, we targeted a key regulatory enzyme of the mevalonate pathway, HMG-CoA reductase (HMGR) 37 . Truncation of HMGR by deletion of its N-terminal regulatory domain renders it insensitive to feedback inhibition by downstream metabolites, augmenting flux through the mevalonate pathway and increasing the availability of limonene precursors. Previous studies in bacteria and yeast engineered to produce limonene have shown that expression of truncated HMGR (tHMGR) can markedly increase limonene production 33,34 . We transfected HeLa-LS cells with a plasmid encoding human tHMGR and turbo red fluorescent protein (tRFP) under the control of an EF1a promoter ( Figs. 2A, 2B). Antibiotic selection and FACS sorting for high expression of tRFP yielded a stable cell line expressing both eGFP and tRFP (Fig. 2C), and containing both LS and tHMGR (HeLa-LS-tHMGR). Solid phase microextraction (SPME) fibers 5,38 were used to sample the culture headspace (i.e. the air above the cells) in flasks containing confluent stably transfected cells ( Fig. 2A). Gas chromatography-mass spectrometry (GC-MS) analysis of the fibers showed a mass spectrum closely matching the limonene standard, with both exhibiting the characteristic ion peaks for limonene (m/z = 68, 93, and 136) 39 at the same relative ratios (Fig. 2D) and identical chromatogram retention times (Fig. 2E).

Quantification of limonene from transfected cells.
We further confirmed the presence of headspace limonene using selected ion flow tube mass spectrometry (SIFT-MS), which affords continuous, real-time VOC detection with quantification down to the parts-per-billion level 3,5,40 .
To obtain quantitative measurements of headspace limonene, we created a calibration curve for limonene (10 pg to 100 µg) spiked into media within a 280 mL T75 flask (Fig. 2F). Headspace concentrations increased as a function of x 0.86 for limonene quantities within the range of 1 ng to 100 µg (R 2 = 0.99) and demonstrated a nearly linear dependence with limonene quantities ranging from 1 ng to 1 µg (R 2 = 0.99). The limit of detection (LOD) for limonene by SIFT-MS was 1.8 ng, corresponding to 0.5 ppb in the headspace. Next, we sought to quantify limonene generated by

Quantification of limonene emitted from limonene-injected and tumor-bearing mice.
Having observed robust limonene expression in transfected HeLa cells in culture, we then tested the feasibility of detecting limonene in exhaled breath from rodents. We created a standard curve relating limonene concentration in chamber headspace to the quantity of limonene spiked into 0.5-L chambers. To determine the fraction of limonene in mice that is emitted into the headspace, we injected mice intraperitoneally with different quantities of a limonene standard solution (from 0.01 µg to 1 mg) and placed individual mice in a closed chamber for 15 minutes, at which point headspace limonene concentrations were measured by SIFT-MS (Fig. 3A, 3B). Using the standard curve, we determined the mass of limonene exhaled by mice at each quantity injected, and calculated the fraction exhaled. At the LOD (0.5 ppb), limonene in the chamber headspace became detectable when 2.3 ng had been spiked into the chamber, whereas limonene evolving from mice only became detectable at an injected dose of 450 ng (Fig. 3B, Supplementary Calculations). A comparison of the graphs for these two conditions shows that only ~0.5% of limonene at each injected dose was emitted into the chamber headspace within 15 minutes of injection. We therefore expected mice bearing limonene-producing tumors to emit a similar fraction into the chamber headspace over this time period. Assuming the limonene production rate in cell culture to be an upper bound on the expected cellular limonene production rate in tumor-bearing mice, we calculated that large tumors with diameters of at least 3.4 cm (4 billion cells) would be required in order to reach the detection limit of SIFT-MS within 15 minutes (Supplementary Calculations).
To test this, we implanted one million HeLa-LS or HeLa-LS-tHMGR cells subcutaneously into each flank of immunocompromised nude mice and monitored them using SIFT-MS at 5 weeks post-implantation. Consistent with our calculations, we found that no limonene was detected in the chamber headspace even when up to 4 mice with a combined tumor burden of ~4 cm 3 were contained in a single chamber.
To increase sensitivity for detecting limonene from tumor-bearing mice, we built a speciallydesigned experimental setup in which highly purified air is continuously flowed through a mouse chamber and exits through an air sampling tube containing a sorbent material (Tenax TA) that traps VOCs, thereby pre-concentrating them for subsequent GC-MS analysis. Compared to SPME fibers, sorbent traps contain much larger quantities of sorbent material and therefore have higher extraction capacities 5 . Six one-liter chambers were set up in parallel to allow for multiple simultaneous experiments (Figs. 3C, S1). We placed groups of HeLa-LS-tHMGR mice and control mice bearing untransfected HeLa tumors at 5 weeks post-implantation into side-by-side chambers, with 4 mice per chamber (average tumor volume per mouse: 1.2 ± 0.2 cm 3 ), and sampled the chamber headspace (100 mL/min airflow) for 1, 4, or 10 hours. In the experimental group, limonene was detectable in chamber air at all sampling durations. Increasing the sampling duration from 1 hour to 4 hours enabled 2.3-fold greater limonene collection (10 ng to 23 ng), and an increase to 10 hours enabled 9.4-fold greater limonene collection (10 ng to 94 ng) (Fig. 3D).
Limonene levels for control mice were below 1 ng at all sampling durations. Therefore, we showed that increased signal-to-background was achievable simply by sampling the chamber headspace for a longer time. By integrating limonene signal over a number of hours, the sorbent trap method improves detection sensitivity 100-fold compared to direct SIFT-MS measurements in sealed unventilated chambers (Supplementary Calculations), where measurements are limited to only a few minutes before mice become hypoxic. To maximize our sensitivity, we chose 10-hour sampling times for all subsequent mouse experiments.
We next sought to determine the minimum tumor size at which limonene is detectable and to evaluate whether tumor growth could be monitored via exhaled limonene alone. HeLa-LS, HeLa-LS-tHMGR, and control mice (bearing untransfected HeLa tumors) were monitored over a 5-week period. Groups of four mice per chamber (n=3 chambers per cohort) were tested once a week for total limonene released into chamber air during a 10-hour period. At week one post-implantation of tumor cells, total evolved limonene from the HeLa-LS-tHMGR cohort (11 ± 2 ng) was statistically higher compared to the HeLa-LS (6 ± 1 ng, p = 0.049) and control mouse groups (4 ± Limonene emitted from HeLa-LS and HeLa-LS-tHMGR mice increased linearly with tumor volume over 4 and 5 weeks post-implantation, respectively (Fig. 3F). Limonene evolution was higher in HeLa-LS-tHMGR mice than in HeLa-LS mice throughout the study, though this difference was statistically significant only in weeks 1 and 5. Limonene evolution from HeLa-LS and HeLa-LS-tHMGR mice peaked in weeks 4 and 5 at 60 ± 16 ng and 94 ± 14 ng, respectively (when tumor burden per mouse was 0.6 ± 0.1 cm 3 and 0.8 ± 0.2 cm 3 , respectively). This plateau in HeLa-LS mice corresponded with a leveling off in tumor growth (i.e. no statistical change) from weeks 4 to 5 (Fig. 3F). At week 5, mice were humanely euthanized due to tumor size. Tumor growth rate, k, was slightly greater in control mice (k = 0.54) than in HeLa-LS-tHMGR (k = 0.48, p = 0.049), whereas it was not statistically different between HeLa-LS-tHMGR and HeLa-LS mice (k = 0.53, p = 0.13) or between HeLa-LS and control mice (p = 0.51) (Fig. 3G). Limonene quantities collected from HeLa control mice at each time point were very similar to blank chambers without mice, with a range of <1 ng to 4 ng (Fig. S3). These values likely represent ambient limonene that was degassing from the chamber walls, given that limonene levels both from control mice and blank chambers were below the detection limit by the end of the 5-week study. Moreover, limonene was not detected above background in chambers containing only mouse diet gel or bedding. Therefore, we concluded that the only sources of limonene in HeLa-LS-tHMGR and HeLa-LS mice were the tumors. The average percentage of tumor limonene exhaled in the breath over all weeks was calculated at 5.2% ± 1.5% and 7.6% ± 3.1% for HeLa-LS-tHMGR and HeLa-LS mice, respectively (Supplementary Calculations, Tables S3-S7).

Whole-body physiologically-based pharmacokinetic (PBPK) model for VOC detection.
To predict the smallest tumor size that would be detectable in humans using a sorbent trap, we developed and simulated a whole-body physiologically-based pharmacokinetic (PBPK) model for limonene disposition in humans (Fig. 4A, Tables S1-S2). Our multi-compartment model shows that in humans, 1.3% of tumor-derived limonene would be emitted in the breath (Fig. 4B). This result agrees with prior literature showing that, due to rapid biodistribution (particularly to adipose tissue) and metabolism, ~1% of absorbed limonene is eliminated unchanged in exhaled breath 31,32,41,42 . Given the LOD of the sorbent trap method (2.3 ng), we calculated that a human tumor would need to produce at least 177 ng (1.3% of 177 ng is 2.3 ng) of limonene over a 10hour period to be detectable in the breath. Assuming a limonene production rate similar to HeLa-LS-tHMGR cells, tumors would become detectable when they reached 7 mm in size (Supplementary Calculations). Substituting limonene with a VOC that has low fat-to-blood (Kf:b) or blood-to-air (Kb:a) partition coefficients could further improve detection sensitivity. For example, our PBPK model shows that if instead of limonene (Kf:b = 140, Kb:a = 42) 2 , a VOC reporter with partition coefficients like those of 2-butanone (Kf:b = 0.75, Kb:a = 215) 43 was chosen, 30% of it would be exhaled in the breath (Fig. 4C); this could permit detection of tumors that are ~30-fold smaller by volume (2.3 mm diameter) at the same VOC expression level. Our PBPK modeling approach can also be extended to other potential VOC candidates for cancer detection.

Discussion
Here, we report a novel strategy for sensitive and specific breath-based cancer detection that uses limonene, a plant terpene, as an exogenous VOC reporter. First, we demonstrated stable heterologous expression of limonene, as validated by mass spectrometry, in a cultured HeLa human cervical cancer cell line transfected with a plasmid encoding the plant enzyme limonene synthase. We also demonstrated that genetically co-expressing a modified key mevalonate pathway enzyme, tHMGR, can double limonene expression in HeLa cells, thereby improving detection sensitivity for these cells in culture and in vivo. We then validated limonene as a sensitive and specific volatile reporter of tumor presence and growth in a xenograft mouse model after subcutaneous implantation of limonene-expressing HeLa cells, and showed that limonene can be detected when tumors are as small as 120 mm 3 (~5 mm diameter). Using human whole-body PBPK modeling, we predicted that tumor-derived limonene would be detectable in human breath from a tumor as small as 7 mm in diameter.
The attributes of an ideal breath-based cancer reporter include: 1) safety; 2) low background (little or no endogenous production in the body); 3) specificity (high expression in cancer cells relative to nonmalignant cells; 4) good deliverability from tumor to the lungs (high air-to-plasma and plasma-to-tissue partition coefficients, and low fat-to-plasma partition coefficient); 5) little or no metabolism to non-volatile compounds; 6) abundance of biochemical precursors; and 7) few enzymatic steps (facilitating ease of design and efficient production in mammalian cells).
Identification of a VOC reporter that meets all of these criteria is challenging. Limonene is an attractive candidate because it is safe, not endogenously produced in human tissues, and requires only a single enzymatic step for expression in mammalian cells. For any given VOC, factors such as deliverability to the lungs and rate of metabolism are dependent on the VOC's structure and intrinsic molecular properties and cannot easily be controlled, except by switching to a different VOC reporter altogether. However, various strategies can be implemented to increase detection sensitivity, mitigate other sources of background, increase the availability of biochemical precursors, and achieve tumor-specific expression, as we discuss below.
A key finding from our work was that sensitivity for detecting small tumors could be dramatically improved via sorbent trapping of limonene from a mouse chamber over extended periods of time.
A 10-fold increase in sampling time from one to 10 hours yielded a concomitant increase in limonene signal from HeLa/LS mice, while background limonene levels for HeLa control mice remained below the detection limit. Further integrating limonene signal over even longer periods of time may enable detection of even smaller tumors. The breath can provide an essentially limitless sample volume for this purpose since the total volume sampled can be increased simply by breathing into a sorbent trap for a longer period of time. By contrast, there are practical limits to the amount of blood that can be drawn from a patient in a given time (10% of total blood volume per month) 44 .
While extending sampling time increases limonene signal, sensitivity can also be improved by minimizing background noise. Due to widespread industrial use and production of the Senantiomer by many plant species, limonene is present at 0.1 to 2 ppb in ambient air 31,32 . In this study, background limonene levels in control and empty chambers decreased each week as the chambers were cumulatively exposed to highly pure air over a longer period of time, further removing residual limonene from the chamber walls. Thus, further improvements in LOD may be possible by pre-purging chambers with air for longer periods of time (e.g., a few weeks) or running experiments in glass chambers, which do not readily adsorb hydrophobic VOCs. In the clinical scenario, human subjects would be placed in a room with highly pure air or would breathe through a one-way filter cartridge to prevent contamination of inhaled air by ambient limonene. Exhaled air would pass through an exhaust valve directly into a sorbent tube, which would subsequently be analyzed offline by GC-MS. The small filter cartridge/sorbent tube assembly could be worn portably to passively collect limonene over a few hours as the subject goes about their day or at night while sleeping. Subjects would need to avoid wearing perfumes or consuming citrus prior to undergoing testing. The presence of limonene in the breath at screening or surveillance would then prompt clinical imaging studies, such as PET or MRI, in an attempt to spatially localize the tumor.
Monitoring of VOC reporter levels could also be used to assess response to therapy inexpensively and more frequently than is practical or economical with in vivo imaging in patients with metastatic disease or large disease burden.
For cancer screening and early detection, we envision targeting expression of the limonene synthase gene to cancer cells using novel and clinically relevant gene delivery approaches including minicircles 17,24 . Smaller tumors could then be detected by employing strategies that increase cellular reporter gene production. Engineering 10-or 100-fold increases in tumor limonene production could permit detection of human tumors as small as 3 mm (below the detection limit of clinical PET imaging) or 1. 5

mm, respectively (Supplementary Calculations).
High cancer reporter expression can obviate time-consuming pre-concentration steps by achieving VOC levels in the breath that are directly detectable by rapid breath analysis platforms such as SIFT-MS, an electronic nose, or eventually a portable breath analyzer for routine at-home cancer monitoring. Clinically, certain aromas in breath have long been recognized by unaided olfaction as telltale signs of various pathologic states, including the ketone odor in diabetic ketoacidosis and the ammonia odor in chronic kidney disease 3 . Very high expression levels (>10 ppb) could likewise allow limonene to be detectable in the breath by the human nose (i.e. cancer diagnosis by a quick "sniff test" at the doctor's office).
Metabolic engineering strategies that have been used for large-scale microbial biosynthesis of limonene and other natural products 27,[33][34][35][36]45,46 can potentially be adapted to scale up tumor cell production of VOC reporters in human cells in culture and in vivo. The production of limonene is mainly limited by the availability of its direct biochemical precursor, geranyl diphosphate (GPP), in the cytosol 29,34 . In mammalian cells, GPP is a relatively short-lived metabolic intermediate because the synthase enzyme that converts isopentenyl diphosphate (IPP) to GPP then rapidly converts GPP to farnesyl pyrophosphate (FPP). Truncated forms of GPP-and FPP-synthases have been investigated in engineered microbes that predominantly produce GPP, increasing monoterpene production [33][34][35]47 . Another strategy that has successfully increased terpene production in microbes is to introduce additional copies of mevalonate pathway genes into cells.
Moreover, since the methylerythritol 4-phosphate (MEP) pathway provides most of the carbon flux for terpene production in plants, heterologous expression of these enzymes (without the PSP) could potentially increase limonene production in human cells as well 35,[47][48][49] . Combinations of genes could in principle be delivered on the same vector as part of a comprehensive in vivo gene delivery approach.
Future work will test strategies for in vivo gene delivery and expression of our VOC reporter. Lung cancer would intuitively be an ideal initial clinical target for sensitive breath-based detection because of the close proximity of lung tumors to exhaled air. Further, using our PBPK model, we plan to identify additional candidate VOC reporters with improved deliverability to the lungs. For a VOC reporter that is not consumed in the diet or present environmentally, its background levels would be limited only by the tumor-specificity of the gene delivery approach or gene expression method. A potential hurdle to clinical implementation of any exogenously-derived reporter is the development of host immunity to the foreign enzyme, which could render subsequent delivery of the gene ineffective. However, techniques for de-immunization, in which epitope sequences are modified to minimize immunogenicity without altering protein structure or function, have long been used in biologic drug design and can similarly be applied to limonene synthase 50 . Tolerogenic sequences can also be introduced to promote immune tolerance to exogenous enzymes 51,52 .
Collectively, we anticipate that these efforts will help advance this technology to facilitate cancer screening and early detection in the clinic.

Study approval. All procedures performed on animals were approved by Stanford University's
Institutional Animal Care and Use Committee (Stanford, CA, USA) and in compliance with the guidelines for humane care of laboratory animals. Transfection of limonene-producing cells with a tHMGR-encoding vector (HeLa-LS-tHMGR) was accomplished in a similar manner, with hygromycin B (ThermoFisher, Waltham, MA) used for antibiotic selection of stable cells, and with FACS selection performed by gating on RFP ( Fig.   2A, B).  Louis, MO). The flasks were manually agitated for 10 seconds and the screw cap septum was punctured by a needle. The flask headspace was sampled for 20 seconds at least 3 times per concentration using selected ion flow mass spectrometry (SIFT-MS, Syft Technologies, Christchurch, New Zealand) with a helium gas carrier. Limonene detection was performed by softionization using H3O + (m/z, 137; branching ratio, 68%; reaction rate, 2.6x10 -9 cm 3 /s), NO + (m/z, 136; branching ratio, 88%; reaction rate, 2.2x10 -9 cm 3 /s) and O2 + (m/z, 93; branching ratio, 29%; reaction rate, 2.2x10 -9 cm 3 /s) to calculate limonene concentration in real-time. After establishing the calibration curve, HeLa-LS and HeLa-LS-tHMGR cells were spiked into 10 mL media (DMEM with 10% FBS) in varying numbers ranging from 20,000 to 10 million cells in T75 flasks.

Vector design (HeLa-LS and HeLa-LS-tHMGR
The flasks were incubated at 37 o C for 24 hours, after which headspace limonene concentrations were measured using SIFT-MS. The cells were then harvested and counted with cell numbers at harvest ranging from ~45,000 to 25 million.

Quantitation of limonene evolution from limonene-injected mice.
Prior to mouse studies, a calibration curve was generated. Known limonene quantities (10 pg to 100 µg) were added to 10 mL of water in 0.5-mL chambers (Kent Scientific, Torrington, CT). The chambers were capped, briefly agitated, and allowed to sit for 15 minutes to equilibrate. The chamber inlet was then