Due to the disrupted metabolisms of COVID-19 patients, their sweat volatile compounds are hypothesized to be different compared with the healthy population. This study thus developed volatile compound analysis approaches for COVID-19 screening using SPME GC–MS and direct pressurized injection onto PID coupled with the selective filter as illustrated in Fig. 1. This figure also illustrated the sweat chromatograms belonging to the PCR proven COVID-19 positive patients (both symptomatic and asymptomatic) which are different from that of the negative control confirming the hypothesis above.Figure 1Diagram showing the sweat based COVID-19 screening approaches using SPME GC–MS, with the total ion chromatograms of two COVID-19 positive samples and the negative control, and the direct analysis with PID-selective filter.Whilst the early study reported with the investigated samples obtained from July 2020 to March 2021 relied on use of electronic nose to perform the screening based on the key markers of aldehydes and ketones26, our later investigation with GC–MS analysis of the cotton samples containing armpit sweat during the Alpha and Beta variants revealed additional markers of acetophenone, (E)-2-octenal, 1-chloro-octane and 2-ethylhexyl acrylate investigated with the COVID-19 positive patients in King Chulalongkorn Memorial Hospital, Bangkok, Thailand, unpublished data. Further investigation towards a larger set of armpit sweat samples led to discovery of the potential COVID-19 markers of nonanal (88% accuracy analyzed with a fluorescence-based sensor, n = 85, covering the Alpha, Beta, Delta and Omicron variants)28, p-cymene (96% accuracy using GC-flame ionization detector, n = 368 covering the Alpha, Beta and Delta variants)33, linalool (98% accuracy) and 2,6,11-trimethyldodecane (94% accuracy) both with SPME prior to analysis with comprehensive two dimensional GC (n = 66 covering the Alpha, Beta and Delta variants)34, and the combined feature of styrene, xylene, ethylbenzene, nonanal, and 2-ethylhexyl acrylate (94% accuracy with SPME GC–MS, n = 140)29. In addition, the study in 2023 reported the Delta and Omicron variants sharing the markers of diacetone alcohol, styrene, 2-pentylfuran, phenylacetaldehyde, undecane, methyl caprylate, trans-2-nonenal, 1-nonanol, decanal, 2-phenoxyethanol, dodecanal, 6,10-dimethyl-5,9-undecadiene-2-one-(e), 1-dodecanol and dodecanoic acid35.Among these compounds, the benzene derivatives could be sensitively detected with a PID with the incorporation of a suitable filter for enhanced selectivity in the sweat sample matrices. The SPME GC–MS data of 125 samples obtained from Chulalongkorn hospital (the training data during the research phase) were analyzed focusing on benzene derivatives with the selected examples with their accuracy data and p-values (< 0.05) from the t-test to differentiate the COVID-19 positive and negative samples provided in Table 1. These compounds were also targeted by the PID analysis as well as the other compounds with lower accuracy such as ethylbenzene and xylene. The identified benzene derivatives could be produced by several potential bacteria reviewed and some were identified in our study, in Table S1. Further results and discussion related to relationship between benzene derivatives and skin microbiomes and the volatile markers for COVID-19 screening investigated with the GC–MS analysis are provided in supporting information.
Table 1 Selected compounds with the statistical parameters to differentiate the positive and negative samples during the research phase (n = 125).PID incorporated with a modified benzene selective filter as a portable COVID-19 sensorThe overall analysis process with the PID is illustrated in Fig. 1. With this device, a SPME vial containing cottons containing sweat was initially equilibrated at room temperature. This is followed by taking the sample headspace (10 mL) and the subsequent manual injection (thumbed by the researcher and pulled by the pump of the PID instrument) onto the filter. The eluting pulse was then transferred onto a PID showing the signal converted into ppm relative to benzene. The 9.8 eV PID is considerably nonselective which requires separation techniques or selective materials to selectively detect the target volatile markers. A commercial benzene selective filter (Filter A) has been modified using the treatment trademark into Filter B31 with selectivity towards different set of the compounds for improved COVID-19 screening performance. Filter A was analyzed by a portable Raman instrument. The characteristic wave numbers for these materials were 443, 591, 903, 988, 1050 and 1200 cm−1 (Fig. 2) which indicate the presence of sulfuric acid in aqueous (enabling catalytic oxidation of non-aromatic polar compounds in the sample headspace) as well as the supporting material (carbon based porous polymer Tenax@). The modification process led to the peak intensity being reduced at 801, 988 and 1798 cm−1 and increased at 682 cm−1 in Filters B. This may be related to conversion of sulfuric acid and the material as well as addition of chemical species in the treatment process. Note that the purpose behind the modification was to selectively alter the catalytic oxidation/reaction selectivity as well as adjusting the polarity of the material towards releasing more marker compounds into the PID after passing the sample through Filter B.Figure 2Average portable Raman results of (A) Filter A and (B) Filter B (n = 3) obtained using Resolve, with the ROC curves obtained from the responses of PID-Filter A and PID-Filter B provided in C and D, respectively. The dots show the optimal cutoff thresholds of the portable sensor responses to discriminate the COVID-19 patients during the period of the delta variant (n = 130).For the performance tests, the plots of recoveries of different compounds in sweat from these materials are provided in Fig. 3. The unmodified material (Filter A) expectedly provides high selectivity towards only benzene under atmospheric sampling condition, e.g. with the other derivatives showing < 1% recoveries from the filter36. However, under pressurized injection (10 mL) applied in this study, different selectivity was observed with the enhanced recoveries towards ethylbenzene, toluene, styrene and the long chain alcohols. Although applications of Filter B reduced the recoveries in average, this filter enabled several marker compounds (2-ethyl hexyl acrylate, tridecanal and p-cymene) to pass through the filter and reach the PID with the higher recoveries (Fig. 3).Figure 3Recoveries of different compounds measured by the PID coupled with Filter A (yellow) and Filter B (blue): left and right, respectively, relative to the signal measurement without the filter. The relatively standard deviation data of Filter A and Filter B were within the range of 3–22% and 4–34%, respectively.Performance of COVID-19 screening approach using the developed sensorThe portable PID coupled with Filter A or B was initially applied to perform COVID-19 screening tests for the set of 130 sweat samples collected during the period of the Alpha, Beta and Delta variants (72 positive and 58 negative samples). These samples were the subset of that described in Sect. “Sample collection”. The collected positive and negative samples were validated according to “detected by RT-PCR or rapid antigen test” and “not detected by RT-PCR or rapid antigen test within 14 days after the sweat test”, respectively. With the criteria that the positive samples showing the signals of ≥ 0.01 ppm, Filters A and B showed accuracies of 86% and 100%, respectively, see also the corresponding ROC curves in Fig. 1C and D, respectively. The greater performance of Filter B corresponds to the relatively high recoveries of p-cymene and 2-ethylhexyl acrylate (Fig. 3) which are the potential markers supported by the high areas under the ROC curves of these compound in Fig. S1. Filter B was then applied to perform further screening tests. It should be noted that this material is not exclusively selective for all the marker compounds. However, this is sufficient to extract the target compounds within the matrix of the heated headspace of the sweat sample. A challenge is thus to develop material tailored made with the greater recoveries of the expected marker compounds in Fig. 3 in order to improve the screening performance in the future.Portable PID-Filter B was applied for the screening test of the same sample set during the period of the alpha–beta variants as that investigated with GC–MS above. This resulted in sensitivity of 100% and specificity of 100% (Fig. S2). This performance was greater than that offered from the GC–MS analysis of the individual compound in Fig. S1A–E (≤ 92% accuracy). This can be explained in the way that more marker compounds could pass through the applied filter as confirmed by the result in Fig. S1F where combination of the marker compounds showed the improved screening performance29. It is also possible that the sweat matrix reacted with the chemicals in Filter B producing other marker products detectable by the PID.Application of the developed sensor for screening of COVID-19 positive populations in Bangkok, ThailandIn this section, the collected COVID-19 positive samples; RT-PCR detected or rapid antigen test positive and COVID-19 negative samples; RT-PCR not detected or asymptomatic and rapid antigen test negative, were validated. The related numbers were 64 positive & 61 negative, 156 positive & 197 negative, 684 positive & 855 negative and 253 positive & 62 negative samples during the periods of research phase, Alpha–Beta, Delta and Omicrons variant of SARS-CoV-2 outbreak, respectively. They were investigated with the portable PID-Filter B with the signal distributions plotted in Fig. 4 and the ROC curves showing the performances provided in Fig. S2. This indicates ≥ 95% accuracies of the developed approach for screening of the investigated COVID-19 samples in Bangkok. Thus far, more than 8000 cases have been screened.Figure 4Distribution plots of the developed PID-Filter B signals for the COVID-19 screening tests during the periods of different variants (Research phase, Alpha–beta, Delta and Omicron, A–D, respectively). The %values from left to right in each figure indicate sensitivity, specificity and accuracy, respectively.Acceptable repeated screening capability was also achieved as illustrated by the same positive samples showing the average signal of ≥ 0.01 ppm (Fig. S3) as well as the negative controls showing 0.00 ppm over > 240 screening tests. Note that a positive control can be a set of PCR positive samples; whilst the negative controls were the PCR negative samples. Each sample can be reused for several days without significant signal drops albeit with the requirement of ≥ 5 min of equilibrium time prior to each repeated sampling. Discussion related to false positive and false negative is provided in Supporting information. Limitation of our screening approach is that microbial distribution could be geographically different affecting distribution of the skin volatile compounds. This approach is thus not expected to be effectively applicable everywhere in the world. It should also be noted for an outdoor screening test that some environment such as air pollution, an area with strong smell of food or a new construction with the new painting should be avoided. Moreover, we could not exactly identify the variant of SARS-CoV-2 of all samples due to limited resource and budget. The application of the results among all variants had to be proposed just by national epidemiological data. However, volatilome represented from metabolism and biological reaction of SARS-CoV-2 infection would generally not be affected by spike mutation in theory.
A large scale study of portable sweat test sensor for accurate, non-invasive and rapid COVID-19 screening based on volatile compound marker detection
