Introduction

The analysis of urine is a non-invasive method widely used in diagnosis. Accordingly, numerous works addressing the study of compounds existing in this matrix are currently ongoing with the aim of improving or accelerating the detection of certain diseases [1]. As a result of these studies, some compounds have been described as possible biomarkers of different disorders. Biomarkers are therefore compounds whose presence or absence could be related to a certain disease or whose concentration differs between samples from healthy people and from patients suffering from a given disease [2, 3]. Usually, a group of biomarkers is used, rather than a single one, for the characterization of a sample. In this work, five possible volatile biomarkers of lung cancer in urine samples were studied (2-butanone, 2-pentanone, pyrrole, 2-heptanone, and 2-ethyl-1-hexanol) [2, 4].

As mentioned, urine has the great advantage of being readily available and so, generally, volume is not an important issue. However, urine is a rather complex matrix whose analysis usually requires thorough sample preparation for analyte isolation and preconcentration [3–9]. Headspace sampling [3–10] is the technique of choice for the analysis of volatile biomarkers in urine samples owing to ease of sample preparation, automation, speed, and the absence of interferences from non-volatile compounds in the matrix. Nevertheless, this technique does not always guarantee that certain problems such as the matrix effect can be overcome [5, 10].

Gas chromatography has been the main technique of choice for the analysis of volatile compounds in urine [3, 4, 6–10], but chromatographic procedures are generally slow and time-consuming. However, several alternatives can be employed to speed up GC separations. These include the use of shorter capillary columns with narrower bores and fast temperature programming. Additionally, the development of non-separative methods for the resolution of different analytical problems is currently of great interest, mainly owing to their fast analysis speed. Some non-separative methods based on sensors [11–14] and even on trained dogs [15] have been proposed for the discrimination of patients suffering from different diseases and healthy controls, using breath, saliva, blood, and urine samples. Electronic nose technology has developed significantly, along with advances in different sensor platforms and complex microarray devices. Electronic noses based on MS, where each mass to charge ratio (m/z) acts as a “sensor” that detects any ion fragment with that ratio, have been used for several tasks including biomedical applications [9, 16] and analysis of beverages [17, 18] among others. Besides electronic noses, other sensitive and non-separative methods such as desorption electrospray ionization-mass spectrometry (DESI/MS) [19], selected ion flow tube mass spectrometry (SIFT/MS) [20], and NMR [19] are often used for the determination of biomarkers in urine samples.

In the present work, a rapid, simple, sensitive, and non-separative method (HS-PTV-MS) for the determination of volatile biomarkers of lung cancer in urine samples is proposed. Five biomarkers have been selected as trial or test compounds to check the possibilities of the proposed methodology. This set of compounds includes some biomarkers (2-pentanone and 2-ethyl-1-hexanol) which have provided very good results in the discrimination of control and lung cancer samples according to the bibliography [2, 4]. Multivariate calibration models were employed to quantify the biomarker concentration in the samples, and the Wilcoxon signed-rank test was used to classify the samples in the group of healthy individuals or in the group of patients with lung cancer, comparing the sample concentrations with the median of the concentrations of a reference group of healthy individuals. To date, this is the first time that multivariate calibration and non-parametric methods have been combined to decide on whether a biological sample belongs to one set or another from its profile signal obtained with an electronic nose based on MS. For the samples assigned to the patient group, a new analysis was performed using a chromatographic method (HS-PTV-GC/MS) for confirmatory purposes.

Material and methods

Reagents, standards, and samples

2-Butanone, 2-pentanone, pyrrole, 2-heptanone, and 2-ethyl-1-hexanol were supplied by Sigma-Aldrich (Steinheim, Germany). Methanol was supplied by Merck (Darmstadt, Germany). The purities of the compounds were at least 98 %. Ultra-high-quality (UHQ) water was obtained with a Wasserlab Ultramatic water purification system (Noain, Spain).

Stock solutions (490–5000 mg/L in methanol) of each compound were prepared and stored at 4 °C in a refrigerator. They were diluted in methanol and used to prepare the working solutions (0.048–12 mg/L) containing the five compounds in ultrapure water.

Urine samples were obtained from 20 adults of both sexes and were stored at −20 °C. Sample nos. 1–13 (four women, nine men) were from healthy individuals apparently unaffected by diseases; sample nos. 14–20 (two women, five men) were from patients at the Internal Medicine Unit of the Virgen de la Vega Hospital in Salamanca. Patient no. 14 had a metastatic infiltrating urothelial carcinoma, no. 15 had a metastatic multicentric hepatocarcinoma, and nos. 16–20 had lung diseases (four patients had lung neoplasms and the other had lung metastasis from melanoma). For the analyses, the samples were left at room temperature until they had thawed completely. Then, they were transferred to a 15-mL glass centrifuge tube with a screw cap (Scharlau, Spain) and were centrifuged for 10 min at 5000 rpm. To perform the measurements of the samples (healthy volunteers and patients), 4.0 mL of urine, 2.0 g of NaCl, and 50 μL of ultrapure water were transferred to a 10-mL headspace vial (Agilent Technologies, DE, Germany) which was sealed with a Teflon®/silicone septum (Agilent Technologies, DE, Germany). To measure the spiked samples, 4.0 mL of urine, 2.0 g of NaCl, and 50 μL of the working solution described above were transferred to a 10-mL headspace vial which was sealed with a Teflon®/silicone septum. Each sample was analyzed in triplicate (three vials, one injection per vial).

The study was authorized by the Hospital Ethics Committee.

The concentration of endogenous compounds may also vary because of the irregularity of the volume of urine excreted. Because of this variability, normalization of the analyte concentrations to the total creatinine concentration (mmol) is required. This practice allows a random sampling of urine instead of a 24-h collection. For this purpose, a modification of the Jaffè method has been used [21, 22]. These measurements were made on a Shimadzu UV/Vis-160 spectrophotometer equipped with a processing and recording unit, using a quartz cell (Suprasil) with a 10-mm path length. Normalized concentration values are expressed as nanomoles of analyte/millimole of creatinine.

Measurements with the screening method (HS-PTV-MS)

HS sampling was performed with a PAL autosampler (CTC Analytics AG, Zwingen, Switzerland). This sampler is equipped with a tray for 32 consecutive samples and an oven with positions for six sample vials. The oven temperature was kept at 90 °C and the equilibration time was set at 5 min. During this time, the vials were shaken at 750 rpm in the oven. A 2.5-mL syringe at 110 °C was used.

The experiments were carried out with a programmed temperature vaporizer (PTV) inlet (CIS-4: Gerstel, Baltimore, MD, USA), with a liner (71 mm × 2 mm) packed with Tenax-TA®. Solvent-vent injection mode was used. Initially, the sample from the headspace was injected into the PTV injector, which was at 35 °C, such that the analytes were retained in the liner while the split valve was open, allowing solvent elimination (0.10 min). The vent flow was adjusted to 50 mL/min and a venting pressure at 5.0 psi (34,474 Pa) was imposed. In the second step, involving the transfer of the sample to the column, the split valve was closed and the PTV was heated rapidly (12 °C/s) until it reached 250 °C. Thus, the analytes were desorbed and transferred to the column (1.5 min). Finally, the split valve was opened again to clean the system (purge flow 150 mL/min) and the liner temperature was held at 250 °C. Cooling was accomplished with liquid CO2.

The interface between the PTV and the MS was a low-polarity DB-VRX capillary column (20 m × 0.18 mm × 1 μm) from Agilent J&W, which was maintained at 240 °C in an Agilent 6890 GC device throughout the time of analysis. Thus, the separation capacity of the column was eliminated, and it behaved as a simple transfer line from the PTV to the mass detector. This instrumental configuration made it unnecessary to change the interface when the chromatographic analyses were performed.

The detector was a quadrupole mass spectrometer (HP 5973 N) equipped with an inert ion source. It was operated in electron-ionization mode using an ionization voltage of 70 eV. The ion source temperature was 230 °C, and the quadrupole was set at 150 °C. The analyses were performed in full-scan mode (1.06 scan/s). The m/z range was 35–120 amu.

The signal-recording time was 2.5 min. An interval of 6.0 min between sample injections was necessary in order to allow the PTV to cool from the final (250 °C) to the initial (35 °C) temperature.

Measurements with the confirmatory method (HS-PTV-GC/MS)

The experimental conditions for the HS autosampler and the PTV inlet were the same as those used for the screening methodology.

The initial oven temperature was 45 °C, held for 1.5 min; this was increased at a rate of 30 °C/min up to 81 °C, then at a rate of 60 °C/min up to 175 °C, and finally further increased to 45 °C/min up to 240 °C and held for 0.50 min. The carrier gas was helium (99.999 % pure; Air Liquide) and the flow rate was 2.0 mL/min. The total chromatographic run time was 6.21 min. Additionally, about 7 min was necessary before the next sample could be measured, since the column had to be cooled down from the final temperature (240 °C) to the initial condition of 45 °C. An interval between sample injections of 14 min was chosen.

The analyses were performed in synchronous selected ion-monitoring (SIM)/scan mode, which allowed the acquisition of both SIM and full-scan data in a single run. Three full-scan (m/z 35–120) groups with different data acquisition speeds were used for compound identification by comparison of the experimental spectra with those of the NIST’08 database (NIST/EPA/NIH Mass Spectral Library, version 2.0). The number of scans per second varied between 2.94 and 9.30. SIM was used for quantification, choosing the characteristic ions in each case (Table S-1, Supplementary material).

Data analysis

Data collection was performed with an Enhanced ChemStation [23] from Agilent Technologies. Partial least-squares (PLS1) multivariate calibration was performed using the Unscrambler® v10.2 statistical package [24]. Box plots, the Mann-Whitney U test, and the Wilcoxon signed-rank test were implemented using the OriginPro v8.0724 statistical package [25].

Results and discussion

Study of the signals obtained

The volatiles generated in the HS produced a total ion current profile as shown in Fig. 1a, corresponding to a sample of urine with the concentration of the analytes ranging from 0.12 to 124 μg/L. The mass spectrum that represents the sum of the intensities of all the ions detected during the data acquisition period is shown in Fig. 1b. Some of the most abundant m/z ratios correspond to the characteristic ions in the mass spectrum of some of the studied analytes (43, 58, and 86 m/z).

Fig. 1
figure 1

a HS-PTV-MS total ion current profile of a urine sample with concentrations of the studied analytes ranging from 0.12 to 124 μg/L. b Mass spectrum that represents the sum of the intensities of all the ions detected

Full size image

Evaluation of the HS-PTV-GC/MS method

In order to check for possible matrix effects, a comparison between the signals obtained with spiked UHQ water samples and urine samples (eight different subjects) was carried out. The samples were spiked with the five analytes at concentrations ranging from 1.6 (2-heptanone) to 88 μg/L (pyrrole). In some urine samples, some studied analytes showed differences in peak areas between the urine samples and between these samples and the spiked UHQ water sample. In order to prevent this matrix effect, the possibility of diluting the urine samples with UHQ water was evaluated. For this, two different dilution ratios were tested, namely 1:1 (2 mL urine + 2 mL UHQ water) and 1:3 (1 mL urine + 3 mL UHQ water). After dilution, the matrix effect was not eliminated. No stronger dilutions were tested owing to the loss of sensitivity.

In view of the matrix effect, quantification in the chromatographic method was performed with the standard addition method. Furthermore, 4-heptanone, an endogenous compound present in all the urine samples, was used as an internal standard. Since each calibration model is performed using the same urine to prepare the standards, the concentration of 4-heptanone in each model is the same in all the samples and it is not necessary to know its value. This calibration strategy, known as standard addition-internal standard and which integrates the advantages of both the internal standard method and the standard addition method, has been used previously [5, 10] but with the addition of an external compound as internal standard. The concentration value for the standard addition set ranged between 0.80 and 150 μg/L. The variables used in the univariate calibration models were the area under the curve of the compounds in the extracted ion chromatogram (SIM mode) for the base peak (Table S-1, Supplementary material). In all cases (20 samples: 13 healthy volunteers and 7 patients), calibrations with no lack of fit were obtained, with values of the coefficient of determination (R 2) equal to or greater than 0.99. Repeatability (n = 5) was evaluated at two different levels using samples of urine and urine spiked with the biomarkers (0.80–60 μg/L). The relative standard deviation (RSD) was lower than or equal to 6 % in all cases, indicating the good repeatability of the method (Table S-2, Supplementary material). The limits of detection depended on each urine sample; as an example, the values for one of the urine samples ranged between 0.050 and 3.1 μg/L (Table S-3, Supplementary material). After the first sample has been analyzed, this method allowed the measurement of four samples per hour with the separative approach.

Evaluation of the HS-PTV-MS method

The strategy of individual calibration for each urine sample based on standard addition-internal standard cannot be used with the non-separative method. In this work, PLS1 multivariate calibration was used. The quality of the results when this technique is used depends on the calibration set having samples in which the variability of the compounds that contribute to the signals used for the determination is modeled. Consequently, calibration models with different types of urine were used. We considered using 10 urine samples as calibration standards: 8 samples from healthy volunteers (nos. 1–8) and 2 samples from patients (nos. 14 and 15; see “Reagents, standards, and samples”). Independent variables in the PLS1 models were the sum of the intensities of all the ions detected during data acquisition with the non-separative method, and the dependent variables were the biomarker concentrations determined by the chromatographic method.

Some compounds showed overlapping mass spectra, and this allowed us to check the non-separative methodology under complex situations. For example, m/z 43 is the base peak of 2-butanone, 2-pentanone, and 2-heptanone; m/z 57, which is the base peak of 2-ethyl-1-hexanol, was also seen for 2-butanone. PLS1 models were obtained for each compound. Cross-validation (leave one out) was used to select the optimum number of PLS1 factors, and the Martens uncertainty criterion (included in the Unscrambler® statistical package v. 10.2) was used as the m/z variable selection technique. This eliminates all the variables whose regression coefficients have uncertainty values greater than their absolute value from the model. The optimum number of PLS1 factors and the root mean square error (RMSE) in the calibration and cross-validation steps, expressed as relative values considering the average of the concentration for each biomarker studied, are shown in Table 1. The m/z ratios selected in each model are shown in Table S-4 (Supplementary material). Repeatability (n = 5) was evaluated at two different levels using a urine sample and a urine sample spiked with the biomarkers (0.80–60 μg/L). The area under the curve of the profile signals corresponding to different m/z ratios was studied. The most abundant m/z ratios of all the compounds were chosen, with the exception of 2-pentanone and 2-heptanone, for which the second most abundant ratios were selected because their base peak coincided with that of 2-butanone (m/z = 43). The relative standard deviation (RSD) was lower than or equal to 10 % in all cases, except for m/z 58, whose value was equal to 20 % (Table S-2, Supplementary material). The multivariate detection limits in the urine samples were obtained following a strategy based on the variance of the concentration predicted by the model [26]. The values ranged between 0.16 and 21 μg/L (Table S-3, Supplementary material). The average concentration of these analytes in urine samples corresponding to lung cancer patients is in the high ppb range [4] which shows that the non-separative technique could be considered suitable for diagnostic purposes. After the first sample has been analyzed, this method allowed the measurement of 10 samples per hour.

Table 1 Relative prediction error in the calibration and cross-validation steps when the PLS1 models were used
Full size table

Sample analysis

The multivariate calibration models corresponding to the screening method (HS-PTV-MS) were used to predict the concentrations in a set of urine samples from five healthy volunteers (sample nos. 9–13) which had not been used in any calibration step. The non-separative method was also applied to the analysis of five samples (nos. 16–20) from patients with lung diseases; predicted concentrations of the studied biomarkers are shown in Table 2. Table S-5 (Supplementary material) offers a general overview of the situation of these patients.

Table 2 Predicted concentration (nmol analyte/mmol creatinine) of the biomarkers in urine samples from patients with the screening method
Full size table

In order to know whether there were significant differences in the concentration of the biomarkers between the group of patients with lung neoplasms (sample nos. 16–20) and the group of healthy individuals (sample nos. 9–13), the Mann-Whitney U test was implemented with the concentrations predicted with the non-separative method of the 10 samples of the external validation set. Significant differences (p < 0.05) were found for four of the five compounds present in the samples (2-butanone, 2-pentanone, 2-heptanone, and 2-ethyl-1-hexanol). Figure 2 shows the box plot representing the concentration of the analytes for both groups of samples.

Fig. 2
figure 2

Box plot graphics showing the normalized concentration distributions of the biomarkers found in the external validation samples with the non-separative method, distinguishing between healthy subjects and patients

Full size image

With a view to assigning the 10 external validation samples to the class of healthy individuals or patients with lung neoplasms, the Wilcoxon signed-rank test was used for the four biomarkers that afforded significant differences. Calculation of the median was done with the biomarker concentrations measured chromatographically in the eight samples from healthy individuals (sample nos. 1–8). This reference value was compared with the concentration of biomarkers predicted by the non-separative method in the 10 external validation samples. Table 3 shows the results obtained. In the case of the healthy individuals, all the samples were classified correctly since the median obtained for each biomarker (reference set) was not significantly lower (p > 0.05) than the concentration obtained for each sample. Regarding the patient samples, in all cases, two or three biomarkers with concentrations higher than those of the reference set were found. The content in 2-pentanone and 2-ethyl-1-hexanol was higher in patient samples than in the reference set in all cases.

Table 3 Results obtained in the classification of the external validation samples using the Wilcoxon signed-rank test
Full size table

Sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) for 2-butanone, 2-pentanone, 2-heptanone, and 2-ethyl-1-hexanol are shown in Table 4. PPV measures the likelihood that a positive result had been correctly predicted. NPV measures the likelihood that a negative result had been correctly predicted. Unlike sensitivity and specificity, which are inherent to the biomarker itself, PPV and NPV depend heavily on the population tested (prevalence). The samples that were assigned to the group of patients were measured chromatographically, and the concentration of the biomarkers found in the screening method was confirmed.

Table 4 Sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) of the four compounds for which significant differences in concentration between healthy individuals and patients were found
Full size table

Figure 3 shows the chromatograms obtained for three different subjects: one of a healthy individual and two of lung cancer patients (sample nos. 16 and 20). Since chromatographic analysis was performed in simultaneous SIM/scan mode, the chromatograms shown in Fig. 3a, c, e are the total ion chromatograms in scan mode and the chromatograms in Fig. 3b, d, f are also total ion chromatograms but in SIM mode. Differences in intensity can be seen in the chromatographic peaks of the compounds of interest between the two urine samples belonging to the patients and the urine of a healthy individual.

Fig. 3
figure 3

Total ion chromatogram in full-scan (a, c, and e) and SIM (b, d, and f) modes of urine samples from a healthy individual and two lung cancer patients (nos. 16 and 20) in which the numbered peaks correspond to 1 2-butanone, 2 2-pentanone, 3 pyrrole, 4 2-heptanone, and 5 2-ethyl-1-hexanol. I.S. endogenous internal standard (4-heptanone)

Full size image

Conclusions

The general analytical procedure proposed here consists of an initial analysis of a urine sample with the screening method (HS-PTV-MS). Following this, the Wilcoxon signed-rank test was used to compare the concentrations obtained with the non-separative method for the sample of interest with the median of the concentrations of a group of healthy individuals (reference group). When there were significant differences in some biomarker, the sample was assigned to the group of patients with lung cancer and was subjected to chromatographic analysis to confirm the results obtained previously. In case a false positive is obtained with the non-separative method due to other compounds with the same m/z ratios as the studied biomarkers, it will be ruled out with the separative method.

Electronic olfactometry based on mass spectrometry coupling with PLS1 calibration offers a reliable technique for the determination of volatile biomarkers in urine samples. Once the sample vial has been introduced into the system, all the other steps proceed automatically, such that the method can be considered suitable for routine analyses. The proposed methodology, with headspace sampling, does not require prior sample treatment, which reduces the errors associated with this step of the analysis. In addition, the method is rapid and simple, has good precision and sensitivity, and in view of the results could be considered a suitable first option for determining volatile biomarkers in a large number of patients. After screening, for confirmatory purposes, GC/MS may only be used in samples with concentrations significantly different from those seen in the healthy population.