Aspergillus Species Discrimination Using a Gas Sensor Array

Abstract

The efficiency of electronic noses in detecting and identifying microorganisms has been proven by several studies. Since volatile compounds change with the growth of colonies, the identification of strains is highly dependent on the growing conditions. In this paper, the effects of growth were investigated with different species of Aspergillus, which is one of the most studied microorganisms because of its implications in environmental and food safety. For this purpose, we used an electronic nose previously utilized for volatilome detection applications and based on eight porphyrins-functionalized quartz microbalances. The volatile organic compounds (VOCs) released by cultured fungi were measured at 3, 5, and 10 days after the incubation. The signals from the sensors showed that the pattern of VOCs evolve with time. In particular, the separation between the three studied strains progressively decreases with time. The three strains could still be identified despite the influence of culture time. Linear Discriminant Analysis (LDA) showed an overall accuracy of 88% and 71% in the training and test sets, respectively. These results indicate that the presence of microorganisms is detectable with respect to background, however, the difference between the strains changes with the incubation time.

Keywords: Aspergillus species; electronic nose; porphyrins; volatile organic compounds.

Figure 1

Picture showing the experimental system set-up under a biohazard hood. (A) Petri dish covered with customized lid suitable to ensure a stable headspace; (B) filter used to block spore migration from Aspergillus culture to sensors; (C) the electronic nose; (D) the CaCl₂ trap.

Figure 2

Principal Component Analysis (PCA) of characterization data. (A) Scores plot, (B) bi-plot which shows the loadings of the sensors.

Figure 3

Signal recorded by the Zn-TBPP sensor. It shows the sensor behavior for the sequence of exposures to the headspace of A. niger, A. flavus and A. fumigatus in order. Arrows indicates the frequency shift considered as the sensor response and used in the data analysis. The response time is 135 s.

Figure 4

Distribution of the sensor responses to uncontaminated culture media and media inoculated with fungi. Each group contains data obtained at days 3, 5, and 10 after the first inoculation. The distribution of the responses is represented by a box-plot calculated with the Matlab embedded function. In each box, the central mark is the median, the edges are the 25th and 75th percentiles, the whiskers extend to the most extreme data points not considered outliers. Outliers, labeled with a cross, are plotted individually.

Figure 5

The distributions of the responses of sensors in the nine groups are represented by box-plots. Groups are labeled with the initials of the strain (Ni, Flu, and Fu) and a number (3, 5, 10) indicating the days since inoculation. For a description of the box-plot details, see the caption of Figure 4.

Figure 6

Sensors’ data related to Aspergillus headspaces were normalized and projected onto the scores plot of the PCA calculated with the characterization data and shown in Figure 2. Microorganisms’ data form a close cluster between aromatics and polar compounds.

Figure 7

Plot of the logarithm of the probability of null hypothesis (p-value) returned by a non-parametric Kruskal–Wallis rank sum. The p-value is related to the largest separation between at least two of the nine groups.

Figure 8

PCA of Aspergillus headspace data. (A) scores plot, (B) bi-plot where the loadings of sensors are indicated. In Figure 8A, Aspergillus species are marked with the same symbol whose colors identifies the time after incubation.

Figure 9

(A) Plot of accuracy of the Linear Discriminant Analysis (LDA) classifier for 100 random partitions of the data in training and test sets. (B) Frequency of occurrence of the multivariate variable made by the accuracy in training and test sets. The ellipse in Figure 9A shows the covariance matrix of the bi-variate normal distribution. The mean value, corresponding to the center of the covariance matrix, was chosen to represent the results.