LAP-MALDI MS coupled with machine learning: an ambient mass spectrometry approach for high-throughput diagnostics

Abstract

Large-scale population screening for early and accurate detection of disease is a key objective for future diagnostics. Ideally, diagnostic tests that achieve this goal are also cost-effective, fast and easily adaptable to new diseases with the potential of multiplexing. Mass spectrometry (MS), particularly MALDI MS profiling, has been explored for many years in disease diagnostics, most successfully in clinical microbiology but less in early detection of diseases. Here, we present liquid atmospheric pressure (LAP)-MALDI MS profiling as a rapid, large-scale and cost-effective platform for disease analysis. Using this new platform, two different types of tests exemplify its potential in early disease diagnosis and response to therapy. First, it is shown that LAP-MALDI MS profiling detects bovine mastitis two days before its clinical manifestation with a sensitivity of up to 70% and a specificity of up to 100%. This highly accurate, pre-symptomatic detection is demonstrated by using a large set of milk samples collected weekly over six months from approximately 500 dairy cows. Second, the potential of LAP-MALDI MS in antimicrobial resistance (AMR) detection is shown by employing the same mass spectrometric setup and similarly simple sample preparation as for the early detection of mastitis.

Fig. 1. (a) Schematic of the LAP-MALDI source. (b) Close-up image of the LAP-MALDI sample irradiation, showing the ion transfer tube heated by a resistance wire opposite the sample plate with several liquid MALDI sample droplets. The irradiated sample droplet shows strong fluorescence. (c) LAP-MALDI MS spectrum (m/z 100–2000) obtained by acquiring data over one minute. (d) Workflow for the sample preparation and LAP-MALDI MS analysis of raw cow milk (TCA: trichloroacetic acid; ACN: acetonitrile; IPA: isopropanol; LSM: liquid support matrix). (e) Total ion chromatogram recorded for 50 separate milk samples analysed by LAP-MALDI MS.

Fig. 2. (a) LAP-MALDI mass spectrum of dairy milk, displaying the m/z range of 450–850. (b) Putative identification details for the most abundant lipid species detected in the mass spectrum shown in panel a (based on mass accuracy, literature and occurrence in the MCDB database at mcdb.ca). (c) LAP-MALDI mass spectrum of dairy milk, displaying the m/z range of 900–2000, where most proteins and peptides are detected as multiply charged ions. (d) Zero-charge mass spectrum obtained from the deconvolution of the mass spectrum shown in panel c. (e) m/z-vs.-charge ion intensity signal plot showing the charge train distributions for the ion signals of the mass spectrum shown in panel c.

Fig. 3. (a) Graphic representation (biplot) of the LDA model and the confusion matrix for its prediction of clinical mastitis. (b) Graphic representation (biplot) of the LDA model and the confusion matrix for its prediction of pre-clinical mastitis. The observations and the multivariate means of each group are represented as points on the biplot. They are expressed in terms of the first two canonical variables. The point corresponding to each multivariate mean is denoted by a plus (“+”) marker. A 95% confidence level circle is plotted for each mean. If two groups differ significantly, the confidence circles tend not to intersect. A circle denoting a 50% contour is also plotted for each group. This depicts a region in the space of the first two canonical variables that contains approximately 50% of the observations, assuming normality. (c and d). Normalised signal intensities of the AMX mass bin at m/z 724.5, which contains the [M + 6H]⁶⁺ ion species of isracidin-like peptide, using the data from the classification of clinical (c) and pre-clinical (d) mastitis. * p-value: 8.04 × 10⁻¹⁸; ** p-value: 4.84 × 10⁻³⁴, fold change 5.81; *** p-value: 3.43 × 10⁻¹⁷; **** p-value: 4.09 × 10⁻²⁶, fold change: 2.12. See text for more details.

Fig. 4. (a) LAP-MALDI mass spectrum of bovine milk (m/z 300–400). Sodiated lactose [M + Na]⁺ is detected at m/z 365.11 and potassiated lactose [M + K]⁺ at m/z 381.08. (b) LAP-MALDI mass spectrum of ampicillin (m/z 350.12; [M + H]⁺), spiked into milk at a concentration of 50 μg mL⁻¹ (c) LAP-MALDI mass spectrum of ampicillin (m/z 350.12; [M + H]⁺) in water at a concentration of 50 μg mL⁻¹. (d) LAP-MALDI mass spectrum of decarboxylated hydrolysed ampicillin (m/z 324.14; [M + H]⁺), obtained after 120 min incubation of ampicillin in milk (50 μg mL⁻¹) with 3–6 units of penicillinase. (e) LAP-MALDI mass spectrum of decarboxylated hydrolysed ampicillin (m/z 324.14; [M + H]⁺), obtained after 120 min incubation of ampicillin in water (50 μg mL⁻¹) with 3–6 units of penicillinase.

Fig. 5. LAP-MALDI MS signal intensity ratio of the protonated decarboxylated hydrolysed ampicillin (m/z 324.14) and the protonated intact ampicillin (m/z 350.12) for various amounts of penicillinase spiked into raw healthy milk (first 8 columns/sample groups) and for clinical mastitis samples (last column/sample group). The clinical mastitis group consisted of 60 samples of raw mastitis milk analysed in duplicate. Each dot represents one technical replicate. Both duplicates of three samples (in red) show elevated levels of lactamase activity, clearly separating them from the bulk of the other mastitis cases. Routine bacteriological testing confirmed that one of the 3 samples contained E. coli bacteria resistant to ampicillin. The other two samples were not tested by routine bacteriological testing. Bacteriological testing of dairy milk is typically bacterium-specific and not all bacteria are routinely tested for AMR. Thus, the other two samples identified by LAP-MALDI with elevated lactamase activity can fill this gap and show the potential for multiplex testing of AMR. SI: signal intensity.