Investigation of the impact of DESI sprayer geometry on its performance in imaging of biological tissue
Jocelyn Tillnera,b, James S.McKenziea, Emrys A.Jonesa,d, Abigail V. M. Spellera, James L.Walshc, Kirill A.Veselkova,Josephine Bunchb,e, Zoltan Takatsa*, Ian S.Gilmoreb*
aBiomolecular Medicine, Department of Surgery and Cancer, Faculty of Medicine, Imperial College London, Sir Alexander Fleming Building, South Kensington, London SW7 2AZ, United Kingdom;
bNiCE-MSI, National Physical Laboratory (NPL), Hampton Road, Teddington, Middlesex, TW110LW, United Kingdom;
cUniversity of Liverpool, Department of Electrical, Engineering and Electronics, Brownlow Hill, LiverpoolL69 3GJ, United Kingdom;
dCurrent address: Waters Corporation, Altrincham Rd, Wilmslow, Cheshire East SK9 4AX, United Kingdom;
eSchool of Pharmacy,University of Nottingham, University Park, Nottingham NG7 2RD,United Kingdom
Corresponding Authors:
*Zoltan Takats,
*Ian S. Gilmore, , fax: 020 86140573
*Josephine Bunch,
Abstract
In this study, the impact of sprayer design and geometry on performance in desorption electrospray ionisation mass spectrometry (DESI-MS) is assessed, as the sprayer is thought to be a major source of variability. Absolute intensity repeatability, spectral composition, and classification accuracy for biological tissues are considered. Marked differences in tissue analysis performance are seen between the commercially available and a lab-built sprayer. These are thought to be associated with the geometry of the solvent capillary and the resulting shape of the primary electrospray. Experiments with a sprayer with a fixed solvent capillary position show that capillary orientation has a crucial impact on tissue complex lipid signal and can lead to an almost complete loss of signal. Absolute intensity repeatability of the most intense fatty acid and phospholipid ion peaks is compared for five lab-built sprayers using pork liver sections. Repeatability ranges from 0.4 to 29.0 % for the individual sprayers. Between sprayers, the repeatability is 8 % for the phospholipid ions and 25 % for fatty acid ions. To assess the impact of sprayer variability on tissue classification using multivariate statistical tools, nine human colorectal adenocarcinoma sections are analysed with three lab-built sprayers and classification accuracy for adenocarcinoma versus the surrounding stroma is assessed. It ranges from 80.7 to 94.5 % between the three sprayers and is 86.5 % overall. The presented results confirm that the sprayer set-up needs to be closely controlled to obtain reliable data and a new sprayer set-up with a fixed solvent capillary geometry should be developed.
Desorption electrospray ionisation(DESI) is an ambient ionisation technique first describedin 2004 1. Since then it has rapidly gained popularity and found a large array of applications in environmental chemistry 2, analysis of pharmaceuticals3,4, forensics5, food chemistry 6, and histopathology throughmass spectrometric imaging (MSI) of tissues7,8.Its minimal sample preparation and ESI-like spectra make DESI a strong contender in the emerging field of molecular pathology9.Integration into everyday clinical workflows, however, necessitates a high level of robustness and reproducibility.
In a recent study,the repeatability and reproducibility of DESI were examined by imaging human oesophageal cancer tissue sections10. Both were calculated as the variation in normalised intensities of 25 lipid species in the mass range of 600 to 900 m/z. After careful optimisation of the set-up, a repeatability (same instrument, same operator) of 22% and(different instrument, different operator)of 20% could be achieved. However, a larger scale VAMAS inter-laboratory study conducted by the National Physical Laboratory 11 using rhodamine-coated slides and double-sided tape as reference standards, illustratedthat DESI-MS can show substantial inter-laboratory variability.Day-to-day absolute intensity variability ranged from 14 to 140 %for rhodamine B, and spectral constancy, i.e. the relative intensity between different regions of the mass spectrum, ranged from 9 to 83 % for the adhesive tape.
A large part of the variability in DESI is thought to be associated with the sprayer.Ionisation has been shown to occur through solubilisation of ions into a thin film of solvent on the sample surface, charge build-up on the surface, and subsequent formation of secondary droplets through the impact of the primary dropletsfacilitated by the nebulising gas12,13. The size, velocity, solvent composition and impact angle of the primary droplets is known to have a significant effect on desorption efficiency14. Changes in the nature of the primary electrospray can thus have a direct impact on ionisation efficiency.
The sprayer setup used for the initial DESI experiments was originally developed as a simple sonic spray ionisation source15, which usesa supersonic nebulising gasand no high voltage.An identical sprayer equipped with variable spray potential capability was used for subsequent electrosonic spray ionization (ESSI) studies16.Since the original publication of DESI, the design of the sprayer has remained virtually unchanged, apart from the commercially available sprayer, whichis marketed by ProsoliaInc. The majority of research laboratories still assemble their own DESI sources 11 and most studies on biological tissuehave been performed using in-house built sprayers (see supplementary Table S117-28).
While a number of studies have examined the influence of geometrical (distances, angles) and operating (flow rates, solvent composition, voltages) parameters on DESI performance 10,19,20,29-31, the impact of the sprayer design itself is poorly understood. Like for most relatively new methods, optimisation of DESI-MS experiments commonly relies on trial and error and empirical testing of parameters, without an understanding of the underlying mechanisms. In this study, we compare different sprayers built according to the same design in order to identify uncontrolled sprayer parameters and illuminate theireffect on the quality and reproducibility of DESI-MS data in a tissue imaging setting.
Experimental Section
Materials.Methanol and water were Chromasolv®LC-MS grade purchased from Sigma-Aldrich (St Louis, MO, USA).Dilauroylphosphatidyl choline was obtained from Avanti Polar Lipids (Alabaster, AL, USA). Rhodamine-coated slides were the same as thoseused in the VAMASinter-laboratory study11. Food grade pork liver was bought at a local supermarket (Sainsbury’s, London, UK). The liver was cut into small blocks and frozen on dry ice. Human colorectal samples were obtained from patients undergoing colorectal tumour resection at St. Mary’s Hospital, Paddington, London as part of a larger study thathas received favourable opinion from Cambridge East REC, and patients were consented for the study prior to surgical intervention. Both the pork liver and the colorectal samples were sectioned to a thickness of 10 µm using a Microm HM 550 cryomicrotome (Thermo Fisher Scientific Inc., Waltham, MA, USA) and sections were stored at -80°C until analysis.Sections were removed from the freezer 10 minutes prior to analysis and allowed to stabilise at room temperature.
DESI sprayers.Nine DESI sprayers were used in this study: seven conventional in-house built sprayers (denoted Sp1-7), an altered in-house built sprayer with a fixed capillary (Sp 8), and commercial sprayer (Sp 9) obtained from ProsoliaInc (Indianapolis, IN, USA). In-house built sprayers were composed of a 1/16” stainless steel tee (Swagelok, Kings Langley UK) with a fused silica gas capillary (363 µm outer diameter, 220 µm inner diameter), and a fused silica solvent capillary (150 µm outer diameter, 50 µm inner diameter) (both from SGE analytical, Milton Keynes, UK), held in place using NanoTightTM sleeves (Idex, Lake Forest, IL, USA). Observation during operation using an optical microscope showed that the initial alignment of the inner capillary was variable between sprayers and often changed once the gas was flowing. We therefore also produced a modified sprayer, where cyanoacrylate glue (Loctite, Henkel, Aachen, Germany) was used to attach the inner solvent capillary to the outer gas capillary, in order ensure consistent alignment (for optical images see supplementary Figure S1). For the experiments with the fixed sprayer, the sprayer holder of the 2D DESI stage was altered so that the sprayer was fully rotational around the central axis of the sprayer body and different orientations could be studied (see Figure 1).
Figure 1: Schematic of the DESI sprayer set-up and the geometric parameters examined. Solvent capillary positions are relative to the central axis of the gas capillary as viewed when looking down the sprayer at the sample.
Mass Spectrometry Instrumentation.Comparisons of the in-house built and commercial sprayer on pork liver were performed on an LTQ Orbitrapmass spectrometer (Thermo Fisher Scientific Inc., Waltham, MA, USA), equipped with a 2D sample stage from ProsoliaInc (Indianapolis, IN, USA) and on a Xevo G2-XS QToF (Waters Corporation, Milford, MA) also equipped with 2D sample stage from ProsoliaInc (Indianapolis, IN, USA). The latter was used when high scan rates were required.All comparisons of in-house built sprayerswere performed on an ExactiveOrbitrap mass spectrometer (Thermo Fisher Scientific Inc., Waltham, MA, USA) equipped with a customized 2D sample stage (Newmark Systems Incorporated, Rancho Santa Margarita, CA, USA). Details of instrumental parameters can be found in supplementaryTable 1.
Table 1: Experimental parameters for DESI-MS analyses.
Experiment / Lab-built vs. commercial sprayer experiment 1 - coated rhodamine and pork liver / Lab-built vs. commercial sprayer experiment 2 - coated rhodamine and pork liver / Comparison of 4 lab-built sprayers (pork liver); Comparison of 3 lab-built sprayers (colorectal samples)Mass spectrometer / LTQ Orbitrap (Thermo Fisher Scientific Inc., Waltham, MA, USA) / Xevo G2-XS QToF (Waters Corporation, Milford, MA) / Exactive (Thermo Fisher Scientific Inc., Waltham, MA, USA)
Mass spectrometric parameters / Scan time: 1s; sampling cone voltage: 50 V; source T: 100°C; source offset: 80 V / scan time: 0.033s; / Scan time: 1s;
2D sample stage / 2D sample stage from ProsoliaInc (Indianapolis, IN, USA) / 2D sample stage from ProsoliaInc (Indianapolis, IN, USA) / custom-built 2D sample stage ((Newmark Systems Incorporated, Rancho Santa Margarita, CA, USA)
Solvent composition / Methanol/ water, 90:10, v:v / Methanol/ water, 95:5, v:v / Methanol/ water, 95:5, v:v
Solvent flow rate / 1.5 µl/min / 1.5 µl/min / 1.5 µl/min
High voltage / 4.5 kV / 4.5 kV / 4.5 kV
Gas inlet pressure / 7 bars / 7 bars / 7 bars
Sprayer-to surface distance / 1.5 mm / 1.5 mm / 1.5 mm
Sprayer-to-inlet distance / variable; optimal distance for lab-built sprayer: 12 mm; optimal distance for commercial sprayer: 10 mm / 8 mm / 12 mm
Sprayer incidence angle / 70° / 70° / 70°
Optical equipment.For laser light sheet visualisation, a laser sheet was created using a 532nm continuous wave DPSS laser (CNI Optoelectronics Technology Co., Changchun, China) at 50 mW and a cylindrical lens (ThorlabsInc, Newton, NJ, USA).The mass spectrometer inlet was emulated with a stainless steel capillary of the same dimensions and a pumping rate to give the same pressure difference, although no voltage was applied to the simulated inlet capillary.
Sprayer tips and rhodamine traces were examined using an optical microscope (Leica, Wetzlar, Germany). Positioning of the fixed sprayer during DESI-MS analysis was monitored using a Dino-Lite digital microscope (Dino-Lite Europe, Naarden, Netherlands).
Data acquisition and data processing.The commercially available sprayer from Prosolia Inc. and the original, in-house built sprayer design (see Figure 2 A&B) were compared for their performance in DESI-MS using a neat lipid standard deposited on a glass slide, rhodamine-coated glass slides, and pork liver sections.Spectral datawith individual peak intensities was extracted directly from the instrument software.
To examine the impact of solvent capillary orientation, the sprayer with the fixed solvent capillary was mounted in four different orientations (see schematic Figure 1). The orientations tested were: forward, directed towards the mass spectrometer inlet, which was expected to be the best orientation; 45° left and 45° right, which corresponded with the positions observed for the sprayers used in the sprayer comparison; backwards, directed away from the mass spectrometer inlet. All orientations were tested on the same three pork liver sections with two line scans per orientation and scan direction. For repeatability measurements, the forward orientation and a left-to-right scanning pattern (standard operating conditions) were used and three line scans were acquired per section.
The resulting raw files were cropped using the recalibrate offline tool in Excalibur (v2.2 SP1.48; Thermo Fisher Scientific Inc.), converted into mzML files using the ProteoWizardmsConvertGUI (Vanderbilt University, Nashville, TN, USA), and then into an imzML image file using imzML Converter v1.3 32. The imzML file was imported into MatLab (R2014a; MathWorks, Natick, MA, USA) environment using an in-house written function. For the investigation of the impact of sprayer orientation on intensity, the mean spectrum for the standard orientation was calculated, and peaks were ranked with regard to their intensities. The top 5 fatty acid-related ions and top 10 phospholipid ions were selected. Identification was performed using the Lipid Maps database ( with an m/z tolerance of 5 ppm for accurate mass data. The presence and correct abundance of the 13C isotope peak was also verified. The list of selected peaks can be found in supplementary Table S2.
For the first sprayer comparison experiment, four in-house built sprayers (Sp1-4) were compared using 12 adjacent sections from the same pork liver sample analysedin randomised order in negative ion mode. Three sections were analysed with each sprayer. Five lines were scanned across each section at a speed of 200µm/s and with a line-to-line distance of 500µm to avoid overlap. For principal component analysis (PCA), the data werepre-processed including peak detection, alignment and quantile normalisation 33 using the mean spectrum, binning to m/z-dependent 10ppm bins, and log transformation for variance stabilisation 34 using theMatlab-based toolbox 34. Multivariate analyses and cross-validations were performed using the same toolbox.The full acquired m/z range from 150 to 1500 and two smaller mass ranges from 150 to 350 m/z and from 600 to 1000 m/z were analysed. The lower m/z range in DESI-MS is dominated by fatty acids, both from the tissue and from contamination of the solvent, as well as other solvent-related peaks. It is therefore subject to higher variability as it is strongly impacted by small changes in solvent quality or composition. The m/z region from 600 to 1000 contains phospholipids and triglycerides, which are commonly used for tissue identification and classification.
For repeatability measurements, the mean spectrum across all four conventional sprayers was calculated and the region from 50 to 950m/z was segmented into 50 m/z sections. From each section, one high intensity peak, one mid-intensity peak, and one low intensity peak relative to the highest peak in that region were selected and tentatively assigned an identity using the Lipid Maps database ( and the Metlin Scripps database ( A summary can be found in supplementary Table S3. Relative standard deviations (RSDs) of absolute intensity were calculated for all selected peaks.Peaks were grouped according to their relative intensity in the overall mean spectrum, with >40 % being classed as high relative abundance (5 peaks), 3 to 40 % medium relative abundance (10 peaks), 1 to 3 % low relative abundance (8 peaks), and <1% very low relative abundance (25 peaks) and RSDs were averaged for each group. The same peak list was used for the fixed capillary sprayer, though not all peaks could be detected.
To assess the relative magnitude of instrument-related and biologicalvariability of the data (adenocarcinoma versus adjacent stroma), three in-house built sprayers (Sp5-7) were used.Nine adjacent sections from the same sample of human colorectal tissue, containing regions of adenocarcinoma and neighbouring healthy tissue, were analysed in negative ion mode. Three sections were imaged with each sprayer, using a pixel size of 100x100µm. Sections were then HE stained and approximately 100 to 150 pixels per section and tissue type were annotated by a histopathologist. The raw data was converted into imzML format using imzML Converter v1.1.4.5i beta 35 and processed as for the PCA above.
Results and discussion
Comparison of the commercial and the in-house built sprayer.The performance of an in-house built sprayer (Sp7) and a commercially available sprayer from Prosolia Inc. were initially compared usinga neat lipid standard deposited on a glass slides. Similar performance was exhibited with regard to spectral intensity and composition(see supplementary Figure S2). However, their performance for analysis of a pork liver section was markedly different as shown in Figure 2 B. The in-house sprayer provided mass spectra with a high abundance of phospholipids in contrast to the commercial sprayer, which showed very low abundance of phospholipids and tissue-related fatty acids (e.g. m/z 303.23 - fatty acid 20:5). Surprisingly, subsequent repeat experiments with the commercial sprayer resulted in similar spectra to the in-house sprayers, as shown in Figure 2 C. We later propose a possible explanation for this extreme variability.
In order to deconstruct the sources of variance, the desorption rate and desorption footprint of both sprayers were studied on deposited rhodamine films as done previously by Green et al. 31 and Gurdak et al. 11. The desorption rate was measured astime dependence of spectral intensity and was found to be similar for both sprayers (see supplementary Figure S3). However, the desorption footprint left behind on the rhodamine surface was markedly different.The footprint of the in-house built sprayer showed a clear directionality, with rhodamine being re-deposited on the upper edge facing the mass spectrometer inlet capillary and a diffuse ring from the outer droplet of the electrospray forming mainly in that same direction, similar to that described by Green et al. 31.The footprint of the commercial sprayer showed a homogenous edge all around with a diffuse ring in all directions with only a slight sideways bias, which suggests a less directional and less focussed electrospray (see Figure 2B). Such differences could potentially have an impact on desorption efficiency of molecules from tissue and spatial resolution during DESI-MS imaging.
Laser lightsheet visualisation was used to investigate the primary electrospray of each sprayer and images are shown in Figure 2 D. It was observed that the droplet emission from the inner capillary is similar for both, with the emission concentrated towards the edge furthest from the spectrometer inlet (or, in this case, the simulated inlet) and closer to the surface (right edge in Figure 2 B). For the in-house sprayer, the inner capillary is set in the forward direction, which is presumably enhanced by the gas flow. Since the capillary is asymmetrically aligned there is a higher gas flow along the far edge. This has the effect of pushing in, or focusing, the edge of the emission cone. In contrast, the commercial source has a smaller diameter nebulising gas exit orifice and has a more symmetric alignment and consequently a more symmetric emission cone, but also a broader spray cone. The experiments were performed at relatively low solvent flow rate (1.5µl/min) and high gas pressure(7 bars). The spreading effect is likely to be enhanced if the solvent flow rate is increased or the gas pressure is lowered.