Supplementary Information: A Taguchi Design of Experiments Approach for Untargeted Metabolomics Sample Preparation Optimization

The content is available as a PDF (1.5 MB).

SUPPLEMENTARY METHODS

Sample growth and preparation

To generate a large and consistent population of PD1074 Caenorhabditis elegans, animals were fed Escherichia coli IBAT RM,¹ and cultures were grown on large-scale culture plates (LSCP) and aliquoted according to Shaver et al.² Samples used for the Taguchi L'9 array (run 1–9) were aliquots from one biological LSCP removing any differences due to sample growth conditions between runs. Samples extracted using the experimentally determined optimum conditions were generated from a separate LSCP. Samples extracted for nuclear magnetic resonance (NMR) spectroscopy were similarly generated from a separate individual LSCP to demonstrate the validity of the experiment using an orthogonal analytical approach. The extraction protocol described herein is shown in Supplementary Figure 1. Frozen C. elegans samples containing 50,000 worms each in 2.0-mL screwcap microcentrifuge tubes were retrieved from a -80°C freezer, covered with Fisher brand aluminum foil, and punctured with a small gauge needle. Samples were lyophilized for 24 h or until complete dryness. Three 3.5-mm glass beads (BioSpec Products, Inc., Cat. No. 11079135) and approximately 200 µL volume of 1.0-mm Zirconia beads (BioSpec Cat. No. 11079110zx) were added to each sample tube. Samples were homogenized using a FastPrep-96TM instrument (MP Biomedicals) equipped with ConeFlex LegacyTM and CoolPrepTM adapters for 24 x 2.0 mL samples at 1,800 revolutions per minute for 90 s. Homogenization was repeated three times for a total of 270 s. Samples were placed on dry ice between rounds to minimize the overheating of the lyophilized powder and breakdown of heat-sensitive compounds.

Supplementary Figure 1. Taguchi DOE workflow.

Ultra-high performance liquid chromatography analysis

The mass spectrometer was operated in positive and negative ionization mode at 240,000 resolution, automatic gain control target of 1e5, and a maximum injection time of 150 ms. Non-polar and polar extracts were analyzed by reversed-phase (RP) and hydrophilic interactive liquid chromatography (HILIC), respectively. An m/z scan range of 70 to 1,050 and 150 to 2,000 was used for HILIC and RP, respectively. RP chromatography was performed with a Thermo Scientific AccucoreTM C30, 2.1 x 150 mm, 2.6-µm particle column. Mobile phase A was water/acetonitrile (40:60 v/v), and mobile phase B was acetonitrile/isopropanol (10:90 v/v). Both mobile phases included 10 mM ammonium formate and 0.1% formic acid additives to improve peak shape and ionization efficiency. HILIC chromatography was performed with an ACQUITY BEH Amide, 2.1 x 150 mm, 1.7-µm particle column. Mobile phase A was water/acetonitrile (80:20 v/v) with 10 mM ammonium formate, and mobile phase B was acetonitrile (100%). Both mobile phases included 0.1% formic acid. Chromatographic gradients are described in detail below. The column temperature was 55°C and 40°C for RP and HILIC, respectively, and samples were maintained at 5°C in the autosampler.

RP chromatography methods

Non-polar extracts (i) were separated using a Vanquish (ThermoFisher Scientific) fitted with a ThermoFisher Scientific Accucore C30 UPLC RP column (2.1 x 150 mm, 2.6-µm particle size). The compounds were eluted with the following gradient: 60:40 acetonitrile:water (ACN:H₂O) with 10 mM ammonium formate and 0.1% formic acid (mobile phase A) and 90:10 isopropanol:acetonitrile with 10 mM ammonium formate and 0.1% formic acid (mobile phase B) using the following gradient program: -5.0 20% B; 0.0 min 20% B; 1.0 min 60% B; 5.0 min 70% B; 5.5 min 85% B; 8.0 min 90% B; 8.2 to 10.5 min 100% B; and 10.7 to 12.0 min 20% B. A curve five value was set for -5.0 and 0.0 minutes and a curve six for the remainder of the gradient. The flow rate was set at 0.400 mL min^-1. The column temperature was set to 50°C, and the injection volume was 2 µL.

HILIC methods

Polar extracts (ii) were separated using a Vanquish (ThermoFisher Scientific) fitted with a Waters Acquity UPLC BEH Amide column (2.1 x 150 mm, 1.7-µm particle size). The compounds were eluted with the following gradient: 80:20 H₂O:ACN with 10 mM ammonium formate and 0.1% formic acid (mobile phase A) and 100% ACN with 0.1% formic acid (mobile phase B) using the following gradient program: -5.0 min 95% B; 0.0 to 0.5 min 95% B; 8.0 to 9.4 min 40% B; and 9.5 to 11.0 min 95% B. A curve five value was set for -5.0 and 0.0 minutes, a curve six at 0.5 min, curve seven at 8.0 min, and a curve six for the remainder of the gradient. The flow rate was set at 0.400 mL min^-1. The column temperature was set to 40°C, and the injection volume was 2 µL.

Mass spectrometer settings and methods

A Q Exactive HF (ThermoFisher Scientific) equipped with a HESI ion source was used for all mass spectrometry data collection. The mass spectrometer was run in full mass spectrometry mode at a resolution of 240,000 (at m/z 200) for the duration of the chromatographic gradient. An automatic gain control target of 1e5 was set, with a maximum injection time of 150 ms. A tune file with the following source conditions was used for positive and negative mode: spray voltage (+) 3,500, spray volage (-) 2,500, capillary temperature: 262.50°C, sheath gas: 50, aux gas: 12.50, spare gas: 2.50, max spray current: 100, probe heater temperature: 425.0°C, and S-lens radio frequency level: 50. An m/z scan range of 70 to 1,050 and 150 to 2,000 was used for HILIC and RP, respectively. Calibration was conducted using ThermoFisher Pierc Negative Ion Calibration Solution and Pierc LTQ Velos ESI Positive Ion Calibration Solution prior to the collection of negative and positive mode data, respectively.

NMR spectroscopy data collection and processing

A separate batch of lyophilized C. elegans PD1074 samples were extracted for NMR analysis using the same extraction conditions as liquid chromatography–mass spectrometry, excluding reconstitution solvent ((ref?)). Non-polar and polar samples were reconstituted in CDCl3 (99.96%; Cambridge Isotope Laboratories, Inc.) and D2O (99%, Cambridge Isotope Laboratories, Inc.) in a 100-mM sodium phosphate–buffered solution with 0.11 mM sodium 2,2-dimethyl-2-silapentane-5-sulfonate (DSS-D6; 98%; Cambridge Isotope Laboratories, Inc.), respectively. The final volume of 45 μL of each sample was transferred into 1.7-mm NMR tubes (Bruker SampleJet). 1H NMR data were acquired using the noesypr1d (one-dimensional, 1D) and jresgpprqf (two-dimensional, 2D, J-resolved) pulse sequences on a NEO 800 MHz Bruker NMR spectrometer equipped with a 1.7-mm TCI CryoProbe. The autosampler (Bruker SampleJet) was kept at 6°C during all data acquisition. A total of 32,768 complex datapoints were collected using 32 scans (and four additional dummy scans) for the data acquisition and 8,192 complex points, and 40 increments were collected using 8 scans (and 4 additional dummy scans) for the J-resolved data acquisition. Spectral width was set to 12 ppm (1D) and 16 ppm for f1 and 77.98 Hz for f2 (J-resolved). All data were processed using zero filling to double the acquisition complex points, and their phase was corrected automatically; only fine phasing was done for some of the samples manually. An exponential function (LB = 1 Hz) was applied to the 1D data, and a sine bell function (LB = 1Hz) was applied to the 2D J-resolved data. Chemical shift reference was done using the DSS peak at 0 ppm. 2D J-resolved data were Fourier transformed, tilted, and symmetrized (as J-resolved spectrum) using Bruker’s commands “tilt” and “symm.” 2D J-resolved data was converted into horizontal positive projections using the topspin function “rhpp,” and these 1D projections of the 2D data were used for downstream comparisons. After data acquisition and processing, the total area under the curve was integrated from 0.5 to 8.0 ppm and 0.9 to 8.7 ppm for non-polar and polar J-resolved data, respectively. Chemical shift regions corresponding to artifacts or solvents were excluded from the integration. Among those, the region between 4.23 to 5.010 ppm (water) and 3.34 to 3.39 ppm (methanol) were excluded from the polar samples, and the region between 7.12 to 7.44 ppm (chloroform and satellite peaks), 3.48 to 3.55 ppm (methanol), and 1.22 to 1.33 ppm (water) from the non-polar samples were excluded.

SUPPLEMENTARY TABLES AND FIGURES

Supplementary Table 1. Selected Factors and Respective Levels.

		Factors
		(A) Extraction solvent (i and ii)	(B) Volume (mL)	(C) Time (h)	(D) Reconstitution solvent (i and ii)
Levels	1	(i) IPA (ii) 80/20 MeOH/H2O	0.5	0.5	(i) IPA ii) H2O
	2	(i) MeOH (ii) 80/20 MeOH/ H2O	1	2	(i) 60/40 H2O /ACN (ii) MeOH
	3	(i) 1:1 1-Butanol: MeOH (ii) 80/20 MeOH/ H2O	3	12	(i) 50/50 IPA/ACN (ii) 80/20 MeOH/ H2O

Supplementary Table 2. MZmine 2.53 Pre-processing Steps, Input Parameters, and Set Values Used for Liquid Chromatography–Mass Spectrometry Data Listed in the Order of Execution.

Pre-processing step (in order of execution)	Input parameters	Set value
Mass detection	mass detector	exact mass
	noise level	5.00E+03
ADAP chromatogram builder	min group size in # of scans	8
	group intensity threshold	1.00E+05
	min highest intensity	1.00E+05
	mz tolerance	0.005 or 5 ppm
Chromatogram deconvolution	algorithm	wavelets (ADAP)
	S/N threshold	3
	S/N estimator	intensity window SN
	min feature height	100,000
	coefficient/area threshold	110
	peak duration range	0.03 to 0.5
	RT wavelet range	0.01 to 0.03
	m/z center calculation	MEDIAN
Isotopic peak grouper	m/z tolerance	0.005 or 5 ppm
	RT tolerance	0.08
	monotonic shape	FALSE
	maximum charge	2
	representative isotope	most intense
RANSAC aligner	m/z tolerance	0.005 or 5 ppm
	RT tolerance	0.08
	RT tolerance after correction	0.03
	RANSAC iterations	0
	minimum number of points	40%
	threshold value	4
	linear model	FALSE
	require same charge state	FALSE
Feature list rows filter	minimum peaks in an isotope pattern	1
	m/z	148.5076 to 1,500
Duplicate filter	filter mode	NEW AVERAGE
	m/z tolerance	0.005 or 5 ppm
	RT tolerance	0.1
	require same identification	FALSE

Supplementary Table 3. Taguchi orthogonal experimental results for the number of extracted RP(+) and HILIC(+) mass spectral features and total ion count (TIC), as well as the respective calculated signal-to-noise (S/N) values. Opt. indicates the conditions for the final optimized parameters.

						Results
Run	No. Features		S/N No. Features		TIC	S/N TIC
	RP	HILIC	RP	HILIC	RP
(E+09)	HILIC
(E+09)	RP	HILIC
1	981	3384	59.82	70.58	2.29	3.76	187	191.47
2	962	4595	59.56	73.23	1.7	5.28	184	194.44
3	1524	1300	63.65	62.18	2.84	1.39	189	182.51
4	1428	1762	62.74	64.79	4.98	2.06	193	186.15
5	1540	1365	63.72	62.53	5.18	1.72	194	184.70
6	1989	620	65.95	55.79	4.37	4.96	193	173.77
7	2067	2247	66.29	67.00	6.19	2.72	196	188.60
8	1997	1891	65.97	65.33	4.37	2.28	193	186.77
9	2315	818	67.29	58.18	4.92	1.09	194	180.71
Opt.	4754	385	73.53	51.7	7.09	0.704	197	176.9

Supplementary Table 4. ANOVA for the number of mass spectral features in reverse phase (RP) positive (+) mode.

Number of mass spectral features - Analysis of variance (RP+)
Source	Degree of Freedom (DF)	Sum of Squares (SS)	Variance (MS)	Percentage Contribution
A) Extr. Solv. (i & ii)	2	45.8	22.9	73.3%
B) Volume	2	13.7	6.8	21.9%
C) Time	2	2.8	1.4	4.4%
D) Recon. Solv. (i & ii)	2	0.4	0.1	0.4%
Total	8	62.6	31.2	100.0%

Supplementary Table 5. Two-way Analysis of Variance of RP and HILIC Signal-To-Noise Values Based on the Number of Liquid Chromatography–Mass Spectrometry Features in Positive Mode.

term	df	sumsq	meansq
RP_setup	8	60.21	7.53
Extraction	1	1.18	1.18
RP_setup:Extraction	8	247.70	30.96

Supplementary Figure 2. Principal components analysis (PCA) 2D scores plot of the nine individual RP(-) Taguchi method experiments. Each color represents a Taguchi run. Color-filled ellipses represent the 95% confidence interval (CI). Run number (1-9) is shown adjacent to the appropriate set of data points. Detailed information on these experiments can be found in Table 2.

Supplementary Figure 3. Principal components analysis (PCA) two-dimensional scores plot of the nine individual HILIC(+) Taguchi method experiments. Each color represents a Taguchi run. Color-filled ellipses represent the 95% CI. Run number (1–9) is shown adjacent to the appropriate set of data points. Detailed information on these experiments can be found in Table 2.

Supplementary Figure 4. Principal components analysis (PCA) two-dimensional scores plot of the nine individual HILIC(-) Taguchi method experiments. Each color represents a Taguchi run. Color-filled ellipses represent the 95% CI. Run number (1–9) is shown adjacent to the appropriate set of data points. Detailed information on these experiments can be found in Table 2.

Supplementary Figure 5. Negative mode: Number of features main effect box plots for (A) extraction solvent, (B) extraction volume, (C) extraction time, and (D) liquid chromatography–mass spectrometry reconstitution solvent. Each boxplot has three data points overlayed representing each individual sample’s contribution to the overall boxplot. On each box, the red asterisk indicates the mean, and the bottom and top edges of the box indicate the 25th and 75th percentiles, respectively. The whiskers extend to the most extreme data points not considered outliers. Y-axis displays larger-the-better signal to noise with the grand mean signal to noise for each chromatography represented by a black dashed line. RP and HILIC data are represented in blue and orange, respectively, in each plot. Levels for each factor (A–D) are represented on the x-axis as 1, 2, and 3. Detailed information on these values can be found in Table 1.

Supplementary Figure 6. Positive mode: Total ion count main effect box plots for (A) extraction solvent, (B) extraction volume, (C) extraction time, and (D) liquid chromatography–mass spectrometry reconstitution solvent. Each boxplot has three data points overlayed representing each individual sample’s contribution to the overall boxplot. On each box, the red asterisk indicates the mean, and the bottom and top edges of the box indicate the 25th and 75th percentiles, respectively. The whiskers extend to the most extreme data points not considered outliers. Y-axis displays larger-the-better signal to noise with the grand mean signal to noise for each chromatography represented by a black dashed line. RP and HILIC data are represented in blue and orange, respectively, in each plot. Levels for each factor (A–D) are represented on the x-axis as 1, 2, and 3. Detailed information on these values can be found in Table 1.

Supplementary Figure 7. Negative mode: Total ion count main effect box plots for (A) extraction solvent, (B) extraction volume, (C) extraction time, and (D) liquid chromatography–mass spectrometry reconstitution solvent. Each boxplot has three data points overlayed representing each individual samples contribution to the overall boxplot. On each box, the red asterisk indicates the mean, and the bottom and top edges of the box indicate the 25th and 75th percentiles, respectively. The whiskers extend to the most extreme data points not considered outliers. Y-axis displays larger-the-better signal to noise with the grand mean signal to noise for each chromatography represented by a black dashed line. RP and HILIC data are represented in blue and orange, respectively, in each plot. Levels for each factor (A–D) are represented on the x-axis as 1, 2, and 3. Detailed information on these values can be found in Table 1.

Supplementary Figure 8. Percent contribution results calculated from the one-way analysis of variance. Percent contribution was calculated by dividing the variance/mean squares for each factor (A–D) by the total variance. (A) Percent contribution for RP/HILIC chromatography in positive and negative mode calculated using the number of features (No. Features) as input. (B) Percent contribution for RP/HILIC chromatography in positive and negative mode calculated using the total ion count (TIC) as input. Yellow highlighted boxes indicate the factor with the highest percent contribution for each analysis.

Supplementary Figure 9. Number of features for all nine experimental runs (1–9) and the optimized extraction method (Opt.) for RP (-) and HILIC (-). RP is represented in blue and HILIC in orange. Individual data points for the three replicates of each run are shown as a scatter plot, with a line connecting the average value of each run.

Supplementary Figure 10. Total ion count (TIC) for all nine experimental runs (1–9) and the optimized extraction method (Opt.) for RP (+) and HILIC (+). RP is represented in blue and HILIC in orange. Individual data points for the three replicates of each run are shown as a scatter plot, with a line connecting the average value of each run.

Supplementary Figure 11. Total ion count (TIC) for all nine experimental runs (1–9) and the optimized extraction method (Opt.) for RP (-) and HILIC (-). RP is represented in blue and HILIC in orange. Individual data points for the three replicates of each run are shown as a scatter plot, with a line connecting the average value of each run.

Supplementary Figure 12. Integrated area under the curve of J-resolved NMR for all nine experimental runs (1–9) and the optimized extraction method (Opt.) for polar and non-polar NMR extracts. Extracts were reconstituted in D₂O and CDCl₃ for polar and non-polar samples, respectively. Non-polar (CDCl₃) data is represented in blue and polar (D₂O) in orange. Individual data points for the three replicates of each run are shown as a scatter plot, with a line connecting the average value of each run.

Graphical abstract. A Taguchi design of experiments approach was used to optimize a sequential liquid chromatography–mass spectrometry extraction using an L’9 (4 factor/3 level) orthogonal array. Factors (A) solvent, (B) volume, (C) time, and (D) reconstitution solvent were investigated.