HPLC-Electrospray Ionization-Mass Spectrometry (LC-ESI-MS) Optimization by High-Performance Design of Experiments (DOE) for Astrocyte Glutamine Measurement

Abstract

The amino acid glutamine (Gln) is a likely source of energy in the brain during neuro-glucopenia. Effects of glucose deficiency on astrocyte Gln homeostasis remain unclear as analytical tools of requisite sensitivity for quantification of intracellular levels of this molecule are not currently available. Here, a primary hypothalamic astrocyte culture model was used in conjunction with Design of Experiments (DOE)-refined HPLC-electrospray ionization-mass spectrometry (LC-ESI-MS) methodology to investigate the hypothesis that glucoprivation alters astrocyte Gln content in a sex-specific manner. Critical mass spectrometric parameters for Gln derivative chromatographic response were identified by comparing performance of Central Composite Design, Box-Behnken Design, and Optimal Design (OD)-A, -D, -I, -Distance, and -Modified Distance DOE models. Outcomes showed that the OD-A generated response was superior relative to other design outcomes. Forecasted surface plot critical mass spectrometric parameters were maximized by OD-A, OD-Distance, and OD-Modified Distance designs. OD-A produced a high-performance method that yielded experimental run and forecasted surface plot maximal responses. Optimized mass spectrometric analysis of male versus female astrocyte Gln content provides novel evidence that glucoprivation significantly depletes this amino acid in female, but not male, and that this sex-specific response may involve differential sensitivity to estrogen receptor signaling. This technological advance will facilitate efforts to ascertain how distinctive physiological and pathophysiological stimuli impact astrocyte Gln metabolism in each sex.

Introduction:

In the brain, the amino acid glutamine (Gln) is utilized in metabolic pathways that support neuro-metabolic stability and neurotransmitter signaling [1–4]. The Gln-glutamate cycle (GGC), involving amino acid exchange between astrocytes and neurons, is a critical regulator of glutaminergic neurotransmission [5]. Astrocyte-synthesized Gln is trafficked to neurons via dedicated transporter function, and it converted in that cell compartment to glutamate (Glu) by glutaminase enzyme action [6]. Neuronal Glu is packaged within synaptic neurotransmitter vesicles for release into a synapse by exocytosis, or, alternatively, is utilized as a precursor to other signaling molecules, e.g. γ-aminobutyric acid and aspartate. In astrocytes, Gln is metabolized to Glu to support tricarboxylic acid (TCA) cycle function via the pyruvate recycling pathway [7–9]. Complete glutamine oxidation in the TCA cycle involves entrance and exit of Glu in the form of alpha-ketoglutarate and malate, respectively, followed by processing of pyruvate to acetyl-CoA [10]. At present, it is unclear how astrocyte Gln homeostasis may be impacted by various physiological and pathophysiological stimuli, as current analytical methods lack sufficient sensitivity for measurement of astrocyte Gln. Gln and Glu are potential cerebral energy substrates during glucose deficiency as their total brain levels decline during hypoglycemia [11, 12]. Recent studies in our laboratory show that hypoglycemia causes dissimilar adjustments in expression of key pyruvate recycling pathway enzyme proteins, e.g. glutaminase and malic enzyme-1, in brain tissue samples containing multiple brain cell types [Uddin et al., 2020]. Current research utilized an established primary astrocyte cell culture model [Ibrahim et al., 2020], in conjunction with optimized high performance liquid chromatography-electrospray ionization-mass spectrometry (LC-ESI-MS) methods to investigate the hypothesis that glucoprivation alters astrocyte Gln content in a sex-dimorphic manner.

There is increasing recognition of the value of Design of Experiments (DOE) implementation for optimization of mass spectrometric response variables [13, 14]. Response Surface Methodology (RSM) is a broadly-used collection of mathematical and statistical modeling tools for assessment of effects of multiple factors on one or more dependent variables, based upon the fit of a polynomial equation to experimental data, to simultaneous optimize those factors to achieve best system performance. Critical parameters identified by factorial designs, such as Central Composite Design (CCD; Box-Wilson Central Composite Design), Box-Behnken Design (BBD); and Optimal Designs (OD)-A (OD-A), -D (OD-D), -I (OD-I), -Distance (OD-DIST), and -Modified Distance (OD-MOD), can be refined by RSM [15, 16]. Algorithms produced by ODs referenced above are respectively distinguished by minimization of average variance of polynomial coefficients (OD-A), enhancement of factor effect estimations through minimization of coefficient confidence ellipsoid volume (OD-D), reduction of average prediction variance over the experimental region (OD-I), addition of points from a starting vertex that maximize Euclidean distance from original points (OD-DIST), and augmentation of design matrix rank through point selection (OD-MOD) to enable estimation of all model terms. Our previous studies showed that a CCD-generated quadratic equation improved fluorenylmethoxycarbonyl (FMOC)-Gln derivative mass chromatographic response [17]. This sought to determine if inherent outcomes of BBD and/or OD designs can enhance Gln chromatographic response compared to that achieved by CCD, thereby enhancing screening speed, efficiency, and cost-effectiveness by decreasing number of experimental runs while maintaining minimal process variability. Here, Design Expert software-generated experimental runs were created for BBD, OD-A, OD-D, OD-I, OD-DIST, and OD-MOD screening designs. Maximizing mass spectrometric parameter inputs identified by experimental runs were used to generate response surface plots to forecast percent relative response of a High-Performance DOE method for improvement of analytical sensitivity. A crucial objective of this work was to obtain proof-of-principle that analytical sensitivity improvements guided by natural outcomes of these designs is sufficient to permit accurate and precise quantification of Gln in primary male versus female hypothalmic astrocyte cultures.

Materials and Methods:

Reagents:

Ammonium acetate was purchased from J.T. Baker, Radnor, PA. Sodium bicarbonate was obtained from Spectrum Chemicals Mfg. Corp, New Brunswick, NJ. LC-MS grade acetonitrile was acquired from VWR BDH Chemicals, Radnor, PA. Methanol was purchased from ThermoFisher Scientific, Waltham, MA. 1-Adamantanamine hydrochloride 99% (AD), 9-fluorenylmethyl chloroformate 98+% (FMOC), L-Glutamic acid 5-benzyl ester 99% (IS), L-Glutamic acid 5-benzyl ester 99%, and L-glutamine 99% (Gln) were obtained from Alfa Aesar, Haverhill, MA. Inserts, 350 μL, small volume, flat bottom, borosilicate glass, 6 × 31 mm, AQ Brand were purchased from Microsolv Technology Corporation, Leland, NC. National C5000–1W 2 mL Clear Glass ID Surestop vials were purchased from ThermoFisherSci.

Pre-column FMOC derivatization of Gln:

A 0.4 mg/mL Gln stock solution was prepared in a 1.5 mL plastic microcentrifuge tube with ultrapure water, as previously described [18]. FMOC (1.5 mg) was dissolved in acetonitrile (1 mL), then vortexed to yield a clear solution. AD (1 mg) and IS (0.2 mg) were separately dissolved in 50% acetonitrile (1 mL). Accurately-weighed sodium carbonate (10 mg) was transferred to ultrapure water (200 mL); after stirring to produce a clear solution, sodium bicarbonate solids (80–100 mg) were added under magnetic stirring to attain pH 9.0. Gln reaction with FMOC was initiated by combining Gln stock (100 μL), IS stock (50 μL), carbonate buffer, pH 9.0 (100 μL), and FMOC (100 μL); the mixture was then vortexed (30 s) and allowed to stand at 25 °C (40 min). After addition of AD (50 μL), contents were vortexed (30 s), allowed to stand at 25 °C (5 min), and centrifuged to yield a clear supernatant, which was transferred to 350 μL inserts for placement in 2 mL Surestop vials positioned in an autosampler tray.

Liquid chromatography-mass spectrometry system:

Sample analysis was performed using a system comprised of a UHPLC Vanquish pump, Vanquish Autosampler, Vanquish UHPLC+ column compartment, and ISQ EC mass spectrometer (ThermoFisherSci.), utilizing ThermoFisherScientific Dionex Chromeleon 7 Chromatography Data System software, as described [17, 18]. A C18 column (4.6 mm ID X 100 mm L, 5 μm, 120 Å; Acclaim 120; ThermoFisherSci.) was used with a 0.25 mL/min. flow rate and 0.01–1 μL injection volume; the autosampler needle was washed with 10% (v/v) methanol (10 s) after each sample injection. Mobile phases A and B consisted of 10 mM ammonium acetate and acetonitrile, respectively. The isocratic mobile phase flow was characterized by a uniform concentration in acetonitrile 50% between 0 to 7 or 14 min.

Input parameters for DOE modeling:

Prior CCD analyses identified ranges of critical independent and interactive mass spectrometric variables, e.g. sheath gas pressure (SGP; 22–50 psig), auxiliary gas pressure (AGP; 5–10 psig), sweep gas pressure (SWGP; 0.7–1.3 psig), ion transfer tube temperature (ITT; 215–266 °C), and vaporizer temperature (VT; 220–320 °C), that affected the Gln-FMOC extracted chromatogram area-under-the-curve response variable (Table 1). These ranges were suitable for application to BBD and OD designs for further improvement of the Gln response. Pre-screening for potential elimination of one or more of these variables [13] was not required here due to their individual indispensable impact [18]. Minimum and maximum values of each parameter (Supplementary Table 1), chosen to meet minimum and maximum safe operative limits of the instrument, were utilized by Design Expert Version 12.0.6.0, 64-bit, S/No. 5606-0315-4331-1011 (Stat-Ease, Minneapolis, MN, US) to create a series of experimental runs for LC-ESI-MS analysis of Gln-FMOC (Suppl. Data; Table 2). Implementation of those experimental runs generated by minimum and maximum limits (two levels) of each critical variable (Supplementary Figure 1; Table 1) produced a maximum response from each DOE in scatter plot (Supplementary Figure 2), surface plot (Supplementary Figure 3), and desirability chart (Supplementary Figure 4) domains. Importantly, surface plots predicted optimum dependent variable response (Supplementary Figure 3), which could be validated by comparison between designs.

Table 1:

Minimum and maximum input parameters for BBD, OD-A, OD-D, OD-I, OD-DIST, and OD-MOD Design of Experiments (DOE).

Code	Variable	Low	High
A	SGP (psig)	22	50
B	AGP (psig)	5	10
C	SWGP (psig)	0.7	1.3
D	VT (°C)	220	320
E	ITT (°C)	215	266

Table 2:

Model Summary Statistics for BBD, OD-A, OD-D, OD-I, OD-DIST, and OD-MOD DOE.

Design	Source	Std. Dev.	R2	Adjusted R2	Predicted R2	PRESS	Recommendation
BBD	Linear	5.71	0.8687	0.8538	0.8209	1958.7
	2FI	6.2	0.8803	0.8275	0.6751	3552.14
	Quadratic	4.1	0.9554	0.9246	0.8216	1950.33	Suggested
	Cubic	2.69	0.9907	0.9676	0.4072	6481.53	Aliased
ODA	Linear	8.18	0.844	0.8127	0.7521	2658
	2FI	3.64	0.9815	0.963	0.8657	1440.18	Suggested
	Quadratic	3	0.9916	0.9748	0.6896	3327.21
	Cubic	0	1	1		*	Aliased
ODD	Linear	3.76	0.9666	0.9599	0.9481	550.44	Suggested
	2FI	3.04	0.9869	0.9738	0.8743	1332.11
	Quadratic	1.41	0.9981	0.9944	0.9386	650.43	Suggested
	Cubic	0	1	1		*	Aliased
ODI	Linear	6.76	0.8845	0.8614	0.8229	1752.21	Suggested
	2FI	7.59	0.9125	0.8251	0.2977	6947.2
	Quadratic	3.77	0.9856	0.9568	0.3558	6371.73	Suggested
	Cubic	0	1	1		*	Aliased
OD-DIST	Linear	6.31	0.9045	0.8854	0.8591	1469.74
	2FI	5.42	0.9577	0.9154	0.8378	1692.28
	Quadratic	2.92	0.9918	0.9755	0.8656	1402.19	Suggested
	Cubic	0	1	1		*	Aliased
OD-MOD	Linear	4.21	0.9573	0.9488	0.9355	669.15
	2FI	4.6	0.9695	0.9389	0.8242	1823.97
	Quadratic	1.33	0.9983	0.9949	0.9631	382.92	Suggested
	Cubic	0	1	1		*	Aliased

The chief purpose of replicate selection is to achieve maximum reduction in experiment numbers. CCD and BBD each design 50 experiments runs (CCD has 42 non-center points with 8 center points, i.e. 8 runs have same variable parameters, whereas BBD has 40 model points with 10 center points per block, i.e. 10 runs have same variable parameters), whereas OD models have 21 model points, 5 Lack of fit points, 5 replicates, for a total of 31runs. One unique center point featured in experimental runs designed here was included due to documented variability of responses [18].

OD designs were observed to produce markedly fewer experimental runs compared to CCD or BBD. While Gln chromatographic responses from each DOE featured an optimal outcome in each domain, those maximum responses differed among individual designs as well as between domains for a single design (Figure 5–7). Thus, a comprehensive comparison of optimal Gln response of various designs was performed to identify maximal performance DOE, wherein Design Expert software-generated patterns of experimental responses (residuals) were converted to a normal plot of non-transformed residuals (Supplementary Figures 5 and 9) to identify closest residual alignment. DOEs utilized software-recommended models, namely quadratic only (BBD; OD-DIST; OD-MOD), 2FI only (OD-A; Supplementary Figure 6), or linear or quadratic (OD-D and OD-I), A quadratic equation was produced for each DOE to identify synergistic and antagonistic factors (indicated by a positive and negative sign) that affect the response (Supplementary Figures 7 and 8). Surface plot forecasts of highest % relative response for optimum mass spectrometric parameter settings identified by desirability chart for each DOE were compared by software (Supplementary Figure 4, using a desirability value of 1, to discern the design that produced the greatest dependent variable outcome (Supplementary Figure 3), guided by published methodology [13, 16, 31].

Figure 5:

Critical mass spectrometric parameters derived from CCD, BBD, OD--A, ODD, OD-I, OD-DIST, OD-MOD experimental runs (Refer Table 4: Type (A)). Parameters were obtained by CCD [1] or BBD (#20), OD-A (#30), OD-D (#4), OD-I (#29), OD-DIST (#15) and OD-MOD (#1) experimental run numbers (Refer Supplementary Table 1–6). *p<0.05, **p<0.01

Figure 7:

Critical mass spectrometric parameters derived from solutions of desirability provided by Design Expert software (Refer Table 4: Type (C)). *p<0.05, **p<0.01

Experiments created for BBD by Design Expert contained 10 center points of the parameters SGP=36 psig, AGP=7.5 psig, SWGP=1 psig, VT=270°C, and ITT=241°C were available in runs #7, 8, 9, 11, 12, 13, 26, 33, 42, 50 (Supplementary Tables 1).

Experiments created for OD-A included replications of a) SGP=22 psig, AGP=5 psig, SWGP=0.9 psig, VT=232°C, and ITT=266°C in runs 1 and 15; b) SGP=50 psig, AGP=10 psig, SWGP=1.3 psig, VT=220°C, and ITT=215°C in runs 5 and 25; c) SGP=26 psig, AGP=10 psig, SWGP=0.7 psig, VT=220°C, and ITT=266°C in runs 13 and 18; d) SGP=43 psig, AGP=10 psig, SWGP=1.3 psig, VT=320°C, and ITT=266°C in runs 17 and 29; and e) SGP=50 psig, AGP=5 psig, SWGP=1 psig, VT=320°C, and ITT=266°C in runs 27 and 30 (highlighted rows; Supplementary Table 2).

Replicate experiments produced for OD-D included a) SGP=22 psig, AGP=5 psig, SWGP=1.0 psig, VT=320°C, and ITT=215°C: runs 9 and 18; b) SGP=22 psig, AGP=5 psig, SWGP=1.3 psig, VT=220°C, and ITT=266°C: runs 12 and 19; c) SGP=50 psig, AGP=5 psig, SWGP=1.3 psig, VT=220°C, and ITT=215°C: runs 15 and 28; d) SGP=35 psig, AGP=7 psig, SWGP=1.3 psig, VT=320°C, and ITT=215°C: runs 20 and 25; and e) SGP=22 psig, AGP=5 psig, SWGP=0.7 psig, VT=220°C, and ITT=215°C: runs 23 and 30 (highlighted rows; Supplementary Table 3).

Supplementary Table 4 depicts replicate experiments (highlighted rows) created for OD-I: a) SGP=23 psig, AGP=6.5 psig, SWGP=1.3 psig, VT=287°C, and ITT=244°C: runs 4 and 9; b) SGP=38 psig, AGP=7.7 psig, SWGP=1 psig, VT=320°C, and ITT=238°C: runs 5 and 27; c) SGP=35.6 psig, AGP=7.3 psig, SWGP=1.2 psig, VT=253°C, and ITT=215°C: runs 7 and 10; d) SGP=38.7 psig, AGP=5 psig, SWGP=0.7 psig, VT=276°C, and ITT=243°C: runs 11 and 13; and e) SGP=50 psig, AGP=6.6 psig, SWGP=1.0 psig, VT=278°C, and ITT=263°C: runs #0 and 30.

Supplementary Table 5 shows replicate experiments created for OD-DIST involving a) SGP=22 psig, AGP=9.9 psig, SWGP=0.7 psig, VT=230°C, and ITT=215°C: runs 2 and 25; b) SGP=22 psig, AGP=5 psig, SWGP=1.3 psig, VT=320°C, and ITT=229°C: runs 14 and 24; c) SGP=32 psig, AGP=10 psig, SWGP=0.7 psig, VT=320°C, and ITT=235°C: runs 18 and 30; d) SGP=35.6 psig, AGP=7.3 psig, SWGP=1.2 psig, VT=253°C, and ITT=215°C: runs 19 and 26; and e) SGP=40 psig, AGP=9.5 psig, SWGP=0.7 psig, VT=220°C, and ITT=265°C runs: 21 and 27.

Lastly, Supplementary Table 6 illustrates replicate experiments created for OD-MOD [a) SGP=32 psig, AGP=8 psig, SWGP=1.3 psig, VT=285°C, and ITT=215°C: runs 2 and 3; b) SGP=22 psig, AGP=10 psig, SWGP=0.7 psig, VT=320°C, and ITT=215°C: runs 4 and 25; c) SGP=25 psig, AGP=5 psig, SWGP=0.7 psig, VT=220°C, and ITT=266°C: runs 8 and 28; d) SGP=41 psig, AGP=10 psig, SWGP=0.7 psig, VT=320°C, and ITT=266°C: runs #10 and 30; e) SGP=22 psig, AGP=6.2 psig, SWGP=0.7 psig, VT=306°C, and ITT=246°C: runs 18 and 19].

Application of high-performance DOE for mass spectrometric analysis of hypothalamic astrocyte Gln content:

Adult male and female Sprague-Dawley rats (3–4 months of age) were housed under a 14 h light:10 h dark lighting schedule (lights on at 05.00 h), and allowed free access to standard laboratory rat chow (Harlan Teklad LM-485; Harlan Industries, Madison, WI) and tap water. Rats were acclimated to daily handling over a seven day period prior to experimentation. All animal protocols were conducted in accordance with NIH guidelines for care and use of laboratory animals, and approved by the ULM Institutional Animal Care and Use Committee. After sacrifice, the hypothalamus was dissected from each brain as a single block at the following boundaries: anterior: rostral border of optic chiasm; posterior: rostral margin of mammillary bodies; lateral: lateral border of tuber cinereum; dorsal: dorsum of third ventricle. Highly-purified primary astrocyte cultures were obtained as described [19, 20]. Briefly, enzymatically-digested hypothalamic tissue was dissociated, by vigorous pipetting, into a single-cell suspension in astrocyte plating media (DMEM high glucose media; prod. no. 12800–017; ThermoFisherSci.) supplemented with 10.0% heat-inactivated fetal bovine serum (FBS; GE Healthcare Bio-Sciences, Pittsburgh, PA) and 1.0% penicillin/streptomycin (prod. no. 15140–122, ThermoFisherSci.). For each astrocyte collection, cells from three hypothalami were pooled by suspension in astrocyte plating media, plated in Poly-D-lysine (prod. no. A-003-E, MilliporeSigma, Burlington, MA) - coated T75 culture flasks at a concentration of 50 μg/mL, and incubated at 37 °C in a humidified environment in the presence of 5% CO₂ with media changes every 2–3 days post-plating. Approximately 7–8 days after plating, microglia and oligodendrocyte precursor cells were removed from cultures. Confluent astrocytes were washed twice with 10 mM phosphate buffered saline, pH 7.4 (PBS), then incubated at 37 °C (7–10 min) with 5.0 mL 0.05% trypsin-EDTA (prod. no. 25300–062, ThermoFisherSci.). Detached astrocytes were centrifuged at 180×g (5 min) and re-suspended in 40 mL fresh media. Cells derived from each confluent layer were plated in two T75 culture flasks, and incubated at 37 °C for 12–14 days with media changes at 2–3 day intervals. Astrocytes were counted, plated, and grown for two weeks prior to experimentation. Astrocyte culture purity was evaluated by immunofluorescence staining of GFAP-immunoreactivity (-ir), as described. Relative abundance of astrocyte glial fibrillary acid protein (GFAP; 1:1,000, prod. no. 3670D; Cell Signaling Technology, LLC, Danvers, MA), microglial (AIF-1/Iba1; prod. no. NBP100–1028, 1:1,000; Novus Biologicals, LLC, Littleton, CO) and oligodendrocyte (O4 marker; 1:1,000, prod. no. MAB345; MilliporeSigma, Burlington, MA) protein marker expression was determined by Western blot. After initial plating at a density of 1 × 10⁶ cells/100 mm², astrocytes were maintained for 3 days in high glucose DMEM media until approximately 70% confluent. Cells were subsequently steroid-starved for 18 h by incubation in high glucose media containing 5.0% charcoal-stripped FBS (prod. no. 12676029; ThermoFisherSci.) (21–23). Cells were then incubated for four h with glucose free-Hanks balanced salt solution (HBSS) media containing 10 nM 17β-estradiol [Ibrahim et al., 2020] to achieve glucoprivation (24–27). Our previous studies showed that male and female astrocytes exhibit divergent adjustments in cellular energy balance, e.g. more negative in male versus more positive in female, through estrogen receptor (ER)-dependent mechanisms. Over the final 4 hr of the experiment, astrocytes were incubated with glucose free-HBSS media containing 10 nM estradiol and one of the following ER antagonists (24, 28): non-selective ER antagonist ICI-182,780 (ICI) (prod. no. 1047; 100 nM; Tocris/Bio-Techne Corporation, Minneapolis, MN), selective ERα antagonist 1,3-Bis(4-hydroxyphenyl)-4-methyl-5-[4-(2 piperidinylethoxy) phenol]-1H-pyrazole dihydrochloride (MPP) [prod. no. 1991; 100 nM; Tocris); selective ERβ antagonist 4-[2-Phenyl-5,7-bis(trifluoromethyl)pyrazolo[1,5-a] pyrimidin-3-yl] phenol (PHTPP) (100 nM) (prod. no. 2662, Tocris); selective transmembrane G protein-couple estrogen receptor-1 (GPER/GPR30) antagonist G15 (100 nM; prod. no. 1161002-05-6, Sandia Biotech, Albuquerque, NM), dissolved in dimethyl sulfoxide (DMSO), as described [16]. DMSO was administered in an equal volume to all treatment groups to result in a media concentration of 0.01%. Other glucose-deprived astrocytes were treated with DMSO alone (C₀). Non-glucoprivic controls (C_5.5) were incubated with HBSS media containing 10 nM 17β-estradiol supplemented with 5.5 mM glucose and DMSO. Astrocytes were detached from plates with 0.05% trypsin EDTA (prod. no. 25300–062; ThermoFisherSci.), washed in PBS, pelleted in Tris buffer. L-Glutamic acid 5-benzyl ester was used as IS, as this derivative exhibited a highly stable mass spectrometric response (m/z 458.2) (9) in standard and astrocyte samples. Cell samples were vortexed (30 s), homogenized, and derivatized with FMOC for detection of Gln-FMOC at m/z 367.1, as described [16]. Astrocyte Gln concentrations were determined using the linear equation y = 2.1722x - 0.0033, over a range of 1.125 – 18 ng/μL, R² = 0.9996. Limits of detection and quantification were 11.7 and 35.5 ng/μL, respectively. Data were analyzed within sexes by one-way analysis of variance and Student Newman Kuels post-hoc test. Differences of p < 0.05 were considered significant.

Results:

Identification of optimum Gln-FMOC signal magnitude across screening designs:

Peak percent relative responses (% RR) were observed in BBD experimental run (Supplementary Figure 10): SGP=50 psig (Panel A), AGP=8 psig (Panel B,), SWGP=1.3 psig (Panel C), VT=270°C (Panel D), and ITT=241°C (Panel E); OD-A run (Figure 1): SGP=50 psig (Panel A), AGP=5 psig (Panel B), SWGP=1 psig (Panel C), VT=320°C (Panel D), and ITT=266°C (Panel E); OD-D run (Figure 2): SGP=50 psig (Panel A), AGP=5 psig (Panel B), SWGP=1.3 psig (Panel C), VT=320°C (Panel D), and ITT=266°C (Panel E); OD-I run (Supplementary Figure 11): SGP=50 psig (Panel A), AGP=10 psig (Panel B), SWGP=0.7 psig (Panel C), VT=320°C (Panel D), and ITT=266°C (Panel E); OD-DIST run (Supplementary figure 12): SGP=50 psig (Panel A), AGP=10 psig (Panel B), SWGP=0.8 psig (Panel C), VT=320°C (Panel D), and ITT=266°C (Panel E); OD-MOD run (Supplementary Figure 13): SGP=50 psig (Panel A), AGP=5 psig (Panel B), SWGP=0.92 psig (Panel C), VT=292°C (Panel D), and ITT=266°C (Panel E).

Figure 1:

Effect of individual mass spectrometric operational variables on percent relative response of standard glutamine derivative in Optimal Design-A (OD-A). A) SGP on %relative response of glutamine derivative: 50 psig showed highest response; B) AGP on % relative response of glutamine derivative: 5 psig showed highest response; C) SWGP on % relative response of glutamine derivatives: 1 psig showed highest response; D) VT on % relative response of glutamine derivatives: 320 °C showed highest area; E) ITT on % relative area of glutamine derivatives: 266 °C showed highest area.

Figure 2:

Effect of individual mass spectrometric operational variables on percent relative response of standard glutamine derivative in Optimal Design-D (OD-D). A) SGP on %relative response of glutamine derivative: 50 psig showed highest response; B) AGP on % relative response of glutamine derivative: 5 psig showed highest response; C) SWGP on % relative response of glutamine derivatives: 1.3 psig showed highest response; D) VT on % relative response of glutamine derivatives: 320 °C showed highest area; E) ITT on % relative area of glutamine derivatives: 266 °C showed highest area.

Model statistical analysis:

The normal percent probability plot of actual residuals showed prominent outliers in all designs except OD-A (Figure 3, Panels A-F). OD-A residuals (Panel B) were mostly aligned along the diagonal line, indicative of a normal distribution. Model summary analyses (Table 2) guide selection of a correct starting point for the final model. The model(s) exhibiting highest Predicted R-squared (R²) and Adjusted R² values are derived from the Fit Summary table; typically, a single model will exhibit both values. Here, the program-selected higher-order model (Quadratic for BBD; 2FI for OD-A; Linear or Quadratic for OD-D and OD-I; Quadratic for OD-DIST and OD-MOD) showed a Predicted R² value that is in reasonable agreement with Adjusted R² value, with a difference of less than 0.2. Main effects, interaction effects, and quadratic effects were optimized and evaluated in the current design. Analysis of variance (ANOVA) yielded a model F-value of 31.06, 53.05, 267.43, 34.23, 60.75, 294.15 for BBD, OD-A, OD-D, OD-I, OD-DIST and OD-MOD, respectively, indicating significance at a 95% confidence level.

Figure 3:

Normal % probability plot of residuals of DOE from Panel A) BBD; Panel B) OD-A; Panel C) OD-D; Panel D) OD-I; Panel E) OD-Dist; Panel F) OD-MOD.

Supplementary Table 8 shows the significant model terms that were identified in BBD (SGP, VT, ITT, SGP², and VT²), OD-A (SGP, VT, ITT, SGP × AGP, SGP × SWGP, SGP × VT, SGP × ITT, SWGP × VT, SWGP × ITT, and VT × ITT), OD-D (SGP, VT, ITT, SGP × AGP, SGP × VT, SWGP × ITT, VT × ITT, SGP² and VT²), OD-I (SGP, ITT, SGP², and ITT²), OD-DIST (SGP, VT, ITT, SGP × AGP, SGP × VT, SWGP × VT, AGP², VT²), and OD-MOD (SGP, VT, ITT, SGP × SWGP, SGP × VT, AGP × VT, AGP × ITT, SGP², AGP², SWGP², VT²) designs. Lack of Fit F-values for BBD, OD-A, OD-D, OD-I, OD-DIST and OD-MOD design were 487.58, 198.36, 19.77, 142.39, 85.18, and 17.6, respectively (Supplementary Table 8).

Parameter estimates show that the variable inflation factor (VIF) of designs has minimum=1 (BBD) and maximum=3.33 (OD-DIST); no values exceeded 10, indicative of a lack of design multicollinearity, which facilitates precise estimation of effects of independent changes in critical mass spectrometric variables (Supplementary Table 7). The final equation incorporating actual factors shows that either the highest positive (BBD: SWGP=17.99 psig; OD-A: AGP=4.486 psig; OD-D: AGP=4.09 psig; OD-I: SWGP=32.52 psig; OD-DIST: SWGP=5.883; OD-MOD: SWGP=27.274) or negative value [BBD: SWGP= −86.6 psig; OD-A: SWGP= −184.9 psig; OD-D: SWGP= −27.784 psig; OD-I: VT= −1.26 °C; OD-DIST: SWGP= −25.972 psig; OD-MOD: SWGP= −31.885 psig) affect Gln-FMOC derivative chromatographic response (Table 3). Antagonistic factors predicted to diminish mass spectrometric response were identified for each design: BBD: AGP, SWGP, VT, ITT, AGP × SWGP, SGP²; OD-A: SGP, SWGP, VT, ITT, SGP × AGP, AGP × SWGP, AGP × ITT; OD-D: SWGP, VT, SGP × AGP, AGP × VT, SGP², AGP², SWGP², ITT²; OD-DIST: SGP, SWGP, VT, ITT, SGP × AGP, AGP × SWGP, SWGP × VT, SGP², AGP²; OD-MOD: VT, ITT, SGP × AGP, SGP × ITT, AGP × SWGP, AGP × VT, AGP × ITT, SGP², AGP², SWGP². Conversely, agonist factors predicted to increase mass spectrometric response were identified for each design: BBD: SGP, SGP × AGP, SGP × SWGP, SGP × VT, SGP × ITT, AGP × VT, AGP × ITT, SWGP × VT, SWGP × ITT, VT × ITT, AGP², SWGP², VT², ITT²; OD-A: AGP, SWGP, VT, ITT, SGP × AGP, SGP × SWGP, SGP × VT, SGP × ITT, AGP × SWGP, AGP × VT, AGP × ITT, SWGP × VT, SWGP × ITT, VT × ITT; OD-D: SGP, AGP, ITT, SGP × SWGP, SGP × VT, SGP × ITT, AGP × SWGP, AGP × ITT, SWGP × VT, SWGP × ITT, VT × ITT; OD-I: SGP, SWGP, ITT, SGP × VT, SGP × ITT, SWGP × VT, AGP², SWGP², VT²; OD-DIST: AGP, SGP × SWGP, SGP × VT, SGP × ITT, AGP × VT, AGP × ITT, SWGP × ITT, VT × ITT, SWGP², VT², ITT²; OD-MOD: SGP, AGP, SWGP, SGP × ITT, AGP × SWGP, AGP × VT, AGP × ITT, SGP², AGP², SWGP².

Table 3:

Final quadratic or 2FI equation for various experimental designs.

		SGP	AGP	SWGP	VT	ITT	SGP * AGP	SGP * SWGP	SGP * VT	SGP * ITT	AGP * SWGP
BBD =	483.67776	1.89394	−16.74767	−86.59707	−1.30505	−2.0744	0.035195	0.84464	0.003095	0.000314	−1.00906
ODA =	507.27962	−3.91678	4.48613	−184.88011	−1.23875	−1.54583	−0.065192	0.974045	0.007285	0.010849	−0.050126
ODD =	152.05102	1.94704	4.09009	−27.7836	−1.23316	0.086589	−0.027069	0.057999	0.003698	0.002578	0.083823
ODI =	−6.7106	3.59522	−0.382049	32.52301	−1.26012	1.06563	−0.041176	−0.40848	0.002436	0.007072	−0.595699
OD-DIST =	559.90591	−0.33137	5.40055	−25.97223	−2.42639	−2.06875	−0.054511	0.305153	0.006021	0.003072	−2.21954
OD-MOD =	124.79952	3.05142	16.15654	27.27386	−0.730927	−0.97622	−0.015926	0.237409	0.002514	−0.000624	−0.406825
		AGP * VT	AGP * ITT	SWGP * VT	SWGP * ITT	VT * ITT	SGP2	AGP2	SWGP2	VT2	ITT2
		0.004146	0.043617	0.024686	0.111382	0.001443	−0.03252	0.333829	17.99209	0.001675	0.003268
		0.005791	−0.012241	0.176617	0.409354	0.003604	NA	NA	NA	NA	NA
		−0.00144	0.000471	0.022376	0.148842	0.000847	−0.027145	−0.194087	−8.9587	0.001679	−0.000972
		−0.0036	−0.004849	0.040103	−0.115943	−0.00096	−0.05116	0.275272	1.4174	0.002569	−0.001548
		0.005382	0.030064	−0.143104	0.242145	0.000963	−0.007597	−0.670539	5.88349	0.004132	0.003018
		−0.00805	−0.026517	0.043685	0.080556	0.000572	−0.030046	−0.460108	−31.88465	0.000838	0.002393

Cook’s distance is useful to identify residual outliers. As shown in Supplementary Figure 14, no observations exceeded the maximum limit of 1 (BBD, OD-A, OD-DIST: Panels A, B, and E, respectively), while there were outliers in the Panel C, D, F). Although residuals showed the normal distribution pattern in the earlier plots, few outliers were found in the cook’s plots (ODD, ODI, and OD-MOD: Panels C, D, F). The present analyses did not negate cook’s outliers. Leverage is a numerical value between 0 and 1 that reveals a potential for a design point to influence model fit. Here, analysis of leverage versus runs showed that residuals did not exceed a value of 0.84 and present in two unique lanes for BBD (Panel A, Supplementary Figure 15), while the rest of the designs showed the influential residuals above the stated line (Panel B-F, Supplementary Figure 15). As shown in Supplementary Figure 15, a small number of responses exhibited a Difference in Fits (DFFITS) value that fell outside limits (Panel A-F) in Supplementary Figure 16.

Surface plot forecasting of high-responsive variable parameters in selected designs:

Figure 4 (Panels A-F) depicts maximum probable responses (%RR) versus SGP, AGP, SWGP, VT, and ITT forecasted by BBD, OD-A, OD-D, OD-I, OD-DIST, and OD-MOD, respectively. High-yield variable parameters based on surface plots are presented in Table 4. Plot curvatures depict variable interactions; for example, SGP (low = 22 psig; high = 50 psig) and AGP (low = 5 psig; high = 10 psig) were used to generate surface plots for each design, whereas the surface plot in Figure 4 utilizes high SWGP=1.3, VT=320; ITT=266 (Panels A, B, C, E, F) as well as medium SWGP=1 and high level of VT=320 and ITT=266 (Panel D). SGP (high) and AGP (high) yielded %RR ≈79%, ≈92%, ≈120%, ≈128% (BBD, Figure 4A); %RR ≈84%, ≈85%, ≈97%, ≈99% (OD-D, Figure 4C); %RR ≈60%, ≈67%, ≈101%, ≈101% (OD-DIST, Figure 4E); %RR ≈91%, ≈90%, ≈94%, ≈93% (OD-MOD, Figure 4F). SGP (high) and AGP (low) yielded %RR ≈88%, ≈92%, ≈105%, and ≈104% (OD-I, Figure 4D); %RR ≈66%, ≈77%, ≈99%, ≈108% (OD-A, Figure 4B). Surface plots that did not project high yields are presented in Supplementary Figures 17–22.

Figure 4:

Response surface plots derived from the DOE for Gln derivatization reaction: The component A: SGP (min=22; max=50), and B: AGP (min=5; max=10) were in common with Panels A, B, C, E, and F that utilize high SWGP=1.3, VT=320; ITT=266; while Panel D utilizes medium SWGP=1 and high level of VT=320 and ITT=266.

Table 4:

Critical mass spectrometric parameters were based on the optimal experimental run for each design (Type A/Domain-I: CCD obtained from [1]; Refer Supplementary Table 1, #20 run for BBD; Supplementary Table 2, #30 run for OD-A; Supplementary Table 3, #4 run for OD-D; Supplementary Table 4, #29 run for OD-I; Supplementary Table 5, #15 run for OD-DIST; Supplementary Table 6, #1 run for OD-MOD). High response forecasted parameters from surface plots (Type B/Domain-II); and desirability=1 with high response parameters were selected from Supplementary Tables 9–14 (Type C/Domain-III).

Type	Run	A:SGP	B:AGP	C:SWGP	D:VT	E:ITT
(A) Maximum Response Parameters Based on Design Experimental Runs	CCDǂ (#13)	36	7.5	1	270	301
	BBD (#20)	50	7.5	1.3	270	240.5
	ODA (#30)	50	5	1	320	266
	ODD (#4)	50	5	1.3	320	266
	ODI (#29)	50	10	0.7	320	266
	OD-DIST (#15)	50	10	0.8	320	266
	OD-MOD (#1)	50	5	0.93	292	266
(B) Surface Plot Forecasted Variable Parameters (Figure 4)	CCDǂ	50	10	1	320	266
	BBD	50	10	1.3	320	266
	ODA	50	5	1.3	320	266
	ODD	50	5	1.3	320	266
	ODI	50	5	1	320	266
	OD-DIST	50	5	1.3	320	266
	OD-MOD	50	5	1.3	320	266
(C) Variables of Desirability with Maximum Response (Refer Supp. Table 9–14)	CCDǂ	50	10	1.3	320	266
	BBD	45.7	5	1.3	310	254
	ODA	43.4	10	1.3	320	266
	ODD	50	5	1.3	320	266
	ODI	50	10	0.7	320	266
	OD-DIST	50	10	0.8	320	266
	OD-MOD	50	5	0.92	292	266

For BBD design (Supplementary Figure 17), SGP (low) and AGP (high) yielded %RR ≈45%, ≈47%, ≈70%, and ≈70% (Panel A); SGP (low) and AGP (low) yielded %RR ≈52%, ≈48%, ≈65%, and ≈67% (Panel B); SGP (high) and AGP (low) yielded %RR ≈82%, ≈89%, ≈110%, and ≈120% (Panel C).

For OD-A design (Supplementary Figure 18), SGP (low) and AGP (high) yielded %RR ≈22%, ≈48%, ≈53%, and ≈54% (Panel A); SGP (low) and AGP (low) yielded %RR ≈56%, ≈40%, ≈46%, and ≈46% (Panel B); SGP (high) and AGP (high) yielded %RR ≈65%, ≈76%, ≈98%, ≈106% (Panel C). For OD-D design (Supplementary Figure 19), SGP (low) and AGP (high) yielded %RR ≈52%, ≈46%, ≈51%, and ≈51% (Panel A); SGP (low) and AGP (low) yielded %RR ≈51%, ≈45%, ≈50%, and ≈50% (Panel B); SGP (high) and AGP (low) yielded %RR ≈87%, ≈87%, ≈99%, and ≈99% (Panel C). For OD-I design (Supplementary Figure 20), SGP (low) and AGP (high) yielded %RR ≈50%, ≈47%, ≈50%, and ≈51% (Panel A); SGP (low) and AGP (low) yielded %RR ≈48%, ≈46%, ≈51%, and ≈53% (Panel B); SGP (high) and AGP (high) yielded %RR ≈86%, ≈87%, ≈98%, ≈96% (Panel C).

For OD-DIST design (Supplementary Figure 21), SGP (low) and AGP (high) yielded %RR ≈39%, ≈33%, ≈57%, and ≈54% (Panel A); SGP (low) and AGP (low) yielded %RR ≈38%, ≈30%, ≈48%, and ≈49% (Panel B); SGP (high) and AGP (low) yielded %RR ≈67%, ≈72%, ≈101%, and ≈104% (Panel C).

For OD-MOD design (Supplementary Figure 22), SGP (low) and AGP (high) yielded %RR ≈55%, ≈48%, ≈50%, and ≈47% (Panel A); SGP (low) and AGP (low) yielded %RR ≈49%, ≈48%, ≈55%, and ≈53% (Panel B); SGP (high) and AGP (low) yielded %RR ≈88%, ≈92%, ≈102%, and ≈102% (Panel C).

Identification of maximum-yielding design parameters:

Maximum response variables identified from screening design experiments and resonance surface plots were evaluated for desirability=1 with optimal %RR (Table 4). The BBD experimental run 20 response did not differ significantly versus CCD run 13 (18). OD-A design run 30 produced a significantly greater response compared to CCD and OD-I, but not BBD, OD-DIST, or OD-MOD. Data show that OD-A design runs were more efficient in maximizing FMOC-Gln chromatographic response relative to OD-D or OD-MOD (Figure 5). Comparison of forecasted critical mass spectrometric variables derived from DOE surface plots are presented in Figure 6. Variable parameters derived from OD-A, OD-D, OD-DIST and OD-MOD plots were identical and displayed as a single cluster (Figure 6; Table 4). Data show that OD-A, OD-D, OD-DIST, and OD-MOD designs produce have higher mean chromatographic response versus CCD or OD-I. Each DOE has specified one hundred solutions with desirability (BBD, Supplementary Table 9; ODA, Supplementary Table 10; ODD, Supplementary Table 11; ODI, Supplementary Table 12; OD-DIST, Supplementary Table 13; OD-MOD, Supplementary Table 14). Mass spectrometric parameters of desirability=1 with maximum %RR were selected to compare outcomes; as OD-I and OD-DIST produced alike mass spectrometric parameters, they were considered as a single cluster (Figure 7; Table 4). The OD-D response was significantly greater compared to other designs, whereas no differences occurred between OD-A, OD-I/OD-DIST, or OD-MOD responses (Figure 7). OD-A optimal response was achieved by experimental runs and surface plots (Figures 5 and 6, whereas OD-D optimal variables were generated from surface plots and desirability (Figure 6 and 7). Data show that OD-A design run 30 (SGP=50, AGP=5, SWGP=1, VT=320, ITT=266; Figure 4, Panel B) and surface plot (SGP=50, AGP=5, SWGP=1.3, VT=320, ITT=266 in Panel B, Figure 4) parameters would produce a greater maximum response in different comparisons (Figure 5 and 6).

Figure 6:

Critical mass spectrometric forecasted parameters derived from surface plots of CCD, BBD, OD-A, OD-D, OD-I, OD-DIST, OD-MOD (Refer Table 4: Type (B)). OD-A, OD-D, OD-DIST and OD-MOD produce high response-forecasted parameters from respective surface plots. **p<0.01

These two mass spectrometric runs varied by SWGP 0.3 psig. Comparison of these runs (Bar A: SGP=50, AGP=5, SWGP=1, VT=320, ITT=266; Bar B: SGP=50, AGP=5, SWGP=1.3, VT=320, ITT=266; Figure 8) disclosed no significant difference; thus, application of either is likely to produce a similar response. Although CCD techniques are highly popular (31), outcomes here show that OD screens are superior in terms of accessibility of factor space (13). Present outcomes identify OD-A as the relative high-performance DOE compared to other designs evaluated here owing to its efficiency in generated an optimal best response with minimum number of experiments.

Figure 8:

Comparison of two highly-performing mass spectrometric variables of OD-A differing by SWGP=0.3 psig. The lack of significant difference (unpaired two tailed t-test, p>0.05) is indicative of a comparable chromatographic response.

Determination of process variability of high-performance critical parameters of OD-A:

Process stability was evaluated using x-bar and R quality control charts, which are used to visualize process average and variation based on samples collected over a specified time. Control limits on each chart document process mean and variation going forward; a value falling beyond control limits denotes that process mean or variation of the process is out-of-control. The x-bar chart y- and x-axis illustrate total mean and control limits versus sample group, in that order. The x-bar chart generated by R-Software (https://www.r-project.org/) provides important information for its interpretation, including the number of groups, control limits, the overall average, the standard deviation, the points beyond the control limits and the violating runs (30). An R-chart is useful for determination of range of the sub-group changes over time; the center line for each subgroup depicts the expected value of the range statistic. OD-A mass spectrometric parameters (Bar A of Figure 8) were implemented on a single-stock Gln-FMOC solution, and fifteen responses (three responses per group) were plotted by x-bar and R-chart (Figure 9). One observation fell outside of the upper control limit (Figure 9A), and the points deviated from the center line (Figure 9B). An internal standard (29) was used to overcome the process out-of-control value by plotting the ratio of analyte to internal standard in both charts (Figure 10). These ratios were confined to control limits (Figure 10A), and points were closely aligned to the center line (Figure 10B).

Figure 9:

Process variability of mass spectrometric responses is shown in xbar and R-chart. A) One observation was outside of the upper control limit; B) points were distanced from the center line.

Figure 10:

Use of internal standard along with standard measurement of Gln derivative produced a response within control limits (A) and alignment of points along the center line (B).

As shown in Figure 11, application of OD-A – optimized LC-ESI-MS technology for analysis of hypothalamic astocyte Gln content disclosed a significant reduction in this amino acid in female due to glucose withdrawal, but not male. Moreover, data reveal that this sex-specific Gln response was averted by pharmacological inhibition of ERα, ERβ, or GPER activity, whereas these ER variants had no demonstrable impact on Gln levels in glucose-deprived male astrocytes.

Figure 11:

Effects of glucoprivation on male (A) versus female (B) hypothalamic astrocyte Gln content. Cells were incubated in the presence (C5.5) or absence of glucose (all other treatment groups), and pretreated with vehicle (C5.5; Co) or one of the following drugs: the non-selective estrogen receptor (ER) antagonist ICI, the ER-alpha antagonist MPP, the ER-beta antagonist PHTPP, or the membrane ER GPR30/GPER antagonist G15. *p<0.05, **p<0.01, **p<0.001

Summary and Conclusions:

Assessment of critical mass spectrometric equipment (AGP, SGP, SWGP, VT, and ITT) variables by CCD, BBD, versus OD DOE models showed that OD-A produced maximal enhancement of LC-ESI-MS sensitivity for Gln measurement. Application of this optimized analytical methodology to primary astrocyte cultures revealed sex differences in effects of glucose deficiency on content of this important intermediate metabolite.

While analytical tools for analysis of Gln levels in bulk-size tissue samples are widely available, current research addressed the need for improved sensitivity for cellular-level measurements. The cost-effective and time-efficient high performance DOE approach described here is advantaged by minimal process variability and very few numbers of experimental runs. This method will advance efforts to understand endogenous Gln utilization in characterized brain cell types in response to physiological and pathophysiological stimuli.

Abbreviations:

AD

adamantanamine hydrochloride

AGP

aux gas pressure

BBD

Box-Behnken Design

CAN

acetonitrile

CCD

Central Composite Design

FMOC

fluorenylmethoxycarbonyl

Gln

glutamine

IS

Internal Standard

ITT

ion transfer tube temperature

LC-ESI-MS

high-performance liquid chromatography-electrospray ionization-mass spectrometry

MeOH

methanol

OD-A

Optimal Design - A

OD-D

Optimal Design - D

OD-I

Optimal Design - I

OD-DIST

Optimal Design - Distance

OD-MOD

Optimal Design - Modified Distance

RSM

Response surface methodology

MPP

1,3-Bis(4-hydroxyphenyl)-4-methyl-5-[4-(2-piperidinylethoxy)phenol]-1H-pyrazole dihydro- chloride

PHTPP

4-[2-phenyl-5,7-bis(trifluoromethyl)pyrazolo[1,5-a]pyrimidin-3-yl]phenol

SGP

sheath gas pressure

SWGP

sweep gas pressure

VT

vaporizer temperature