Understanding Nontraditional Differential Mobility Behavior: A Case Study of the Tricarbastannatrane Cation, N(CH2CH2CH2)3Sn. J Am Soc Mass Spectrom 2020;31:796-802. [PMID: 32129991 DOI: 10.1021/jasms.9b00042]

Towards Generalizable In Silico Predictions of Differential Ion Mobility Using Machine Learning and Customized Fingerprint Engineering

Differential mobility spectrometry (DMS), a tool for separating chemically similar species (including isomers), is readily coupled to mass spectrometry to improve selectivity in analytical workflows. DMS dispersion curves, which describe the dynamic mobility experienced by an ion in a gaseous environment, show the maximum ion transmission for an analyte through the DMS instrument as a function of the separation voltage (SV) and compensation voltage (CV) conditions. To date, there exists no fast, general prediction tool for the dispersion behavior of ions. Here, we demonstrate a machine learning (ML) model that achieves generalized dispersion prediction using an in silico feature addition pipeline. We employ a data set containing 1141 dispersion curve measurements of anions and cations recorded in pure N2 environments and in N2 environments doped with 1.5% methanol (MeOH). Our feature addition pipeline can compute 1591 RDKit and Mordred descriptors using only SMILES codes, which are then normalized to sampled molecular distributions (n = 100 000) using cumulative density functions (CDFs). This tool can be thought of as a "learned" feature fingerprint generation pipeline, which could be applied to almost any molecular (bio)cheminformatics tasks. Our best performing model, which for the first time considers solvent-modified environments, has a mean absolute error (MAE) of 2.1 ± 0.2 V for dispersion curve prediction, a significant improvement over the previous state-of-the-art work. We use explainability techniques (e.g., SHAP analysis) to show that this feature addition pipeline is a semideterministic process for feature sets, and we discuss "best practices" to understand feature sets and maximize model performance. We expect that this tool could be used for prescreening to accelerate or even automate the use of DMS in complex analytical workflows (e.g., 2D LC×DMS separation) and perform automated identification of transmission windows and increase the "self-driving" potential of the instrument. We make our models available as a free and accessible tool at https://github.com/HopkinsLaboratory/DispersionCurveGUI.

Chemical Transformations Can Occur during DMS Separations: Lessons Learned from Beer's Bittering Compounds

While developing a DMS-based separation method for beer's bittering compounds, we observed that the argentinated forms of humulone tautomers (i.e., [Hum + Ag]⁺) were partially resolvable in a N₂ environment seeded with 1.5 mol % of isopropyl alcohol (IPA). Attempting to improve the separation by introducing resolving gas unexpectedly caused the peaks for the cis-keto and trans-keto tautomers of [Hum + Ag]⁺ to coalesce. To understand why resolution loss occurred, we first confirmed that each of the tautomeric forms (i.e., dienol, cis-keto, and trans-keto) responsible for the three peaks in the [Hum + Ag]⁺ ionogram were assigned to the correct species by employing collision-induced dissociation, UV photodissociation spectroscopy, and hydrogen-deuterium exchange (HDX). The observation of HDX indicated that proton transfer was stimulated by dynamic clustering processes between IPA and [Hum + Ag]⁺ during DMS transit. Because IPA accretion preferentially occurs at Ag⁺, which can form pseudocovalent bonds with a suitable electron donor, solvent clustering also facilitated the formation of exceptionally stable microsolvated ions. The exceptional stability of these microsolvated configurations disproportionately impacted the compensation voltage (CV) required to elute each tautomer when the temperature within the DMS cell was varied. The disparity in CV response caused the peaks for the cis- and trans-keto species to merge when a temperature gradient was induced by the resolving gas. Moreover, simulations showed that microsolvation with IPA mediates dienol to trans-keto tautomerization during DMS transit, which, to the best of our knowledge, is the first observation of keto/enol tautomerization occurring within an ion-mobility device.

First-Principles Modeling of Preferential Solvation in Mixed-Modifier Differential Mobility Spectrometry

Differential mobility spectrometry (DMS) separates ions based on mobility differences between high and low electric field conditions. To enhance resolution, solvents such as water and acetonitrile are often used to modify the collision environment and take advantage of differing dynamic clustering behavior between analytes that coelute in hard-sphere environments (e.g., N₂). When binary solvent mixtures are used to modify the DMS environment, one solvent can have a dominant influence over the other with respect to ion trajectories. For example, for quinoline derivatives, a 9:1 water:acetonitrile solvent mixture exhibited identical behavior to an environment containing only acetonitrile as a modifier. It was hypothesized that this effect arises due to the significantly different binding strengths of the two solvents. Here, we utilize a first-principles model of DMS to study analytes in single and binary solvent mixtures and explore the effects governing the dominance of one solvent over the other. Computed DMS dispersion curves of quinoline derivatives are in excellent agreement with those measured experimentally. For mixed-modifier environments, the predicted cluster populations show a clear preferential solvation of ions with the stronger binding solvent. The influence of ion-solvent binding energies, solvent concentration, and solvent molecule size is discussed in the context of the observed DMS behavior. This work can guide the usage of binary solvent mixtures for improving ion separations in cases where compounds coelute in pure N₂ and in single-solvent modifier environments. Moreover, our results indicate that binary solvent mixtures can be used to create a relative scale for solvent binding energies.

Investigation of Zero-/High-Field Ion Mobility Orthogonal Separation Using a Hyphenated DMA-FAIMS System and Validation of the Two-Temperature Theory at Arbitrary Field for Tetraalkylammonium Salts in Nitrogen

Toward greater separation techniques for ions, a differential mobility analyzer (DMA) has been coupled with field asymmetric waveform ion mobility spectrometry (FAIMS) to take advantage of two mobility-related but different methods of separation. The filtering effect of the DMA allows ions to be selected individually based on low-field mobility and studied in FAIMS at variable electric field, yielding mobility separations in two dimensions. Because spectra fully describe ion mobility at variable field strength, results are then compared with a two-temperature theory-predicted mobility up to the fourth-order approximation. The comparison yields excellent results up to at least 100 Td, beyond which the theory deviates from experiments. This is attributed to two effects, the enlargement of the structure due to ion heating and the inelasticity of the collisions with the nitrogen bath gas. The corrected mobility can then be used to predict the dispersion plot through a newly developed implicit equation that circumvents the possible issues related to the more elaborate Buryakov equation. Our results simultaneously show that the DMA-FAIMS coupling yields complete information on ion mobility versus the field-strength to gas-density ratio and works toward predicting such spectra from ion structures and gas properties.

Extending Chirality in Group XIV Metallatranes

The syntheses of the racemic amino alcohol rac-N(CH₂CMe₂OH)(CMe₂CH₂OH)(CH₂CHMeOH) (L^22'1*H₃, 2) and its representative N(CH₂CMe₂OH)(CMe₂CH₂OH)(CH₂C(R)HMeOH) (L^22'1RH₃, 3) with the stereogenic carbon center being R-configured are reported. Also reported are the stannatranes L^22'1*SnOt-Bu (4) L^22'1RSnOt-Bu (6) and germatranes L^22'1*GeOEt (5) and L^22'1RGeOEt (7) as well as the trinuclear tin oxocluster [(μ₃-O)(μ₃-O-t-Bu){SnL^22'1R}₃] (8). NMR and IR spectroscopy, electrospray ionization mass spectrometry (ESI MS), and single crystal X-ray diffraction analysis characterize these compounds. Computational studies accompany the experimental work and help understand the diastereoselectivity observed in the course of the metallatrane syntheses.

Kinetics in DMS: Modeling Clustering and Declustering Reactions

Differential mobility spectrometry (DMS) uses high-frequency oscillating electrical fields to harness the differential mobility of ions for separating complex sample mixtures prior to detection. To increase the resolving power, a dynamic microsolvation environment is often created by introducing solvent vapors. Here, relatively large clusters are formed at low-field conditions which then evaporate to form smaller clusters at high-field conditions. The kinetics of these processes as the electrical field strength oscillates are not well studied. Here, we develop a computational framework to investigate how the different reactions (cluster association, cluster dissociation, and fast conformational changes) behave at different field strengths. We aim to better understand these processes, their effect on experimental outcomes, and whether DMS model accuracy is improved via incorporating their description. We find that cluster association and dissociation reactions for typical ion-solvent pairs are fast compared to the time scale of the varying separation fields usually used. However, low solvent concentration, small dipole moments, and strong ion-solvent binding can result in reaction rates small enough that a lag is observed in the ion's DMS response. This can yield differences of several volts in the compensation voltages required to correct ion trajectories for optimal transmission. We also find that the proposed kinetic approach yields generally better agreement with experiment than using a modified Boltzmann weighting scheme. Thus, this work provides insights into the chemical dynamics occurring within the DMS cell while also increasing the accuracy of dispersion plot predictions.

The hitchhiker's guide to dynamic ion-solvent clustering: applications in differential ion mobility spectrometry

This article highlights the fundamentals of ion-solvent clustering processes that are pertinent to understanding an ion's behaviour during differential mobility spectrometry (DMS) experiments. We contrast DMS with static-field ion mobility, where separation is affected by mobility differences under the high-field and low-field conditions of an asymmetric oscillating electric field. Although commonly used in mass spectrometric (MS) workflows to enhance signal-to-noise ratios and remove isobaric contaminants, the chemistry and physics that underpins the phenomenon of differential mobility has yet to be fully fleshed out. Moreover, we are just now making progress towards understanding how the DMS separation waveform creates a dynamic clustering environment when the carrier gas is seeded with the vapour of a volatile solvent molecule (e.g., methanol). Interestingly, one can correlate the dynamic clustering behaviour observed in DMS experiments with gas-phase and solution-phase molecular properties such as hydrophobicity, acidity, and solubility. However, to create a generalized, global model for property determination using DMS data one must employ machine learning. In this article, we provide a first-principles description of differential ion mobility in a dynamic clustering environment. We then discuss the correlation between dynamic clustering propensity and analyte physicochemical properties and demonstrate that analytes exhibiting similar ion-solvent interactions (e.g., charge-dipole) follow well-defined trends with respect to DMS clustering behaviour. Finally, we describe how supervised machine learning can be used to create predictive models of molecular properties using DMS data. We additionally highlight open questions in the field and provide our perspective on future directions that can be explored.

Synthesis, structure and properties of trivalent and pentavalent tricarbabismatranes

The first trivalent and pentavalent tricarbabismatranes were synthesized by the reaction of N(CH₂{2-LiC₆H₄})₃ with BiCl₃ and subsequent reaction with XeF₂, respectively. The trivalent bismatrane was easily oxidized by air, while the pentavalent bismatrane difluoride was relatively stable to air. A similar pentavalent bismatrance dichloride was prone to C-Cl bond reductive elimination even at room temperature.

Improved First-Principles Model of Differential Mobility Using Higher Order Two-Temperature Theory

Differential mobility spectrometry is a separation technique that may be applied to a variety of analytes ranging from small molecule drugs to peptides and proteins. Although rudimentary theoretical models of differential mobility exist, these models are often only applied to small molecules and atomic ions without considering the effects of dynamic microsolvation. Here, we advance our theoretical description of differential ion mobility in pure N₂ and microsolvating environments by incorporating higher order corrections to two-temperature theory (2TT) and a pseudoequilibrium approach to describe ion-neutral interactions. When comparing theoretical predictions to experimentally measured dispersion plots of over 300 different compounds, we find that higher order corrections to 2TT reduce errors by roughly a factor of 2 when compared to first order. Model predictions are accurate especially for pure N₂ environments (mean absolute error of 4 V at SV = 4000 V). For strongly clustering environments, accurate thermochemical corrections for ion-solvent clustering are likely required to reliably predict differential ion mobility behavior. Within our model, general trends associated with clustering strength, solvent vapor concentration, and background gas temperature are well reproduced, and fine structure visible in some dispersion plots is captured. These results provide insight into the dynamic ion-solvent clustering process that underpins the phenomenon of differential ion mobility.

Rapid separation of cannabinoid isomer sets using differential mobility spectrometry and mass spectrometry

With legalization and decriminalization of cannabis in many parts of the world comes the need for rapid separation and quantitation of the psychoactive ingredients.

Protonation-Induced Chirality Drives Separation by Differential Ion Mobility Spectrometry

Upon development of a workflow to analyze (±)-Verapamil and its metabolites using differential mobility spectrometry (DMS), we noticed that the ionogram of protonated Verapamil consisted of two peaks. This was inconsistent with its metabolites, as each exhibited only a single peak in the respective ionograms. The unique behaviour of Verapamil was attributed to protonation at its tertiary amino moiety, which generated a stereogenic quaternary amine. The introduction of additional chirality upon N-protonation of Verapamil renders four possible stereochemical configurations for the protonated ion: ( R,R ), ( S,S ), ( R,S ), or ( S,R ). The ( R,R )/( S,S ) and ( R,S )/( S,R ) enantiomeric pairs are diastereomeric and thus exhibit unique conformations that are resolvable by linear and differential ion mobility techniques.  Protonation-induced chirality appears to be a general phenomenon, as N -protonation of 12 additional chiral amines generated diastereomers that were readily resolved by DMS.

UVPD Spectroscopy of Differential Mobility-Selected Prototropic Isomers of Rivaroxaban

Two ion populations of protonated Rivaroxaban, [C₁₉H₁₈ClN₃O₅S + H]⁺, are separated under pure N₂ conditions using differential mobility spectrometry prior to characterization in a hybrid triple quadrupole linear ion trap mass spectrometer. These populations are attributed to bare protonated Rivaroxaban and to a proton-bound Rivaroxaban-ammonia complex, which dissociates prior to mass-selecting the parent ion. Ultraviolet photodissociation (UVPD) and collision-induced dissociation (CID) studies indicate that both protonated Rivaroxaban ion populations are comprised of the computed global minimum prototropic isomer. Two ion populations are also observed when the collision environment is modified with 1.5% (v/v) acetonitrile. In this case, the protonated Rivaroxaban ion populations are produced by the dissociation of the ammonium complex and by the dissociation of a proton-bound Rivaroxaban-acetonitrile complex prior to mass selection. Again, both populations exhibit a similar CID behavior; however, UVPD spectra indicate that the two ion populations are associated with different prototropic isomers. The experimentally acquired spectra are compared with computed spectra and are assigned to two prototropic isomers that exhibit proton sharing between distal oxygen centers.

UVPD spectroscopy of differential mobility-selected prototropic isomers of protonated adenine

Two prototropic isomers of adenine are formed in an electrospray ion source and are resolved spatially in a differential mobility spectrometer before detection in a triple quadrupole mass spectrometer. Each isomer is gated in CV space before being trapped in the linear ion trap of the modified mass spectrometer, where they are irradiated by the tuneable output of an optical parametric oscillator and undergo photodissociation to form charged fragments with m/z 119, 109, and 94. The photon-normalised intensity of each fragmentation channel is measured and the action spectra for each DMS-gated tautomer are obtained. Our analysis of the action spectra, aided by calculated vibronic spectra and thermochemical data, allow us to assign the two signals in our measured ionograms to specific tautomers of protonated adenine.

Predicting differential ion mobility behaviour in silico using machine learning

Although there has been a surge in popularity of differential mobility spectrometry (DMS) within analytical workflows, determining separation conditions within the DMS parameter space still requires manual optimization. A means of accurately predicting differential ion mobility would benefit practitioners by significantly reducing the time associated with method development. Here, we report a machine learning (ML) approach that predicts dispersion curves in an N2 environment, which are the compensation voltages (CVs) required for optimal ion transmission across a range of separation voltages (SVs) between 1500 to 4000 V. After training a random-forest based model using the DMS information of 409 cationic analytes, dispersion curves were reproduced with a mean absolute error (MAE) of ≤ 2.4 V, approaching typical experimental peak FWHMs of ±1.5 V. The predictive ML model was trained using only m/z and ion-neutral collision cross section (CCS) as inputs, both of which can be obtained from experimental databases before being extensively validated. By updating the model via inclusion of two CV datapoints at lower SVs (1500 V and 2000 V) accuracy was further improved to MAE ≤ 1.2 V. This improvement stems from the ability of the "guided" ML routine to accurately capture Type A and B behaviour, which was exhibited by only 2% and 17% of ions, respectively, within the dataset. Dispersion curve predictions of the database's most common Type C ions (81%) using the unguided and guided approaches exhibited average errors of 0.6 V and 0.1 V, respectively.

Determining Collision Cross Sections from Differential Ion Mobility Spectrometry

The experimental determination of ion-neutral collision cross sections (CCSs) is generally confined to ion mobility spectrometry (IMS) technologies that operate under the so-called low-field limit or those that enable empirical calibration strategies (e.g., traveling wave IMS; TWIMS). Correlation of ion trajectories to CCS in other non-linear IMS techniques that employ dynamic electric fields, such as differential mobility spectrometry (DMS), has remained a challenge since its inception. Here, we describe how an ion's CCS can be measured from DMS experiments using a machine learning (ML)-based calibration. The differential mobility of 409 molecular cations (m/z: 86-683 Da and CCS 110-236 Å2) was measured in a N2 environment to train the ML framework. Several open-source ML routines were tested and trained using DMS-MS data in the form of the parent ion's m/z and the compensation voltage required for elution at specific separation voltages between 1500 and 4000 V. The best performing ML model, random forest regression, predicted CCSs with a mean absolute percent error of 2.6 ± 0.4% for analytes excluded from the training set (i.e., out-of-the-bag external validation). This accuracy approaches the inherent statistical error of ∼2.2% for the MobCal-MPI CCS calculations employed for training purposes and the <2% threshold for matching literature CCSs with those obtained on a TWIMS platform.