Revealing the Impacts of Chemical Complexity on Submicrometer Sea Spray Aerosol Morphology

Sea spray aerosol (SSA) ejected through bursting bubbles at the ocean surface is a complex mixture of salts and organic species. Submicrometer SSA particles have long atmospheric lifetimes and play a critical role in the climate system. Composition impacts their ability to form marine clouds, yet their cloud-forming potential is difficult to study due to their small size. Here, we use large-scale molecular dynamics (MD) simulations as a “computational microscope” to provide never-before-seen views of 40 nm model aerosol particles and their molecular morphologies. We investigate how increasing chemical complexity impacts the distribution of organic material throughout individual particles for a range of organic constituents with varying chemical properties. Our simulations show that common organic marine surfactants readily partition between both the surface and interior of the aerosol, indicating that nascent SSA may be more heterogeneous than traditional morphological models suggest. We support our computational observations of SSA surface heterogeneity with Brewster angle microscopy on model interfaces. These observations indicate that increased chemical complexity in submicrometer SSA leads to a reduced surface coverage by marine organics, which may facilitate water uptake in the atmosphere. Our work thus establishes large-scale MD simulations as a novel technique for interrogating aerosols at the single-particle level.


Figures
. Different morphologies observed for phase separated or homogeneous mixed organic/aqueous particles. Orange and blue colors represent organics and water, respectively. Figure S2. Molecular structure of glucose oligosaccharide, laminarin. Since laminarin has a variety of molecular weights and branching ratios, this complexity is impossible to recreate computationally. Only structure was generated to represent neutral, branched oligosaccharides. See Main Text for details. Figure S3. NAMD efficiency scaling across UIUC Blue Waters and TACC Frontera. Note scale differences.

Experimental Design
The chemical components of the SSA models were selected based on the most up-to-date molecular analyses of nascent submicron SSA; although, it is worth noting that at the time of writing, only an estimated 25% of the total chemical species found in SSA has been fully characterized. 1 We aimed to reasonably represent in our models 1) the chemical composition of ions; 2) the chemical composition of organic matter; and 3) the mass percentage of water, which is variable based on hygroscopic growth factor and relative humidity.

1) Chemical composition of ions.
Elemental analysis of water-soluble ions by ion-exchange chromatography has identified the major ionic species in small particles with dry diameter dp = 150 nm, 1,2 given in Table S2. We considered the accuracy of the CHARMM36 force field when selecting which ions to incorporate into our models. Sodium is commonly used in MD simulations given its biological relevance, and its interactions with TIP3P water, protein, lipids, and chloride ions have been experimentally validated below concentrations of 1 M. 3 The concentration of sodium in bulk seawater and SSA is approximately 0.46 M, falling within the accuracy range. been validated and incorporated into the CHARMM36 force field parameter set used in this work. 12 Sulfate and bromide anions were not included in our models due to a relative lack of force field validation in the literature.

2) Chemical composition of organic matter.
There is a general consensus on the major classes of organic molecules observed in SSA, although their relative abundances are known to vary with ocean productivity, temperature, and geographic region. 1,2,[13][14][15][16] Molecular and elemental characterization in Cochran et al. identified major submicron SSA types to contain 1) siliceous material (e.g., diatomaceous fragments); 2) long-chain fatty acids; 3) short-chain fatty acids; 4) free saccharides and polysaccharides; and 5) fluorescent humic-like substances. 13 Kirpes et al. reported a category S5 for amino acids, which is consistent with other reports of free and combined amino acid enrichment at the sea surface microlayer and within SSA. 14,[17][18][19] Additional categories of organic matter could be argued to include alcohols, sterols, phospholipids, and hydrocarbons, but are omitted from this work for simplicity. 1,20,21 Thus, the most abundant and consistently-reported components across ambient marine aerosol field observations are fatty acids, protein, saccharides, and humic-like substances. Humic acids are water soluble, macromolecular fluorescent compounds of biological origin, with molecular weights, connectivities, and compositions, all of which vary with biological processing. Since a specific molecular structure for humic acid was unavailable at the time of writing, and due to the complexity of its analysis and parameterization, it was not included in our models. The remaining major categories of organics are, however, represented in this work. Fatty acid chain lengths, including long and short chains, as well as their relative ratios, were selected based on anionic surfactant speciation by Cochran et al. 21 The protein we incorporated is Burkholderia cepacia (BCL) lipase, which has been found to retain its activity in nascent SSA 22 and has been extensively modeled and characterized by our group and others. [23][24][25][26][27] We then incorporated three classes of saccharides: anionic polysaccharides (lipopolysaccharides), neutral oligosaccharides (laminarin), and free monosaccharides (glucose), each of which has specific and unique properties that can influence SSA phase, shape, viscosity, and reactivity.
We note that all fatty acids here are fully protonated to reflect the low pH environment (<6) 28  The water content of SSA depends on a variety of factors, including size, age, hygroscopic growth factor and relative humidity. The water content of the SSA in this work was based on the estimates derived by Bertram et al. for freshly-emitted SSA at a relative humidity of 80% 1 .

4) Timescales.
We are interested in investigating the impact of chemical complexity on SSA morphology and dynamics, specifically for those systems sizes which are too small to reliably experimentally validate. We selected a timescale of 500 ns -1 s for our simulations based on small-scale studies of fatty acid aggregation and mixing.
The time-resolved information gathered from particle shape evolution and clustering analyses (Figures 2 and 3) indicate that these parameters reach consistent values after ~200 ns.

Brewster Angle Microscopy Experimental Details
The following discussion and Figures S5-7 are reproduced from the dissertation of Man Luo. dimensions. It has been widely used for characterizing one molecule thick monolayers at the air/water interface.
The BAM uses p-polarized light at the Brewster Angle of incidence. At the Brewster angle, at the air/water S8 interface there is no reflection from the air/water interface and the background is dark. When a condensed phase of a monolayer with a different refractive index is present at the air/water interface, light reflection will occur, and a high contrast image can be generated. Overall, BAM can be used to investigate the morphology of the monolayer at the air/water interface by examining dark and light regions of the surface. Figure S5: Schematic of the BAM coupled with the Langmuir trough.
A schematic of the BAM setup with the Langmuir trough is shown in Figure S5. Using this setup, BAM images can be obtained during the monolayer compression process. Therefore, the morphology change of the monolayers during compression can be studied. One barrier was taken off from the trough in order to place the BAM instrument and the remaining barrier was used to compress the monolayer.
BAM images obtained during compression provide insights into surface monolayers. Figure S6 shows an example of how mixed FA monolayers (here termed "Mix-FA" and composed of the 1: 2: 4: 3 ratio described in the Experimental Methods section) can be disrupted by LPS in the subphase. In this experiment, BAM images are collected as the monolayer is compressed up until a surface pressure of 30 mN/m. These images are shown at surface pressures before the complete collapse of the monolayer (see Figure S7).

1) Asphericity and Relative Shape Anisotropy.
Here, we use the definition of asphericity derived by Theodorou and Suter: 34 The quantities ! , ! , and ! , are given by the symmetric radius of gyration tensor, such that the eigenvalues are ! ≥ ! ≥ ! . S, derived from the definition where N represents the number of atoms in the system, ∈ {1,2,3}, and ∈ {1,2,3}, is computed directly from atomic coordinates. The squared radius of gyration, ) ! , is given by the first invariant of S, and is used to calculate the relative shape anisotropy, k 2 . This term is a shape descriptor of the symmetry of the particle; that is, a value of 1 indicates a rod-like symmetry, where all atoms lie along a line, while a value of 0 indicates the particle has a higher degree of symmetry, such as that of a perfect tetrahedron or sphere. 1,3 Here, we use the definition of k 2 given by Theodorou and Suter: 1 A close inspection of this equation reveals that the quantity in the numerator is a function of the asphericity and the parameter acylindricity, given by ! − ! , where a value of zero indicates perfect cylindrical symmetry. A more detailed description of these quantities is out of the scope of this work but can be found in the provided references.

S11
To quantify distribution of organic material throughout the particle, we calculate the best-fit ellipsoid to the dataset. Let n, M be the total particle mass, and a, b, and c correspond to the dimensions of each semi-axis, as illustrated by Figure S8. I can be calculated using the internal measure inertia command in VMD, 35 which returns the eigenvalues and the principal axes. Figure S8: Sketch of an ellipsoid with axes of symmetry labeled.
We then subdivide the ellipsoid into three regions by volume, termed "core," "bulk," and "shell," using the approach outlined by Karadima et. al. 36 Drawing concentric ellipsoids along the principle axis each enveloping approximately a third of the volume, we can describe the distribution of mass by type into each region of the particle.

3) Mean Square Deviation and Diffusion.
Here, we computed the diffusion coefficients for water across the various systems using the Einstein formula to calculate the mean squared displacement (MSD), where N is the number of equivalent particles, r is their coordinates, and d = 3 for movement in 3 dimensions.
This, specifically, was done utilizing MDAnalysis's tool EinsteinMSD. [37][38][39] The diffusion coefficient, D, is related to the MSD by equation (SE7). Performing linear regression on the "middle" region of the MSD curve allows for the estimation of the diffusion coefficient from the slope of the linear regression. The "middle" region can be described as the linear region of a log plot of the MSD. 40 For this work, it was defined as from 1-9 ns of S12 the MSD calculation as the MSD above 10 ns can deviate from linearity ( Figure S9). The TIP3P water model tends to over-estimate the rate of diffusion due to the lack of longer-range water structure. Despite this our calculations still yield diffusion coefficients significantly lower than both computationally determined TIP3P values and the experimentally determined values of pure water at 298 K. 41 The spherical shell radii were selected semi-arbitrarily to give similar numbers of water molecules in each shell and to account for the asphericity observed in certain systems. Each of the selections contains at least 40,000 water molecules used for the MSD calculations.

Figure S9
: Example log plot of MSD from C3. The core region deviates from linearity after 10 ns leading to the 1-9 ns region being defined as the "middle" region for diffusion coefficient calculations.

4) Surface Curvature Estimation.
Here, we use the inverse radius of the best fit sphere as our definition of curvature. That is, for the given set of datapoints, we find the best fit sphere using a NumPy linear algebra least squares function. In this case, the data points are the points defined by the fatty acid headgroups. Figure S10 shows some example sphere fits to lipid selected lipid clusters. S13 Figure S10: Spherical fits to selected lipid clusters to demonstrate the sphere fit function.

5) Estimation of Surface Area Coverage.
To estimate the total surface area, we must first know the area per lipid of the particular surface cluster.
Surface clusters were identified manually by visual inspection. The estimation of area per lipid is non-trivial.
There are many methods one could use for this calculation, but the procedure we followed is enumerated below.
1. The radial distribution function (RDF) was calculated between the headgroups (the first carbon atoms) using MDTraj. 42 This allows us to extract the average distance between headgroups.
2. Using the RDF plot, we used NumPy 43 to fit a gaussian curve of the function to the first peak, which corresponds to the distribution of nearest headgroups. We then extract the mean value ( ) as the radius and the standard deviation ( ) as the error associated with the calculation.
3. Using the radius r of the best fit sphere to the lipid cluster calculated above, we can estimate the average area covered by a lipid on a curved surface using the geometric principles of a spherical cap, illustrated in Figure S11, where the distance between two lipid headgroups is defined as 2a. gives the estimate of the total area covered by the lipid cluster.

6) Estimation of Ellipsoidal Surface Area Coverage.
One equation estimating the surface area of an ellipsoid is given by Knud Thomsen's Formula: where the constant p = 1.6075 gives an approximate error ≤1.061%. [44][45][46] The estimated surface area can thus be calculated using the semi-axis values extracted from the best fit bounding ellipsoid calculation above.

7) Error Propagation
Error was propagated using the equations in Table S4 for the analytical propagation of uncertainty. 47 Table S4: Equations for the propagation of uncertainty.

Type of Analysis Example Equation Eq
Addition/Subtraction

Movies
Movie M1: Surface view of System A1 during first 50 ns of simulation. Initial frames show randomly distributed components evolving during minimization, heating, and equilibration, followed by 50 ns of production. Orange, red, and yellow molecules represent fatty acids lauric, myristic, palmitic and stearic acids; water is given by blue dots, and lipase is given in lime green QuickSurf representation (VMD).
Movie M2: Surface view of System A1 during final 50 ns of simulation, showing nearly complete surface saturation by fatty acids (red, orange & yellow), with lipase (lime green) inserting in between fatty acid patches. Orange, red, and yellow molecules represent fatty acids lauric, myristic, palmitic and stearic acids; water is given by blue dots, and lipase is given in lime green QuickSurf representation (VMD).
Movie M3: Cross section view of System B1 during final 50 ns of simulation, showing distribution of organic material throughout the center of the particle. Notable are fatty acids (red, orange & yellow) forming into bilayer and aggregate clusters inside the particle. Lipase (lime green) and LPS (purple) inserts both into fatty acid aggregates in the center and the monolayers at the surface. Orange, red, and yellow molecules represent fatty acids lauric, myristic, palmitic and stearic acids; water is given by blue dots, lipase is given in lime green QuickSurf representation (VMD), and LPS is given in purple by VdW representation (VMD).
Movie M4: Surface view of System B1 during final 50 ns of simulation, showing incomplete complete surface saturation by fatty acids (red, orange & yellow), with lipase (lime green) and LPS (purple) inserting in between fatty acid patches. Orange, red, and yellow molecules represent fatty acids lauric, myristic, palmitic and stearic acids; water is given by blue dots, lipase is given in lime green QuickSurf representation (VMD), and LPS is given in purple by VdW representation (VMD).
Movie M5: Cross section view of System C3 during final 50 ns of simulation, showing impeded molecular diffusion throughout the particle by LPS (magenta), protein (lime), laminarin (pink) and fatty acid (red, orange, & yellow) aggregation. Orange, red, and yellow molecules represent fatty acids lauric, myristic, palmitic and stearic acids; water is given by blue dots, BCL lipase is given in lime green QuickSurf representation (VMD), and LPS, laminarin, and glucose are given in magenta, pink, and indigo, respectively, by VdW representation (VMD).