• autoxidation
• Criegee intermediate
• unsaturated lipids

As a consequence of the Earth’s oxygen-rich atmosphere, reactive oxygen species (ROS) are formed ubiquitously during photochemistry, pollution events, smoking, and in vivo during cellular respiration and as byproducts of biochemical pathways (1, 2). ROS include hydroxyl (OH) and hydroperoxyl radicals (HO₂), hydrogen peroxide (H₂O₂), superoxide, and excited states of O₂. ROS degrade organic molecules by initiating and propagating free-radical chain reactions known as autoxidation (3). In cells, autoxidation of lipids, known as peroxidation (3), disrupts membrane structure and function and induces oxidative stress and rancidity in foods. Oxidative stress (3, 4) is widely implicated in aging, cancer, neurodegeneration, diabetes, and so on (5, 6). Substantial commercial effort is devoted to antioxidant formulations to short-circuit autoxidation to prevent disease and prolong the shelf life of foods, plastics, and other industrial products (7). Rational design of antioxidants in turn depends upon understanding the rates, mechanisms, and intermediates involved in autoxidation.

Autoxidation (i.e., lipid peroxidation) is a free-radical chain reaction that occurs in three phases: initiation, propagation, and termination (3). For lipids that contain C=C bonds, evidence gathered since the 1940s shows that autoxidation is initiated via H abstraction by ROS and other radicals at allylic and biallylic reaction sites (8, 9). Propagation occurs via H-atom transfer to peroxy radicals (RO₂) to form hydroperoxides (ROOH), which maintains the radical pool (Fig. 1A). Decomposition and further reactions of ROOH form additional radical products that amplify the chain reaction. However, a number of studies report reaction products (e.g., epoxides and various scission products) that are not easily explained by this established mechanism (Fig. 1A), suggesting that other reaction networks are involved (10, 11). Here, we present evidence for an additional autoxidation mechanism initiated by OH addition to C=C bonds and propagated by pathways involving Criegee intermediates (CIs).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1.

Two reaction schemes for the OH-initiated autoxidation of unsaturated lipids. (A) The established lipid peroxidation scheme, which is initiated by H abstraction at allylic C-H sites and propagated by peroxy radicals (RO₂) and hydroperoxides (ROOH). (B) CI-driven lipid autoxidation mechanism, which is initiated by the OH addition reactions and propagated by chain reactions involving the CI. The reaction of two RO₂ radicals are common termination steps in both mechanisms. Additional termination reactions of CI include their reactions with aldehydes (RC = O) to produce SOZs, with alcohols (ROH) to produce α-hydroperoxide ethers, and with carboxylic acids (ROOH) to produce hydroperoxide esters.

Previous mechanistic studies of lipid peroxidation were typically initiated by azo compounds (3, 12) to mimic the role of ROS, which are difficult to study due to their high reactivity and complex kinetics. The use of azo initiators, as ROS proxies, preselects the main autoxidation pathways to proceed via the established H abstraction route shown in Fig. 1A. Thus for OH, which can both rapidly abstract H atoms as well as add to C=C double bonds (13), azo initiators may not always capture the operative pathways in the environment or in biology.

The hydroxyl radical (OH) is readily formed in the environment/atmosphere photochemically (1) and in vivo (2). Despite its substantial reactivity toward all major classes of biological molecules, very few studies have focused on isolating OH reactivity from other ROS (14). Early studies suggest that the reaction mechanism of OH with unsaturated lipids might be rather different from other ROS species (13).

The OH-initiated autoxidation of model liquid-phase lipids is examined using a continuous flow stirred tank reactor (15). Lipid nanodroplets are introduced as aerosols and exposed to gas-phase OH, produced in situ (Methods). Reaction kinetics and products are measured in real time using a vacuum UV aerosol mass spectrometer (16, 17). This approach differs substantially from previous experiments but is designed to achieve controlled and quantifiable OH initiation and propagation rates. This is done by precisely measuring the gas-phase OH concentration ([OH]) and droplet surface area, which together yield, via gas kinetic theory, a precise measure of the OH-droplet collision frequency (i.e., initiation rate). From the OH collision frequency and consumption rate of the lipid, an effective reaction probability (γ_eff) can be determined (15). γ_eff is the fraction of OH collisions that yield a reaction. If the lipid is consumed at or near the OH collision frequency, then γ_eff ≤ 1, whereas values larger than 1 indicate the presence of free-radical chain reactions (i.e., autoxidation) (18, 19).

Using this approach, the OH reaction kinetics and products were measured for a series of neat model lipid compounds: squalene (Sqe, C₃₀H₅₀), arachidonic acid (AA, C₂₀H₃₂O₂), linoleic acid (LA, C₁₈H₃₂O₂), oleic acid (OA, C₁₈H₃₄O₂), and cis-9-tricosene (Tri, C₂₃H₄₆). Sqe is a major constituent (∼12%) in human sebum (20), while AA, LA, and OA are important unsaturated fatty acid components of phospholipid cell membranes (3). These compounds differ in the number of C=C, allylic and biallylic C–H bonds, and functional groups enabling insight into how molecular structure alters autoxidation rates and product distributions (3).

Shown in Fig. 2 is γ_eff as a function of [OH] for Sqe, AA, LA, OA, and Tri. [OH] is kept low to mimic environmental and biological conditions where autoxidation occurs. For those molecules with C=C bonds, γ_eff > 1. For Sqe, at [OH] ∼10⁶ cm⁻³, γ_eff approaches ∼70, indicating that for every OH collision, 70 Sqe molecules are additionally consumed by radical chain cycling pathways. At [OH] ∼10⁷ cm⁻³, γ_eff is around 21, 31, 8, 6, and 3 for Sqe, AA, LA, OA, and Tri, respectively. γ_eff is observed to increase with the number of C=C bonds with the exception of AA, which exhibits a slightly larger reactivity than that of Sqe despite having two fewer C=C bonds. This trend points to the key role that C=C bond number plays both in the initiation and propagation steps, as discussed below.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

Effective reaction probability (γ_eff) as a function of OH concentration ([OH]). γ_eff of Sqe, AA, LA, OA, Tri, squalane (Sqa), 2-decyl-1-tetradecanol and 2-octyl-1-dodecanol as a function of [OH] at an RH of 30%. The data for Sqa are from ref. 15. Values of γ_eff ≤ 1 indicate that the reaction is proceeding at or below the OH collision (i.e., initiation) rate. Values larger than 1 are evidence for radical chain reactions and an overall autoxidation rate that exceeds the OH collision frequency.

For the unsaturated lipids, decreasing [OH] increases γ_eff, providing clear evidence for radical cycling (i.e., autoxidation), since higher [OH] produces a larger radical density in the droplet, which favors radical + radical termination reactions (e.g., RO₂ + RO₂, Fig. 1) (18). In contrast, radical chain cycling is not observed (γ_eff < 1) for three saturated compounds: squalane (Sqa, no C=C double bonds, γ_eff ∼ 0.5) (15), 2-octyl-1-dodecanol (γ_eff ∼ 0.2), and 2-decyl-tetradecanol (γ_eff ∼ 0.3).

While the observations in Fig. 2 confirm that autoxidation occurs and is consistent with previous studies, the major lipid reaction products that are detected are not. Shown in Fig. 3 is a mass spectrum recorded for Sqe (m/z = 410) during the OH reaction. A series of peaks are observed at m/z = 248, 316, and 384 and are identified as aldehydes (C₁₇H₂₈O, C₂₂H₃₆O, and C₂₇H₄₄O). Other reaction products are observed at m/z = 322, 390, 458, 526, and 594 and are identified as secondary ozonides (SOZs). The observation of SOZs is unexpected and together with the aldehydes are nearly identical to the products typically observed in alkene ozonolysis reactions. For example, the ozonolysis of Sqe (21) produces the same five SOZs (SI Appendix, Fig. S3), which are formed by reactions of CIs with carbonyl compounds (21, 22). Extensive experimental checks confirmed that the ozone concentration in our reactor is below 1–3 ppb and therefore too small to account for the observed products (SI Appendix, Fig. S1). Thus, the detection of SOZs, instead of hydroperoxides (ROOH, Fig. 1A), formed by OH autoxidation provides a distinctive fingerprint for the presence of CI. This is consistent with recent observations by Beauchamp and coworkers (23), who reported evidence for CI formed during OH reactions.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3.

(A and B) VUV-AMS of the OH + Sqe reaction products. (C) Reaction products and (D) effective reaction probability as a function of alcohol mole fraction. (A) Difference mass spectra (unreacted Sqe – reacted Sqe) showing aldehyde products (C₁₇H₂₈O, C₂₂H₃₆O, and C₂₇H₄₄O) and five SOZs. (B) Difference spectra showing product distribution when the CI scavenger (2-decyl-1-tetradecanol) is added to Sqe. Major products are the same three aldehydes, and six α-hydroperoxide ethers (with their structures and fragments in colors). (C) SOZs, aldehydes, and α-hydroperoxide ethers as a function of alcohol mole fraction (2-decyl-1-tetradecanol). (D) γ_eff of Sqe and 2-decyl-1-tetradecanol as a function of alcohol percentage (2-decyl-1-tetradecanol). All experiments are performed at [OH] (∼1.2 × 10⁷ molecules/cm³) and under RH ∼0%.

As further confirmation for the presence of CI, two liquid alcohols (2-octyl-1-dodecanol and 2-decyl-1-tetradecanol) were separately added to the nanodroplets comprising Sqe (Fig. 3 and SI Appendix, Fig. S2). Alcohols are CI scavengers and upon reaction form distinctive α-hydroperoxide ether products (24). The alcohol scavengers are saturated and thus competing reactions with OH, under our reaction conditions, are very slow (Fig. 2).

As shown in Fig. 3B, the SOZ products disappear when adding the CI scavenger and a series of peaks form and are assigned to the expected α-hydroperoxide ether products. These products often do not appear as molecular ions but rather at m/z values corresponding to the elimination of either H₂O₂, HO₂, H₂O, or OH—likely due to dissociative photoionization. In separate control experiments, the same CI scavenger was used in the O₃ + Sqe reaction where CIs are known to form. The same set of peaks are observed (SI Appendix, Fig. S3) thus confirming the assignments in Fig. 3 for the OH reaction. The six distinct α-hydroperoxide ether products originate from the reactions (SI Appendix, Figs. S4 and S5) of the alcohol with the six possible CI (C₃H₆O₂, C₈H₁₄O₂, C₁₃H₂₂O₂, C₁₇H₂₈O₂, C₂₂H₃₆O₂, and C₂₇H₄₄O₂) formed in Sqe (21).

As shown in Fig. 3C, increasing the alcohol mole fraction decreases the quantity of SOZ, which is accompanied by a corresponding increase in aldehydes and α-hydroperoxide ether products. This is a consequence of the competition between CI + aldehyde reactions to form SOZ and CI + alcohol reactions that form α-hydroperoxide ethers.

Importantly, increasing the concentration of the CI scavenger (2-decyl-1-tetradecanol) decreases γ_eff (i.e., autoxidation rate) for Sqe (Fig. 3D). At an alcohol mole fraction of 48%, the radical chain length decreases by a factor of ∼2 from a value of 21 (pure Sqe). It is accompanied by a substantial increase in the apparent γ_eff of the alcohol (γ_eff > 10 in a 30% mixture with Sqe), as it is increasingly consumed by CI formed in the reaction. Together these observations show that CI reaction pathways are central both for the formation of SOZs and α-hydroperoxide ether products as well as for controlling the overall rate of autoxidation.

An additional set of experiments reveals the general importance of CI in the OH autoxidation of cis-9-tricosene and the three unsaturated lipid acids (AA, LA, and OA). These include reactions with OH (and separate control reactions with O₃ for comparison), with and without the CI scavengers (SI Appendix, Figs. S6–S13). With the addition of the alcohol scavenger to pure cis-9-tricosene, α-hydroperoxide ethers products appear, thus confirming the production of CI during OH autoxidation.

For the three fatty acids, the carboxylic acid group can additionally react with CI to produce hydroperoxide esters (25). These are detected, and confirmed using separate O₃ experiments (SI Appendix, Figs. S8–S13). For AA, α-hydroperoxide ethers are detected with the alcohol scavenger, while without the scavenger the hydroperoxide ester is not and is a subject for future investigation.

From these measurements, we can conclude that the dominant OH-initiated autoxidation products (e.g., SOZs) arise from CI pathways, which appear also to play a central role in controlling autoxidation rates (Fig. 3D). These products are not explained by the established autoxidation mechanism (Fig. 1A). Additionally, the substantial changes in the mass spectra observed when adding alcohol or water (described below) are not consistent with the established autoxidation mechanism, since peroxy radicals (RO₂), unlike CI, do not readily react with H₂O or alcohol.

Based these observations and reported CI pathways (26⇓–28), an autoxidation mechanism is proposed and shown in Fig. 1B. The initiation step involves the OH addition to the C=C bond of the lipid and subsequent O₂ addition to generate a β-hydroxyl peroxyl radical (Fig. 1B). This differs substantially from the established initiation step of H abstraction and is likely unique to OH (vs. other ROS). Two β-hydroxyl peroxy radicals can react to form stable termination products or alkoxy radicals. Alternatively, C–C bond scission of the β-hydroxyl peroxyl radical could produce a CI and α-hydroxyl alkyl radical, consistent with the recent results of Beauchamp and coworkers (23). In the gas phase, rapid unimolecular decomposition of the CI can regenerate OH (27). This OH, in turn, adds to another C=C bond to propagate the chain reaction. We propose that this is the key propagation step that explains the large γ_eff (Fig. 2) and its dependence upon C=C bond number (e.g., Sqe vs. Tri) as well as the decrease in γ_eff with the addition of the CI scavenger (Fig. 3). Bimolecular reactions of CI (Fig. 1B) terminate the chain reaction, such as the reaction with aldehydes to produce SOZs (22), or with alcohol to produce α-hydroperoxide ethers (24), which are consistent with the mass spectra shown in Fig. 3.

Water plays a substantial role in vivo and in the environment. Hence, we examined how water impacts the autoxidation rate of Sqe. As expected (Fig. 4), the concentration of SOZs decrease with increasing relative humidity (RH), reflecting the competition between reactions of CI with water vs. aldehydes as shown in Fig. 1B (21). Importantly, however, γ_eff does not change with increasing RH, indicating that water has no apparent effect on the magnitude of the autoxidation rate. The CI + H₂O reaction produces a carbonyl product and an additional ROS species—H₂O₂ (29, 30), which is likely photolyzed to reform OH thereby sustaining the autoxidation chain reaction. In vivo or in other environments, it is likely that OH would be regenerated from H₂O₂ via other pathways involving Fenton-type chemistry.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 4.

Effective reaction probability (γ_eff) of Sqe, relative fraction of SOZs and aldehydes as a function of % RH in the Sqe + OH reaction. (Left y axis) γ_eff of Sqe and (Right y axis) normalized sum of the five SOZs (C₂₀, C₂₅, C₃₀, C₃₅, C₄₀) and three aldehydes (C₁₇H₂₈O, C₂₂H₃₆O, C₂₇H₄₄O) as a function of RH. All experiments are performed at [OH] ∼1.5 × 10⁷ molecules/cm³. These results indicate that although the product distributions depend upon RH, the autoxidation rate does not.

Here we report evidence that lipid autoxidation initiated by OH is propagated by CI. This mechanism is an additional pathway to the established one (i.e., initiated by H abstraction and propagated by peroxy radicals) and is likely unique to OH. Despite the identity of the initiating ROS, the two pathways are in reality intertwined, since OH can be formed in the established mechanism either by the thermal decomposition of ROOH/H₂O₂ or by Fenton’s chemistry in biological systems. This mechanism generates many of the volatile aldehyde products associated with the onset of rancidity in food products and cellular membrane decomposition. The formation of SOZ products suggests new studies are needed to evaluate their chemistry in biological environments. More generally, these results imply that CIs play a much more general role, beyond atmospheric chemistry, in the oxidative destruction of organic molecules in a variety of natural and industrial environments.