XAS at the Fe K-edge was used to probe the oxidation state and coordination environment of the Fe in the samples. Spectra with a resolution of 0.8 eV were collected on beamlines 4-1, 4-3, and 11-2 at the Stanford Synchrotron Radiation Lightsource for the dust samples and a set of mineral standards (7 ). The monochromator used was Si(220) at 90°, and the energy calibration was carried out using an Fe foil at 7112.0 eV (at the first inflection edge, that is, the maximum in the derivative spectrum). All spectra were collected in fluorescence mode using a 13- or 30-element Ge detector. The mineralogy of the samples was determined using LCF (7 ), and PCA was conducted (75 , 76 (link)) primarily to group samples by oxidation state using principal components #1 and #2. To do this, principal components were determined for the sample set and then specific minerals with known compositions were fit with these principal components, after which the fraction of component #2 was regressed versus Fe(II) content to establish a linear relationship between the two (R2 = 0.96); this relationship was used for each sample to determine Fe(II) content (fig. S2). Errors were reported as 67% confidence intervals based on the calibration curve. This approach is advantageous because it does not require knowledge of specific mineralogy. In contrast, LCF methods require knowledge of component minerals for accurate quantification of Fe(II), which can be difficult for some samples. However, LCF allows us to further characterize the minerals present, for example, to differentiate between Fe(II) carbonates and Fe(II) silicates or to differentiate Fe(III) in hematite from that of goethite. Pyrite, siderite, goethite, hematite, magnetite, biotite, hornblende, ferrihydrite, and glauconite standards were all used for LCF. Iron(II) content was calculated using LCF based on the oxidation state of Fe in pure minerals. We considered hornblende to contain 50% Fe(II) and 50% Fe(III). The spectra from a subset of the standards are plotted with oxidation state in fig. S3 to show similar trends in edge position and oxidation state to Fig. 2. Errors on these estimates are the errors generated by the SIXPack interface (77 ) propagated.