The depressions appear in association with all resonance lines and low-excitation permitted transitions in the spectrum. Notably, no depressions are seen on the blue wings of He II λ1640, O III] λ1663, and N III] λ1750, and N IV] λ1486 (see Fig. 1 bottom and Fig. S1)

There are also weak troughs on the blue sides of weaker lines such as λ1335, Si II λ1302, C III* λ1176, and the P V and Fe III transitions in the 1130-1150 Å observed wavelength range. These troughs mainly affect the continuum, as there is little to no line emission associated with these features that can be modeled with a blue/red emission asymmetry (see Fig. 1 bottom)

By modeling the depressions as absorption, we can fit all strong emission lines with a similar profile that is largely symmetric. In addition, this same symmetric emission model can fit the historical spectra with only slight adjustments in parameters. If the depression is modeled as an emission asymmetry, drastically different line profiles are required IV, and He II, even in the same COS spectrum, and these profiles must change dramatically with respect to the archival spectra (Fig. S2)

The changes in the asymmetry of the broad UV emission lines due to absorption show a weak correlation with the opacity of the absorber in the soft X-ray spectrum, indicating a relation between them (Fig. 2C). To model the UV emission from NGC 5548, for the continuum we used a reddened power law (with extinction fixed at E(B-V) = 0.02 mag) plus weak Fe II emission longward of 1550 Å rest wavelength (32), broadened by 4000 km/s (FWHM), absorbed by a Galactic damped Lyα absorber with fixed NH = 1.45x1024 m-2 (33) and centered at -13 km/s (based on H I emission). For the brightest emission lines Si IV, C IV and He II) we use five Gaussian components — narrow line emission (NLR

~300 km/s FWHM), intermediate line emission (ILR

~860 km/s), modestly broad emission (~2500 km/s), broad emission (~8500 km/s), and very broad emission (~16,000 km/s) Si IV and C IV, each doublet component is modeled separately for the first four of these Gaussians. The relative intensity ratios are fixed at 1:1, assuming the emission is optically thick. Only a single component is used for the very broad emission. For Lyα, C IV, and He II, the best-fit values for the NLR and ILR components are very similar to those in the low-state STIS spectrum of, e.g., see (20). For all other lines, we fix the intensities and widths of the ILR and NLR components at values measured in theSTIS spectrum since these weaker lines cannot be reliably measured independently in our COS spectrum due to their weakness relative to the much brighter continuum and surrounding broader emission. The broad absorption troughs in the spectrum have a noticeable asymmetry. The troughs have a much greater extent to velocities blueward of their deepest point than they do to the red. To model these troughs, we used asymmetric Gaussians with a negative flux profile in which the half-width at half maximum (HWHM) of the blue side of the trough was larger than the HWHM