The compartmentalization of eukaryotic cells presents considerable challenges to the herpesvirus life cycle. The herpesvirus tegument, a bulky proteinaceous aggregate sandwiched between herpesviruses’ capsid and envelope, is uniquely evolved to address these challenges, yet tegument structure and organization remain poorly characterized. Here we use deep-learning–enhanced cryoEM to investigate the tegument of human cytomegalovirus virions and noninfectious enveloped particles (NIEPs; a genome packaging-aborted state), revealing a portal-biased tegumentation scheme. We resolve atomic structures of portal vertex-associated tegument (PVAT) and identify multiple configurations of PVAT arising from layered reorganization of pUL77, pUL48 (large tegument protein), and pUL47 (inner tegument protein) assemblies. Analyses show pUL77 seals the last-packaged viral genome end through electrostatic interactions, pUL77 and pUL48 harbor a head–linker–capsid-binding motif conducive to PVAT reconfiguration, and pUL47/48 dimers form 45-nm–long filaments extending from the portal vertex. These results provide a structural framework for understanding how herpesvirus tegument facilitates and evolves during processes spanning viral genome packaging to delivery.

HCMV strain AD169 particles were mixed with 5-nm gold beads and applied onto freshly glow-discharged Quantifoil Holey Carbon grids, then plunge frozen into liquid ethane. Tilt series were acquired using the SerialEM software suite on a ThermoFisher Titan Krios electron microscope operated at 300 kV and equipped with a Gatan imaging filter, Gatan K2 Summit direct electron detector, and Volta phase plate (VPP). Prior to collecting each tilt series, we advanced and pre-conditioned the VPP by electron illumination at a dose of 12 nC for 60s to ensure a phase shift of ~0.3π. Movies were recorded with the K2 Summit camera operated in counting mode at 5 frames/s with a dose rate of ~8-10 e-/pixel/s. Tilt series were collected from -66° to +60° at 2° intervals at 53,000x nominal magnification (for a calibrated pixel size of 2.6 Å) with a cumulative electron dose of approximately 100 e-/Å2. Defocus was targeted to -0.6 μm.

Collected movies of raw tilt series were aligned and averaged using MotionCor2 to generate a single micrograph for each tilt angle. Tilt series were aligned by tracking gold fiducial beads, and tomograms were reconstructed using the IMOD 4.11 software package implementing the simultaneous iterative reconstruction technique (SIRT) method. We next trained IsoNet on our reconstructed tomograms. Full usage details of IsoNet and a complete description of its workflow have previously been elaborated (Liu et al., Nat Commun 2022). Briefly, we generated tomogram STAR files using the IsoNet function isonet.py prepare_star. Reconstructed tomograms were then downscaled using isonet.py rescale, resulting in 6x-binned tomograms with a pixel size of 15.6 Å/pixel. Because tomograms were acquired using a VPP, we did not perform CTF deconvolution. A tomogram-wide mask was created using isonet.py make_mask, which automatically targets and identifies feature-rich regions of the tomogram for subsequent subtomogram extraction and neural-network analysis. 400 subtomograms (96x96x96 pixels) centered on feature-containing masked regions were then randomly extracted from the 11 tomograms using isonet.py extract. Using isonet.py refine, we fed these subtomograms as raw inputs to generate the initial neural network for dataset-optimized missing-wedge correction and denoising. By default, 30 iterations of neural network and data refinement occur, with each successive iteration implementing both the previous iteration’s network and data output as new inputs. Denoising—in short, artificially introducing noise into our dataset to promote neural network discernment of noise within signal data—was gradually increased from none in the first 10 iterations to 0.2 in the final 5 iterations. The final trained neural network, specifically optimized on our entire set of tomograms, was then applied to binned individual whole tomograms using isonet.py predict to obtain final denoised and missing-wedge-corrected tomograms.