Structural basis of DNA packaging by a ring-type ATPase from an archetypal viral system

Abstract Many essential cellular processes rely on substrate rotation or translocation by a multi-subunit, ring-type NTPase. A large number of double-stranded DNA viruses, including tailed bacteriophages and herpes viruses, use a homomeric ring ATPase to processively translocate viral genomic DNA into procapsids during assembly. Our current understanding of viral DNA packaging comes from three archetypal bacteriophage systems: cos, pac and phi29. Detailed mechanistic understanding exists for pac and phi29, but not for cos. Here, we reconstituted in vitro a cos packaging system based on bacteriophage HK97 and provided a detailed biochemical and structural description. We used a photobleaching-based, single-molecule assay to determine the stoichiometry of the DNA-translocating ATPase large terminase. Crystal structures of the large terminase and DNA-recruiting small terminase, a first for a biochemically defined cos system, reveal mechanistic similarities between cos and pac systems. At the same time, mutational and biochemical analyses indicate a new regulatory mechanism for ATPase multimerization and coordination in the HK97 system. This work therefore establishes a framework for studying the evolutionary relationships between ATP-dependent DNA translocation machineries in double-stranded DNA viruses.

. Biochemical properties of the HK97 DNA packaging system. Dependence on spermidine in the packaging of (A) the HK97 genome and (B) linear cos-containing pUC18 DNA.
Dependence on (C) divalent metal ions, (D) pH and (E) monovalent salts in the packaging of linear cos-containing pUC18 DNA. Reaction components were treated with Chelex-100 resin before addition of divalent metal ions. Samples with different salt levels were ethanol-precipitated before analysis by agarose gel electrophoresis. Ethidium bromide staining was used to visualize DNA in the agarose gel.

Analysis of photobleaching events
DNA particles were identified by extracting connected components from an average of the first 500 frames in the red channel. The intensity-weighted centroids of these components were taken to be DNA coordinates. Coordinates fewer than 10 px apart were rejected. For each coordinate in the red channel, given the translation vector from calibrating with TetraSpeck microspheres (ThermoFisher) emitting in both channels, the corresponding pixel in the green channel was found. A circle 3.5 px in radius, large enough only to enclose an event, was drawn around the center of this pixel. Mean intensity of the circle as a function of time was extracted and analyzed for photobleaching steps using the PIF algorithm (7) with Chung-Kennedy filtering (8). Over-fitted steps were identified by generating a counter-fit with steps between the original steps and comparing chi-squared statistics, as introduced by Kerssemakers et al. (9), and subsequently rejected.
For more robust analysis, a moving two-sample t-test (10) and a chi-squared minimization-based step-finding algorithm (9) were also applied. However, this resulted in a systematic under-fitting or over-fitting of steps, owing to the low signal-to-noise ratio of the experiment.
To build a distribution of the number of photobleaching steps per protein event co-localizing with DNA, an event was accepted only if steps could be fitted and if it had an intensity-weighted centroid within experimental error of the centroid of the DNA particle. A centroid-based measure was used because the density of events in the green channel was high and in turn background pixels surrounding each event could not be reliably defined for the fitting of Gaussian distributions. Adapting from Gelles et al. (11), experimental error was estimated to be the precision (two standard deviations) with which the xand y-position of one GFP-LT monomer could be assigned relative to another, in time. We calculated one standard deviation to be 40.4 nm in both directions.
We note that the number of photobleaching events could be underestimated due to the presence of a dark GFP population as a result of misfolding or photobleaching prior to the experiment; and due to mis-fitting of two photobleaching events as one because the two events occurred within the same video frame, as described in detail by Ulbrich and Isacoff (12). The observed photobleaching rate of the GFP fusion protein was 0.10 s -1 , typical of GFP proteins (13). We also acknowledge the possibility of overestimating the number of photobleaching steps when the event monitored was coincident with other passively absorbed GFP molecules.
If one assumed all molecules observed in the DNA-bound motor experiment were pentamers, a binomial distribution with 72% probability of GFP being fluorescent could be fitted. However, it was likely that lower-order species as observed in the absence of prohead and small terminase also contributed to the distribution. Given the complexity of the experiment, it was reasonable to conclude only that the upper limit for the oligomeric state of TerL in a motor assembly was 5.