Behold! the final gel! Building the cube up and tearing it down. Lane 4 contains the purified cube. That's cool. Lanes 5-11 are various restriction digests of the cube, which show that it breaks down in ways consistent with it having been a cube with 6 knotted faces in the first place. 7/8

They could prove ligation with exonuclease digestion! Two things in this busy gel:
1. Lane 4 losing the top band meant the cube should be 10 bases-per-side, not 11. This wasn't obvious as DNA has a periodicity of 10.5 bp.
2. Lane 2 keeping the top band from lane 1 means ligation was a success! 6/8

Here's the assembly gel where they add 5 strands to their first, circularized strand and then ligate the two halves of the cube together to get an open 'belt'. However they ran into an aggregation problem in the 6th step, which is why they did the purify-and-reconstitute step from figure 1. 5/8

Here's the plan: starting with 10 linear, synthetic DNA strands, Ned's team annealed strands onto a growing belt with intermediate ligation steps, resulting in a final structure which has 6 strands, each one corresponding to one face of the cube 2/8

Some ideas about using redox states of metal atoms as storage later, and we get some circuit diagrams! I have no idea how unique or viable these are, but the idea was to store information in a cubic lattice, which would allow read/write on single bits via external signals on individual "wires". 5/7

There's a problem though. DNA on its own isn't a good conductor, you need to add something else to it. Enter trans-polyacetyline (tPA). This polymer later earned the 2000 Nobel prize as the first conductive polymer (the basis of your OLED TVs today, though tPA itself is relegated to museums). 4/7

The rest of the figures are further validation; showing that the strands need to be mixed in a 1:1:1:1 ratio, the electrophoretic properties of junctions are different than duplexes, and also it melts differently. Junctions are simple, yet have subtle complexities! 5/5

And here's the all-important gel! I added annotations of which strands from Fig 1 are in each well along the top. The key point they're trying to make is that the 4-strand complex makes a structure larger than 3-strand complexes, but electrophoretically smaller than a 64-base duplex would be. 4/5

Behold! The junction! That sequence right at the junction, known as J1, was gospel in the DNA nanotech community for many years because, well, it worked. 3/5

These could then be connected via sticky ends to form 2D and 3D polyhedra and lattices. He originally hoped this could be used as easily-crystallized cages which could host hard-to-crystalize guest molecules. Alas, this was easier said than done, it took another 29 years to crystallize a 3D lattice.

This is the key innovation of the paper that seems blindingly obvious in retrospect. By synthesizing DNA strands with custom sequences, it would be possible to make "immobile junctions" which can't slide. He wrote a FORTRAN program to generate such sequences while avoided competing structures.

The fish in Depth reminded him of Holliday junctions, the intermediate structure in DNA recombination. In recombination, the two duplexes involved have similar sequences, which allows the junction to "slide" by exchanging individual base pairs between duplexes. But what if they couldn't slide...?

The thing you have to understand about Ned is that he was an x-ray crystallographer, a famously finicky method, and in 1982 he wasn't having much luck with things crystalizing. The story goes that he was sitting in the campus pub staring at a print of MC Escher's Depth when an idea hit him.