Disruption refers to the phenomenon in which a schema represented by a particular parent is not passed down into its offspring.
That bit about schemata is important -- disruption isn't really defined except for in terms of schemata. Briefly, a schema is a ternary string containing 0s, 1s, and *s (don't-cares), whereas individuals are only binary (no *s). Each individual is a representative of $2^n$ schemata. For example,
10010011
is an instance of all of the following schemata:
1*******
*00**01*
1******1
...
The idea Holland had was that, for example, if good solutions tended to have a 1 in the leading bit, then the proportion of the population that is an instance of the schema
1*******
would increase over time according to a certain inequality. That's the schema theorem. Now, what if instead of that simple schema being more fit than average, we had a more complicated one, like
1*****0*
That is, suppose that individuals matching that pattern are above average in fitness. According to Holland, the GA should work by increasing the proportion of the population that conforms to that pattern over time.
But look at what happens with one-point crossover. It's going to pick a crossover point somewhere along the string and split it around that point. If both parents are instances of the schema, then both children will be as well. But if either parent is not an instance of the schema, then we have a probability that the children won't inherit that pattern. That failure to inherit the pattern is called disruption. Holland would have said that this was very bad. The presence of that schema in the population is evidence that it's good, and crossover is wrecking it. The modern understanding is much more nuanced than that, but that's the rationale behind considering disruption as a concept at least.
So you question asks whether schemata that have bits that are close together are more or less likely to be disrupted. We should really be more precise -- it's not whether any old bits are close together that matters. It's the distance between the first and last non-* bit that matters (called the defining length). The greater that distance, the more likely we'll put a crossover point in between them, and therefore the more likely we are to disrupt the schema.