By specifying a format string of '%s', you effectively asked textscan for a single output. But then you asked to read a string 5 times. Consequently output is a single cell which contains five things. Those five things are themselves cells, which contain the strings you want. (Confused?)
So to get the third header, do
header{1}{3}
Or, you could do
textscan(fid, '%s%s%s%s%s', 1, ...
In which case you will get back 5 cells. But then each of them contains a single cell (which contains a string)!
Unfortunately this is just the way textscan works. It makes sense for reading whole columns of data, but not so much for a header line.
You can always do what you were doing and add
header = header{1}
Or do it the other way I showed above and add
header = [header{:}]
Either way, then, header will be a 5-element cell array of the strings.
EDIT, attempting to clarify: textscan returns an n-element cell array, where n is the number of format specifiers in the format string. Each cell will contain the contents that were read from the file. If the contents were numeric, you'll just get a numeric array. But if they were strings, you'll get a cell array of strings. That's why you're ending up with cell arrays containing cell arrays.
