Perhaps you should consider writing a parser for this format - always a good idea if you're going to be dealing with it on a regular basis. Example:

#!/usr/bin/perl -w
use strict;

open my $Gpff, '<', 'protein.gpff' or die "protein.gpff: $!\n";

my ($key, %data);
while (<$Gpff>){
    chomp;
    if (/^(?:\s\s)?([A-Z]+)\s+(.*)$/){
        $key = $1;
        $data{$key} = $2;
    }
    else {
        s/^\s+/ /;
        $data{$key} .= $_;
    }
}
close $Gpff;
[download]

Given the above, you can now easily extract the data that you want by its label; that is, printing $data{SOURCE} will output "mitochondrion Ephydatia muelleri". You may need to adjust the parser to suit your exact application, but this should give you a good start.