FWIW, that seems like the wrong approach to take. You should be using something that actually knows how to parse that record format (BioPerl ...).

Now to your code, gene=LOC653505 comes after gene=LOC644591, so you have to parse the file twice, example:

#!/usr/bin/perl --

use strict;
use warnings;

sub scan_gene {
    my $fh = shift;
    my $tell = tell $fh;
    my @gene;
    while(readline $fh){
        push @gene, $1 if /gene=(\S+)/;
    }
    seek $fh, $tell, 0;
    return @gene;
}

my @gene = scan_gene(\*DATA);
my $geneix = 0;

#warn "@gene";

while(<DATA>){
    my( $before, $sign, $after ) = split /([+-])/, $_, 2;
    print "$before $sign $gene[$geneix]\n";
    $geneix++ if index($after, $gene[$geneix]) != -1;
#    warn "before $before\nsign $sign\nafter $after";
}


__DATA__
NT_113797    CDS    122829    123323    -  gene=LOC644591  ProteinID=X
+P_932799.1  
NT_113798    CDS    4457    4636    - 
NT_077932    CDS    9894    9928    - 
NT_077932    CDS    65297    65828    + 
NT_077932    CDS    89196    89690    -  gene=LOC653505  ProteinID=BJD
+ND993
[download]