> cat foo.bed
chr1    123    234    .    .    +
chr1    456    789    .    .    -

> cat bar.bed
chr1    239    456
chr1    800    900
chr3    456    987

> closestBed -D a -id -a foo.bed -b bar.bed 
chr1    456    789    .    .    -    chr1    800    900    -12

It helps if you properly format the BED file. You may also consider the -io option. You can trivially convert your "BED" file into a real BED file with awk '{printf("%s\t%s\t%s\t.\t.\t%s\n", $1, $2, $3, $4)}' incorrectly_formatted.bed > correctly_formatted.bed.

If you're amenable to alternatives, you could use BEDOPS closest-features on strand-separated genes.

For instance, sort the genes and repeats, if not sorted:

$ sort-bed genes.bed > sorted-genes.bed
$ sort-bed repeats.bed > sorted-repeats.bed

Then separate your genes by strand with awk (or your tool of choice):

$ awk '($4=="+")' sorted-genes.bed > sorted-genes-forward.bed
$ awk '($4=="-")' sorted-genes.bed > sorted-genes-reverse.bed

Then run closest-features --dist on each gene set, against the repeats. The --dist operand reports the distances between the gene and the nearest upstream and downstream repeats.

In the forward strand case, we use awk to print out the first and second fields separated by the pipe character (|). The first field is the gene. The second field is the repeat element upstream from the forward-strand gene element:

$ closest-features --dist sorted-genes-forward.bed sorted-repeats.bed \
    | awk 'BEGIN {FS='|'} {print $1"|"$2}' - \
    > genes-repeats-forward-with-distances.bed

In the reverse strand case, we use awk to print out the first and third fields separated by the pipe character. The first field is the gene. The third field is the repeat element upstream from the reverse-strand gene element:

$ closest-features --dist sorted-genes-reverse.bed sorted-repeats.bed \
    | awk 'BEGIN {FS='|'} {print $1"|"$3}' - \
    > genes-repeats-reverse-with-distances.bed

If you want to put results back together at the end, just use BEDOPS bedops with the --everything operand to do a multiset union:

$ bedops --everything genes-repeats-*-with-distances.bed > final-answer.bed

Like this example shows, one nice thing about BEDOPS bedops is that it can do operations on many input files at once, not just two BED files.