refGene.txt sizes and frameshifts of exons not consistent

[jdenvir]

I need to generate amino acid sequences and substitutions from genomic SNPs and reference genome sequences with gene annotations. Currently I'm doing this for mouse but will likely need to do this for human and other organisms. I'm using mm9 as the reference; the chromosome fasta files for the genomic reference sequence and refGene.txt from UCSC for the gene annotation.

I have noticed several examples of places in refGene.txt where the exon frameshifts are not consistent with the exon sizes. For example, in NM_001033875 the second exon runs from 141,401,096 to 141,401,187 which is 91 bases (these are zero-based, inclusive at the start and exclusive at the end, or some equivalent interpretation). The frameshift for this exon is listed as 1, so the frameshift for the next exon should be 2; however it is listed as 0. This means there is an additional nucleotide that is part of the coding sequence.

This example is clearly visible in IGV by loading mm9 and navigating to chr4:141,400,995-141,401,117 - note the two nucleotide overhang at the end of exon 2 (on the right) while there is no frameshift at the beginning of exon 3 (on the left).

Can anyone explain this discrepancy? Is there a way to infer the nucleotide that is "missing" from the coding sequence?