The loss of chromosome 9 open reading frame 72 (C9ORF72) expression, associated with C9ORF72 repeat expansions, has not been examined systematically. Three C9ORF72 transcript variants have been described thus far; the GGGGCC repeat is located between two non-coding exons (exon 1a and exon 1b) in the promoter region of transcript variant 2 (NM_018325.4) or in the first intron of variant 1 (NM_145005.6) and variant 3 (NM_001256054.2). We studied C9ORF72 expression in expansion carriers (n = 56) for whom cerebellum and/or frontal cortex was available. Using quantitative real-time PCR and digital molecular barcoding techniques, we assessed total C9ORF72 transcripts, variant 1, variant 2, variant 3, and intron containing transcripts [upstream of the expansion (intron 1a) and downstream of the expansion (intron 1b)]; the latter were correlated with levels of poly(GP) and poly(GA) proteins aberrantly translated from the expansion as measured by immunoassay (n = 50). We detected a decrease in expansion carriers as compared to controls for total C9ORF72 transcripts, variant 1, and variant 2: the strongest association was observed for variant 2 (quantitative real-time PCR cerebellum: median 43 %, p = 1.26e-06, and frontal cortex: median 58 %, p = 1.11e-05; digital molecular barcoding cerebellum: median 31 %, p = 5.23e-10, and frontal cortex: median 53 %, p = 5.07e-10). Importantly, we revealed that variant 1 levels greater than the 25th percentile conferred a survival advantage [digital molecular barcoding cerebellum: hazard ratio (HR) 0.31, p = 0.003, and frontal cortex: HR 0.23, p = 0.0001]. When focusing on intron containing transcripts, analysis of the frontal cortex revealed an increase of potentially truncated transcripts in expansion carriers as compared to controls [digital molecular barcoding frontal cortex (intron 1a): median 272 %, p = 0.003], with the highest levels in patients pathologically diagnosed with frontotemporal lobar degeneration. In the cerebellum, our analysis suggested that transcripts were less likely to be truncated and, excitingly, we discovered that intron containing transcripts were associated with poly(GP) levels [digital molecular barcoding cerebellum (intron 1a): r = 0.33, p = 0.02, and (intron 1b): r = 0.49, p = 0.0004] and poly(GA) levels [digital molecular barcoding cerebellum (intron 1a): r = 0.34, p = 0.02, and (intron 1b): r = 0.38, p = 0.007]. In summary, we report decreased expression of specific C9ORF72 transcripts and provide support for the presence of truncated transcripts as well as pre-mRNAs that may serve as templates for RAN translation. We further show that higher C9ORF72 levels may have beneficial effects, which warrants caution in the development of new therapeutic approaches.