CDK12 inactivation across solid tumors: an actionable genetic subtype.

Inactivating CDK12 alterations have been reported in ovarian and prostate cancers and may have therapeutic implications; however, the prevalence of these mutations across other cancer types is unknown. We searched the cBioPortal and GENIE Project (public release v4.1) databases for cancer types with > 200 sequenced cases, that included patients with metastatic disease, and in which the occurrence of at least monoallelic CDK12 alterations was > 1%. The prevalence of at least monoallelic CDK12 mutations was highest in bladder cancer (3.7%); followed by prostate (3.4%), esophago-gastric (2.1%) and uterine cancers (2.1%). Biallelic CDK12 inactivation was highest in prostate cancer (1.8%), followed by ovarian (1.0%) and bladder cancers (0.5%). These results are the first (to our knowledge) to estimate the prevalence of monoallelic and biallelic CDK12 mutations across multiple cancer types encompassing over 15,000 cases.

INTRODUCTION

Inactivating CDK12 alterations have been reported in ovarian and prostate cancers; however, the prevalence of these mutations across all cancer types is unknown [1]. While CDK12 was initially thought to be involved in homologous-recombination DNA repair, emerging data suggest a unique role of this gene in DNA replication-associated repair. To this end, it has been suggested that inactivating CDK12 mutations lead to widespread focal genomic duplications that generate gene fusion-induced neoantigens and favorable responses to immune-checkpoint blockade therapy using PD-1 inhibitors [2]. Given this potentially actionable molecular subtype, we sought to determine the prevalence of monoallelic and biallelic CDK12 alterations across tumor types.

RESULTS

Datasets (in cBioPortal and GENIE) from prostate, breast, colorectal, bladder, ovarian, uterine, head-and-neck squamous cell carcinoma, melanoma, and esophago-gastric cancers were included (Table 1); other tumor types did not reach a 1% frequency of CDK12 alterations. The prevalence of at least monoallelic CDK12 mutations was highest in bladder cancer (3.7%); followed by prostate (3.4%), esophago-gastric (2.1%) and uterine cancers (2.1%). Biallelic CDK12 inactivation was highest in prostate cancer (1.8%), followed by ovarian (1.0%) and bladder cancers (0.5%) (Figure 1).

Table 1

Datasets publically available from cBioPortal and GENIE Project that were used, by disease group, with overall sample size

Disease	Dataset	Sample Size	Total
Bladder	BLCA_TCGA_PAN_CAN_ATLAS_2018	408	1,181
	DFCI-ONCOPANEL-3	69
	MSK-IMPACT341	95
	MSK-IMPACT410	326
	MSK-IMPACT468	143
	UTUC_MSKCC_2013	84
	VICC-01-T5A	3
	VICC-01-T7	53
Breast	BRCA_IGR_2015	216	3,442
	BRCA_MBCPROJECT_WAGLE_2017	157
	DFCI-ONCOPANEL-3	304
	MSK-IMPACT341	410
	MSK-IMPACT410	1,021
	MSK-IMPACT468	1,076
	VICC-01-T5A	87
	VICC-01-T7	171
Colorectal	CRC_MSK_2018	1,134	3,272
	DFCI-ONCOPANEL-3	351
	MSK-IMPACT341	209
	MSK-IMPACT410	906
	MSK-IMPACT468	465
	VICC-01-T5A	47
	VICC-01-T7	160
Esophagogastric	DFCI-ONCOPANEL-3	146	1,458
	EGC_MSK_2017	341
	ESCA_TCGA_PAN_CAN_ATLAS_2018	182
	MSK-IMPACT341	122
	MSK-IMPACT410	216
	MSK-IMPACT468	106
	STES_TCGA_PUB	288
	VICC-01-T5A	11
	VICC-01-T7	46
HNSCC	DFCI-ONCOPANEL-3	83	1, 010
	HNC_MSKCC_2016	151
	HNSC_TCGA_PAN_CAN_ATLAS_2018	517
	MSK-IMPACT341	37
	MSK-IMPACT410	132
	MSK-IMPACT468	75
	VICC-01-T5A	6
	VICC-01-T7	9
Melanoma			906
	MSK-IMPACT341	64
	MSK-IMPACT410	364
	MSK-IMPACT468	214
	VICC-01-T7	140
	VICC-01-T5A	37
Ovarian	DFCI-ONCOPANEL-3	125	1,065
	MSK-IMPACT341	88
	MSK-IMPACT410	139
	MSK-IMPACT468	155
	OV_TCGA_PUB	489
	VICC-01-T5A	31
	VICC-01-T7	38
Prostate	DFCI-ONCOPANEL-3	97	2,251
	MSK-IMPACT341	153
	MSK-IMPACT410	569
	MSK-IMPACT468	377
	PRAD_FHCRC	149
	PRAD_MICH	61
	PRAD_MSKCC	194
	PRAD_SU2C_2015	150
	PRAD_MSKCC_2017	501
Uterine	DFCI-ONCOPANEL-3	120	1,085
	MSK-IMPACT341	119
	MSK-IMPACT410	258
	MSK-IMPACT468	326
	UCEC_TCGA_PUB	232
	VICC-01-T7	19
	VICC-01-T5A	11

Figure 1

Prevalence of CDK12 mutations across 9 cancer types

DISCUSSION

In the era of precision oncology, inactivation of CDK12 may represent a new molecular subtype with therapeutic implications [6], although the pan-cancer prevalence of this genomic alteration was previously unknown. These results are the first (to our knowledge) to estimate the prevalence of monoallelic and biallelic CDK12 mutations across nine cancer types encompassing >15,000 cases. This is important as CDK12 alterations may be implicated in favorable responses to immune checkpoint inhibition, with biallelic alterations theoretically expected to respond better than monoallelic alterations. Prospective clinical trials (e.g. NCT03570619) are now needed to adequately assess this therapeutic hypothesis, and our data could be useful in the design of such trials.

Our results are limited to data that were publicly available. In addition, genotyping and mutation calling are sensitive to several factors, e.g. quality of the sample, sequencing depth and platform, and the pipeline used. Additionally, datasets from the GENIE Project revealed overall lower CDK12 mutation rates than datasets retrieved from cBioPortal. The reason for this is unclear but may include different pipelines with different sensitivity and specificity, artifacts due to DNA damage in sample preparation found in the capture-panels used in the GENIE Project, and differing sample quality (all samples from the GENIE Project were formalin-fixed paraffin-embedded while most from cBioPortal were fresh-frozen samples) [3-5]. Because of this, we hypothesize that our reported prevalences are likely underestimates of the true frequency of these mutations. Nevertheless, our analysis suggests that there are at least nine cancer types with a CDK12 mutation prevalence between 1-4%, hopefully prompting further exploration of immunotherapy approaches using a basket-trial design. Given the recent FDA-approval of larotrectinib for NTRK-altered cancers regardless of histologic type, we envision a similar mode of clinical exploration for CDK12-altered tumors.

METHODS

We searched the cBioPortal [3,4] and GENIE Project (public release v4.1) [5] databases for cancer types with ≥200 sequenced cases, that included patients with metastatic disease, and in which the prevalence of at least monoallelic CDK12 alterations was ≥1%. Analyses were restricted to datasets containing both CDK12 mutation and copy-number alteration (CNA) data using hybridization-capture panels from Dana-Farber Cancer Institute, Memorial Sloan-Kettering Cancer Center and Vanderbilt-Ingram Cancer Center. CDK12 mutations were considered inactivating (i.e. resulting in loss-of-function) in the case of homozygous loss, genomic rearrangements, frameshift or nonsense protein-truncating mutations, splice-site mutations, or missense mutations within the kinase domain. Monoallelic alterations were defined as at least one protein-truncating CDK12 variant; biallelic alterations were defined as a protein-truncating variant plus a second protein-truncating variant, a kinase domain missense variant, or loss-of-heterozygosity of the wild-type CDK12 allele. All analyses were performed in R.