Analysis of CDK12 alterations in a pan‐cancer database

Abstract Background CDK12 inactivation leading to increased neoantigen burden has been hypothesized to sensitize tumors to immune checkpoint inhibition. Pan‐cancer data regarding the frequency of CDK12 alterations are limited. We aimed to characterize CDK12 alterations across all cancer types through real‐world clinical‐grade sequencing. Methods This was a single‐center retrospective analysis of 4994 cancer patients who underwent tissue or blood genomic profiling, including CDK12 assessment, conducted as part of routine care from December 2012 to January 2020. Prevalence, clinical characteristics, and treatment outcomes of patients with tumors with pathogenic CDK12 alterations were described. Results In all, 39 (0.78%, n = 39/4994) patients had pathogenic CDK12 alterations. Among CDK12‐altered tumors, the most common organ site was prostate (n = 9, 23.1%) followed by colorectal (n = 5, 12.8%). Adenocarcinoma was the most common histology (n = 26, 66.7%). Median follow‐up from time of diagnosis was 4.02 years. Median overall survival from time of metastasis was 4.43 years (95% CI: 3.11–5.74). Ten patients with CDK12‐altered tumors received at least one immune checkpoint inhibitor‐containing regimen. The majority of patients (n = 6/10, 60%) experienced an objective response. Progression‐free survival for patients who had metastatic disease and received a checkpoint inhibitor‐containing regimen was 1.16 years (95% CI: 0.32–2.00). Conclusion CDK12 alterations are rare events across hematologic and solid tumor malignancies. They represent a clinically distinct molecular cancer subtype which may have increased responsiveness to checkpoint inhibition. Prospective studies are warranted to investigate checkpoint inhibition in CDK12‐altered tumors.


| INTRODUCTION
The growing routine use of molecular profiling in recent years has revolutionized the medical management of advanced hematologic and solid tumor malignancies. One goal of molecular profiling is to identify molecular susceptibilities in different tumors that may be prognostic or predictive of response to treatment, the latter resulting in the practice of precision oncology. Understanding key characteristics of molecular targets present in a variety of tumor types is central to the goal of tailoring effective treatments in the age of precision medicine.
CDK12 alterations have generated recent interest as a potential biomarker for cancer response. CDK12 encodes for the tumor-suppressor protein cyclin-dependent kinase-12, which plays various roles in RNA processing and DNA repair in select genes. [1][2][3][4] CDK12 has been implicated in the homologous recombination repair pathway, although efforts to target these tumors with poly [ADPribose] polymerase [PARP] inhibitors such as olaparib have achieved mixed results. 5, 6 More recently, CDK12 loss of function alterations was found to be associated with increased focal tandem duplications and greater genomewide structural variation in ovarian and prostate cancers. [7][8][9] Growing evidence has shown that structural variations resulting from CDK12 inactivation may result in increased neoantigen burden and increased expression of chemokines, making patients with CDK12-altered tumors promising candidates for immune checkpoint inhibitors. 10 Given its potential as a target for an expanding panel of therapies, further uncovering the clinical and genomic features of the CDK12 alteration genotype in the oncologic landscape is growing in importance. In prostate cancer, CDK12 inactivation has shown more aggressive clinical features including greater proportion of distant metastases and shorter time to PSA progression. 6,11,12 However, there are limited pan-cancer data regarding the frequency of CDK12 alterations, especially in the context of patients with or without metastases. We aimed to characterize the clinical features of CDK12-altered tumors utilizing a pancancer database of patients undergoing clinical-grade genomic profiling.

| Patient population
This was a single-center retrospective analysis approved by the Institutional Review Board at the University of California San Diego. We evaluated patients with a diagnosis of cancer who underwent clinical-grade (i.e., Clinical Laboratory Improvement Amendments-certified) genomic profiling as part of routine care. Eligible patients were patients older than 18 years of age with a diagnosis of invasive cancer who had at least one clinical-grade next generation sequencing test which included analysis of CDK12 on either tissue or blood from December 2012 to January 2020.

| Genomic data
Patients with evidence of a pathogenic CDK12 alteration affecting at least one allele were included. Pathogenic alterations were defined as those that result in a truncated protein (i.e., frameshift, nonsense, splicing mutations) or genomic rearrangements that involve the CDK12 locus (e.g., homozygous deletions, gene fusions, other translocations). Pathogenic mutations were defined using databases such as COSMIC and published literature. A list of pathogenic mutations and classification of alterations is listed in Table S1. A variety of clinical-grade sequencing assays was used to determine CDK12 status in our cohort (tissue-based assays utilizing either primary or metastasis tissue: FoundationOne, FoundationOne CDx, and Tempus xE/xO/xT; cellfree DNA-based assays: Guardant360, FoundationOne Heme, and FoundationOne Liquid CDx). The average depths of coverage for the various next generation sequencing platforms are as follows: 500x for Tempus xT, 250x for Tempux xE, 300x for Tempus xO, and 15000x for Guardant. The median depths of coverage for Foundation sequencing platforms were 500x for Foundation CDx, 500x for Foundation Heme, and deep coverage for Foundation Liquid CDx (Foundation Medicine does not disclose the exact liquid coverage, but it is significantly deeper than tissue due to low levels of circulating tumor DNA in the plasma). Co-occurring alterations in other genes for patients with CDK12-altered tumors were analyzed. Microsatellite instability (MSI) and mismatch repair (MMR) status was not included in the analysis due to limited data provided by the earlier genomic profiling reports. Analysis was only performed on co-occurring alterations that were present on the same assay in which a CDK12 alteration was detected. If patients had more than one genetic test performed, the most recent tissue-based test was utilized for analysis. Matched samples were not analyzed.

| Clinical data
Baseline demographic, pathologic, and clinical characteristics were abstracted from the electronic medical record. Data regarding age, gender, race/ethnicity, smoking history, cancer site of origin, histology, and Charlson Comorbidity Index 13 were captured. The American Joint Committee on Cancer staging system 8th edition was used to define patients with localized, regional, and distant metastases. 14 We tabulated systemic treatments with a focus on immune checkpoint inhibitors.

| Statistical analysis
Patient and disease characteristics were summarized using descriptive statistics. The prevalence of patients with CDK12 alterations was calculated. Overall survival (OS) was defined as the date of diagnosis to death, censored at the time of last follow-up. We also calculated OS from the date of metastasis to death, censored at the time of last follow-up. Progression-free survival (PFS) was defined for patients receiving immune checkpoint inhibitors as the time from treatment initiation with first-line immunotherapy to radiographic progression (using the Response Evaluation Criteria in Solid Tumors [RECIST] version 1.1 principals), clinical progression (defined as disease-related complication or clinical deterioration) or death, whichever occurred first. Response was assessed using RECIST version 1.1 principals. Time-to-event outcomes (PFS, OS) were analyzed using the Kaplan-Meier method, and median values (with 95% confidence intervals [CI]) were reported. As an exploratory analysis, we compared the OS using the log-rank test in patients with metastatic disease having received a checkpoint inhibitor compared to those with metastatic disease not having received a checkpoint inhibitor. Tumor mutational burden for patients having received a checkpoint inhibitor was stratified as low (≤5 mutations/Mb), intermediate (>5 and <20), high (≥20 and <50), and very high (≥50).
Patient and disease characteristics of patients with CDK12-altered tumors are detailed in Tables 1 and 2, respectively. The median age at diagnoses was 64 years (interquartile range: 58-71 years) and the majority of patients were male (n = 25, 64.1%). While 61.5% of patients (n = 24) were white, 38.5% (n = 15) were non-white with Hispanic (n = 6, 15.4%) representing the next most frequent racial group represented in the cohort of patients with CDK12-altered tumors. Most patients (n = 18, 46.2%) had localized disease at diagnosis of whom 15 subsequently developed metastatic disease. A total of 33 patients (84.6%) developed metastatic disease at any time during their disease course. The most common histology was adenocarcinoma (n = 26, 66.7%). CDK12 alterations were observed across a spectrum of malignancies, with the most common primary sites being prostate (n = 9, 23.1%), followed by colorectal (n = 5, 12.8%) and breast (n = 4, 10.3%).

| Genomic characteristics
Among the 39 patients with pathogenic CDK12 alterations, 33 (84.6%) patients were identified using tissue testing: 30 (90.9%) using FoundationOne and three (9.1%) using Tempus xT. Of patients identified on tissue testing (n = 33), 22 (66.7%) were identified from primary tumor testing and 11 (33.3%) from metastasis tumor testing. In the entire cohort of 4994 patients, 0.5% were identified with CDK12 alterations on primary tumor testing and 0.26% on metastasis testing. Six patients were identified using ctDNA testing: three using Guardant360, one using Tempus xF, one using FoundationOne Heme, and one using FoundationOne CDx Liquid. There were a total of 41 CDK12 alterations observed across 39 patients, and the alteration types are summarized in Figure 1. Of patients with pathogenic CDK12-altered tumors, two patients (4.5%) had biallelic inactivating alterations. Both patients with biallelic alterations had prostate cancer. The spectrum of the type of CDK12 alterations observed across primary sites of origin is detailed in Figure 2. The most common types of alterations were frameshift mutations (n = 16), followed by non-sense mutations (n = 11), while a smaller proportion of alterations due to other mechanisms. A complete list of all CDK12 alterations is included in Table S2.
In all, 23 patients had more than one sequencing assay performed throughout their clinical course, and 30.4% (n = 7/23) demonstrated concordance with CDK12 alterations. Of the nine patients who had multiple genetic profiling assays that included at least one prior to treatment initiation, only three patients did not have CDK12 alterations at baseline and acquired it during the course of treatment. We interrogated for the presence of additional concurrent genomic alterations in patients with CDK12 alterations (Figure 2). Co-occurring alterations were the most common in TP53, occurring in 26 patients (66.6%). This was followed by ERBB2 in nine patients (23.0%), KRAS (n = 7/29, 17.9%), and PIK3CA (n = 6/39, 15.3%). Co-occurring BRCA2 alterations were identified in two patients (5.1%), one with gastrointestinal tumors and one with bladder cancer.
Tumor mutational burden (TMB) was evaluated in tumors with CDK12 alterations. Of the samples with reportable TMB (n = 40), the majority had low-or intermediate-range TMB at 45% and 37.5%, respectively. The remaining had high (2.5%) or very high (15%) TMB. The TMB for patients who received immunotherapy are

| Treatment exposure
There were 14 (35.9%) patients who received neoadjuvant/adjuvant systemic therapy for local/regional disease and 33 (84.6%) patients who received systemic therapy for metastatic disease. The median number of lines of systemic therapy administered to patients with CDK12-altered tumors was three. Cytotoxic chemotherapy was the most common treatment, with nine patients (23.1%) receiving therapy with a platinum-containing regimen. Additionally, 17.9% (n = 7) and 7.7% (n = 3) received treatment with tyrosine-kinase inhibitors and PARP inhibitors, respectively. A total of 10 patients (25.6%) received treatment with at least one checkpoint inhibitor-containing regimen, with three patients receiving two checkpoint inhibitor-containing regimens ( Table 2).

| Treatment outcomes
The median follow-up for the population was 4.01 (95% CI: 1.82-6.21) years from diagnosis to last follow-up or death. For the total cohort, OS as calculated from the date of diagnosis was 6.94 years (95% CI: 3.65-10.22) ( Figure 3A) and 4.43 years (95% CI: 3.11-5.74) from the date of metastases development for those with metastatic disease (n = 37) ( Figure 3B). Outcomes for patients treated with checkpoint blockade are delineated in Table 3. The objective response rate to checkpoint inhibition was 60.0% (n = 6). At the time of last follow-up, five patients remained on therapy while five had discontinued treatment due to radiographic progression (n = 2), clinical progression (n = 1), toxicity

| DISCUSSION
The clinical implications of CDK12 mutations have been evolving over the last several years, as these genomic aberrations have been detected in a proportion of malignancies. Our study demonstrates that CDK12-altered cancers have a prevalence of 0.88%, which is consistent with previously published reports on CDK12 genomic alterations. Pan-cancer studies evaluating the prevalence of CDK12 alterations have been limited. A similar pan-cancer analysis, which lacked access to detailed clinical data, demonstrated that the prevalence of CDK12 genomic alterations was 1.1% across all cancer types, with mutations most frequently seen in prostate, gastrointestinal, and gynecologic malignancies. 7 These findings are similar to our cohort, with the most commonly mutated tumors being prostate with a frequency of 20% of all CDK12-mutated patients, followed by colorectal cancer. Knowing the prevalence of CDK12 alterations among various solid tumor types may be helpful for selectively evaluating for these mutations for prognostic and treatment purposes, particularly in the case of prostate cancer where biallelic CDK12 inactivation is associated with shorter time to biochemical progression and distant metastases. 10 The ethnic distribution of our cohort was unique in that approximately half of patients with CDK12-altered tumors were non-White, with Asian/Pacific Islander being the group with the third highest prevalence of CDK12-altered tumors after white individuals. While the prevalence of CDK12 mutations across ethnic groups has not been fully characterized, there have been some studies highlighting a predisposition for CDK12-mutated breast Best objective response to therapy (CR, PR, SD, PD)  15,16 Further studies evaluating the ethnic distribution of genomic alterations are important in bridging disparities gaps in genomic profiling and developing strategies to bring precision medicine to all races and ethnicities. CDK12 alterations have also been associated with an increased response to immunotherapy. The mechanism driving CDK12-altered tumor sensitivity to immune checkpoint inhibitors lies in the association between CDK12-loss-of-function mutations with a higher focal tandem duplication burden; focal tandem duplications lead to increased production of fusion-induced neoantigens that can be responsive to PD-1/PD-L1 blockade. 10 This concept was previously demonstrated in the context of prostate and ovarian cancer, 17 but has more recently expanded to include several different cancer types. In the pan-cancer analysis by Sokol et al., CDK12-loss-of-function was associated with an increased focal tandem duplication burden in all evaluated malignancy subtypes including gastrointestinal, gynecologic, and cancer of unknown primary. While hypothesis generating in nature and not accounting for confounding factors, we observe a potential signal of longer OS in patients with CDK12-mutated cancers having received an immune checkpoint inhibitor-containing F I G U R E 3 (A) Overall survival of all patients with CDK12 alteration (n = 39). Overall survival defined as the date of diagnosis to death or last follow-up, whichever came first. Median Survival 6.91 years (95% CI: 3.65-10.22). (B) Overall survival of all patients with CDK12 alteration who developed metastatic disease at any time (n = 32). Overall survival defined as the date of diagnosis of metastatic to death or last follow-up, whichever came first. Median Survival 4.43 years (95% CI: 3.11-5.74). (C) Overall survival by receipt of checkpoint inhibitorcontaining regimen for all patients with CDK12 alterations who developed metastatic disease at any time (n = 32). Overall survival defined as the date of diagnosis of metastatic to death or last follow-up, whichever came first. Green line represents patients with metastatic disease having received a checkpoint inhibitor-containing regimen (n = 9). Blue line represents patients with metastatic disease not having received a checkpoint inhibitor-containing regimen (n = 23). (D) Progression-free survival of checkpoint inhibitor-containing patients who had metastatic disease (n = 9). Progression-free survival is defined as the time from first-line immunotherapy start to radiographic progression, clinical progression, death, or last follow-up. Median progression-free survival 1. 16  regimen. These data are limited by a relatively small cohort of immunotherapy-treated patients (10 out of 39) as well as the fact that 4 out of 10 patients who received immunotherapy also had concurrent targeted therapy or chemotherapy, which suggests that the achieved response cannot be attributed to checkpoint blockade alone. With these caveats taken into consideration, the results capture immunotherapy efficacy across a broad spectrum of cancer types including lung, liver, and melanoma. Interestingly, a subset of patients had co-occurring BRCA2 and CDK12 mutations, raising the question of whether PARP inhibition may be an effective treatment option in this cohort of patients. It has been hypothesized that PARP inhibitors could be effective in CDK12-deficient cancers, as CDK12 mediates DNA repair via homologous recombination. 5 In BRCA-mutated triple-negative breast cancer cells and patient-derived xenografts, the PARP inhibitor dinaciclib was shown to increase the degree of response for PARP inhibitor-sensitive models, and reverse homologous recombination and PARP inhibitor resistance. 18 However, published reports have not shown a pan-cancer connection between CDK12 alterations and homologous recombination deficiency (HRD) phenotype. The PROfound trial evaluated the efficacy of olaparib in men with metastatic castrate-resistant prostate cancer (mCRPC) and a qualifying alteration in genes involved in HRD. The study demonstrated an improvement in PFS with olaparib compared to abiraterone or enzalutamide. However, the proportion of CDK12-altered tumors was 26.2%. In an exploratory analysis in patients with CDK12altered tumors, PFS was 5.09 months with olaparib compared to 2.20 months with abiraterone or enzalutamide, though statistical comparison between the groups was not performed. 19 In our cohort, there was a lack of response to PARP inhibition, which is in concordance with previously published data that also show poor responses to PARP inhibitors in CDK12-altered prostate cancer. 6 Similarly, HRD has been associated with increased sensitivity to platinum chemotherapy, 20 and it would be of clinical utility to consider an exploratory analysis of the response of CDK12-altered tumors to platinum compounds.

Time to progression or last follow-up (months)
A strength of our study is that it reflects the realworld practice of using a multitude of clinical-grade sequencing assays to identify patients with CDK12-altered tumors. Additionally, our pan-cancer analysis includes a heterogeneous patient population with a wide spectrum of primary malignancies. However, some limitations are present due to the inherent nature of a retrospective pancancer analysis and our relatively small sample of patients with CDK12-altered tumors which prevents more robust conclusions. The non-CDK12-altered dataset had limited granularity in regards to baseline characteristics and clinical outcomes, which prevents comparisons with our CDK12-altered cohort. While variant allele frequency of the CDK12-altered tumors would have been of value to determine its impact on overall tumor biology, this detail was unavailable in the majority of the sequencing assays in our cohort. Evaluating surrogate markers for immunotherapy response, such as MSI/ MMRI status, in our CDK12-altered cohort would have been clinically beneficial; however, this was limited by lack of data from earlier genomic reports. The responses to immunotherapy in our cohort could have been influenced by high tumor mutational burden or MSI high status. Further studies evaluating MSI/MMR and CDK12 alterations are needed to determine whether CDK12 alterations are independent predictors of immunotherapy response. In addition, only a subset of patients in our cohort were treated with immune checkpoint inhibitors and/or PARP inhibitors, limiting robust comparison of outcomes among patient treated with or without checkpoint blockade.
Given its role in human cancers and regulation of genome stability, CDK12 is currently being studied as a potential therapeutic target. CDK inhibitors for cancer treatment are on the horizon, with some drugs having multi-specific CDK inhibitor activity such as dinaciclib, 21 and others having CDK12-specific inhibition such as THZ531 22 and SR-4835. 23 Dinaciclib has the ability to reverse PARP inhibitor resistance by downregulating homologous recombination DNA repair genes, suggesting that combination therapy with PARP and CDK12 inhibitors may be an effective approach. This combination is currently being studied in a Phase I trial with dinaciclib and veliparib (PARP-1 inhibitor ABT-888) for treatment of metastatic solid tumors, and is estimated to complete accrual in December 2021 (available online: http://clini caltr ials.gov, NCT01434316).
In addition, clinical trials evaluating immunotherapy response in CDK12-mutated cancers are underway. The IMPACT trial is an ongoing study investigating whether CDK12-mutated mCRPC is more susceptible to nivolumab and ipilimumab (NCT03570619). There is also the Phase II study of abemaciclib and atezolizumab in mCRPC (NCT04751929), as well as durvalumab and olaparib in prostate cancer patients with high neoantigen load (NCT 04336943). Overall, CDK12 is a promising targetable biomarker that may be predictive of immune checkpoint inhibitor sensitivity, and also play a role in clinical decision making for selective genomic sequencing and its therapeutic implications. The prevalence of CDK12 alterations is rare and conclusions cannot be drawn based on our current data as to whether pan-screening for CDK12 is of clinical utility. Impact trial testing of immunotherapy in patients with CDK12-mutated malignancies will help better guide decision making around CDK12 testing.

| CONCLUSION
This study is one of the few pan-cancer analyses of CDK12 alterations demonstrating that CDK12 alterations are rare events across different cancer types. CDK12 mutations were associated with responses to immunotherapy, suggesting that CDK12 may be a predictive biomarker of immune checkpoint inhibitor response in addition to being a marker with targetable therapeutic potential.