Genomic Markers of CDK 4/6 Inhibitor Resistance in Hormone Receptor Positive Metastatic Breast Cancer

Simple Summary The mechanisms of intrinsic and acquired resistance to CDK4/6 inhibitors are poorly understood. Patients with HR+ MBC treated with CDK4/6i and antiestrogen therapy were grouped into early (<6 months), intermediate (6–24 months for 0–1 lines; 6–9 months for ≥2 lines), or late progressors (>24 months for 0–1 lines; >9 months PFS for ≥2 lines). Among the 109 patients who received CDK4/6i as 1st- or 2nd-line therapy, 17 genes showed association with PFS (p-value < 0.15 and HR ≥ 1.5 or HR < 0.5). RNAseq was analyzed for 24/109 (22%) patients with 0–1 prior lines of therapy and 56 genes associated with PFS (HR ≥ 4 or HR ≤ 0.25 and FDR ≤ 0.15). Genomic biomarkers including FGFR1 amplification, PTEN loss, and DNA repair pathway gene mutations showed significant associations with shorter PFS for patients receiving CDK4/6 inhibitor therapy. Abstract Cyclin-dependent kinase 4/6 inhibitors are the standard of care for hormone receptor-positive metastatic breast cancer. This retrospective study reports on genomic biomarkers of CDK 4/6i resistance utilizing genomic data acquired through routine clinical practice. Patients with HR+ MBC treated with palbociclib, ribociclib, or abemaciclib and antiestrogen therapy were identified. Patients were grouped into early (<6 months); intermediate (6–24 months for 0–1 lines; 6–9 months for ≥2 lines); or late progressors (>24 months for 0–1 lines; >9 months PFS for ≥2 lines). NGS and RNA sequencing data were analyzed in association with PFS, and survival analysis was stratified by prior lines of chemotherapy. A total of 795 patients with HR+ MBC treated with CDK 4/6i were identified. Of these, 144 (18%) patients had genomic data and 29 (3.6%) had RNA data. Among the 109 patients who received CDK4/6i as 1st- or 2nd-line therapy, 17 genes showed associations with PFS (p-value ≤ 0.15 and HR ≥ 1.5 or HR < 0.5). Whole transcriptome RNAseq was analyzed for 24/109 (22%) patients with 0–1 prior lines of therapy and 56 genes associated with PFS (HR ≥ 4 or HR ≤ 0.25 and FDR ≤ 0.15). In this retrospective analysis, genomic biomarkers including FGFR1 amplification, PTEN loss, and DNA repair pathway gene mutations showed significant associations with shorter PFS for patients receiving CDK4/6 inhibitor therapy.

The mechanisms of CDK4/6i resistance are poorly understood in the real-world setting. Broad-panel next-generation sequencing (NGS) has become widely available in routine clinical practice and offers the ability to identify potentially actionable genomic alterations. The utility of the broad panel NGS is yet to be defined for patients with MBC [25,26]. Therefore, the current retrospective analysis was initiated to understand the molecular mechanism of resistance to CDK4/6i by analyzing tumor genomic data acquired through routine clinical care in patients who received CDK4/6i as standard-of-care therapy.

Patients
This study was conducted using an institutional review board (IRB)-approved protocol. Patients with HR+ MBC who had tumor NGS testing and were treated with at least one CDK4/6i (palbociclib, ribociclib, or abemaciclib) and antiestrogen therapy between January 2015 and July 2020 were identified through electronic medical record (EMR) search. Patient characteristics and disease variables were obtained by chart review: age, race/ethnicity, histology type, initial tumor stage; prior lines of therapy for metastatic disease; systemic treatment for metastasis (CDK4/6i and endocrine therapy); and sites of metastases.
Response to therapy was evaluated through routine staging with CT or PET-CT, and PFS and OS were determined based on chart review. The results from germline genetic testing were collected if applicable. PFS on CDK4/6i was calculated from the first day of treatment until the date of disease progression. The response to CDK4/6i was classified as follows: early progressors (PFS ≤ 6 months), intermediate progressors (PFS 6-24 months for 1st line patients and 6-9 months for ≥2 lines of therapy), and late progressors (PFS ≥ 24 months for first line patients and ≥ 9 months for ≥2 lines of therapy)(21).

Tumor Genomics Analysis
Of 144 patients with genomic results, a total of 43 (30%) specimens were sequenced with the Tempus comprehensive xT ® platform, which includes a targeted 648 gene panel. Of those, 79 (55%) samples were sequenced using FoundationOne ® CDx comprehensive profiling, which includes 324 mutations. A total of 22 (15%) of these samples were sequenced using HopeSeq platform, including 89 "hot-spot" genes [27]. An additional 9 (6%) patient samples were sequenced using GEM ExTra ® , which covers all protein coding regions [28]. To study the associations of PFS and genomic alterations, we evaluated the hazard ratio (HR) for each gene associated with PFS using a Cox proportional hazard model.

Expression Analysis
Of the 109 patients who received 0-1 prior lines of therapy, 24 (22%) had whole transcriptome RNAseq via Tempus ® [29]. The Cox proportional hazard model was used to evaluate the HR for PFS for 11,426 protein coding genes after filtering. The false discovery rate (FDR) was used to adjust for multiple comparisons. Only protein-coding genes were evaluated, and genes with low expression or variation, with average expression or standard deviation less than 25% of genome-wide quantile, were filtered.

Survival Analysis
PFS data were plotted as Kaplan-Meier curves. We compared PFS among patients with 0-1 prior line of therapy (n = 109) vs. ≥2 lines of therapy (n = 35) with a log-rank test. We further compared PFS among 0, 1, and ≥2 lines of therapy with n = 80, 29, and 35 patients, respectively. The same analyses were performed for overall survival (OS) with a log-rank test.

mRNA Expression Associated with PFS
Of the 144 patients with genomic sequence data, 109 patients had 0-1 prior line of therapy, and 24/109 (22%) had whole transcriptome RNAseq via Tempus ® [29]. Coding gene fragments per kilobase of exon per million (FPKM) mapped fragments were analyzed in patients with continuous PFS (Figure 2). A total of 11,426 protein-coding genes were studied, and a false discovery rate (FDR) was calculated to adjust for multiple comparisons.  Table S2.

Discussion
CDK4/6i has revolutionized the treatment of advanced HR+ MBC, and the indication has been expanded to early-stage breast cancer with remarkable results from the monarchE trial [31,32]. However, the benefits are not observed across all patients due to intrinsic or acquired resistance. Understanding the mechanism of the resistance to CDK4/6i is critical for personalized treatment in breast cancer. Multiple studies were conducted to understand the molecular mechanisms of resistance to CDK 4/6i in breast cancer [33,34]. Paired baseline and end-of-treatment ctDNA from 195 patients in the PALOMA-3 was studied [16] and showed that acquired mutations (mut) in RB1 occurred in 5% of patients after palbociclib. In addition, new driver mutations in PIK3CA and ESR1 were observed. Pooled ctDNA analysis from the MONALEESA trial evaluating the efficacy of ribociclib plus different endocrine therapies demonstrated potential biomarkers associated with response, including FRS2, MDM2, PRKCA, ERBB2, AKT1, and BRCA1/2, and biomarkers associated with resistance, including CHD4, BCL11B, ATM, or CDKN2A/2B/2C [12]. Li et al. reported associations of RB1 loss, FAT1 loss (Hippo pathway) and de novo resistance to CDK4/6i using MSK-IMPACT [35]. Formisano et al. reported FGFR1/ZNF703 amplification (amp) in 20/427 (4.7%) pre-treatment ctDNA samples associated with reduced clinical benefit from ribociclib in patients enrolled in the MONALEESA-2 trial [36]. Lastly, NGS on 59 tumors with CDK4/6i exposure revealed resistance mechanisms: RB1 loss; activating alterations in AKT1, RAS, AURKA, CCNE2, ERBB2, and FGFR2; and the loss of ER expression [37]. In the current study, despite the limited sample size and the low frequency of many somatic mutations, several statistically significant patterns were identified, including genomic alterations associated with CDK4/6i resistance: FGFR1 amp, ZNF703 amp, CDK4 amp, PALB2 mut, BRCA1 mut, RB1 loss, CDKN2B loss, CREBBP mut, MLL3 mut, ERBB2 mut, PTEN loss, MAP2K4 mut, MDM2 amp, FRS2 amp, ATRX mut, and GPR124 mut. In addition, FAT1 mut/loss was associated with increased PFS. In this retrospective analysis using real-world data, molecular biomarkers such as FGFR1 amplification, PTEN loss, and DNA repair pathway gene mutations showed significant association with shorter PFS with CDK4/6i therapy. CDK 4 and 6 control cellular transition from the G1 phase to S phase of the cell cycle and activated cyclin D and CDK4/6 complex promotes the initiation of DNA synthesis and entry into S-phase by retinoblastoma phosphorylation that de-represses activity of E2F family facilitating cyclin E1 and E2 activity. Cyclin E then activates CDK2, which hyperphosphorylates RB, further increasing the expression of E2F target genes that are critical for the cell to proceed into S phase. The loss or mutation of RB, the main target of CDK4/6, is a well-known driver of the resistance. Condorelli et al. reported acquired somatic RB1 mutations from 3 patients with MBC after exposure to CDK4/6i [15]. The RB1 mutations were identified at 5, 8, and 13 months after initiation of CDK4/6i in each patient from circulating tumor DNA (ctDNA) analyzed by commercially available NGS-based assay. The biomarker study performed in the PALOMA-3 trial evaluating the efficacy of fulvestrant plus palbociclib demonstrated that Cyclin E1 (CCNE1) expression was associated with resistance to Palbociclib, and low expression of CCNE1 mRNA was related to better efficacy of palbociclib [18]. Our finding of reduced PFS in association with CDK4 amp, RB1 loss, and CDKN2B loss is consistent with the above findings. CDKN2A and CDKN2B are endogenous inhibitors of CDK 4/6. High CDKN2 levels may predict reduced CDK4/6 activity and resistance to CDK4/6i [34,38]. Due to our limited sample size, our finding of CDKN2B loss in association with shorter PFS needs to be further verified. Hippo signaling in ER+ breast cancer is an established tumor suppressor in ER+ MBC, and FAT1 loss serves as a mechanism of resistance to CDK4/6i [35]. Our finding of FAT1 mutation/loss in association with better PFS could be due to the small sample size.
Several common, potentially actionable resistance mechanisms involving growth factor pathways have been identified, including the FGF, Aurora Kinase, AKT, and MAPK signaling pathways [39]. The fibroblast growth factor receptor (FGFR) pathway is involved in cell proliferation, differentiation, and growth and plays a role in CDK4/6i therapy resistance. FGFR1 amplification has been shown to trigger the recruitment and activation of STAT3 [40] and may be involved in multiple effector pathways that activate ERK1/2 in different cellular contexts ("additive" signaling) [41]. ctDNA analysis from patients enrolled in MONALEESA-2(3) demonstrated shorter PFS in patients with FGFR1 amplification compared with patients with wild-type FGFR1 [36,42]. The ablation of PTEN, through increased AKT activation, was sufficient to promote resistance to CDK4/6 inhibition in breast cancer cells in vitro and in vivo [14]. PTEN loss resulted in the exclusion of p27 from the nucleus, leading to the increased activation of both CDK4 and CDK2 [14]. Our finding of PTEN loss and the trend of reduced PFS is consistent with the above findings. ZNF703 amplification, predominantly identified in Luminal B subtype of breast cancer, is associated with poor clinical outcomes [43]. In the ctDNA analysis of the MONALEESA-2 trial, Formisano et al. identified that5% (20/427 patients) patients had copy number alterations of the 8p11.23 genomic locus, which harbors FGFR1 and ZNF703 genes and is associated with reduced clinical benefit for ribociclib (median PFS 10.61 vs. 24.84 months, p = 0.075) [36]. ERBB2 mutations were identified in 5/41 CDK4/6-resistant tumor [37], and preclinical work demonstrated that ERBB2 mutation activates downstream MAPK/AKT/mTOR and confers resistance to CDK4/6 blockade in breast cancer cells [44]. MAP2K4 encodes a member of the MAPK family, and MAP2K4 mutation was identified in breast cancer resistant to AI [45]. Fibroblast growth factor receptor substrate 2 (FRS2) acts upstream of the FGF signaling pathway, is an adaptor protein expressed in a small subset of epithelial cells and triggers a cytokine-rich inflammatory microenvironment that promotes breast cancer carcinogenesis [46]. DNA repair defect (DRD) gene mutations may confer resistance to CDK4/6i. Safonov et al. recently reported gBRCA2 mutation in association with significantly inferior PFS (HR 2.17, 95% CI 1.46-3.22, p < 0.001) for 1st-line treatment with CDK4/6 I [47]. No significant association between PALB2 mutation and PFS with CDK4/6i was identified in this study. In a real-world study of CDK4/6i including 2698 patients, 9.9% had gBRCA2 mutation, which is associated with shorter time to subsequent therapy or death (HR1.24, 95% CI 0.96-1.59) [48]. In our study, gBRCA1 mutation was reported in 2/144 (1.4%) patients and was associated with a trend of inferior PFS (HR3.93, 95% CI 0.94-16.5, p = 0.06). 7/144 (5%) patients had gBRCA2 mutation, but no association with PFS was observed (HR 1.34, 95% CI 0.62-2.90, p = 0.46). MDM2 and MDM4 inhibit TP53, and amplification renders resistance in ER+ BC patients. MDM2 inhibits DNA break repair through association with the Mre11/Rad50/Nbs1 DNA repair complex [49]. Approximately 30-50% of HR+ MBC showed mutation of P53 or regulators MDM2 and MDM4 [37]. MDM2 inhibitor could provide a therapeutic opportunity, particularly for those malignancies that have lost functional p53. Our finding of MDM2 mutation associated with CDK4/6i resistance is consistent with this finding. Alpha Thalassemia/Mental Retardation Syndrome X-Linked (ATRX) loss is frequently identified in gliomas and may be a molecular marker for DNA damage response defects and ATRX knock-out led to PARP inhibitor sensitivity in glioma cells [50]. The role of ATRX in breast cancer remains unclear.
Mixed lineage leukemia gene (MLL3), a histone monomethylase that is known to interact with nuclear hormone receptors such as ERα, is frequently mutated in multiple cancer types. MLL3 is the sixth most mutated gene in ER+ breast cancer: 8.5% in AACR GENIE [51] and 9% of ER+ breast cancer patients in TCGA [52]. MLL3 mutation is a de novo cause of endocrine therapy resistance in ER+ breast cancer and the mutation pattern of MLL3 in breast cancer is most consistent with a haploinsufficient tumor suppressor [53]. G-protein coupled receptor 124 (GPR124) is significantly expressed in tumor vasculature and play a critical role in VEGR-induced tumor angiogenesis, which involves cell-cell interaction, permeability, migration, and invasion [54]. CREBBP binds to CREB, which is known to play a critical role in embryonic development. CREBBP/EP300 HAT inhibition suppresses EGR-dependent transcription in breast cancer in vitro and in vivo [55].
Analyzing mRNA expression, Turner et al. reported that high CCNE1 mRNA expression predicted worse response to palbociclib in the PALOM-3 trial. High CCNE1, CCND3, and CDKN2B mRNA expression levels were associated with resistance to palbociclib in the NeoPalAna trial (18). Guerrero-Zotano et al. reported that high E2F4 signature activity correlated with response to palbociclib [56]. In the current study, mRNA expression predicting worse response to CDK4/6i included increased expression of PARP1 (average expression 1.98, HR 6.30 and p = 0.00007) and STAT1 (average expression 0.94, HR 4.42, and p = 0.002), and decreased expression of BCROL1 (average expression −0.39, HR 0.21, p = 0.001) and NFRKB (average expression 0.58, HR 0.12, p = 0.002). Due to the limited sample size, these findings need to be further verified through larger studies.
In our study, 16% of patients underwent tumor genomic testing. The genomic tests were utilized significantly less frequently in community network sites compared with the main campus (7% vs. 18%). The patients included in this study were treated prior to 2020, when broad-panel NGS was commonly adopted as routine clinical practice. NCCN Guidelines for Breast Cancer V.1.2021 include NGS and comprehensive genomic profiling (CGP) as a method of detecting actionable mutations and fusions such as BRCA1/2 mutations, PIK3CA mutations, ESR1 mutation, HER2 mutation, MSI-H status, and deficient mismatch repair (dMMR) [57]. Integrating community oncologists into the academic paradigm of personalized medicine, including continued development of interpersonal relationships between community oncologists and academic site physicians through molecular tumor board teams, will play a major role in personalized therapy advancements in oncology [58].
There were several limitations to our study including a limited sample size, with only a fraction of patients receiving CDK4/6i genomic sequencing, which limits the statistical power of the current analysis. The recently updated NCCN 2021 guidelines recommend routine NGS testing for patients with metastatic breast cancer, which likely will increase utilization in community practice. Since this study was retrospective in nature, the genomic tests were performed through a variety of platforms per the oncologists' preference. The adoption of institutional practice guidelines will also facilitate an increased number of patients receiving genomic sequencing, the appropriate utilization of repeat biopsy upon disease progression, the increased availability of adequate tissue specimens for NGS, and the minimized inter-test variance by minimizing the variation among different nextgeneration sequencing platforms. Currently, an institutional effort is underway to optimize a precision medicine approach and mitigate the aforementioned challenges. It is also noted that the median PFS in this study for patient who received 0-1 line of prior therapy (N = 109) appears to be shorter than that in the real-world study reported by DeMichele et al. (N = 1430) [59]. Prior endocrine therapy and the relatively high percentage (72%) of patients with visceral metastasis may explain the shorter PFS in our study.

Conclusions
In this retrospective analysis, genomic biomarkers such as FGFR1 amplification, PTEN loss, and DNA repair pathway gene mutations showed significant associations with shorter PFS for patients receiving CDK4/6 inhibitor therapy.
Supplementary Materials: The following supporting information can be downloaded at: https:// www.mdpi.com/article/10.3390/cancers14133159/s1, Figure S1: Tile plots of genomic alterations of each patient's tumor specimen.; Figure S2: Kaplan-Meier survival analysis stratified by prior lines of chemotherapy; Figure S3. Kaplan-Meier survival analysis stratified by 0, 1, and >2 prior lines of chemotherapy: Table S1. Tumor genomic analysis of patients with 0 or 1 line of prior therapy (n = 109) and mutations that are present in >2 patients showed association with impaired PFS; Table S2. RNA analysis for patients with 0 or 1 line of prior therapy (n = 24) patients with 0-1 prior lines of therapy. Funding: The authors thank the COH Biostatistics Core, which was supported by the National Cancer Institute of the National Institutes of Health (P30CA033572). We also acknowledge Tempus for research collaboration. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NCI.

Institutional Review Board Statement:
The study was approved by City of Hope IRB. All procedures were performed in accordance with the Code of Ethics of the World Medical Association (Declaration of Helsinki). Written informed consent was obtained, and the privacy rights of patients have been observed.
Data Availability Statement: Data supporting the reported results can be found at 10.5281/zenodo.6599679.

Conflicts of Interest:
Yuan has contracted research sponsored by Merck, Eisai, Novartis, Genentech, Celgene, and Pfizer; is a consultant for Puma, Pfizer, Gilead, and is on the Speakers Bureau for Eisai, Novartis, Genentech, AstraZeneca, Daiichi Sankyo, and Gilead. The other authors declare that they have no competing interests.