Deep sequencing of variants in all coding regions of MUC16, BRCA2, ERBB3, AR, EGFR, PTEN, ERBB2, PIK3CA, BRCA1, ESR1, PTGFR, TGFB1, AKT1, FGFR1, ERCC4, PIK3R1, and KRAS in cfDNA samples of 40 HR+ HER2– MBC patients revealed that each patient displayed at least one pathogenic or likely pathogenic variant and the highest frequency of likely pathogenic or pathogenic alterations occurred in the AR gene. The prevalence of MUC16, BRCA2, ERBB3, and AR variants was > 90% in this cohort. We here identified single variants with significant correlation with increased survival after diagnosis of metastasis. On the other hand, a BRCA1 variant as well as a number of pathogenic or likely pathogenic variants of greater 3.5 were significantly associated with worse survival after diagnosis of metastasis. The longitudinal monitoring of cfDNA variant evolution of one HR+ HER2– patient revealed the prominence of PIK3CA and ESR1 variants before staging the disease to be progressive under everolimus and exemestane.
Implications of input amount, UMI integration, high coverage, and IVA filter
The required blood volume of 9 ml is comparable to the obligatory blood volume for other liquid biopsy tests used in clinical practice already (such as Guardant360® ctDNA  or OncoBEAM™-RAS-Test) yielding in minimally 4 ml plasma. The high plasma input of 4 ml enabled isolation of cfDNA amounts usable for reliable targeted PCR-based cfDNA library preparation (library yield > 4 nM). The integration of UMIs was exemplified for identification of PIK3CA hotspot variants in this study and guarantees the identification of only true positive variants, but requires high cfDNA input and deep sequencing qualities. Exactly those deep sequencing qualities were met in all 40 cases used for variant analysis, since 18 million read fragments were sequenced on average in each sample, the mean coverage was 22,000×, and the UMI coverage was 6351. This high coverage enabled calling variants with low AFs and resulted in the identification of a high prevalence of variants, which has rarely been described before [20,21,22]. Many studies [2, 23,24,25] that described a low frequency of variants were conducted with a lack of high coverage, missing descriptions of the input amount or sequencing of hotspots rather than all exonic regions of a gene. We here ensure specificity and relevance of called variants by usage of UMIs and stringent IVA filter settings  and variant prevalence detected by analysis of tissue or liquid biopsy should be differentiated. Among other settings, only predicted deleterious variants were called and variants with a prevalence of > 3% in the healthy reference population (Allele Frequency Community (gnomAD&CGI), 1000 Genomes Project, ExAC and NHLBI ESP exomes) were excluded. Classification of the variants was conducted according to the guidelines of the association for molecular pathology . Although efforts ensuring a high sensitivity and specificity in variant calling were undertaken, a verification of variants by independent methods in the future would be desirable, which was not conducted here due to the limited cfDNA amount available and the plenty of variant locations to be verified. The detected sequence alterations were separately listed with different variant nomenclature when different isoforms were concerned. This can be one reason for the large number of reported variants. Although we did not compare the data against cfDNA of healthy donors that underwent the same workflow, we excluded variants with a high prevalence in the reference populations. Matched germline samples were not available, thus, we do not claim the described variants to be somatic per se, but low AFs might indicate the variants to be somatic rather than germline and sequencing of matched germline samples has been initiated for future studies.
MUC16 variants were reported to be frequently found in most cancer types [27, 28] which is in line with the high prevalence of MUC16 variants in our patient cohort. However, due to the fact that MUC16 consists of ~ 22,000 amino acids, the variant frequency corrected by sequence length and mutational heterogeneity resulted in exclusion of MUC16 as one of the most altered genes in most cancer types, including BC [28, 29]. MUC16 variants were found to induce MUC16 overexpression in BC and thus caused increased cancer growth and migration as well as decreased sensitivity to cisplatin . The presented new highly significant association of three MUC16 variants (R10996S, S836R, P835A) with increased survival after diagnosis of metastasis challenges the knowledge about the negative effect of MUC16 variants in BC. In accordance with the statement of decreased sensitivity to cisplatin , preliminary data of our cohort showed that the mean number of MUC16 variants was reduced in the cohort that had received platinum-based therapy sometime before blood draw (n = 4) compared to those patients who had never received any platinum-based therapy (n = 34) (15 versus 28 MUC16 variants), but no significant reduction in survival time comparing both cohorts (107 months versus 117 months) was found (data not shown). A cluster analysis of all MUC16 variants further separated the patients into two distinct groups, but these data are too preliminary to draw any clinical conclusions.
Nuclear steroid hormone receptor variants
ESR1 variants were mostly found in ER+ patients and were induced by endocrine treatment [2, 31]. ESR1 variants in cfDNA of MBC patients were reported to have a prevalence of 25.3% (PALOMA3), 28.8% (BOLERO-2) and 39.1% (SoFEA) . The 55% ESR1 variant prevalence in the presented study may be explained by the inclusion of a majority of HR+ cases in late treatment lines. In this regard, it is to mention that 62% of all included patients received two or more different endocrine treatment regimens before blood draw including non-steroid aromatase inhibitors, steroid aromatase inhibitors, selective estrogen receptor modulators, or selective estrogen receptor degraders (Online Resource 1).
Longitudinal cfDNA monitoring in one of the patients identified the appearance and increase in AF of ESR1 Y536S, ESR1 Y537S, and ESR1 Y539S which might be one potential mechanism of resistance causing progressive disease after 2 years of endocrine treatment. These three variants are located at positions encoding for the ligand-binding domain of ER and cause resistance to aromatase inhibitor therapy and, thus, were already described to be predictive [2, 8, 32,33,34].
AR expression on the mRNA- and protein level was studied in BC cases after realization of a complex interplay of androgens and estrogens . However, to the best of our knowledge, we are one of the first studying the genomic alterations in the AR gene in BC in more detail and strikingly found that 88% of all patients displayed likely pathogenic and pathogenic (defined by guidelines of the association for molecular pathology ) AR variants. Almost half of all likely pathogenic and likely pathogenic variants found in all tested 17 genes were localized in the AR gene, as previously described . Such high prevalence of AR variants should be further studied to elucidate the functional influence of AR variants on the AR expression, subsequently, on AR-dependent pathways and, finally, on BC pathogenesis. DNA regions encoding for glutamine repeats, also called polyQ stretches, are frequently found in the AR gene  and were described to be of high functional relevance . However, variant calling in polyQ stretches was found to be challenging . AR splice variants were already detected in BC patients  and AR-V7 was reported to cause resistance to anti-AR therapies in prostate cancer . Consequently, the functional analysis of AR variants might indicate new treatment options for BC patients with AR variants in the future.
PIK3CA variants showed a high prevalence in HR+ HER2– BCs (45%/29% of luminal A/B BC patients) . Here, we detected PIK3CA variants in 73% of HR + HER2-BC cases. This increased prevalence may be due to the high number of patients included who had received more than three treatment lines, since a higher PIK3CA variant prevalence was described for heavily pretreated cohorts .
PIK3CA E726K has already been detected in BC tissue and was predicted to cause a gain of function as well as increased oncogenic transformation, which was also described for PIK3CA H1047R located in the activation loop of the kinase domain [42,43,44]. With regard to conferred mechanisms of resistance, sensitivity to everolimus was not affected by PIK3CA variants . However, PIK3CA variants affected the PIK3/AKT pathway activation and the latter caused endocrine resistance . Thus, besides the detected ESR1 variants, the here described variants PIK3CA E726K and PIK3CA H1047R with dramatically increased AF under everolimus and exemestane in the serial cfDNA samples of one patient might mirror another resistance mechanism to exemestane.
Variants of the ErbB protein family
We found high prevalence of variants in the genes of the three ErbB protein family members ERBB2, ERBB3, and EGFR. This underlines the importance of EGFR for BC growth and progression in HER2– BC cases . Interestingly, however, ERBB2 P582L, ERBB2 P597L, and ERBB2 P612L, all causing the change of proline to leucine, were found to be significantly associated with increased survival after diagnosis of metastasis. This finding questions the clinical conclusion of the genomic variants found in plasma, since the increased survival observed in ERBB2 altered cases might be due to effective targeting of ERBB2 altered tumor cells by therapy to result in detection of these genomic variants in the plasma .
Variants of the BRCA protein family
Alterations in proteins involved in the DNA repair pathways have a high impact on carcinogenesis. Individuals with germline BRCA2 variants have an incidence of 57%–65% to develop BC . BC patients with BRCA1 or BRCA2 mutations, however, were demonstrated to be sensitive to platinum agents and poly (ADP-ribose) polymerase (PARP) inhibitors . Using deep sequencing of cfDNA, we here described a prevalence of BRCA1/2 variants in 95%/60% of the cohort. This high prevalence of BRCA variants in cfDNA of HR+ HER2– MBC patients has never been described before, probably also due to limitations of coverage capacities [51,52,53]. It is of note that the mean AF of BRCA1/2 variants found in the cohort was 1.5%/3.9%, indicating the detection of somatic rather than germline variants.
Prognostic value of cfDNA characteristics
The accumulation of likely pathogenic and pathogenic variants might cause survival disadvantages for the patients, since we found a significant correlation of the number of pathogenic and likely pathogenic variants with survival after diagnosis of metastasis.
Longitudinal monitoring across treatment
The longitudinal cfDNA variant monitoring in one patient does not mirror the entire cohort, but was chosen as an example for the evolution of variants in the course of the disease, which might also be taken as an indicator of certain selective pressure that seems to happen during the observation time. The high AFs of PIK3CA and ESR1 variants at progression under eribulin are consistent with the knowledge about the persistence of aromatase inhibitor-selected ESR1 variants throughout subsequent therapies  and a general increase of variant AFs at therapy failure . Due to the fact that all variants detected at the time point of progression under eribulin had already been detected at the previous sampling time points, the mechanism for acquired resistance to eribulin might not be caused by the observed variants themselves.
Interestingly, the appearance of ESR1 and PIK3CA variants with AFs > 4% was already observed 8 months earlier than progression of the disease by visual staging. Thus, these variants appear more sensitive in monitoring the evolution and potential resistance than contemporary staging methods. The promising effect of cfDNA monitoring, in general, to predate treatment response changes by weeks to months was described before [11, 12].
In conclusion, comprehensive analysis of cfDNA variants by targeted deep sequencing in HR+ HER2– BC cases not only confirmed the benefit of UMIs in the variant verification, but also identified new promising variants with clinical relevance for monitoring, prognosis, and therapy stratification. More comparative research needs to be done, focusing on other BC subtypes and also on other BC relevant genes to further emphasize the advantages of deep sequencing of cfDNA for BC therapy management.