Tissue and liquid biopsy profiling reveal convergent tumor evolution and therapy evasion in breast c

A growing armamentarium of molecular diagnostics are offering an opportunity to enable precision therapies for patients with advanced breast cancer. While testing for hormone receptor status and HER2 amplification have long been standard, next-generation sequencing-based diagnostics are now available to guide the use of precision therapies, including PI3K inhibition1 and PARP inhibition2. Furthermore, recent advances for patients with breast cancer, including antibody-drug conjugates3,4 as well as immune checkpoint inhibitors5,6 have required a reconsideration of the optimal diagnostic approach for advanced breast cancer. While these therapies have demonstrated promise in terms of breadth and durability of response with some therapy options moving from metastatic to early disease settings3, cancer continues to evolve and evade therapeutic pressure over time7,8,9,10,11,12,13,14.

Previous studies have examined the evolution of tumors following therapeutic intervention, especially for endocrine therapy and CDK4/6i interventions9,11,12,13. In these contexts, resistance alterations are common and tend to occur via on-pathway or bypass mechanisms. For endocrine therapy resistance, for instance, Razavi and colleagues identified a 18% prevalence of on-target resistance via the acquisition of ESR1 point mutations in addition to recurrent alterations in bypass pathways such as MAPK and MYC, although the diversity of resistance mechanisms has yet to be fully described9. Clinically, resistance alterations have the potential to inform treatment decisions. For example, the acquisition of ESR1 point mutations may motivate use of existing or emerging selective estrogen receptor degraders (SERDs) and acquired PIK3CA alterations are potentially targetable with PI3K inhibitors.

Multifocal sequencing has revealed significant heterogeneity in breast cancer during development, evolution, and therapy resistance15,16,17. Different clonal populations may respond differently to therapy and independently acquire cancer-driving and resistance alterations. Zundelevich and colleagues identified significant variability in the genomic alterations in primary, multiple relapse, and metastatic biopsies in patients resistant to endocrine therapy17. Such heterogeneity poses a challenge clinically for ideal treatment of patients with multifocal disease and diagnostically to identify the relevant drivers and resistance mutations.

Liquid biopsies (LBx), which interrogate circulating tumor DNA (ctDNA) in the blood, have emerged as a diagnostic and monitoring tool in breast and other cancer types. This platform provides genomic information on dozens to hundreds of genes implicated in cancer, with some next-generation sequencing (NGS)-based LBx now having received regulatory approval both inside (FoundationOne® Liquid CDx, Guardant360® CDx) and outside the US (FoundationOne® Liquid CDx). LBx has particular utility in the metastatic setting where high shed of ctDNA in the plasma may represent an option when tissue biopsy (TBx) material may not be available or representative of multifocal disease states in which metastatic subclones may acquire distinct resistance mutations18,19.

In this study, we examined breast tumors profiled with TBx and LBx comprehensive genomic profiling (CGP) as part of routine clinical care to characterize the diversity of acquired mutations and their associations with therapeutic interventions in a real-world setting (Fig. 1).

Liquid biopsies may be utilized in identifying targetable gene alterations and biomarkers for specific therapies across different subtypes of breast cancer as well as in monitoring therapy-associated implications on tumor evolution and treatment evasion. (SERD Selective Estrogen Receptor Degrader, SERM Selective Estrogen Receptor Modulator, TMB tumor mutational burden, MSI microsatellite instability, ICPI immune checkpoint inhibitor, ADC antibody-drug conjugate).

Full size image

Mutational spectrum of breast tumors profiled with tissue and liquid biopsy

We examined the landscape of genomic mutations identified in a large, real-world database of breast cancers profiled with TBx (n = 29,704) and LBx (n = 3339) CGP (Fig. 2, Supplementary Data 1, Fig. 3a, b). The genes commonly targeted in both the assays are presented in Supplementary Data 2. LBx and TBx identified largely similar prevalence for short variants (point mutations and short indels) in most genes (p > 0.05; Supplementary Data 3; Fig. 3a, b). Both platforms identified an appreciable frequency of activating alterations in PIK3CA mapping to alpelisib companion diagnostic variants1 (27.5% in TBx and LBx; Fig. 3c). Further, 4.5% of TBx and 7.2% of LBx samples harbored 2 or more PIK3CA mutations, possibly sensitizing to PI3K inhibitors, as described previously20. Deleterious BRCA1/2 alterations were also observed at similar frequencies across the two platforms (7.1% TBx, 7.4% LBx; Fig. 3d). Notably, there was a higher prevalence in LBx for ESR1 (10.6% TBx, 25.9% LBx), NF1 (4.5%, 8.1%), KRAS (2.0%, 4.1%), RB1 (4.6%, 8.0%), and ERBB2 mutations (3.4%, 4.3%), alterations known to be associated with therapeutic resistance9,10,11,12,14,21 (Supplementary Data 3; Fig. 3a, b). A comparison of point mutations and indels detected across different receptor subgroups between the two assay platforms showed largely similar patterns of gene alterations; statistically significant differences included elevated prevalence of ESR1 and FGFR1 in ER+ LBx, and PTEN in TNBC LBx (Supplementary Fig. 1, Supplementary Fig. 2, Supplementary Data 4). Overall, LBx showed a lower sensitivity for copy number alterations, with differences in MYC, FGFR2, and ERBB2 amplifications (Supplementary Fig. 3; Supplementary Data 5). Both platforms also identified similar patterns of co-occurring gene alterations, such as the co-occurrence of TP53 with BRCA1 and PIK3CA with ESR1 (Supplementary Data 6; Supplementary Fig. 4). In an analysis of the most recent assay targeting 324 genes (n = 1430)22, patterns of gene alteration prevalence were similar to those observed in the larger LBx cohort (n = 3339), particularly for samples with tumor fraction of ≥10%. (Supplementary Fig. 5, Supplementary Data 7).

This study cohort comprised patients who received comprehensive genomic profiling (CGP) as part of routine clinical care (top). A total of 29,704 patients with tissue biopsy (TBx)-based CGP, 3339 patients with liquid biopsy (LBx)-based CGP and 712 patients with longitudinal TBx and LBx CGP were examined. Patient and specimen-level characteristics have been provided for each of these cohorts. Additional breakdown of histological subtype, site of biopsy and receptor subtype information is provided for the 712 patients with longitudinal TBx and LBx CGP. Treatment information was retrospectively reviewed in an independent de-identified clinico-genomic database (CGDB), comprising 6,757 patients with TBx CGP and 1150 patients with LBx CGP (bottom). Patient and specimen-level characteristics have been provided for each of these cohorts. Additional information such as ECOG status, lines of therapy are available for samples in the CGDB cohort. (ILC invasive lobular carcinoma, IDC invasive ductal carcinoma, TNBC triple-negative breast cancer, NOS not otherwise specified, IQR interquartile range).

Full size image

a Prevalence of the 20 most frequently altered genes. Alteration prevalence between the two assays was compared using a two-sided Fisher’s exact test with FDR correction (p-value thresholds *: 0.05, **: 0.01, ***: 0.001). b Comparison of prevalence for all genes assessed in TBx (y-axis) and LBx (x-axis), with a zoomed in view of alterations seen at frequency <10%. Genes with statistically significant alteration prevalence between the two assays, based on a two-sided Fisher’s exact test with FDR correction (p-value ≤ 0.05) have been shown in orange. Pie chart showing the breakdown and prevalence of different classes of. c PIK3CA and d BRCA1/2 alterations across the two biopsy platforms. Only short variants (e.g., point mutations and short indels) were analyzed here.

Full size image

Examination of 6757 patients profiled with TBx and 1150 patients profiled with LBx in a clinico-genomic database (CGDB) revealed differences in the ordering patterns for these CGP tests (Fig. 2). Compared to TBx, patients profiled with LBx were older (63 y vs. 60 y, p < 10−5), more likely to have metastatic disease (86.7% stage IV vs. 74.3% for TBx, p < 10−5), and were treated with more lines of prior therapy (median 3+ vs. 2, p < 10−5), suggesting that LBx is frequently used in more advanced setting. Consistent with these differences, LBx showed higher similarity with metastatic-biopsied TBx relative to breast-biopsied TBx in the overall cohort, especially for genes like ESR1, NF1, and RB1 (Supplementary Fig. 6; Supplementary Data 3). This suggests that some of these differences in alteration prevalence may be attributed to disease stage. However, since most resistance-associated alterations were still observed at a significantly higher prevalence in LBx compared to metastatic-biopsied TBx, other factors are likely influencing these findings, including the number of lines of prior therapy.

712 patients with longitudinal TBx and LBx profiling reveal a spectrum of acquired tumor resistance mechanisms and actionable targets

712 patients were profiled with LBx following a TBx, during the course of care (Fig. 2, Supplementary Data 8). Samples were temporally heterogeneous, with a median of 495 days between biopsy collection (range 0–6455 days), included a mix of local (35%) and metastatic TBx (44%), and represented a diversity of subtypes (histological: IDC 84%, ILC 14%, metaplastic 2%; pathological: ER+/HER2− 64%, HER2+ 15%, TNBC; 22%) (Fig. 2). For pathogenic mutations detected in TBx, positive percent agreement (PPA) was 72% for LBx, heavily influenced by estimated tumor fraction and predicted germline status (Supplementary Fig. 7, Supplementary Data 9). PPA was highest in samples with a tumor fraction of >10%, especially for potentially actionable alterations in PIK3CA (93%) and BRCA1/2 (95%) (Supplementary Fig. 7a–c, Supplementary Data 9). PPA was higher for predicted germline alterations relative to somatic alterations (100% vs 70%; Supplementary Fig. 7d, Supplementary Data 9), such as in cases with founder BRCA1/2 alterations (Supplementary Fig. 7e). Of note, germline alterations were detected, regardless of tumor fraction, as expected since the former can be detected in non-tumor DNA even when ctDNA shed is low (Supplementary Fig. 7d, Supplementary Fig. 8). Tissue biopsies were further examined based on the site of TBx: local (breast biopsy), metastatic (distant site biopsy) and lymph node. Due to increased ctDNA shedding, PPA was significantly higher in metastatic and lymph node biopsies compared to breast-biopsied tumors (80% vs. 78% vs. 58%, respectively, p < 10−4, Supplementary Fig. 7f). PPA was not impacted by receptor subtype or time between tests (Supplementary Fig. 7g, Supplementary Fig. 7h, Supplementary Data 9).

We examined the landscape of acquired mutations in follow-up LBx and identified events most commonly in TP53 (246 alterations; 26.8% of acquired alterations), ESR1 (181; 19.7%), NF1 (69; 7.5%), PIK3CA (65; 7.1%), BRCA2 (48; 5.2%), PTEN (39; 4.2%), RB1 (32; 3.5%), and ERBB2 (21; 2.3%) (Fig. 4a). However, there was also a long tail of acquired mutations with acquired short variants seen in 48 genes, including in KRAS (n = 13), AKT1 (n = 9), EGFR (n = 7), FGFR2 (n = 7), BRAF (n = 5), MTOR (n = 4), FGFR1 (n = 4), NRAS (n = 3), HRAS (n = 2), FGFR3 (n = 1), and RET (n = 1) (Supplementary Data 10). Acquired short variants were more prevalent in patients with a longer time between biopsies, with LBx collected 3+ years after TBx harboring at least one acquired alteration in 69.5% of cases (Fig. 4b, Supplementary Data 11). A longer duration between biopsies likely represents a larger number of intervening therapies, which may result in the development of resistance mutations and more dynamic tumor evolution.

a Acquired short variants (e.g., point mutations, short indels) in patients with longitudinal tissue biopsy (TBx)/liquid biopsy (LBx), where LBx was collected at the time of or following TBx (n = 712). The fraction of samples with acquired short variants based on b collection time difference and c receptor subtype. d Acquired copy number amplifications detected in follow-up LBx samples (number of detected events for each gene is provided within parenthesis). e Acquired rearrangements detected in follow-up LBx samples (number of detected events for each gene is provided within parenthesis). f Reversion mechanisms impacting BRCA1 and BRCA2. Each patient is represented by a unique color, with shared BRCA1/2 alterations shown in red and identified reversions shown in black. Variant allele frequency for each identified BRCA1/2 short variant is provided. g Distribution of the estimated clonal fraction for shared and acquired short variants in LBx. Each box plot displays the interquartile range (IQR), with the lower and upper boundaries representing the 25th and 75th percentile; the line within the box represents the median and the whiskers extend to ±1.5 × IQR. h Polyclonality of acquired short variants identified in LBx. The fraction of a single (#1) or multiple (#2, #3+) detected alterations within each gene is shown.

Full size image

When stratified based on molecular subtype, acquired alterations were most common in ER+/HER2− samples (60%) with lower frequencies in TNBC (51%) and HER2+ subgroups (51% in ER+/HER2+ , 33% in ER−/HER2+ ) (Fig. 4c, Supplementary Fig. 9, Supplementary Data 8, Supplementary Data 11). These findings are consistent with previous reports identifying high frequencies of resistance alterations in patients with ER+ subtype treated with endocrine therapy8,9,23.

Although less common, liquid biopsies also detected acquired amplifications and rearrangements (n = 77). These included recurrent amplifications (Fig. 4d) in FGFR1 (n = 9), CD274 (PD-L1; n = 5), PDCD1LG2 (PD-L2; n = 4), MYC (n = 3), RET (n = 3), EGFR (n = 2), ERBB2 (n = 2), ESR1 (n = 2), KRAS (n = 2), and PIK3CA (n = 2) and rearrangements (Fig. 4e) involving NF1 (n = 5), BRCA1 (n = 4), BRCA2 (n = 4), TP53 (n = 4), RB1 (n = 3), CDK12 (n = 2), FGFR2 (n = 2), and PTEN (n = 2) (Supplementary Data 10). Of note, liquid biopsy samples with acquired amplifications and rearrangements had a significantly higher tumor fraction compared to samples where these large, acquired events were not detected (28.6%, 26.5% vs. 2.0% respectively, each p < 10−5; Supplementary Fig. 10), thereby highlighting the potential of high tumor fraction LBx samples in detecting these complex events. When including all classes of mutations, acquired alterations were seen in 56% of cases, including 70% of ER+/HER2− cases (Supplementary Data 8).

Many of the acquired alterations represent potentially targetable entities. Although acquired PIK3CA mutations were less often found in canonical codons 542/545/1047 (35/65, 54% acquired vs 134/184, 73% shared; p = 0.008) and included E726K (9/65), E418K (3/65), and E53K (3/65) mutations, most acquired alterations were observed in contexts qualified for PI3K inhibitors (54%) (Supplementary Data 12). For the ER+/HER2− population, acquired PIK3CA alterations were observed in 8% of patients. Previous reports have associated acquired PIK3CA alterations with APOBEC mutagenic processes18; consistent with this, 77% (48/62) of acquired missense PIK3CA alterations were observed in canonical APOBEC trinucleotide contexts (T[C > T/G]N) (Supplementary Data 12). Acquired ERBB2 short variants, which are potentially targetable with HER2 TKIs or with HER2 antibody-drug conjugates, were common in the dataset (n = 21 alterations in 15 patients) and were observed in both the ER+/HER2− and HER2+ subsets (2% and 4%, respectively)24,25,26,27. ERBB2 mutations have been previously reported as an endocrine therapy resistance mechanism, but their presence in the HER2+ population may represent resistance to HER2-targeted therapies9,28. In addition to short variants, a patient with ER−/HER2+ breast cancer also exhibited an acquired ERBB2-GRB7 fusion in the follow-up LBx, which retained the extracellular, juxtamembrane, and kinase domains.

Follow-up LBx also detected rare, potentially targetable acquired alterations. Alterations in FGFR1-3 family members were observed in 3% (23/712) of cases, predominantly in ER+ disease (Supplementary Fig. 11), and included amplifications of FGFR1 (n = 9 cases) and FGFR1/2 fusions (n = 3 cases). KRAS alterations were acquired in 14 cases, including nine with an acquired alteration in codon 12. A RET-BAIAP2L1 fusion was identified in LBx of patient 411, 1591 days after TBx (Fig. 4e, Supplementary Data 8, Supplementary Data 10); previous studies have demonstrated a role of RET in endocrine resistance with possible responses to RET inhibitors29,30,31.

Recent studies have shown the emerging role of reversions in BRCA1/2-mutant cancers, thereby conferring resistance to PARPi or platinum therapy32,33. In our cohort, acquired BRCA1/2 reversions were found in 21% of patients with a shared BRCA1/2 mutation. Reversion mechanisms were diverse, including short variants such as splice alterations, frameshifts, deletions of the original frameshift, and missense alterations (n = 8 cases; Fig. 4f) as well as more complex structural deletions (n = 7 cases). Of note, patient 345 exhibited 23 reversion BRCA2 alterations in the follow-up LBx, collected 132 days after the TBx, with a truncal BRCA2 A938fs*21 mutation (Fig. 4f). Similarly, most patients had multiple subclonal BRCA1/2 reversion events, likely representing convergent evolution in the context of acquired tumor resistance through multiple subclones. There were similar reversion rates in BRCA1 and BRCA2 (25% v 19%, p = 0.7). BRCA2 reversions often comprised a larger number of events (median = 3.5 reversion events) compared to BRCA1 (median = 1 reversion event). Notably, five of the 10 cases with a predicted BRCA1/2 reversion mutation also exhibited a large structural rearrangement event (Fig. 4f), highlighting the importance of broad LBx profiling in the resistance setting. Of note, we also examined patients with ≥2 BRCA1 or BRCA2 mutations in the overall LBx cohort (n = 3339), and identified 8 out of 10 cases with ≥2 BRCA1, and 16 out of 23 cases with ≥2 BRCA2 showing mechanisms of reversions (Supplementary Data 13).

Consistent with tumor heterogeneity and clonal evolution dynamics, acquired short variants in LBx revealed a lower clonal fraction relative to shared alterations with TBx (Fig. 4g). These findings highlight the utility of LBx in identifying the repertoire of subclonal events, present in only a portion of the tumor or a subset of metastatic lesions. In addition, acquired alterations tend to be polyclonal (Fig. 4h), across different receptor subtypes (Supplementary Fig. 12a), with over 50% of PIK3CA and TP53-mutant samples exhibiting 2 or more alterations. Similarly, polyclonal alterations in BRCA1, ESR1, ERBB2, and BRCA2 were also commonly identified. In contrast, alterations in KRAS and CDH1 often comprised single events. Presence of polyclonal mutations was also frequent in the overall LBx cohort, especially in ESR1, TP53, PIK3CA, and NF1 (~20%+), whereas, other gene alterations, such as in AKT1, KRAS, and EGFR were commonly monoclonal (Supplementary Fig. 12b). These polyclonal alterations allude to mechanisms of multiclonal convergent evolution in acquiring resistance. To test for sub-threshold evidence in the matched TBx, we examined 381 variants detected uniquely in the LBx; 363 (>95%) had ≤ five supporting reads, within the margin of sequencing error (Supplementary Data 14). These findings further support that a majority of these alterations are truly acquired, likely resistance mutations detected in the follow-up LBx sample.

To confirm the patterns observed in the matched TBx/LBx setting, we independently assessed 1328 patients with breast cancer who received longitudinal TBx CGP. A similar landscape of acquired alterations (TP53, ESR1, NF1, BRCA1/2, ERBB2), associations with time between biopsies, and elevated prevalence of acquired alterations in the ER+/HER2− subtype were identified in the second TBx (Supplementary Fig. 13, Supplementary Data 15, Supplementary Data 16). Interestingly, a higher rate of acquired events in multiple genes was observed in follow-up LBx compared to TBx (150 follow-up LBx samples, 21% vs. 98 follow-up TBx samples, 7.3%; p < 10−5, Supplementary Data 11, Supplementary Data 16). These results suggest that sequential biopsies, whether from liquid or tissue, can better inform mechanisms of tumor evolution and potential resistance.

Clonal hematopoiesis (CH), an age-related process of accumulating somatic mutations in hematopoietic stem cells, is a known layer of complexity in LBx-based profiling. Alterations frequently associated with CH, such as in JAK2 (specifically, V617F; n = 4 cases), ATM (n = 44 alterations in 34 cases), and CHEK2 (n = 46 alterations in 36 cases; Supplementary Data 9), were also observed to be unique to LBx in our cohort, albeit rare. These alterations were observed at a higher frequency in LBx relative to TBx (1.4% v 0.1% JAK2, 9.9% v 2.3% ATM,12.8% v 2.2% CHEK2 each p < 0.05; Supplementary Data 2), at lower allelic fractions (median 0.4% v 1.2% for all mutations, p < 0.05), and trended older (63 y vs 58 y overall), consistent with CH origin34. Though TP53 could also be associated with CH, small differences were seen with patient age (median 59 y v 58 y) and allelic fraction (median 0.9% v 1.2% overall; p > 0.05). Furthermore, TP53 prevalence was lower in LBx relative to TBx (43.8% v 52.2%, p < 10−5; Supplementary Data 2). NF1 has also been reported as a CH-associated gene35. While NF1 alterations were higher in prevalence in LBx compared to TBx (8.1% v 4.5%), NF1 is also a known resistance mechanism to endocrine therapy9,10,36. Consistent with this, NF1 was frequently acquired in follow-up tissue biopsies in the overall cohort (Supplementary Fig. 13), and was observed frequently in samples taken post-endocrine therapy in our clinical cohort, as will be described below. Notwithstanding these rare, potential CH-associated mechanisms, our findings showcase the diversity of the genomic landscape derived from LBx, thereby providing insights to better understand the tumor evolution dynamics and guide therapy selection by offering promising targets for patients with advanced breast cancers.

Acquired alterations are associated with therapeutic intervention

We next examined the association of acquired alterations with treatment patterns using a clinico-genomic database (CGDB) comprising 196 patients with longitudinal genomic tests, eligible for analysis (Fig. 5a). Similar to the CGP cohort, acquired alterations were most commonly observed in TP53 (30.1%), ESR1 (20.3%), PTEN (15.4%), NF1 (12.6%), PIK3CA (10.5%), ATM (3.5%), BRCA1 (4.2%), RB1 (3.5%), KRAS (2.8%), ERBB2 (2.1%), and BRCA2 (2.1%) (Fig. 5b, Supplementary Data 17).

a A schematic representing the cohort of 196 patients receiving multiple, longitudinal comprehensive genomic profiling tests, with real-world clinical and outcome data matched from Flatiron Health evaluated in this study. b Patterns of acquired short variants (e.g., point mutations, indels) in the overall cohort (top), and based on specific treatment interventions (bottom), revealed treatment-specific patterns of acquired alterations. Acquired alterations (y-axis) are colored based on the type of alteration for each gene (x-axis). The therapy groups include the following—endocrine therapy (ET) and Cyclin-dependent kinase inhibitors (CDK4/6i), endocrine therapy (ET), HER2-targeted therapy (antibody (Ab)-drug conjugates, tyrosine kinase inhibitors (TKI)), Poly adenosine diphosphate-ribose polymerase inhibitors (PARPi) mono/combo therapy. c Three representative patient journeys, displaying the utility of LBx during the course of care, are shown (time shown in months, mo). Shared alterations across biopsies are shown in red. Alterations unique to a specific biopsy are shown in blue while alterations not targeted on other platforms are shown in gray.

Full size image

Patients were further stratified by intervening treatments (Fig. 5b, Supplementary Data 17). Patients treated with endocrine therapies exhibited frequent acquisition of ESR1, NF1, and PIK3CA alterations. PARPi use was associated with acquired BRCA1/2 reversion mutations. CDK4/6 inhibitor treatment was associated with RB1 alterations and, likely due to the co-treatment with endocrine therapy, also presented with acquired ESR1, NF1, and PIK3CA alterations. HER2-targeted therapies were associated with acquired alterations in ERBB2, FGFR2, KRAS, and PIK3CA. These findings further elucidate the treatment-specific spectrum of acquired alterations and clonal evolution, inferred through longitudinal profiling. When leveraging the overall clinical cohort to examine the prevalence of gene alterations in patient exposed or not exposed to a given therapy class, similar associations were observed, such as elevated levels of ESR1 and RB1 in patients receiving a combination of endocrine therapy and CDK4/6 inhibitor therapy, and BRCA1/2 reversions following PARPi therapy (Supplementary Fig. 14, Supplementary Data 18). A number of additional, rare gene alterations were elevated in post-treatment samples of specific therapy classes (Supplementary Data 18, Supplementary Fig. 14).

We more closely examined the treatment journeys of three patients with multiple acquired alterations (Fig. 5c).

Case 1

A 67-year-old woman presented with Stage IV ER+/PR+/HER2− IDC. Tissue CGP from a breast mass identified IGF1R amplification and a GATA3 frameshift. Following progression on letrozole/palbociclib the patient was switched to fulvestrant/abemaciclib without progression. Liquid CGP during fulvestrant/abemaciclib treatment detected CDK4/6 inhibitor resistance mutation in RB1 (variant allele fraction, VAF 3.8%) and endocrine therapy resistance alteration ESR1 Y537S (0.4%). Other variants detected included ATM (VAF 12%), CDK12 (1.4%), TP53 (0.2%), and EGFR A1118T (0.2%). Mutations in ATM are common in solid tumors but are also in the bone marrow niche, raising the possibility of CH origin, although it is uncommon to detect CH variants above 10% of the cell-free DNA content37.

Case 2

A 41-year-old woman presented with Stage IIB ER+/HER2− ILC and progressed with skin metastases at age 51 following bilateral mastectomy and adjuvant tamoxifen/letrozole. Tissue CGP revealed a L1620*fs in exon 10 of BRCA1 (predicted somatic, with loss of heterozygosity of the wild type allele), and amplifications of MDM2 and FRS2. The patient progressed on fulvestrant/palbociclib, anastrozole/abemaciclib, and talazoparib with metastatic spread to bone and lymph nodes. Liquid CGP after progression on talazoparib detected the truncal BRCA1 L1620*fs at a VAF of 20%, as well as five additional BRCA1 mutations: 3 indels, a splice site, and a rearrangement that skips exons 8–9 (VAFs ranged between 0.3 and 2.6%), each predicted to skip the frameshift and restore at least partial BRCA1 function.

Case 3

A 67-year-old woman presented with Stage IV ER+/HER2 IHC negative (0–1+) ILC and was treated with letrozole/palbociclib; liquid CGP at progression identified PIK3CA E542K mutation and ERBB2 A775_G776insYVMA. After progression on lapatinib/capecitabine, a metastasis tested ER+/HER2 IHC-positive (3+). The patient was switched to paclitaxel/pertuzumab/trastuzumab, and then gemcitabine/pertuzumab/trastuzumab. A TBx (liver metastasis) and a LBx were collected and profiled during this last treatment. TBx identified mutations in PIK3CA, ERBB2, and TBX3. LBx identified the truncal PIK3CA and ERBB2 mutations as well as potential resistance alterations FGFR2 K569E (VAF 1.3%), KRAS G12V (0.5%), ERBB2 L755S (0.4%), and ESR1 D538G (0.1%). The ESR1 and ERBB2 L755S mutations are commonly seen after exposure to letrozole and lapatinib, respectively38,39. It is worth noting the ERBB2 exon 20 YVMA insertion may represent resistance to lapatinib at baseline40,41. KRAS mutations have also recently been described as a form of potential resistance to HER2-targeted therapy42. FGFR2 mutations have recently been described as a mechanism of resistance to hormonal therapy and CDK4/6 inhibitors43.

Collectively, these cases highlight the role of serial biopsy in identifying tumor evolution, with implications on emerging resistance and therapy selection.

With the rapid development of targeted therapies for breast cancer treatment, circulating tumor DNA profiling has shown potential in identifying actionable alterations, monitoring response, predicting outcome, and assessing treatment evasion44,45. The interplay of clonal heterogeneity and tumor evolution is critical for early stages of tumor initiation and progression, as well as late-stage relapse in advanced tumors46. Liquid biopsy captures both the spatial and temporal clonal heterogeneity in tumors, and can therefore be informative in monitoring treatment-associated response, especially in the metastatic setting. Here, we characterized the baseline and acquired alterations in a large breast cancer cohort comprising patients who underwent TBx and LBx CGP as part of routine clinical care.

Acquired alterations, potentially associated with resistance, were observed in a majority of patients with matched TBx/LBx, with the highest rates in ER+ disease and those with longer biopsy intervals. Analysis of CGDB revealed that liquid biopsies are used later in the course of patient care, typically in later stage tumors treated with more lines of prior therapy. This observation, in part, likely explains the higher prevalence of acquired alterations and clonal heterogeneity in the liquid biopsy setting. Consistent with prior studies, stratifying patients by intervening treatment, identified known therapy-related associations including acquired RB1 alterations in the CDK4/6i setting13, and BRCA1/2 reversions with PARPi32,33. Beyond the well-characterized ESR1 resistance mutations in the ER+ tumors, MAPK pathway alterations, including NF1 loss of function alterations and those in RAS pathway receptors (KRAS, BRAF, NRAS, HRAS) were also commonly identified in our cohort. Some of these alterations, including KRAS G12C, are currently being investigated in solid-tumor basket trials47, allowing for emerging options for these patients. BRAF V600E was recurrently observed as an acquired alteration, notable since the FDA granted accelerated approval for debrafinib/trametinib combination therapy48,49. Acquired PIK3CA and ERBB2 mutations, potentially targetable with small molecule inhibitors1,4 were also common.

Additional rare but actionable alterations that we observed in follow-up LBx may have implications on potential treatment options and combinatorial treatment strategies in these patients. The various FGFR inhibitors in development, taken together with our findings on FGFR1/2/3 acquired alterations in LBx, might be suggestive of patients with relapsed ER+ breast cancer being an appropriate population for targeted enrollment43. Albeit rare, an ERBB2 fusion identified in a patient with HER2+ breast cancer and an acquired RET fusion identified in a patient with an acquired ESR1, offer interesting insights into subtype-specific resistance mechanisms and warrant further investigation. These findings highlight that tumor evolution is more of a rule rather than an exception during disease progression, consistent with previous studies showing frequent acquisition of mutations in the resistance setting9,10,11,12,14,21,43.

Tumor heterogeneity, the basis of tumor evolution, has been associated with worse outcome in several tumors8,50,51,52. Thus, understanding what drives heterogeneous adaptation to intervening therapies is critical in devising better therapeutic approaches. Consistent with heterogeneous tumor evolution, acquired LBx alterations in our cohort were frequently subclonal and often exhibited polyclonal patterns. This was particularly true for BRCA1/2 reversions which were almost universally polyclonal, suggesting convergent evolution of resistance across multiple independent subclones may promote resistance to therapy. Similar findings were seen for ESR1, PIK3CA, and

ERBB2, with patients exhibiting multiple resistance-associated alterations in one or more of these genes. While therapies provide one source of evolutionary pressure, mutations that promote oncogenicity, prevent cell death, or provide a clonal advantage may also be selected for. TP53 mutations were commonly observed across breast cancer subtypes and may represent a mechanism of selection-driven clonal evolution. While this finding highlights the clinical challenges of therapeutic resistance, these characteristics could aid in the discovery of additional resistance mechanisms, even in the absence of longitudinal or paired data. In a sufficiently sized dataset, alterations frequently observed to be subclonal and/or polyclonal could be flagged as possible resistance alterations and prioritized for additional study and identification of targets for drug discovery. It is also plausible that the elevated presence of alterations identified in LBx may result from an increased evolutionary burden in certain tumors.

TBx and LBx CGP revealed a high prevalence of actionable alterations, including those in well-characterized drivers such as BRCA1/2 and PIK3CA. However, the extent to which LBx captured the mutational landscape derived from matched TBx profiles was strongly influenced by the estimated tumor fraction, with higher rates in sufficiently shedding samples. LBx CGP with no identified mutations may likely represent low ctDNA shedding; in such cases, reflex testing to TBx, if feasible, may be warranted. This underscores the need for consideration of LBx tumor fraction in the clinical setting where reporting of an accurate estimate of tumor fraction, could enable interpretation of a negative result. Further, the patient-matched paired biopsy analysis in this study was limited to samples profiled on an older LBx technology comprising 62–70 genes, in comparison to the larger TBx panel. As a result, the extent of acquired alterations and consequently, the extent of tumor evolution, may be understated in this cohort. However, the recent approval of larger LBx CGP profiling panels, may help overcome this limitation by providing a broader landscape of the tumor genomics, as observed in our preliminary analysis of 1430 patients who underwent testing on a 324+ gene panel. These tests have the potential to identify a greater breadth of potentially actionable alterations. Notwithstanding these limitations, our study showcased the abundance and diversity of tumor evolution and resistance mechanisms, under selective pressures of routine care and treatment. With the recent FDA-approval of NGS-based LBx CGP assays, and the associated uptake in payor reimbursement, our expectation is that liquid biopsy may grow to be used earlier in the care of breast cancer patients, potentially to enable tumor profiling at initial recurrence.

A limitation of this study was the lack of matched white blood cell data. Mutations that are rare in solid tumors such as JAK2 V617F can be assumed to be non-tumor derived and their presence may even be a reason for clinical follow-up and monitoring for development of myelodysplasia. Mutations that are more common in solid tumors and considered biomarkers for certain therapies pose more of a problem. Mutations in ATM, CHEK2 have been observed in white blood cells37. CH variants have even been observed in BRCA2 and KRAS, albeit rarely and not in patients with breast cancer53,54. Somatic mutations with low VAFs thus need to be interpreted with some level of caution. It is worth noting that PARPi treatment is not currently recommended by any regulatory agency guidelines for breast cancer with ATM2 and CHEK2. Our study is also limited based on the clinical and molecular (e.g., ER status) available in the CGP cohort. While acquired alterations are observed in paired-sample biopsies, especially those with longer biopsy intervals, the tumor features and clinical interventions are unknown for most of these cases.

Collectively, findings from our study reveal the dynamic evolution of breast tumors over time. Since both baseline and acquired alterations are potentially targetable (e.g., activating PIK3CA, ERBB2, KRAS, BRAF, FGFR1/2/3), longitudinal profiling throughout the course of care may provide additional treatment options, through combinatorial therapies or switching therapy regimens based on the evolving tumor landscape, to improve outcomes for patients with breast cancer. With poor outcomes in the relapsed and metastatic setting, clinical trials are urgently needed to assess whether targeting these alterations can improve outcomes. This could be achieved through an umbrella trial in the relapsed setting, matching patients to targeted therapies. Similar approaches have been taken for the CUPISCO and TAPISTRY trials which match patients to targeted therapy arms based on their genomic profiles55,56. Our study highlights the potential for liquid biopsy approaches to better characterize the evolutionary landscape, guide clinical decision making, monitor treatment response/resistance and ultimately aid in personalized care for patients with breast cancer.