Article https://doi.org/10.1038/s41467-023-36801-9 Circulating tumor DNA reveals complex biological features with clinical relevance in metastatic breast cancer Received: 2 September 2022 Aleix Prat 1,2,3,4,5,6,16 , Fara Brasó-Maristany 1,16, Olga Martínez-Sáez 1,3,4,16, Esther Sanfeliu1,7, Youli Xia2, Meritxell Bellet 6,8,9, Patricia Galván1, Accepted: 14 February 2023 Débora Martínez1, Tomás Pascual 1,3,6,10, Mercedes Marín-Aguilera2, Anna Rodríguez1,3, Nuria Chic1,3, Barbara Adamo1,3,4, Laia Paré2, Maria Vidal 1,3,4, Mireia Margelí11, Ester Ballana12, Marina Gómez-Rey13, Mafalda Oliveira 6,8,9, Check for updates Eudald Felip11,12, Judit Matito13, Rodrigo Sánchez-Bayona14, Anna Suñol8,9, 1234567890():,; 1234567890():,; Cristina Saura 6,8,9, Eva Ciruelos14, Pablo Tolosa 14, Montserrat Muñoz1,3,4,6, Blanca González-Farré1,7, Patricia Villagrasa2, Joel S. Parker15, Charles M. Perou15 & Ana Vivancos13 Liquid biopsy has proven valuable in identifying individual genetic alterations; however, the ability of plasma ctDNA to capture complex tumor phenotypes with clinical value is unknown. To address this question, we have performed 0.5X shallow whole-genome sequencing in plasma from 459 patients with metastatic breast cancer, including 245 patients treated with endocrine ther- apy and a CDK4/6 inhibitor (ET + CDK4/6i) from 2 independent cohorts. We demonstrate that machine learning multi-gene signatures, obtained from ctDNA, identify complex biological features, including measures of tumor proliferation and estrogen receptor signaling, similar to what is accomplished using direct tumor tissue DNA or RNA profiling. More importantly, 4 DNA- based subtypes, and a ctDNA-based genomic signature tracking retino- blastoma loss-of-heterozygosity, are significantly associated with poor response and survival outcome following ET + CDK4/6i, independently of plasma tumor fraction. Our approach opens opportunities for the discovery of additional multi-feature genomic predictors coming from ctDNA in breast cancer and other cancer-types. Sequencing of tumor DNA has brought many new biomarkers and and henceforth called “ctDNA”) allows an easy access to some tumor- possibilities to precision oncology1. Detection of somatic gene muta- based genetic information at any given timepoint and can replace a tions, amplifications, and gene fusions allows the delivery of targeted tumor tissue biopsy in some cases, thus avoiding delays and compli- therapies in multiple cancer-types, such as lung cancer, colorectal, cations of a solid tumor invasive biopsy procedure, which can be quite melanoma and breast cancer1. In addition, detection of a high number challenging sometimes in the metastatic setting. of somatic mutations (i.e., tumor mutational burden), or a micro- Identification of single tumor DNA alterations can be clinically satellite instability-high phenotype, can help identify candidates for useful1. However, cancer is highly complex and additional biological anti-PD1/PDL1 immune checkpoint inhibitors2,3. Importantly, sequen- information is likely needed to refine the prediction of patients´ prog- cing of tumor DNA in blood samples (i.e., the so-called liquid biopsy, nosis and/or treatment benefit4. Breast cancer is the perfect example A full list of affiliations appears at the end of the paper. e-mail: alprat@clinic.cat Nature Communications | (2023)14:1157 1 Article https://doi.org/10.1038/s41467-023-36801-9 since RNA-based profiling tests provide clinical and biological useful were applied10 on 0.5X shallow whole genome sequencing (shWGS) information beyond individual somatic gene mutations or amplifications data from a first dataset, hereafter Plasma-1 cohort (Fig. 1a and Sup- such as PIK3CA or ERBB2 5–8. In early disease, multi-gene RNA-based plementary Fig. 1a), composed of 246 plasma samples. Most samples prognostic assays (e.g., OncotypeDX, Mammaprint and Prosigna) are profiled (n = 209) were from 174 patients with advanced HR+/HER2- available and recommended by clinical guidelines. In advanced disease, breast cancer. Samples from the same patient were obtained at dif- RNA-based profiling is becoming a promising prognostic and predictive ferent timepoints. Additional samples from the other clinical subtypes tool5–7,9. Unfortunately, tissue samples in patients with advanced disease were also assayed including 19 plasma samples from 16 patients with are not readily available and, even so, the type of metastatic organ or site advanced HER2-positive (HER2+) disease, 17 plasma samples from 16 can compromise the expression patterns obtained from bulk RNA and patients with advanced triple-negative breast cancer (TNBC) and 1 might not reflect the intra-patient tumor heterogeneity. plasma sample from 1 patient with advanced disease but unknown HR To help overcome these challenges, we previously reported a and HER2 status. supervised learning integrative computational approach to predict RNA- based tumor expression signatures values using data from DNA copy- Plasma tumor fraction number alterations10. Specifically, we found 150 multi-feature DNA- Of 246 plasma samples, 178 (72.4%) had a Tumor cell Fraction (TF) of based signatures tracking a variety of breast cancer biological processes, ≥3% (range 4–84%; median 9.4%), indicating presence of tumor, including proliferation status, clinical ER tumor status, tumor histology, according to the ichorCNA tool12 (Supplementary Fig. 1b). The TFs and estrogen-signaling pathway activity, with a high predictive perfor- detected in our study are in line with those reported in other breast mance (i.e., area under the ROC curve ≥0.75)10. This finding opened the cancer studies13,14. In plasma samples with a TF ≥3%, we measured the possibility to use DNA sequencing of ctDNA in blood to capture clinically signals of 514 DNA segments and subsequently determined the scores relevant information beyond single genetic alterations and tumor frac- for each of the 150 previously reported DNA signatures designed to tion, the latter being associated with patient’s prognosis11. Our approach predict tumor RNA- and protein-based phenotypes10 (Fig. 2a). Of note, could be highly relevant in the metastatic setting, where ctDNA might be all signatures/models were applied exactly as previously reported10 the only readily available genetic material from tumors. (Supplementary Data 1 and 2); thus, in this study, no new models were Here, we hypothesized that (1) DNA-based signatures tracking developed and the samples analyzed can be considered ‘test/valida- breast cancer biological processes can be detected in ctDNA and tion’ datasets. provide clinically useful information and (2) DNA-signatures detected As expected, TF as a continuous variable was found strongly cor- in plasma and tissue can help identify tumor subtypes within hormone related with the number of altered DNA copy-number segments in each receptor-positive and HER2-negative (HR+/HER2-) breast cancer. Our sample (Pearson’s rho = 0.76) (Fig. S2). In addition, TF was found work demonstrates that complex tumor phenotypic traits can be strongly correlated (i.e., Pearson’s rho ≥0.70 or ≤−0.70) with 46 of 150 identified in ctDNA and might be clinically relevant in advanced breast (31.0%) ctDNA-based signature scores (Supplementary Data 3). The cancer. Our ctDNA-based multi-gene signature approach opens ctDNA-based signatures highly correlated with TF were mostly tracking potential avenues for discovering and implementing biomarkers in biological processes associated with Luminal B disease (i.e., positively oncology, especially in the advanced disease setting where liquid correlated with TF) or Luminal A disease (i.e., negatively correlated with biopsies are becoming commonplace. TF). To compare the ctDNA-based signature scores across patients, we used the TF-adjusted tumor copy-number signal, as provided by the Results ichorCNA tool12. As expected, this normalization step decreased the To demonstrate that ctDNA can capture complex tumor phenotypes, strength of association between TF and the number of altered copy- 150 previously defined multi-feature DNA-based signatures (Table 1) number segments, and between TF and each ctDNA-based signature Table 1 | Description of selected DNA-based signatures Description DNA-based signatures Original RNA-based signatures Name DNA segments N genes PMID References 46 Research-based Mammaprint NKI70 149 60 11823860 47 Estrogen-regulated gene expression Scorr_IE 166 754 16505416 48 TP53 mutational status P53_Mut_Correlation 240 48 17150101 49 Genes up-regulated in Basal-like GSEA_SMID_Basal_UP 232 648 18451135 50 Proliferation/cell cycle Wirapati_Proliferation 112 355 18662380 16 Retinoblastoma loss of heterozygosity RB-LOH 236 345 18782450 15 Research-based PAM50 Luminal A Scorr_LumA_Correlation 197 50 19204204 15 Research-based PAM50 Luminal B Scorr_LumB_Correlation 184 50 19204204 15 Research-based PAM50 HER2-enriched Scorr_Her2_Correlation 66 50 19204204 15 Research-based PAM50 Basal-like Scorr_Basal_Correlation 229 50 19204204 51 Glycolysis gene expression Glycolysis_Signature 123 5 19291283 18 Luminal-related gene expresion Luminal_Cluster 236 78 21214954 18 HER2 amplicon gene expression HER2_Amplicon 34 19 21214954 52 TP53 mutational status in ER + P53_ER + 89 29 21248301 20 P53null luminal mouse model MM_p53null_Luminal 166 224 24220145 20 MYC amplified mouse model MM_Myc 184 228 24220145 19 Estrogen receptor signaling GP7_Estrogen_signaling 255 701 25109877 The information for each of the 150 ctDNA-based signatures, including the weights of each DNA segment, can be found in Supplementary Data 11. Nature Communications | (2023)14:1157 2 Article https://doi.org/10.1038/s41467-023-36801-9 a Plasma-1 Plasma-2 h h n=207 patients n=252 patients n=246 plasma samples n=357 plasma samples HR+/HER2- HR+/HER2- h h n=174 patients n=209 plasma samples n=242 patients n=346 plasma samples HER2+ HER2+ h h n=16 patients n=19 plasma samples n=6 patients n=7 plasma samples TNBC TNBC h h n=16 patients n=17 plasma samples n=4 patients n=4 plasma samples unknown h n=1 patient n=1 plasma sample TF>3% h n=155 patients n=192 plasma samples TF>3% h n=156 patients n=178 plasma samples n=107 RNA samples (nCounter BC360 panel) n=54 DNA samples (NGS VHIO300 panel) b CDK-Validation-1 CDK-Validation-2 h h n=124 patients* n=124 plasma samples n=121 patients* n=121 plasma samples TF>3% TF>3% h h n=87 patients n=87 plasma samples n=65 patients n=65 plasma samples * 57.3% 1st line, 53.2% visceral metastasis, * 59.5% 1st line, 55.4% visceral metastasis, median follow-up=12.5 months median follow-up=12.3 months Fig. 1 | Summary of plasma samples used in the study. a Description of plasma HR+/HER2- breast cancer treated with endocrine therapy and a CDK4/6 samples obtained from patients with advanced breast cancer across Plasma-1 inhibitor (CDK-Validation-1 and CDK-Validation-1 cohorts). TF tumor fraction. and Plasma-2 cohorts. b Description of samples from patients with advanced score (Supplementary Data 3). Of note, no ctDNA-based signature was (Fig. 2b and Supplementary Data 4). Two of the highly correlated DNA found moderately or strongly correlated with TF (i.e., Pearson´s rho signatures were the UNC_8q_Amplicon and the UNC_Scorr_P53_Muta- ≥0.50) after the adjustment (Supplementary Data 3). tion. Overall, these results suggest a moderate association between ctDNA-based and tumor DNA-based signatures across timepoints and Plasma versus tissue DNA-based signatures DNA sequencing approaches (i.e., ctDNA shWGS versus capture-based We next explored the correlation of each of the 150 DNA-based sig- using FFPE DNAs). Of note, immune-related DNA-based signatures natures determined using plasma ctDNA (adjusted by TF) versus tumor showed low correlation coefficients (Supplementary Data 4). tissue DNA across 54 patients with available paired sample-types To further demonstrate that copy number alterations (CNA)- obtained within a timeframe of 8-weeks (n = 27) or more than 8 weeks based data from plasma allows the identification of similar biolo- (n = 27). Genome-wide copy number data was obtained from formalin- gical states as in tissue, we calculated the intra-patient correlation fixed paraffin-embedded (FFPE) tumors using a capture-based panel, of the CNA-based signals of 514 DNA segments across 54 paired and from ctDNA using shWGS (see methods). Across all 150 signatures, samples (tumor tissue versus plasma). Overall, 57% of patients had the average correlation coefficient between tumor and plasma was a correlation coefficient between plasma and tissue >0.50, which 0.40 (range 0.02 to 0.66) and 40 signatures (26.7%) had a correlation increased to 83% when we evaluated 29 patients with plasma coefficient ≥0.50. When the correlations were evaluated in the 27 cases TF >10%. In these 29 patients with a plasma TF >10%, 59 and 24% had where plasma and tumor were obtained within a timeframe of a correlation coefficient of >0.70 and >0.80, respectively (Fig. 2c 8.0 weeks, the number of signatures with a correlation coefficient and Supplementary Data 5). Overall, these results strongly suggest ≥0.50 was 63 (42% versus 19.3% in the 27 cases where plasma and that plasma ctDNA can reliably capture CNA-based signals from tumor were obtained within >8.0 weeks; p < 0.001), including 36 of the tumor tissue, although the amount of ctDNA might impact the 40 (90%) signatures with a correlation coefficient ≥0.50 in all patients ability to accomplish this (Fig. 2d and Fig. S3). Nature Communications | (2023)14:1157 3 Article https://doi.org/10.1038/s41467-023-36801-9 a b Advanced UNC HER1 Cluster2 score RB-LOH score breast cancer ctDNA shallow WGS 3 ● Cor= 0.55 ● Cor= 0.67 ● 8 ctDNA-based ctDNA-based 2 p<0.001 p<0.001 ● h 6 1 ● ● ● ● ● ● ● ● ●● ● ● 4 ● ● plasma ● 150 DNA-based 0 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● signatures ● ● ● ● ● ●● ● ● 2 ● ● −1 ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● 0 ● ●● ● ●● ● ● Tumor phenotypes (DNA CNA, mRNA −2 ● ● ● ● ● ● ● ● ● expression, IHC protein expression) −0.5 0.0 0.5 1.0 0 1 2 3 FFPE tumor tissue Tumor DNA−based Tumor DNA−based c Signal of 514 segments d HR+/HER2-, PAM50 HER2-enriched, High RB-LOH signature Case#1 Case#2 Plasma 1.5 1.5 2 Copy Number (log2 ratio) Cor=0.928 ● ● ●● ● ● ● ●● ● ● Cor=0.942 1.0 p<0.001 1.0 p<0.001 ●● ctDNA-based ctDNA-based ● ● ● ● ● ● ● ● ● ● ● ● 1 0.5 0.5 ● ● ● ● ● ● ● ●●● ●● ● ● ● ● ● ● ● ● ●● ●●● ● ● ●● ● ● ●●●● ● ●● ● ● ● ●● ● ● ●● ● ●● ●●● ●●● ●●●● ●●●● ● ●●●● ●●● ●●●●●● ●●● ● ● ● ● ●●●●●●●●● ●● ●● ● ●● ●● ● ●● ● ●●●● ●●●●●●●●●●● ● ●●●●●●●●●●●●●●●● ●●● ● ●● ● ● ● ● ●● ● ● ● ● ● ●●●●●●● ● ● ● ● ● ● ●● ● ●●● ●● ● ● ● ●●●●● ●● ●●●●● ●● ●● ●●● ●●● ●● ● ● ● ● ●● ●●●●●●● ●●●●● ● ●● ● ●● ● ●● ●●●● ●● ●●● ●●● ● ●●●●● ●●●●●● ●● ●● ●● ●●●● ●●●●●●●● ● ●●●●●● ● ●●●●● ●●●●●●● ●●●● ●●● ●● ● ● ●● ● ● ●●● ● ● ●● ● ● ●●● ●● ●● ●● ●● ● ●●●●●●●●●● ●● ● ●● ● ● ● ● ●● ● ● ●● ●●● ● ● ● ●●● ●●●●●● ●●●● ● ● ● ●●●● ●● ● ● ●●●●● ●●● ● ●●●● ●● ●● ● ●● ●●● ●●●●● ●●● ●●●●● ●● ●● ● ●● ● ●●●● ●● ●●● ● ● ●●●●● ● ●● ●●●●●● ●●● ● ● ●●● ● ● ● ●● ●● ●●● ● ●● ●● ●●● ● ●●●●●● ● ● ●● ●●●●● ●●● ●● ●●●●●●●●●●● ●● ● ● ● ●● ●●●●●●●● ●● ●●●●●●●●● ●● ●●●●●● ● ● ● ● ● ●● ●● ● ●● ● ●●● ●●●●●● ●●●●●●●●●●● ●●●● ●● ●● ● ● ●● ●● ● ●●● ● ● ● ●●●● ● ●● ● ●● ● ●●●●● ●● ● ● ● ●● ●● ● ●● ● ● ● ● ●●●● ●● ● ● ●● ●● ● ● ●● ● ● ● ●●●●●● ● ● ●● ●● ●● ● ●● ●● ●●● ●●●● ●●●● ●●●● ●● ● ● ● ● ●● ●● ● ● ●● ●●●●●● ● ● ●● ●●● ●●● ●● ● ● ● ●●●●● ● ● ●●●●● ●●●●● ●●● ●●●● ●● ●● ●●●● ●●● ●● ●● ●●●●● ●● ●●●●●●● ●●●● ●● ● ● ●●●●●●●●●●● ●●● ●●●● ●●● ● ● ●● ●● ●● ● ●●● ● ●● ● ●● ●●●●●●●●●●●●● ●●●● ●● ●● ●●●●● ●●● ● ●●●● ●● ● ● ● ●●●●● ●●●● ● ●● ●●●● ●● ●● ●●●●●●●●● ●●●●●● ●●●●●●●● ● ●●●●●●●●●●●●●●●●● ●●● ●●●●●●● ●●● ● ●● ●●●●●●●●●●●●●●●● ● ●●●●●●●●● ● ● ●●●●● ●●●●●●●●●●●● ● ●●● ● ●●●●●●●●●●●● ●●●● ●●●● ●●●●●● ● ●● ●● ● ●●●● ●● ●●●●● ● ●● ●●●●●●●●●●●●● ●● ●●●●●●●●●●●●●●●● ●●●●●● ● ●●●●● ● ●●● ●● ●● ●● ●● ●●●●●●●● ●● ●●● ●●●●●●●●● ● ● ●●●● ● ●●●● ●● ●● ● ● ●●●● ●● ● ● ●●●●● ● ●●●●●●●●●●●●● ● ●● ●●●●● ●● ●● ●● ●● ● ● ● ●● ● ● 0.0 0.0 ●● ●● ●●●●●● ●● ● ●●●● ●●● ●● ●●●● ● ● ●● ● ● ●●●●●●●● ●● ● ●●●●●●●● ● ● ● ● ●● ● ●●● ●● ●● ●● ●●●● ●● ● ●● ● ●●●●●●● ●● ● ●●●●●●● ● ●●●● ●● ●●● ● ●● ●●●●●●●●● ● 0 ●● ●● ● ●●●●● ●● ● ●●●● ● ●●●● ● ●● ●●●● ●● ●●●●● ● ● ● ●●● ●●● ●● ●●● ●●●●●●●● ● ● ● ●●●●●● ● ●●●● ●● ●● ●●●●●● ●●●● ●●●●●●●●●●●● ● ●●●● ●●●● ● ●●●● ● ●● ● ● ●● ● ● ●●●●●●● ● ●● ● ●● ●●●● ● ●●●●●●●●● ●●●● ● ●● ● ●●●●●● ● ●● ●●●● ● ●●● ●● ● ● ● ● ●● ● ● ● ● ●● ●●● ● ●●●●●● ● ●●●● ●●●●● ●●●● ●●●●●● ●●● ●● ● ●●● ●● ● ●●● ●● ●●● ●●● ● ● ●●●●●●● ● ●●●●●●● ● ● ●● ● ●●●●●●●●●●● ●●●● ●● ●●●●●●●● ●●● ●● ● ●●● ●●● ●● ● ● ●●●●● ●●●●●●●●●●●● ●●● ●●● ●●●● ●● ●●●●●●●●●● ●●●●● ●●●●● ●● ●●●●●● ●●●●●● ●●●●●●●●●●●● ●●●● ●● ●●●●●●●●●●●● ●●● ●● ●● ● ●●●●●●●●●●●●●● ● ●●●●●● ● ● ●●●●●●●● ●● ● ● ● ● ● ●●●●● ●●● ● ●●● ● ●●●●●●● ● ●●●●●● ●●●●●●●●●● ●● ●● ● ●● ●● ●●●●● ●● ● ●● ●●● ●●●● ●● ●●● ●●●●●●● ●● ● ● ●● ●●● ●● ●●●●●●● ● ●●●●●●●●●●● ●● ● ● ● ● ●● ●● ● ●●●●●● ●●● ●●●●● ●●●●●●●● ● ●● ● ●●● ●● ● ●●● ● ● ● ●● ● ●● ● ● ● ● ●●●●●●● ● ● ●●●●●●● ● ●●● ● ●●● ● ●●● ● ● ● ●●● ●● ● ●● ● ●●● ● ●●● ●●●●●● ● ● ●●●●●●●●● ●● ●● ●●● ● ●●● ● ●● ● ● ● ● ● ● ●●●●● ● ●● ●●● ● ●● ●● ● ● ●●●● ●● ● ● ● ● ●●● ● ●● ●●●●●● ●●●●●●● ● ●●● ●●●●●●●●● ● ● ● ● ●● ● ●●● ●●●●● ● ●●●●● ● ●●● ● ●●●● ●●●● ●● ●●● ● ● ●●●●●●●●●●●● ● ●● ●●● ●●●●● ●●●●●●●● ● ●●●● ●● ●●●●●●●●●●●●●●●● ●●●●● ● ●●●●●● ● ●●●●●●● ●●●● ●● ● ● ●● ● ● ● ● ●● ● ●●●● ●● ● ●● ●●●●●●● ●●●●●●● ●●●●● ● ● ●●● ●●●●●●● ● ●●● ● ●●●●● ●●● ● ●●●●●●●●●●● ● ● ●●● ●●●●● ● ● ●●●●●●●●● ●● ●● ●●●●●●● ● ● ● ●●● ●●●●●● ●●● ● ● ●● ●● ●● ● ● ● ● ● ● ●● ● ● ● ●● ●●●●●● ●●● ● ●● ●●●●●●●●●●● ●●● ● ●●●●● ●●●●●●● ●●●● ●●● ● ● ● ● ●●● ● ●●●● ● ●● ● ● ● ● ●●●●● ● ●●●●● ●●● ●● ●● ●●● ● ● ● ● ●● ● ●●●● ● ● ●●● ●● ●● ● ●●● ●● ● ● ●● ●● ● ●● ●●● ●● ●●● ●●●●●● ●●●● ● ●●● ●●●●● ● ●●●● ●● ●● ● ●● ●● ●● ● ●● ● ● ●●●●●● ● ● ●● ●●● ●● ● ●●●● ●● ● ●●●●●●● ●● ●●●●● ● ●● ●● ●●●●●● ● ● ●●●●●●●●● ●● ●●●●● ●●●●●●●● ●●●●●● ●●● ● ●● ●●● ●●● ●●●●● ● ●● ●● ●●●● ● ●●● ●●●●●●●●●●●●● ●●●●● ● ●●● ● ● ● ●● ● ●●● ● ●● ● ● ● ●● ● ● ●●● ● ●●●●●●● ● ● ● ● ●●● ● ●● ● ●● ● ● ● ● ●●●●● ●●● ●● ●● ●●● ●●●●●● ●●●● ●●●●● ●● ●●● ● ●● ● ● ●● ●● ●●●●● ●●●● ●●●●●● ● ● ● ● ●● ● ●●●●● ● ● ●●●● ● ●● ● ●●● ● ●● ●● ● ●● ● ● ● ● ●● ● ●●●●●● ●● ● ●●●● ●●●● ●● ●● ● ●●● ●● ● ●● ● ● ●● ● ●● ●●●●●●● ● ●● ●●● ● ●●●●●● ●●● ●● ● ●●●●● ●●●●●●● ● ●●●● ●● ● ●● ●● ● ● ● ●●● ● ●●●● ●●●● ● ●●● ●●● ●●●● ●●●● ● ●●●● ● ●● ● ●●●● ●●● ●●● ● ●●● ● ●●● ● ● ●● ●● ●● ● ● ●● ●●● ● ● ●●●●●●●●●●●●●●● ● ●●●●● ●●●●●●●● ●●● ● ● ●●● ● ●●●● ●●●● ●● ●●●● ● ●●●● ●●● ● ●● ●●●●●●●●●●● ● ●● ● ●●●● ●●●●●●●●● ●● ●● ●●●●● ●●●●●●●●● ● ●● ● ● ●●●●●●●●● ● ●●●●●● ●●● ● ● ●●●●●●●●● ●●● ● ● ● ●●●●●●● ●● ●●●●● ●●●●● ●●● ●●●●●●●● ●● ●●●●● ● ● ●●●● ●● ●●●●●●●●●●●●● ●● ●●●● ●● ●●● ● ●●●●● ●●●● ●●● ●● ● ●●●●●●●●●● ●●●● ● ●●●● ●● ●●●●●●●● ●● ● ● ●●●●●● ● ● ●●●●● ●●● ●●●●●● ●● ●● ●●● ● ●● ● ●●● ●●● ●●●●●● ●●●●●●●●●●●● ●●●● ● ●● ●● ●● ●●●●● ●● ● ●●● ●● ●●●●●● ●●●●●● ●● ●● ●●●●●●●● ● ● ●● ●●●●●●● ● ●●●● ●●●●●●● ●● ● ●● ●●●●●●●●●●● ● ●● ●●●●●● ●● ● ●●●●●● ● ●● ● ● ● ●● ● ● ● ● ●● ● ●● ● ●●● ● ● ● ● ● ●● ● ●●● ● ●● ●●●●● ●●●●●● ● ●●●●●● ●● ● ●● ● ● ● ●● ● ●●●●● ●●● ●● ●●●● ● ●● ●●● ●●●●● ●●● ●●●● ● ●● ●●●●●● ● ●● ● ●●● ●● ● ●●●●●●●●●● ●●●●●●●●● ● ● ●● ●●●●●●● ●●●●●●●●●●● ●●●● ●● ● ● ● ●●● ●●● ●● ●● ● ●● ●● ● ● ● ●●●● ●●●●● ● ● ●●●●●●●●●●● ●●●●●● ●● ●●● ●●●●● ●●● ●●● ●●●●● ● ● ● ● ●●● ●●●●●●●●● ● ●●●●●● ●● ●● ● ●●●●● ●●●●●●●● ● ●● ● ●●●● ●●● ●● ●●●●● ●● ●● ●● ● ● ●●●●●●●●●●● ●●● ●●● ●● ●● ●●●●●●●●●●● ●●● ● ● ●●●●●●●● ●● ●●●● ●●●● ●●●●●● ● ●●● ● ●● ●●●● ● ● ● ● ●● ●● ● ●●●●●●● ●● ●● ●●●●● ●●●●●●● ●● ●● ● ●● ● ●●● ● ●● ●● ●●●● ● ●●●●●●● ● ●●●● ●●●● ●●● ●●● ● ● ●●● ● ●● ● ●● ● ● ●● ●●● ● ●●● ●●●● ● ● ●● ●●● ● ● ●●● ● ●● ●● ● ●●●● ●●●●●●●●●●●●● ●●● ●●●●●●●● ●●●●●●●● ● ● ●● ●●● ● ● ●●●● ●●●●● ● ●● ● ●●●● ● ●●●● ●●●●●●● ●●●●● ●● ●●● ●●●● ●●● ● ● ●●●●●●● ●●●●●● ●●●● ●● ● ●●● ●●●●● ● ●●●●●●●●●● ●●●●●●● ●●● ● ● ●●● ●●● ● ●●● ●●●●● ● ● ● ●● ● ● ●●●●●●●●●●●●●●● ●●●●●●●● ●● ●●●●●●● ● ●●●●●●● ●●●●●●●●●●●●●●●● ●●●●●●● ●●● ●●●●● ● ●●●●●●●●●●● ●●●●●●● ●●●●●●● ●●●●●●● ●● ● ●●●●● ●●●●●●● ● ●●●●● ●● ● ●●●●●●● ●● ●●●● ● ●●●●●●●●●●● ●● ●● ●●●●● ●●●●●● ●●●● ● ● ● ●● ● ● ●● ●● ●●● ●● ●● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ●●● ●●●●●●●●●●●● ●● ●●●● ● ● ● ●● ●●●● ●●●●●● ●●●● ●● ●●●●●●●●●●●●●● ●● ●●●●● ●●●● ● ● ● ●● ●●●●●●●●●●●● ●● ● ● ● ● ● ●● ● ● ● ●● ● ●● ● ●● ● ● ●●●●●●●●●●●●●●●●● ● ●●● ● ●●●●●● ● ●● ● ● ●● ●●● ●● ●● ● ● ● ● ● ● ● ● ●●● ●●●●●● ● ● ● ●● ● ● ● ● ● ● ●●●● ●●●●●●●● ●●●●●● ●●● ●●●●●●●●● ●● ●●●●●●●●●● ●●●● ●●●●●●●●●●● ●●●●●● ● ●●●●● ●●● ●● ●●● ● ● ●●●●●●●●●●● ●● ● ● ●● ●● ●●● ● ● ●●● ●● ●●● ● ●●●●●●● ● ●● ●●●●●●●● ● ●● ● ●●● ●● ●●●●●●● ●● ●●● ●● ●●●●●●●●●●●●● ●●●●●●●●● ●● ● ●●● ●●●●●●●●●●● ●● ●●● ● ●●●● ● ● ●●●●● ●●● ● ●● ●●●●● ● ●● ● ●●●● ●● ●●●●● ●●● ● ●● ● ●●● ●●●●● ●● ●●●● ●●●●● ● ●● ● ●●● ● ●●●● ●●●● ● ●● ● ●●● ●● ● ● ● ●●●●● ● ●● ●●●●●● ●●● ● ● ●●●●●● ●●● ● ●● ● ●● ● ● ● ● ● ● ● ●●●●● ●● ● ●●●● ●●●● ● ●● ● ● ●● ●●● ● ● ●● ● ●● ● ●● ● ●● ● ●●●● ● ● ● ● ●●● ●● ●● ●● ● ●● ● ●● ●● ●● ●●● ●●●●● ● ●● ●● ● ● ● ●● ●●●●●● ●● ●● ● ● ●● ● ●●● ●● ● ● ●● ●●●● ●● ●● ● ● ●● ● ●● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● −0.5 ●● ● ● ● ● ● ● −0.5 ● ● ● ●● −1 ● ● ● −1.0 ● ● −1.0 ●●●●●●●● ● ● ● ●●●● ● ●●●●●● ●●●●●●●●● ● ●● ●●●●●●●● ● ● ● ●●● ● ● ● ● ● −1.5 ● ● ● ● ● ● ●● ● −1.5 −2 ● 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 18 20 22 X −1.0 −0.5 −1.0 −0.5 Tumor 0.0 0.5 1.0 0.0 0.5 1.0 2 Copy Number (log2 ratio) Tumor DNA−based Tumor DNA−based 1 e Tumor RNA-based PAM50 signatures 0 −1 UNC_MM_C3Tag_1pFDR_UP −2 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 18 20 22 X UNC_MM_C3Tag.2012 ctDNA-based signatures UNC_MM_WAPTag_1pFDR_UP UNC_MM_Myc.2012 f UNC_Duke_Module03_betacatenin ctDNA-based estrogen signaling score Pcorr_CSR_Activated 4 p<0.001 p<0.001 ctDNA-based HER2 signaling score RB_LOSS 4 UNC_RB_LOH 2 UNC_HER1_Cluster2 3 Tumour_hypoxia_DNA_hypermethylation UNC_Scorr_P53_Mut 0 2 ER+ HER2+ UNC_Scorr_LumA Chemo_Endocrine_Scor 1 −2 Pcorr_NKI70 0 rma A Pro R2-e l-like e Lum score Lum nal B d correlation l-lik rati riche −4 No inal 0.50 0.33 ER-negative HER2-negative HE Basa i 0.17 −1 on n 0.00 -0.17 AUC=0.772 AUC=0.720 -0.33 -0.50 life ER- ER+ HER2− HER2+ Fig. 2 | Circulating tumor DNA (ctDNA) in metastatic breast cancer. a Plasma by Pearson correlation. d Example of copy number alterations (CNA) plots of samples were obtained from 207 patients (174 with HR+/HER2-, 16 HER2+, 16 TNBC, matched timepoint plasma and tissue samples of a HR+/HER2- tumour, PAM50 1 N/A). After purification of plasma cell-free DNA, shallow whole genome sequen- HER2-enriched and high RB-LOH signature score. e Unsupervised cluster heatmap cing (shWGS) was performed. Using the ctDNA-based sequencing data from 514 analysis of Pearson´s correlation coefficients obtained by comparing the scores of DNA segments, 150 previously developed DNA copy number-based signatures10 the top individual ctDNA-based signature versus the scores of each individual tracking a variety of biological processes were applied in patients with a Tumor cell PAM50 RNA-based tissue signature across 58 matched-timepoint paired plasma- Fraction (TF) ≥3%. Individual scores for each signature were obtained. The com- tissue cases. Correlation values can be found in Supplementary Data 6. f Boxplots of plete names of the signatures, references and gene lists can be found in Supple- a ctDNA-based signatures tracking ER-related biology in ER-negative (n = 23) and mentary Data 1. b Examples of correlation between the scores of the ER+ tumors (n = 155) (p < 0.001) (left) and boxplots of a ctDNA-based signatures UNC_HER1_Cluster2 (p < 0.001) and the RB-LOH (p < 0.001) signatures when tracking HER2-related biology in HER2- (n = 165) and HER2+ tumors (n = 13) determined in plasma versus tumor tissue (the correlations for the other signatures (p < 0.001) (right). P-value (P) was determined by two-tailed unpaired t-tests. For can be found in Supplementary Data 4). Of note, tissue samples were obtained at the boxplot, center line indicates median; box limits indicate upper and lower different timepoints than the plasma samples. c Examples of correlations of 514 quartiles; whiskers indicate 1.5× interquartile range. Examples of ER and HER2 DNA signals between plasma and tissue in two single patients (case#1 p < 0.001, immunohistochemistry (ICH) stainings (20X) are provided. Source data are pro- case#2 p < 0.001). Correlation coefficients (Cor) and p in b and c were determined vided as a Source Data file. ctDNA-based data versus tissue RNA-based expression data enrichments and RNA expression data, we evaluated the correlation RNA-based expression data from FFPE tissue using a research-based of 6 PAM50 RNA-based tissue signatures with each of the 150 PAM50 intrinsic subtype assay15 was available for 108 cases with a ctDNA-based signatures (adjusted by TF) (Supplementary Data 6). TF ≥3% in plasma. Tissue samples were obtained at various time- To summarize these results, we plotted in an unsupervised cluster points. To further explore the association between ctDNA analysis, the correlation coefficients of the most correlated Nature Communications | (2023)14:1157 4 Article https://doi.org/10.1038/s41467-023-36801-9 a b h Endocrine therapy + p=0.008 CDK4/6 inhibitor p=0.140 RB−LOH signature score 4 n=124 n=124 HR+/HER2-neg 2 c RB-LOH signature score RB-LOH signature score 0 1.0 1.0 Log Rank p= 0.007 PFS probability 0.8 0.8 OS probability −2 0.6 0.6 CR/PR SD PD 0.4 0.4 0.2 0.2 Low Low Log Rank p= 0.032 Medium Medium e 0.0 0.0 High High Features of the RB-LOH signature 0 5 10 15 20 25 30 0 5 10 15 20 25 30 8q24.21 (MYC) 5.13 0.16 0.26 ctDNA Time (months) Time (months) 21q22.2 4.05 0.40 0.21 average signal d 21q wholearm 3.81 0.17 0.19 6 4 3q26.2 (PRKCI) 3.03 0.15 0.19 2 0 RB-LOH signature score 10p14.68 2.68 0.22 -0.24 -2 -4 Plasma -- univariate 10p13.69 2.55 0.30 -0.36 3q wholearm 2.51 0.30 0.19 DNA Plasma adjusted by TF segment weights Plasma adjusted by PAM50 3q28 (PIK3CA) 2.38 0.07 0.24 0.4 chr20 2.18 0.2 Plasma adjusted by TF+PAM50+clinical 0.33 0.19 0.0 Tissue -- univariate 4p15.31 -1.58 -0.36 -0.29 -0.2 -0.4 Tissue adjusted by PAM50 13q14.2 (RB1) -1.77 -0.40 -0.45 Plasma adjusted by tissue 4p15.2 -1.80 -0.21 -0.28 Mean change Tissue adjusted by plasma post-CDK 2q33.2 (PARD3B) -1.94 -0.35 -0.30 0.3 2q22.1 (LRP1B) -2.17 -0.33 -0.21 0.1 -0.1 0.63 1.00 1.58 2.51 3.98 0.50 0.79 1.26 2.00 3.16 19p13.3 -2.53 -0.35 0.17 -0.3 -0.5 PFS Hazard Ratio OS Hazard Ratio 5q13.2 -3.40 -0.21 0.02 f RB-LOH signature score 0.0 LumA signature score −1 Tumor fraction (TF) 13q14.2 RB1 locus 0.5 8 −2 0.4 −0.5 −3 6 −4 0.3 4 −5 −1.0 0.2 −6 2 0.1 −7 −1.5 p=0.035 p=0.022 p=0.010 p=0.512 Baseline Post-CDK Baseline Post-CDK Baseline Post-CDK Baseline Post-CDK Patient 1 Patient 2 Patient 3 Patient 4 Patient 5 Patient 6 Patient 7 signatures across 58 matched-timepoint paired plasma-tissue cases with a TF ≥3% in plasma. Correlation coefficients of the mRNA (Fig. 3c). In general, ctDNA-based signatures were positively and expression of each individual gene with each 150 ctDNA-based sig- negatively correlated with the known biology that each nature scores were also determined. Like the PAM50 RNA tumor PAM50 subtype signature is hypothesized to be tracking. For signatures, the mRNA expression of luminal genes (e.g., ESR1 and example, a ctDNA-based signature tracking RB-LOH gene expres- GATA3) was positively correlated with luminal ctDNA-based sig- sion signature16, which is enriched in E2F target genes and tracks natures, while proliferation and cell cycle-related genes by mRNA tumor proliferation rates (Supplementary Fig. 4), was found posi- (e.g., MKI67, AURKA, TTK, E2F1 and CCNE1) were positively corre- tively correlated to the PAM50 RNA-based Basal-like, HER2- lated with proliferation-related ctDNA-based signatures (Supple- enriched and proliferation tumor signatures, and negatively corre- mentary Fig. 5). Overall, these findings confirm that DNA-copy lated to the PAM50 RNA-based Luminal A tumor signature (Fig. 2e). number-based signatures coming from ctDNA can track the main Beyond PAM50, we evaluated the expression of 771 genes in breast cancer phenotypes and their known gene expression tumors using the nCounter Breast Cancer 360 Panel in 107 cases features. Nature Communications | (2023)14:1157 5 Article https://doi.org/10.1038/s41467-023-36801-9 Fig. 3 | The ctDNA-based RB-LOH signature predicts clinical outcome in plasma when adjusted for TF (n = 87), in plasma when adjusted for PAM50 RNA- advanced HR+/HER2- breast cancer treated with endocrine therapy and a based subtypes (n = 53), in plasma when adjusted for TF + PAM50 + clinical vari- CDK4/6 inhibitor. a A plasma sample was obtained from 124 patients within 48 h ables (n = 53), in tissue alone (i.e., tissue–univariate; n = 63) and in plasma when prior to starting endocrine therapy and CDK4/6 inhibition (CDK-Validation-1 adjusted for tissue and vice-versa (n = 28). Data are presented as the hazard ratios cohort). ctDNA-based signatures were applied in plasma samples with a TF ≥3% (HR) with error bars showing 95% confidence intervals. e Average ctDNA signal of 16 (n = 87). b Boxplots of RB-LOH ctDNA-based signature score in patients with features of the RB-LOH DNA-based signature (column on the left), weight and complete or partial response (CR/PR, n = 26), stable disease (SD, n = 32) and pro- direction of each feature (column in the middle) in the original signature as gressive disease (PD, n = 18). For the boxplot, center line indicates median; box reported in Xia et al.10 and mean change of the 16 features (column on the right) in 7 limits indicate upper and lower quartiles; whiskers indicate 1.5× interquartile range. patients with paired plasma samples (baseline pre-treatment vs post-CDK4/6 P-values (p) were determined by two-tailed unpaired t-tests. c Kaplan-Meier curves inhibitor treatment). f ctDNA-based signature scores of the RB-LOH signature, the of PFS (left) and OS (right) of the RB-LOH ctDNA-based signature. Each patient Luminal A signature, the 13q14.2 RB1 locus, and TF across 7 patients with paired group is based on tertiles. P-values (p) were determined by Log Rank Test. d Forest plasma samples (baseline vs post-CDK4/6 inhibitor treatment). P-values (p) were plots of hazard ratios (HRs) for PFS (left) and OS (right) of the RB-LOH DNA-based determined by two-tailed paired t-tests. Source data are provided as a Source signature when evaluated in plasma alone (i.e., plasma–univariate; n = 87), in Data file. ctDNA-based signatures versus tissue ER and HER2 status therapy and a CDK4/6 inhibitor in the first-line setting (Supple- ER expression by immunohistochemistry (IHC), and HER2 over- mentary Data 7 and Supplementary Fig. 5). expression by IHC and/or amplification by in-situ hybridization, are A high score of a ctDNA-signature tracking RB-LOH(19) was key biological features of breast cancer4,17. To evaluate the relationship found one of the top biomarkers associated with poor outcome and between ctDNA-based information and ER or HER2 tumor clinical treatment response (Fig. 3b, c). Since loss of RB is a known biomarker status, we evaluated the association of each of the 150 mechanism of resistance to CDK4/6 inhibitors21–25, we decided to adjusted ctDNA-based signatures with either ER clinical status (i.e., focus on this signature, which is composed of 224 copy number positive versus negative) or HER2 status (i.e., positive versus negative) features, including amplification of 2p (e.g., ETV6), 3q (e.g., PIK3CA), across 177 of 178 samples with a TF ≥3% with known ER and HER2 IHC 8q (e.g., MYC), 20q (e.g., AURKA) and 21q (e.g., TMPRSS2 and ERG), status from tumor tissue. As expected, ctDNA-based signatures and deletion of 2q (e.g., PARD3B), 4q, 5q, 12q, 13q (e.g., RB1), 15q and tracking luminal biological processes (e.g., luminal_cluster_signature18 17p. Interestingly, multivariable analysis showed that the associa- and GSEA_Median_GP7_Estrogen_signaling19) were found enriched in tion of the ctDNA RB-LOH signature with PFS and OS was indepen- ER-positive (ER+) disease (p < 0.001; false discovery rate [FDR] < 1%; dent of TF (as a continuous variable), type of CDK4/6 inhibitor, line highest AUC = 0.77) compared to ER-negative disease. Similarly, of treatment (first-line versus second-line versus later lines), pre- ctDNA-based signatures tracking HER2 expression or amplification sence of visceral disease and number of metastatic sites (Fig. 3d). (e.g., HER2-signature15 and HER2-amplified-HER2-amplicon18) were Finally, the direction (i.e., amplification or deletion) and strength found significantly enriched (p < 0.001; FDR < 1%; highest AUC = 0.72) (i.e., coefficient) of the 48 main features of the original tissue-based in HER2+ disease compared to HER2- disease (Fig. 2f and Supplemen- DNA RB-LOH signature10 were properly detected in ctDNA (corre- tary Fig. 6). Overall, these results suggest that ctDNA-based profiling lation coefficient between the original coefficient weights of the 48 captures and predicts specific phenotypic tumor traits. main DNA segments of the RB-LOH signature and the actual ctDNA signals measured in plasma = 0.72, p < 0.001; Fig. 3e and Supple- ctDNA-based signatures in metastatic HR+/HER2- disease trea- mentary Fig. 10). ted with endocrine therapy and a CDK4/6 inhibitor To evaluate the association of each ctDNA-based signatures with RB-LOH in tumor tissue versus plasma in metastatic HR+/HER2- patient’s outcome, we analyzed baseline pre-treatment plasma disease treated with endocrine therapy and a CDK4/6 inhibitor samples from 124 patients with advanced HR+/HER2- breast cancer We analyzed tissue samples of 63 patients with advanced HR+/HER2- treated with endocrine therapy and a CDK4/6 inhibitor (CDK-Vali- breast cancer treated with endocrine therapy and a CDK4/6 inhi- dation-1 cohort). Eighty-seven plasma samples had a TF ≥3% (Fig. 1b bitor. RB-LOH signature determined in tumor tissue DNA was sig- and Fig. 3a). The median follow-up was 12.5 months (range 1.0 to nificantly associated with PFS and OS (Fig. 3d and Supplementary 56.7 months), and most patients were defined as endocrine- Fig. 11). Twenty-eight patients had paired tumor tissue and plasma sensitive (83.9%) and treated in the first-line setting (59.8%) samples, these were interrogated to compare the ability of the (Table 2). TF was associated with poorer progression-free survival tissue-based signatures versus the ctDNA-based signatures to better (PFS) but was not associated with overall survival (OS) (Supple- predict PFS and OS. Of the 150 DNA-based signatures, 17% and 13% mentary Fig. 7). Among the different clinical variables, treatment were statistically significantly associated with PFS and OS, respec- line (i.e., first versus second/third) and bone-only disease were tively, when evaluated in plasma. When the same signatures were associated with better PFS, while ECOG performance status was the evaluated in tissue, only 2% and 1% were statistically significantly only variable associated with OS (Supplementary Fig. 8). associated with PFS and OS, respectively (Supplementary Data 8). From the 150 ctDNA-based signatures (adjusted by TF), 36 Thus, ctDNA-based signatures were better in predicting survival (24%) and 37 (25%) were found significantly associated with PFS and outcomes in advanced breast cancer treated with endocrine ther- OS, respectively (Supplementary Data 7). Twenty-seven (18%) sig- apy and CDK4/6 inhibition than the same signatures when evaluated natures were found significantly associated with both PFS and OS. In in tissue, suggesting that plasma-based signatures capture “the general, signatures associated with poor survival outcome were most up to date” biological state of the disease before starting those hypothesized to be tracking proliferation and non-ER+/non- therapy. Furthermore, when the RB-LOH signature was evaluated luminal-related biological processes, such as the MM_p53null.- head-to-head in the 28 patients with paired tumor tissue and plasma Luminal (i.e., TP53-deficient) and MM_Myc signatures (i.e., high samples in a bivariate cox model, the RB-LOH ctDNA plasma sig- c-MYC amplification)20. Conversely, ctDNA-signatures associated nature was found significantly associated with PFS, but not the RB- with better outcome were tracking luminal A-related biological LOH tumor signature (Fig. 3d). Overall, baseline pre-treatment processes. Similar results were obtained when the analysis was ctDNA better captured the prognosis of patients than archival performed in the subset of 52 patients treated with endocrine tumor tissue DNA. Nature Communications | (2023)14:1157 6 Article https://doi.org/10.1038/s41467-023-36801-9 Table 2 | Baseline clinical characteristics of patients with HR+/HER2- advanced disease treated with endocrine therapy and a CDK4/6 inhibitor Cohort CDK-Validation-1 CDK-Validation-2 All patients Patients with TF ≥3% All patients Patients with TF ≥3% N % N % N % N % Number of patients 124 87 121 65 Median age (range) 61 (34-86) 62 (34-86) 60 (31-88) 60 (41-83) Sex Female 122 98.4% 87 100.0% 121 100% 121 100% Male 2 1.6% 0 0.0% 0 0% 0 0% CDK4/6 inhibitor Palbociclib 70 56.5% 48 55.2% 48 39.7% 22 33.8% Ribociclib 50 40.3% 36 41.4% 52 43.0% 26 40.0% Abemaciclib 4 3.2% 3 3.5% 21 17.4% 17 26.2% Setting in advanced disease 1st line 71 57.3% 52 59.8% 72 59.5% 37 56.9% 2nd line 27 21.8% 15 17.2% 18 14.9% 10 15.4% ≥3rd line 26 21.0% 20 23.0% 31 25.6% 18 27.7% Number of metastases <3 74 59.7% 50 57.5% 56 46.3% 44 67.7% ≥3 50 40.3% 37 42.5% 64 52.9% 21 32.3% Unknown 0 0.0% 0 0.0% 1 0.8% 0 0.0% Type of metastasis Visceral metastasis 66 53.2% 34 39.1% 67 55.4% 38 58.5% De novo metastasis 30 24.2% 19 21.8% 34 28.1% 22 33.9% Bone only 12 9.7% 0 0.0% 20 16.5% 12 18.5% Prior endocrine therapy sensitivity Sensitive 103 83.1% 73 83.9% 94 77.7% 48 73.8% Resistant 19 15.3% 14 16.1% 26 21.5% 17 26.2% Unknown 2 1.6% 0 0.00% 1 0.8% 0 0.0% ECOG Performance status 0 50 40.3% 38 43.7% 62 51.2% 33 50.8% 1 61 49.2% 39 44.8% 49 40.0% 27 41.5% 2 12 9.7% 10 11.5% 5 4.1% 4 6.2% Unknown 1 0.8% 0 0.0% 5 4.1% 1 1.5% *Previous systemic treatments are available for the CDK-Validation-1 cohort (Supplementary Data 17). ctDNA RB-LOH signature versus ctDNA RB1 individual region in independent cohort of 357 plasma samples, hereafter Plasma-2 metastatic HR+/HER2- disease treated with endocrine therapy cohort, which included 121 patients with advanced HR+/HER2- and a CDK4/6 inhibitor breast cancer treated with endocrine therapy and a CDK4/6 inhi- The DNA-based RB-LOH signature considers the signal of the RB1 locus bitor at Hospital Clínic of Barcelona and the Vall d’Hebron Institute (13q14.2) among 224 other features10 (Fig. 3e). The correlation coeffi- of Oncology (VHIO) and assayed using our ctDNA WGS-assay. cient between the ctDNA signal of 13q14.2 and the ctDNA RB-LOH sig- Baseline characteristics of the 121 patients of CDK-Validation-2 nature score was −0.12 across the 178 samples with TF ≥3%, reinforcing cohort are reported in Table 2. Sixty-five patients (54.0%) had a the concept that the ctDNA RB-LOH signature is different from mea- TF ≥3%. With a median follow-up of 12.3 months, the RB-LOH sig- suring only the individual 13q14.2 region. Indeed, the ctDNA signal of nature was significantly associated with worse PFS when evaluated the individual 13q14.2 segment was not significantly associated with PFS as a continuous variable (hazard ratio=1.42 [1.03–1.96], p = 0.032), (p = 0.061) but was significantly associated with better OS (p = 0.020), and using the previously defined cutoffs (i.e., tertiles) derived from while the RB-LOH signature was strongly associated with PFS and OS in the CDK-Validation-1 cohort (Supplementary Fig. 13a, b). Finally, the patients with advanced HR+/HER2-negative breast cancer treated with prognostic value of RB-LOH signature was confirmed in the com- CDK4/6 inhibitors. In a bivariate cox model with both variables, only the bined CDK-Validation-1 and CDK-Validation-2 cohorts (n = 152) ctDNA RB-LOH signature was prognostic. Overall, the RB-LOH ctDNA- (Supplementary Fig. 13c, d). Moreover, we evaluated if the ctDNA- based signature better captured the clinical behavior than an individual based RB-LOH was significantly associated with PFS and OS in 71 DNA region looking only at RB1, thus highlighting the power of a multi- patients with a TF 3–10% before starting endocrine therapy in feature algorithm for sensing pathway activity (Supplementary Fig. 12). combination with a CDK4/6 inhibitor. Like the overall population, we observed a statistically significant association with both clinical Independent prognostic validation of the RB-LOH signature endpoints (PFS: hazard ratio = 1.32, p = 0.023; OS: Hazard ratio = To further validate the prognostic value of the RB-LOH signature, 1.54, p = 0.011), suggesting that our approach can work in patients we evaluated pre-treatment baseline plasma samples from a second with low TF (3–10%). Nature Communications | (2023)14:1157 7 Article https://doi.org/10.1038/s41467-023-36801-9 To provide more evidence of the prognostic value of the RB-LOH Visually, cluster 2 could be further subdivided (minimum 20 samples signature, we evaluated publicly available data from an independent and a correlation coefficient >0.75) into Cluster2A and Cluster2B, both cohort of 381 patients with advanced HR+/HER2- breast cancer treated of which showed differences in the enrichment of ctDNA-based pro- with endocrine therapy and a CDK4/6 inhibitor at the Memorial Sloan liferation features, and luminal A-related signatures. Consistent with Kettering Cancer Center (MSK)26. RB-LOH signature scores were the luminal A-related biology identified in Custer 2 A, this group was obtained from next-generation-sequencing data of tumor tissue DNA characterized by 16p amplification and 16q deletion (Supplementary profiled under the MSK-IMPACT platform, which includes copy- Data 10 and Fig. S18), both of which are known features of low-grade number data from 468 cancer-associated genes27. With a median and low-proliferative breast cancers28–30. Finally, Cluster 1 showed low follow-up of 71.0 months, the RB-LOH signature was significantly enrichment of proliferation and luminal B-related signatures and high associated with PFS (hazard ratio = 1.41, 95% CI 1.21–1.64, p < 0.001) enrichment of luminal A subtype related signatures. Plasma TF in (Supplementary Fig. 14a, b). Similar results were obtained in the subset Cluster 1 was significantly lower compared to the other clusters com- of 223 patients where the MSK-IMPACT assay was performed in biop- bined (average 6.8% vs. 9.8%), p < 0.001; Fig. 4a and Supplementary sies taken within 1 year of starting therapy (PFS hazard ratio = 1.61, 95% Fig. 19, which might be predicted for slow growing luminal A tumors. CI 1.32–1.96, p < 0.001) (Supplementary Fig. 14c). However, RB-LOH RNA-based expression data from FFPE tissue using a research- was not prognostic when assessed in 158 biopsies taken more than 1 based PAM50 intrinsic subtype assay15 was available for 108 cases with year before starting therapy with a CDK4/6 inhibitor (Supplementary a TF ≥3% in plasma. Tissue samples were obtained at various time- Fig. 14d). points. As expected, Cluster 3 was enriched for tumors with a PAM50 non-luminal subtype (i.e., HER2-enriched or Basal-like) compared to Capturing biological features before and after CDK4/6 the other clusters (85.7% versus 25%, p < 0.001) (Supplementary inhibition Fig. 20). Concordant with this finding, the PAM50 Luminal A and HER2- Scores from the 150 ctDNA-based signatures (adjusted by TF) were enriched signatures (as a continuous variable) were found differen- evaluated in paired plasma samples (i.e., baseline versus post-treatment tially expressed in Cluster 3 versus the other clusters (Fig. 4b). In after progressive disease) in 7 patients with advanced HR+/HER2- addition, we observed that Cluster 2B was enriched for PAM50 Luminal breast cancer treated with endocrine therapy and a CDK4/6 inhibitor B tumors compared to Cluster 2A (53.3% versus 18.8%, p = 0.044). (Fig. 3f). Among them, 103 signatures (68.7%) were found differentially enriched between the two time-points (FDR < 5%). As might be expec- Prognosis of ctDNA-based tumor subtypes in patients with ted, enrichment of signatures tracking non-luminal/proliferation-rela- metastatic HR+/HER2- disease treated with endocrine therapy ted biological processes (e.g., RB-LOH) and luminal A-related biological and a CDK4/6 inhibitor processes were found significantly increased and decreased, respec- This work demonstrates that tumor profiles identify samples with tively, in post-treatment samples compared to pre-treatment samples similar patterns of expression, and these patterns are associated with (Fig. 3f and Supplementary Data 9). Of note, TF did not significantly clinically relevant genotypes. We then hypothesized that ctDNA- change between the two timepoints across the 7 patients (Fig. 3f), and 1 defined subtypes, representing repeatably observed combinations of patient with a substantial decrease in TF still showed an increase in the these patterns, may explain variation in clinical outcomes. Regarding RB-LOH score and a decrease of the luminal A signature. Importantly, the relationship between clusters 1–4 and treatment response, we the signal from 12 of the top 16 DNA segments (75%) of the RB-LOH observed an overall response rate (ORR) of 52.7%, 34.0%, 7.1% and signature changed in the expected direction in post-treatment samples 16.7% in Cluster 1, Cluster 2, Cluster 3 and Cluster 4, respectively compared to pre-treatment samples (Fig. 3f), providing evidence of our (p < 0.001). Regarding the relationship between cluster 1–4 and prior approach providing additional information beyond changes of TF and endocrine sensitivity, this was observed in 91.7%, 80.0%, 57.1%, and signals from individual DNA regions. These biological changes identi- 55.6% in Cluster 1, Cluster 2, Cluster 3 and Cluster 4, respectively fied in ctDNA are concordant with similar biological changes identified (p = 0.004) (Fig. 4c). across 18 patients with paired tumor-based RNA expression before and Next, we evaluated the prognostic value of the 4 ctDNA-based at progression to endocrine therapy and a CDK4/6 inhibitor (Supple- tumor groups in the CDK-Validation 1 and 2 cohorts combined mental Fig. S15). Specifically, PAM50 Luminal A and proliferation sig- (n = 152). Compared to Cluster 1 (median PFS = 18.3 months), Clusters natures were found significantly decreased and increased, respectively, 2, 3 and 4 were found significantly associated with worse PFS (median in tumor tissue progressing to endocrine therapy and CDK4/6 PFS of 10.2, 2.9, and 6.4 months, respectively). Median OS was not inhibition. reached for Cluster 1 and 2, while median OS were 16.8 and 17.8 months for Clusters 3 and 4, respectively. Cluster 3, which has the highest Identifying ctDNA-based tumor subtypes ctDNA-based RB-LOH signature score (p < 0.001; Supplemental Phenotype-based classification in metastatic breast cancer using Fig. S21), was significantly associated with worse PFS and OS compared tumor tissue RNA expression profiling such as intrinsic subtyping (i.e., to Cluster 1 (PFS hazard ratio = 4.76, 2.29–9.91, p < 0.001; OS hazard Luminal A, Luminal B, HER2-enriched and Basal-like) is prognostic and ratio = 5.21, 1.68–16.19, p < 0.001) (Fig. 4d). might predict treatment benefit5–7. To evaluate if the biology displayed by the 150 ctDNA-based signatures (adjusted by TF) can identify sub- DNA-based tumor subtypes in tissue samples types with clinical relevance, we performed an unsupervised hier- To explore how tumor subtypes identified in ctDNA data perform archical cluster analysis of all 150 signatures across 178 plasma samples when applied to tumor tissue DNA, the scores of the 150 DNA-based with a TF ≥3% (Fig. 4a). Four main clusters/groups of samples were signatures were determined and evaluated using tumor DNAs from identified using consensus clustering plus (Supplementary Fig. 16) and 1,689 patients with early-stage breast cancer from the publicly avail- then validated in an independent cohort of 357 plasma samples able METABRIC dataset29, and 381 patients with advanced HR+/HER2- (Plasma-2 cohort), including 193 with a TF ≥3% (Supplementary Fig. 17). breast cancer from the MSK-IMPACT dataset (Fig. 5a and Supplemental Clusters 3 and 4 showed high scores of ctDNA-based proliferation- Fig. 22a). We developed a 4-class subtype classifier from the ctDNA related signatures and low scores of differentiation status and of groups identified in Plasma-1 cohort (Fig. 4a) and applied this predictor luminal A-related signatures. Compared to Cluster 4, Cluster 3 showed onto METABRIC’s and MSK´s tumor DNA data. Overall, the 4 main high expression of basal-like-related gene expression (p < 0.001). clusters were identified in both cohorts (Fig. 5a and Supplemental Cluster 2 showed high enrichment of differentiation and luminal Fig. 22a), and these clusters showed the expected DNA signal across B-related signatures, and low enrichment of basal-like related biology. chromosomes (Fig. 5b). Concordant with this finding, the PAM50 non- Nature Communications | (2023)14:1157 8 Article https://doi.org/10.1038/s41467-023-36801-9 Cluster-2 IHC subtype a Cluster-1 Cluster-3 HR+/HER2- Cluster-4 HER2+ TNBC N/A IHC subtype PAM50/RNA subtype PAM50 Basal-like HER2-enriched 150 ctDNA-based signatures (scores) Luminal A-related Luminal A Luminal B Normal-like Luminal B-related N/A Differentiation-related ctDNA signature score 3.0 1.5 0.0 -1.5 -3.0 Basal-related Tumor fraction 90% 75% 60% 45% 30% Proliferation-related 15% 3% Tumor fraction 1 2A 2B 3 4 CLUSTERS b c p<0.001 p=0.004 Prior endocrine sensitivity (%) RNA-based tissue signatures 100 100 Best response (%) 80 PAM50 HER2-enriched score 0.6 p<0.001 80 0.4 PAM50 Luminal A score 0.4 60 60 0.2 0.2 40 40 0.0 0.0 20 20 −0.2 −0.2 0 0 lu 1 lu 2 lu 3 4 lu 1 lu 2 lu 3 4 −0.4 −0.4 C ter C ter C ter er C ter C ter C ter er st st s s s s s s lu lu −0.6 p=0.006 C Cluster 3 Other clusters Cluster 3 Other clusters CR/PR SD/PD C sensitive resistant d 1.00 1.00 ||| || || || | | | | ||| | | ||||||| | || | | | || ||| | | || | | || | ||| Cluster 3 vs Cluster 1 | | || || | | | | | | | | || || | | | 0.75 0.75 | PFS probability | || | OS probability || HR (95%CI)=4.76 (2.29-9.91) | | || | | | | | ||||| | p<0.001 || || || || | | | | | ||| | 0.50 0.50 | | | || | | | | | | | Cluster 3 vs Cluster 1 || || | | HR (95%CI)=5.21 (1.68-16.19) | p=0.004 0.25 0.25 | | | | | | | | | Cluster 1 | | Cluster 1 Cluster 2 Cluster 2 Cluster 3 Cluster 3 0.00 0.00 Log Rank p<0.001 Log Rank p<0.001 Cluster 4 Cluster 4 0 5 10 15 20 25 30 0 5 10 15 20 25 30 Time (months) Time (months) luminal subtypes in the METABRIC cohort were enriched in Cluster 3 patients with advanced HR+/HER2- breast cancer in the MSKCC dataset compared to the other clusters (79.6% versus 11.6%, p < 0.001) (Fig. 5c). (Supplemental Fig. 22b). Similar results were obtained in the subset of In addition, TP53 somatic mutations in METABRIC were also enriched 223 patients where the MSK-IMPACT assay was performed in tumor in Cluster 3 compared to the other clusters (80.7% versus 51.2%, p < tissue biopsies obtained within 1 year of initiating therapy (Supple- 0.001) (Fig. 5d). Finally, the 4 main clusters were found significantly mental Fig. 22c, d). associated with disease-free survival (DFS) and OS in all patients, and in Finally, to understand whether the identification of the 4 clusters/ patients with early-stage HR+/HER2- breast cancer (Fig. 5e), and PFS in subtypes is independent of the tissue profiled (i.e., primary versus Nature Communications | (2023)14:1157 9 Article https://doi.org/10.1038/s41467-023-36801-9 Fig. 4 | ctDNA-based profiling of metastatic breast cancer. a Unsupervised tissue samples were obtained at different timepoints prior to obtaining the plasma cluster analysis of 178 plasma samples with a TF ≥3% (columns) and the scores of sample. P-values (p) were determined by two-tailed unpaired t-tests. For the box- 150 ctDNA-based signatures (rows). Orange and violet colors represent scores plot, center line indicates median; box limits indicate upper and lower quartiles; above and below the median score of the signature across the dataset. Below the whiskers indicate 1.5× interquartile range. c Association of ctDNA-based clusters array tree, the IHC subtype and the PAM50 molecular subtype are shown for each with response to endocrine therapy and CDK4/6 inhibition (p < 0.001) and prior sample. Four clusters of samples (clusters 1 to 4) were identified. Within cluster 2, endocrine sensitivity (p < 0.001) in 152 patients from the combined CDK-Validation- two subgroups of samples were also identified (clusters 2A and 2B). b Expression of 1 and CDK-Validation-2 cohorts. P-values (p) were determined by Fisher’s exact test. 2 tissue PAM50 RNA-based signatures (i.e., Luminal A [p = 0.0005] and HER2- d Kaplan-Meier curves of PFS (Log Rank p < 0.001; Cluster 3 vs Cluster 1 p < 0.001) enriched [p = 0.006]) in cluster 3 versus the other clusters. This analysis was per- (left) and OS (Log Rank p < 0.001; Cluster 3 vs. Cluster 1 p = 0.004) (right) of the 4 formed in 107 paired plasma and tumor tissue samples. Of note, 58 tumor tissue ctDNA-based clusters in 152 patients of the combined CDK-Validation-1 and CDK- samples were obtained at the same timepoint as the plasma sample and 49 tumor Validation-2 cohorts. Source data are provided as a Source Data file. metastasis), we analyzed the 150 DNA-based signatures across 158 The RB-LOH mRNA-signature was originally developed by com- tumor tissues (either primary tumors or metastatic tumors). The pro- paring gene expression between DNA assessed RB-LOH-positive breast portion of primary tumors falling in Clusters 1, 2, 3 and 4 was 47.6%, tumors versus LOH-normal tumors16. Expression of 452 genes varied 27.0%, 11.1% and 14.3%, respectively. The proportion of metastatic with RB-LOH status and 423 were highly expressed in tumors with RB- tumors falling in Clusters 1, 2, 3 and 4 was 27.0%, 39.3%, 19.1% and LOH. These genes were associated with cell cycle-related biological 14.6%, respectively (Supplementary Fig. 23). Thus, an enrichment for processes, including cell division, DNA metabolism, spindle organiza- cluster 1 (better prognosis) was observed in primary tumors compared tion and biogenesis and response to DNA damage, and many were to metastatic tumors (p = 0.010). This enrichment for a more aggres- known E2F-regualted genes16. If one closely examines the selected sive tumor biology in the metastatic setting is consistent with previous features of the DNA-copy number based RB-LOH signature, it includes studies31,32. 10s of features that encompass many chromosomal regions known to harbor unambiguous acknowledged players of the cell cycle including Discussion RB1, E2F1,3, CCND1,2,3, CDK6 and MYC; thus an important quality of Through in-silico supervised learning analysis of The Cancer Genome these multi-feature Elastic Net predictors is that they have been trained Atlas (TCGA) early-stage breast cancer dataset we previously identified to ‘sense’ many parts of a given pathway and to combine these into one 150 DNA copy-number-based multi-feature predictors able to track a objective machine learning trained predictor, which links these toge- variety of biological processes at the RNA and protein level10. Here, we ther into a single quantitative score. As expected, the frequency of the addressed if our approach could be applicable in liquid biopsies (i.e., RNA-based RB-LOH signature varied by intrinsic subtype with the plasma ctDNA). In this report, we predict complex tumor phenotypes lowest and highest LOH frequencies observed in the Luminal A (15%) like tumor cell proliferation rates using blood samples from patients and Basal-like (72%) subtypes, respectively, concordant with the with metastatic cancer. Our discovery should lead to opportunities for known lack of RB function in Basal-like tumors and thus a lack of biomarker discovery, validation, and implementation in oncology. benefit of CDK4/6 inhibitors biology5. In metastatic breast cancer, phenotype-based classification using The ability to detect and monitor tumor phenotypes in blood tumor tissue RNA expression profiling (i.e., Luminal A, Luminal B, samples opens opportunities for precision oncology. On one hand, HER2-enriched and Basal-like) is prognostic and might predict tumor tissue biopsies of metastatic disease are difficult to perform, treatment benefit5–7. Specifically, both non-luminal subtypes within pose risks to the patient and do not fully capture the intra-patient HR+/HER2- disease have been associated with a poor outcome in the tumor heterogeneity36. On the other hand, tumor biology evolves over context of endocrine-based treatment. In addition, the Basal-like sub- time and the biology captured in a tissue sample might not fully type within HR+ disease is associated with a lack of benefit from recapitulate the biology present at the latter times decisions are often endocrine therapy and CDK4/6 inhibition5, a finding highly concordant made in the metastatic setting. For example, ∼1/3 of primary breast with ours. Another example within metastatic HER2+ disease is the tumors switch molecular subtype when relapse occurs31,32,37, a pro- result of a phase II PATRICIA trial (NCT02448420), where the benefit of portion that increases to ∼50% if the primary tumor was Luminal A. In endocrine therapy and palbociclib seemed restricted to ∼50% of addition, large relative changes in tissue gene expression have been patients with a Luminal A or B subtype33. Finally, in advanced TNBC, the found between primary and metastatic breast cancer. For example, 47 identification of the Basal-like versus non-Basal-like subtypes revealed of 105 breast cancer-related genes (45%) were found differentially significant differences in response rates to carboplatin versus doc- expressed across 123 paired primary-metastatic tumors31. Interest- etaxel in non-Basal-like disease in the TNT phase III trial9. Based on ingly, expression of proliferation-related genes was better at predict- these findings, several prospective clinical trials are currently selecting ing OS in metastatic disease when analyzed in metastatic tissue rather patients based on their tumor’s RNA-based phenotype (e.g., TATEN than primary tissue31. This is concordant with our findings that the RB- (NCT04251169), PATRICIA II (NCT02448420), NEREA (NCT04460430) LOH ctDNA-signature is better associated with prognosis when mea- and HARMONIA (NCT05207709)). sured in ctDNA prior to initiating therapy compared to the same sig- Among the different phenotypic features identified in ctDNA, the nature evaluated in archival tissue samples coming from the primary RB-LOH signature16 was found as one of the top signatures associated tumor, which typically predate the metastatic specimen by many years. with survival outcomes in patients with advanced HR+/HER2- breast Our findings should facilitate the development of prospective cancer treated with endocrine therapy in combination with a CDK4/6 trials in specific clinical scenarios selecting patients with a more inhibitor. CDK4/6 kinases associate with cyclin D proteins during homogenous tumor biology. For example, patients with HR+/HER2- transition from G1 to S phase of the cell cycle21. The cyclin D/CDK4/6 advanced breast cancer and a high score in the ctDNA-based RB-LOH complex phosphorylates RB, dissociating it from the E2F transcription signature could be randomized to a chemotherapy-based treatment factors, which are ultimately responsible for cell cycle progression34. In strategy versus the standard-of-care of endocrine therapy and CDK4/6 this scenario, loss of RB is an evident cause of resistance to CDK4/6 inhibition. In addition, tracking RB-LOH and other ctDNA-signatures inhibitors21, and various preclinical studies support this hypothesis22,23. might better help to understand why tumors progress and make In addition, mutations in RB, which are rare, are responsible for resis- treatment decisions based on that information. For example, a trial tance to CDK4/6 inhibition in a few patients24,25,35. could evaluate endocrine therapy alone in patients with HR+/HER2- Nature Communications | (2023)14:1157 10 Article https://doi.org/10.1038/s41467-023-36801-9 a b IntClust 11p 19p PAM50 14q Cluster 1 IntClust 3p Cluster 2 Luminal A-related 10q 1 15q Cluster 3 2 3 4p Cluster 4 4ER− 5q 4ER+ 16p 20p Luminal B-related 5 1q Tumor tissue 6 5p DNA signal 150 DNA-based signatures (scores) 7 8 12q 17q 3.00 9 2.00 10 20q 4q 1.00 8q 0.00 PAM50/RNA subtype -1.00 11q Basal-like 2q -2.00 HER2-enriched 18p -3.00 Luminal A 9p Luminal B 1p 22q Basal-like-related 13q DNA signature score 17p 18q 1.50 8p 0.75 9q 0.00 16q -0.75 6p -1.50 10p Proliferation-related 21q 2p 6q 19q 7q 3q 3 4 1 2A 2B CLUSTERS c 120 e All tumors HR+/HER2- tumors 1.0 1.0 100 Log Rank p<0.001 Log Rank p<0.001 PAM50 subtype (%) 0.8 0.8 DFS probability DFS probability 80 0.6 0.6 60 Basal-like HER2-enriched Cluster 1 vs Cluster 3 Cluster 1 vs Cluster 3 0.4 0.4 40 Luminal A HR (95%CI)=1.817 (1.4361-2.299) HR (95%CI)=2.351 (1.592-3.472) Luminal B p<0.001 0.2 p<0.001 0.2 20 Cluster 1 Cluster 1 Cluster 2 Cluster 2 Cluster 3 Cluster 3 0.0 0.0 0 Cluster 4 Cluster 4 1 2 3 4 0 20 40 60 80 100 120 0 20 40 60 80 100 120 er er er er st st st st lu lu lu lu Time (months) Time (months) C C C C All tumors HR+/HER2- tumors d 1.0 1.0 120 Log Rank p<0.001 Log Rank p<0.001 0.8 0.8 100 OS probability OS probability TP53 status (%) 0.6 0.6 80 TP53 WT Cluster 1 vs Cluster 3 Cluster 1 vs Cluster 3 60 TP53 mut 0.4 0.4 HR (95%CI)=1.762 (1.451-2.195) HR (95%CI)=2.270 (1.579-3.263) p<0.001 p<0.001 40 0.2 0.2 Cluster 1 Cluster 1 Cluster 2 Cluster 2 20 Cluster 3 Cluster 3 0.0 0.0 Cluster 4 Cluster 4 0 0 20 40 60 80 100 120 0 20 40 60 80 100 120 1 2 3 4 Time (months) Time (months) er er er er st st st st lu lu lu lu C C C C Fig. 5 | DNA-based tumor profiles in tissue samples and association with clinical were determined using a multi-class SAM analysis (FDR < 5%) c PAM50 molecular outcomes. a Unsupervised cluster analysis of 1,689 tumor samples (columns) from subtype distribution across the 4 DNA-based clusters in 1689 breast tumors of the METABRIC dataset29 and the 150 DNA-based signatures scores (rows). Orange and METABRIC dataset29. d TP53 mutation distribution across the 4 DNA-based clusters violet colors represent scores above and below the median value of the signature in 1,517 breast tumors of the METABRIC dataset29. e Kaplan-Meier curves of DFS across the dataset. Below the array tree, the InctClust classification29 and the (left) and OS (right) of the 4 DNA-based clusters assessed in all tumors (n = 1689) PAM50 molecular subtypes are shown for each sample. The predictor of the 4 (DFS Log Rank p = 3.51e-10; Cluster 1 vs 3 p < 0.001; OS Log Rank p = 1.56e−09; clusters, which was derived from plasma ctDNA data in the Plasma-1 cohort (see Cluster 1 vs 3 p < 0.001) and HR+/HER2- tumors (n = 1131) (DFS Log Rank p < 0.001; Fig. 4), was applied in METABRIC and the results for each sample are shown below Cluster 1 vs 3 p < 0.001; OS Log Rank p < 0.001; Cluster 1 vs 3 p < 0.001) of the the data matrix. b Heatmap of signals from individual chromosomic regions found METABRIC database29. Source data are provided as a Source Data file. significantly associated with each cluster 1-4. Differentially represented segments Nature Communications | (2023)14:1157 11 Article https://doi.org/10.1038/s41467-023-36801-9 advanced breast cancer with a high score of the ctDNA-based Luminal treatment blood plasma samples from 124 patients with HR+/HER2- A signature, and then add CDK4/6 inhibition when the ctDNA-based advanced breast cancer treated with endocrine therapy in combina- Luminal B signature increases during therapy versus initiating endo- tion with a CDK4/6 inhibitor (i.e., palbociclib, ribociclib or abemaci- crine therapy and CDK4/6 inhibition upfront. Another clinical situation clib) between the years of 2018 and 2021 (CDK-Validation-1 cohort). All is when one is forced to change treatment after a single metastatic plasma samples were obtained before the start of treatment (Fig. 1b). lesion progresses while the other metastatic lesions are radiologically In 7 patients, we obtained an additional plasma sample after progres- under control and the patient is doing clinically well. In this context, sing while on therapy. detection of a low proliferation status in ctDNA might open opportu- Additionally, we included a second independent cohort of 357 nities such as considering local therapy approaches on the progressing plasma samples of patients with advanced breast cancer (Plasma-2 lesion and continue with the same systemic therapy. cohort) (Fig. 1A), including 121 baseline samples of patients with Our study has limitations. First, ∼39% of patients had a TF < 3%, HR+/HER2- advanced breast cancer treated with endocrine therapy and ctDNA-signatures could not be evaluated. In addition, ∼30% of in combination with a CDK4/6 inhibitor (i.e., palbociclib, ribociclib patients had a TF of 3–10% and this might limit the detection of the or abemaciclib) at Hospital Clinic of Barcelona (n = 50) and Vall ctDNA-based signatures. Further studies evaluating deeper ctDNA d’Hebrón Institute of Oncology (n = 71) (CDK-Validation-2) sequencing strategies and signature detection in patients with very low (Fig. 1b). Only baseline pre-treatment plasma samples were avail- TF, including those with early-stage disease, is warranted. For example, able from this cohort. Moreover, we analyzed plasma samples from expressed genes might be inferred by evaluating nucleosome foot- 14 healthy individuals. prints from whole-genome sequencing of plasma DNA38. Second, this In addition, 185 FFPE tumor tissues from patients treated at Hos- was an exploratory study, and we did not use a sample size calculation pital Clinic of Barcelona were collected, including samples from 110 but we used all the samples available for the correlative analysis. The patients with available plasma samples (in Plasma-1 cohort), and lack of formal design through pre-planned analysis prohibits inference samples from 17 patients with HR+/HER2- advanced breast cancer who of negative results. Third, we did not address if any of the 150 sig- did not have plasma samples but were treated with endocrine therapy natures predict clinical outcome in advanced HER2+ or TNBC or ben- and a CDK4/6 inhibitor. efit from other targeted therapies. Fourth, we decided to focus on Publicly available DNA, RNA and clinical data from METABRIC and signatures previously identified and validated. Fifth, although both the MSKCC datasets were obtained from cBioportal. independent datasets collected plasma samples prospectively from patients treated with endocrine therapy and a CDK4/6 inhibitor, the DNA-sequencing of plasma samples cohorts are not from clinical trials and are prone to potential biases Approximately 30 mL of peripheral blood was collected into K2-EDTA such as patient selection, inconsistent evaluation of the disease during Vacutainer tubes (Becton Dickinson) and plasma isolation was per- therapy and subjectivity in determining drug response and progressive formed within 2 h of blood collection through two centrifugation steps. events. Sixth, we did not compare the prognostic value of the ctDNA- Centrifugation at 1600 x g for 10 min at 4 °C separated plasma from based signatures measured at baseline with the prognostic value of peripheral-blood cells. Approximately 12 mL of plasma were obtained early ctDNA dynamics after 2–4 weeks of initiating therapy11,39. per patient, which were subsequently centrifugated at 16,000 x g for Seventh, ctDNA-signatures might not capture well tumor immune- 10 min at 4 °C to remove the residual supernatant and any remaining related biology as tumor tissue DNA-signatures. contaminants including cells. Plasma samples were then aliquoted in To date, DNA sequencing has identified few FDA-approved 1.5 mL tubes and immediately stored at −80 °C. cfDNA was obtained actionable genetic alterations in cancer, especially breast cancers. from 3 mL of plasma using the QIAamp Circulating Nucleic Acid Kit Phenotypic characterization using multi-gene RNA-based expression (QIAGEN Inc.) according to the manufacturer’s instructions and quan- might add biological and clinically relevant information. However, tified with a Qubit dsDNA high-sensitivity assay kit and the Qubit 4.0 implementation of tumor-based RNA-based gene expression profiling fluorometer (Life Technologies, Carlsbad, CA, USA). cfDNA was con- in the metastatic setting is a major challenge since tumor tissue is often centrated using SpeedVac to fulfil the requirements for library pre- not readily available. Here, we demonstrate that complex tumor phe- paration. Library preparation was performed by ligating unique dual notypic traits can be identified in ctDNA and may provide clinical indexes (UDI) custom adapters to a minimum of 10 ng of the isolated value. Our ctDNA-based multi-gene signature approach opens avenues cfDNA (10–50 ng dsDNA). More specifically, the fragment ends of for discovering and implementing biomarkers in oncology, especially cfDNA were blunted and 5′ phosphorylated and, after that, 3′ ends were in the advanced disease setting where liquid biopsies are becoming A-tailed to favour adapter ligation. Adapters were 10 bp – UDI as commonplace. recommended to mitigate errors introduced by index-hopping or switching in Illumina instruments with patterned flow cells, such as the Methods NovaSeq 6000. Indexed libraries were quantified by qPCR using the The Hospital Clínic Barcelona institutional ethics committee approved KAPA Library Quantification Kit (Roche Sequencing Solutions), pooled, the study in accordance with the principles of Good Clinical Practice, and sequenced in a NovaSeq 6000 Illumina at 0.5X mean coverage with the Declaration of Helsinki, and other applicable local regulations read length of 2 ×150 bp. ShWGS was analyzed with hmmcopy_utils (HCB/2019/0666). Written informed consent was obtained from all (https://github.com/shahcompbio/hmmcopy_utils) and ichorCNA patients before enrolment. The medical records were retrospectively v0.2.0 (https://github.com/broadinstitute/ichorCNA), with a bin size of reviewed to obtain the necessary clinical data. 500 kb and default parameters12. ichorCNA is a previously reported tool by Adalsteinsson et al.12 for estimating the fraction of tumor in cfDNA Study participants and samples from ultra-low-pass WGS. ichorCNA uses a probabilistic model, imple- We collected 246 plasma samples from 207 patients (Plasma-1 cohort) mented as a hidden Markov model, to simultaneously segment the treated at Hospital Clinic of Barcelona (n = 190), Hospital 12 de Octubre genome, predict large-scale copy number alterations, and estimate the in Madrid (n = 15) and Institut Català d’Oncologia (n = 2) at different TF of an ultra-low-pass whole genome sequencing samples. ichorCNA is stages of their metastatic disease: 209 plasmas from 179 patients with optimized for low coverage (~0.1X) sequencing of samples and has advanced HR+/HER2- breast cancer, 19 plasmas from 16 patients with been benchmarked using patient and healthy donor cfDNA samples. HER2+ advanced breast cancer, 17 plasmas from 16 patients with Adalsteinsson et al. reported that, using a TF cutoff of 3%, ichorCNA advanced TNBC and 1 plasma from 1 patient with unknown ER and achieves a sensitivity of 0.95 for detecting presence of tumor and a HER2 status (Fig. 1a). Plasma-1 cohort included 124 baseline pre- specificity of 0.91 for correctly classifying a healthy donor12. Nature Communications | (2023)14:1157 12 Article https://doi.org/10.1038/s41467-023-36801-9 Analytical validation analysis of the methodological approach 10% MAF for INDELs. Frequent single nucleotide polymorphisms To validate the approach, the following analytical analyses were per- (SNPs) in the population were removed based on the gnomAD data- formed. First, since the coverage used in this study was at the higher base (allele frequency ≤0.0001). CNA were calculated from an in-built end of what is considered ultra-low pass WGS by definition (0.1–0.5X genome-wide SNP backbone targeting 20000 SNPs using CNVkit coverage), we performed an in-silico downsampling of sample cover- (v0.9.6.dev). Data was manually curated, and classification of identi- age for 12 samples in order to assess performance at different coverage. fied variants was performed using publicly available databases (COS- Different TFs were represented within these 12 samples. For each MIC, cBioPortal, ClinVar, VarSome, OncoKB). sample, a subset of aligned reads was selected to account for: 2X (40 M reads, 151 nucleotide [nt] long), 1X (20 M reads, 151 nt long), 0.5X (10 M DNA-based signature estimation reads, 151 nt long) and 0.1X (2 M reads, 151 nt long) coverages. Reported For both tumor DNA sequencing and plasma cell-free ctDNA sequen- TFs were consistent across coverages (Supplementary Data 12). cing, segmentation files from CNVkit output (for tumor DNA) and Second, correlation analyses of the bin-to-bin log2 values of these ichorCNA output (ctDNA) were first mapped to gene-level feature. samples indicated a decrease in correlation of log2 values reported by Values from 514 DNA segments were then determined as described in iChorCNA for the 0.1X condition in samples of TF < 10%, although 0.5X Xia et al.10. Briefly, each segment score was calculated as the mean copy (the coverage used in the presented study) performs similarly to 1X number score across genes within the segment. The coefficients of (Supplementary Data 13). Of note, the 0.5X coverage samples raise DNA segments for predicting gene signatures were obtained from Xia similar signature score values to 1X and 2X. For TFs below 20%, 0.1X et al. DNA-based signature scores were calculated as the weighted coverage may raise different scores and is not recommended for this average of DNA segment values for each sample. approach. Moreover, we applied our CNA signatures on the 11 samples For ctDNA, TF and tumor ploidy were estimated by ichorCNA. For with TF ≥3%. Again, 0.5X coverages raise similar profiles across all ctDNA samples with TF > 0, TF and tumor ploidy adjusted signature signatures to 1X and 2x (Supplementary Data 14). As an example, RB- scores were calculated by first adjusting copy number values in LOH signature score using different coverage is shown in Supple- ichorCNA segmentation file: adjusted_copy_number_ratio = mentary Fig. 24. log2(logR_copy_number/tumor_ploidy). Then DNA-based signature Third, regarding determining the sensitivity of each signature scores were derived the same as described for tumor tissue. For cal- according to TF, we performed ‘serial’ dilutions of TF by in-silico culating the number of altered segments, we used arbitrary gain/loss mixing reads from the 6 cases with TF > 50% with reads from the threshold of +/− 0.07 for unadjusted segment values and 0.32/−0.42 pooled 14 healthy control samples to generate TFs of: 50%, 20%, 10%, for adjusted segment values10. Segments with values above the gain 5%, and 1%. The resulting samples were analyzed with iChorCNA, and threshold or below the loss threshold were called altered. CNA signatures run on each (Supplementary Data 15). At 1% TFs, iChorCNA may fail to detect tumor profiles in the shWGS data (5/6 fails Gene expression analysis of FFPE tumor samples at 1% in-silico TF). According to our data, 5–10% TF has an acceptable RNA was extracted using the High Pure FFPET RNA isolation kit (Roche, overall failure rate (<20%). With regards to applying the CNA sig- Indianapolis, IN, USA) following manufacturer’s protocol. One to five natures, there is a good correlation between samples >10% TF. As an 10-μm FFPE slides depending on tumor cellularity were used for each example, RB-LOH signature score is shown in Supplementary Fig. 25. tumor sample, and macrodissection was performed, when needed, to Next, we evaluated, from a prognostic perspective, the effect of avoid normal tissue contamination. A minimum of ∼100 ng of total removing an increasing number of segments of a determined sig- RNA was analyzed on the nCounter platform42 (Nanostring Technolo- nature. For instance, we randomly removed 20, 40, 80, or 160 DNA- gies, Seattle, USA) using the 770-gene Breast Cancer 360TM Gene Panel, based segments of the RB-LOH signature (measured in plasma) and which includes the 50 PAM50 genes. Gene expression for each sample evaluated its association with PFS. We repeated this analysis 500 times was independently normalized to the geometric mean of 5 house- and estimated the average hazard ratio and p. Removing features of keeping (ACTB, MRPL19, PSMC4, RPLP0, and SF3A1), and research- the RB-LOH signature affected its prognostic ability (i.e., hazard ratio based PAM50 subtyping was determined15. decreased, and p increased) (Supplementary Fig. 26). Finally, shWGS was performed in 14 healthy individuals, and the METABRIC and MSK breast cancer datasets log2 values by ichorCNA estimates were determined to be on average Clinical-pathological data was obtained from cbioportal43. Processed 0 (Supplementary Data 16). DNA segment values were downloaded, and DNA-based signature scores were calculated as the weighted average of DNA segment values DNA-sequencing of FFPE tumor samples for each sample. Treatment and clinical outcome information from the DNA obtained from FFPE-derived tissues was purified with the QIAamp MSK dataset was obtained from Table S2 from Razavi et al.26. DNA FFPE Tissue kit (QIAGEN Inc.) for all samples available, following manufacturer’s instructions. Quantification was performed with a A DNA-based 4 subtype predictor Qubit dsDNA broad-range assay kit and the Qubit 4.0 fluorometer (Life To identify the 4 subtype clusters using DNA-based data, we selected Technologies, Carlsbad, CA, USA). A minimum of 100 ng of extracted signatures that were significantly differentially expressed across the 4 DNA was processed for library preparation using a custom clusters identified in ctDNA using a multi-class significance analysis of hybridization-based capture panel targeting 435 genes with reported microarrays (SAM)44 with <5% FDR. Then we used the selected gene list somatic mutations in different tumor types (VHIO-300 v4 panel) per- and calculated 4 centroids from the training data. For every new formed with Agilent SureSelectXT Low Input Target Enrichment Sys- sample in METABRIC29, we calculated the Euclidean distances to the 4 tem (Agilent Technologies, Inc). Indexed libraries were quantified by centroids and assigned a cluster class to each sample based on the qPCR using the KAPA Library Quantification Kit (Roche Sequencing nearest centroid. Solutions), pooled, and sequenced in a HiSeq 2500 Illumina (2 x 100 bp) at an average coverage of 500X. Reads were aligned to the General statistical procedures hg19 reference genome with BWA40, applied GATK41 base quality score Categorical variables were expressed as number (%) and compared by recalibration, indel realignment, and duplicate removal. Variant calling χ2 test or Fisher’s exact test. Differentially expressed signatures (VarScan2 v2.4.3) required a minimum of 7 reads supporting the var- between two groups were identified using a two-class unpaired SAM iant allele to call a mutation. The sensitivity of the technique is 5% with an FDR < 5%. Differentially expressed signatures between two Minor allele frequency (MAF) for Single nucleotide variant (SNVs) and timepoints (i.e., baseline versus post-progression to endocrine therapy Nature Communications | (2023)14:1157 13 Article https://doi.org/10.1038/s41467-023-36801-9 and a CDK4/6 inhibitor) were identified using a two-class paired SAM 10. Xia, Y., Fan, C., Hoadley, K. A., Parker, J. S. & Perou, C. M. Genetic with an FDR < 5%. Estimates of survival were from the Kaplan–Meier determinants of the molecular portraits of epithelial cancers. Nat. curves and tests of differences by the log-rank test. Univariate and Commun. 10, 5666 (2019). multivariable Cox models for PFS and OS were used to test the prog- 11. Martínez-Sáez, O. et al. Circulating tumor DNA dynamics in nostic significance of each variable. The Bonferroni correction method advanced breast cancer treated with CDK4/6 inhibition and endo- was used to control the family-wise error rate in case of multiple crine therapy. npj Breast Cancer 7, 8 (2021). comparisons. PFS was defined as the period from initiation of endo- 12. Adalsteinsson, V. A. et al. Scalable whole-exome sequencing of crine therapy and a CDK4/6 inhibitor until disease progression or date cell-free DNA reveals high concordance with metastatic tumors. of last follow-up. OS was defined as the period from initiation of Nat. Commun. 8, 1324 (2017). endocrine therapy and a CDK4/6 inhibitor until death or date of last 13. Reichert, Z. R. et al. Prognostic value of plasma circulating tumor follow-up. All cluster analyses were displayed using Java Treeview DNA fraction across four common cancer types: a real-world out- version 1.1.3. Average linkage hierarchical clustering was performed comes study. Ann. Oncol. https://doi.org/10.1016/j.annonc.2022. using Cluster v3.045. Two-sided p < 0.05 were considered statistically 09.163. significant. Statistical computations were carried out in R 4.0.3 (http:// 14. Husain, H. et al. Tumor Fraction Correlates With Detection of cran.r-project.org). Actionable Variants Across > 23,000 Circulating Tumor DNA Sam- ples. JCO Precis. Oncol., e2200261 (2022). https://doi.org/10.1200/ Reporting summary po.22.00261 Further information on research design is available in the Nature 15. Parker, J. S. et al. Supervised risk predictor of breast cancer based Portfolio Reporting Summary linked to this article. on intrinsic subtypes. J. Clin. Oncol. 27, 1160–1167 (2009). 16. Herschkowitz, J. I., He, X., Fan, C. & Perou, C. M. The functional loss Data availability of the retinoblastoma tumour suppressor is a common event in The data collected for the study are not available, as participants of this basal-like and luminal B breast carcinomas. Breast Cancer Res. 10, study did not agree for their data to be shared publicly. However, we R75 (2008). encourage investigators interested in data access and collaboration to 17. Martínez-Sáez, O. & Prat, A. Current and future management of contact the corresponding author (Aleix Prat, alprat@clinic.cat). HER2-positive metastatic breast cancer. JCO Oncol. Pract. 17, Access can be obtained for academic use only under a data transfer 594–604 (2021). agreement and upon Ethics Committee approval. The timescale for 18. Fan, C. et al. Building prognostic models for breast cancer patients this process is approximately 6 months and the data will be available using clinical variables and hundreds of gene expression sig- for 3 years. The data generated in this study and presented in the natures. BMC Med. Genom. 4, 3 (2011). figures are provided in the Supplementary Data/Source Data 19. Hoadley, K. A. et al. Multiplatform analysis of 12 cancer types files. Source data are provided with this paper. reveals molecular classification within and across tissues of origin. Cell 158, 929–944 (2014). References 20. Pfefferle, A. D. et al. Transcriptomic classification of genetically 1. Mateo, J. et al. A framework to rank genomic alterations as targets engineered mouse models of breast cancer identifies human sub- for cancer precision medicine: the ESMO Scale for Clinical type counterparts. Genome Biol. 14, R125 (2013). Actionability of molecular Targets (ESCAT). Ann. Oncol. 29, 21. O’Leary, B., Finn, R. S. & Turner, N. C. Treating cancer with selective 1895–1902 (2018). CDK4/6 inhibitors. Nat. Rev. Clin. Oncol. 13, 417–430 (2016). 2. Marabelle, A. et al. Association of tumour mutational burden with 22. Konecny, G. E. et al. Expression of p16 and Retinoblastoma deter- outcomes in patients with advanced solid tumours treated with mines response to CDK4/6 inhibition in ovarian cancer. Clin. Cancer pembrolizumab: prospective biomarker analysis of the multicohort, Res. 17, 1591–1602 (2011). open-label, phase 2 KEYNOTE-158 study. Lancet Oncol. 21, 23. Finn, R. S. et al. PD 0332991, a selective cyclin D kinase 4/6 inhibitor, 1353–1365 (2020). preferentially inhibits proliferation of luminal estrogen receptor- 3. Marabelle, A. et al. Efficacy of Pembrolizumab in Patients With positive human breast cancer cell lines in vitro. Breast Cancer Res. Noncolorectal High Microsatellite Instability/Mismatch 11, R77 (2009). Repair–Deficient Cancer: Results From the Phase II KEYNOTE-158 24. Condorelli, R. et al. Polyclonal RB1 mutations and Study. J. Clin. Oncol. 38, 1–10 (2020). acquired resistance to CDK 4/6 inhibitors in patients with metastatic 4. Prat, A. & Perou, C. M. Deconstructing the molecular portraits of breast cancer. Ann. Oncol. 29, 640–645 (2018). breast cancer. Mol. Oncol. 5, 5–23 (2011). 25. O’Leary, B. et al. The Genetic Landscape and Clonal Evolution of 5. Prat, A. et al. Correlative biomarker analysis of intrinsic subtypes Breast Cancer Resistance to Palbociclib plus Fulvestrant in the and efficacy across the MONALEESA Phase III studies. J. Clin. Oncol. PALOMA-3 Trial. Cancer Discov. 8, 1390–1403 (2018). 39, 1458–1467 (2021). 26. Razavi, P. et al. The genomic landscape of endocrine-resistant 6. Prat, A. et al. Prognostic value of intrinsic subtypes in hormone advanced breast cancers. Cancer Cell 34, 427–438.e426 receptor–positive metastatic breast cancer treated with Letrozole (2018). With or Without Lapatinib. JAMA Oncol. 2, 1287–1294 (2016). 27. Zehir, A. et al. Mutational landscape of metastatic cancer revealed 7. Prat, A. et al. Everolimus plus Exemestane for hormone receptor- from prospective clinical sequencing of 10,000 patients. Nat. Med. positive advanced breast cancer: A PAM50 intrinsic subtype ana- 23, 703–713 (2017). lysis of BOLERO-2. The Oncologist 24, 893–900 (2019). 28. Kikuchi-Koike, R. et al. Array comparative genomic hybridization 8. Llombart-Cussac, A. et al. HER2-enriched subtype as a predictor of analysis discloses chromosome copy number alterations as indi- pathological complete response following trastuzumab and lapa- cators of patient outcome in lymph node-negative breast cancer. tinib without chemotherapy in early-stage HER2-positive breast BMC Cancer 19, 521 (2019). cancer (PAMELA): an open-label, single-group, multicentre, phase 2 29. Curtis, C. et al. The genomic and transcriptomic architecture of trial. Lancet Oncol. 18, 545–554 (2017). 2,000 breast tumours reveals novel subgroups. Nature 486, 9. Tutt, A. et al. Carboplatin in BRCA1/2-mutated and triple-negative 346–352 (2012). breast cancer BRCAness subgroups: the TNT Trial. Nat. Med. 24, 30. Koboldt, D. C. et al. Comprehensive molecular portraits of human 628–637 (2018). breast tumours. Nature 490, 61–70 (2012). Nature Communications | (2023)14:1157 14 Article https://doi.org/10.1038/s41467-023-36801-9 31. Cejalvo, J. M. et al. Intrinsic subtypes and gene expression profiles Acknowledgements in primary and metastatic breast cancer. Cancer Res. 77, Instituto de Salud Carlos III PI19/01846 to AP, Breast Cancer Now 2213–2221 (2017). 2018NOVPCC1294 to AP, Fundació La Marató TV3 201935-30 to AP, 32. Aftimos, P. et al. Genomic and transcriptomic analyses of breast RESCUER, funded by European Union’s Horizon 2020 Research and cancer primaries and matched metastases in AURORA, the Breast Innovation Programme under Grant Agreement No. 847912 to AP, and International Group (BIG) Molecular Screening Initiative. Cancer Fundación CRIS contra el cáncer PR_EX_2021-14 to AP. Fundación cien- Discov. (2021). https://doi.org/10.1158/2159-8290.Cd-20-1647. tífica AECC Ayudas Investigador AECC 2021 INVES21943BRAS to FBM. 33. Ciruelos, E. et al. Palbociclib and Trastuzumab in HER2-positive advanced breast cancer: results from the Phase II SOLTI-1303 Author contributions PATRICIA Trial. Clin. Cancer Res. 26, 5820–5829 (2020). A.P., A.V., F.B.M., C.M.P., and J.S.P. designed the study. A.P., A.V., 34. Sherr, C. J. Cancer cell cycles. Science 274, 1672–1677 (1996). F.B.M., O.M.S., E.S., M.B., P.G., D.M., T.P., A.R., N.C., B.A., M.V., M.M., 35. Bardia, A. et al. Genomic profiling of premenopausal HR+ and E.B., M.O., E.F., J.M., R.S.B., A.S., C.S., E.C., P.T., M.M., B.G.F., and P.V. HER2– metastatic breast cancer by circulating tumor DNA and contributed to data collection and assembly. A.P., A.V., C.M.P., J.S.P., association of genetic alterations with therapeutic response to F.B.M., Y.X., L.P., and M.G.R., interpreted and analyzed the data. All endocrine therapy and Ribociclib. JCO Precis. Oncol., 1408–1420 authors wrote and reviewed the report and approved the final version (2021). https://doi.org/10.1200/po.20.00445. for submission. 36. Berthelet, J. et al. The site of breast cancer metastases dictates their clonal composition and reversible transcriptomic profile. Sci. Adv. Funding 7, eabf4408 (2021). The study was designed and performed by investigators from Reveal 37. Jørgensen, C. L. T. et al. PAM50 intrinsic subtype profiles in primary Genomics, Hospital Clinic/IDIBAPS and VHIO. All authors had full access and metastatic breast cancer show a significant shift toward more to all data in the study and had final responsibility for the decision to aggressive subtypes with prognostic implications. Cancers 13, submit for publication. 1592 (2021). 38. Ulz, P. et al. Inferring expressed genes by whole-genome sequen- Competing interests cing of plasma DNA. Nat. Genet. 48, 1273–1278 (2016). Potential conflicts of interest are the following: A.P. reports advisory 39. O’Leary, B. et al. Early circulating tumor DNA dynamics and clonal and consulting fees from Roche, Pfizer, Novartis, Amgen, BMS, Puma, selection with palbociclib and fulvestrant for breast cancer. Nat. Oncolytics Biotech, MSD, Guardan Health, Peptomyc and Lilly, lec- Commun. 9, 896 (2018). ture fees from Roche, Pfizer, Novartis, Amgen, BMS, Nanostring 40. Li, H. & Durbin, R. Fast and accurate short read alignment with Technologies and Daiichi Sankyo, institutional financial interests from Burrows–Wheeler transform. Bioinformatics 25, 1754–1760 (2009). Boehringer, Novartis, Roche, Nanostring, Sysmex Europa GmbH, 41. McKenna, A. et al. The Genome Analysis Toolkit: A MapReduce Medica Scientia inno. Research, SL, Celgene, Astellas and Pfizer; framework for analyzing next-generation DNA sequencing data. stockholder and consultant of Reveal Genomics, SL; a patent PCT/ Genome Res. 20, 1297–1303 (2010). EP2016/080056 and the DNADX patent filed (EP22382387.3). F.B-M. 42. Geiss, G. K. et al. Direct multiplexed measurement of gene has the DNADX patent filed (EP22382387.3). P.V. has the DNADX expression with color-coded probe pairs. Nat. Biotechnol. 26, patent filed (EP22382387.3). C.M.P is an equity stockholder and con- 317–325 (2008). sultant of BioClassifier LLC, and for Reveal Genomics. C.M.P is also 43. Cerami, E. et al. The cBio Cancer Genomics Portal: An open plat- listed as an inventor on patent applications for the Breast PAM50 form for exploring multidimensional cancer genomics data. Cancer assay, and a patent application on DNA-based predictors of breast Discov. 2, 401–404 (2012). tumor phenotypes. J.S.P is an equity stockholder and consultant for 44. Tusher, V. G., Tibshirani, R. & Chu, G. Significance analysis of Reveal Genomics and is also listed as an inventor on patent applica- microarrays applied to the ionizing radiation response. Proc. Natl tions for the Breast PAM50 assay, and a patent application on DNA- Acad. Sci.s 98, 5116–5121 (2001). based predictors of breast tumor phenotypes. A.V. has the DNADX 45. Eisen, M. B., Spellman, P. T., Brown, P. O. & Botstein, D. Cluster patent filed (EP22382387.3). The remaining authors declare no com- analysis and display of genome-wide expression patterns. Proc. Natl peting interests. Acad. Sci. 95, 14863–14868 (1998). 46. van ‘t Veer, L. J. et al. Gene expression profiling predicts clinical Additional information outcome of breast cancer. Nature 415, 530–536 (2002). Supplementary information The online version contains 47. Oh, D. S. et al. Estrogen-regulated genes predict survival in hor- supplementary material available at mone receptor–positive breast cancers. J. Clin. Oncology 24, https://doi.org/10.1038/s41467-023-36801-9. 1656–1664 (2006). 48. Troester, M. A. et al. Gene expression patterns associated with Correspondence and requests for materials should be addressed to p53 status in breast cancer. BMC Cancer 6, 276 (2006). Aleix Prat. 49. Smid, M. et al. Subtypes of breast cancer show preferential site of relapse. Cancer Res. 68, 3108–3114 (2008). Peer review information Nature Communications thanks Vessela Kris- 50. Wirapati, P. et al. Meta-analysis of gene expression profiles in breast tensen, Vered Stearns and the other, anonymous, reviewer(s) for their cancer: toward a unified understanding of breast cancer subtyping contribution to the peer review of this work. Peer reviewer reports are and prognosis signatures. Breast Cancer Res. 10, R65 available. (2008). 51. Hu, Z. et al. A compact VEGF signature associated with distant Reprints and permissions information is available at metastases and poor outcomes. BMC Med. 7, 9 (2009). http://www.nature.com/reprints 52. Coutant, C. et al. Distinct p53 gene signatures are needed to predict prognosis and response to chemotherapy in ER-Positive and ER- Publisher’s note Springer Nature remains neutral with regard to jur- negative breast cancers. Clin. Cancer Res. 17, 2591–2601 (2011). isdictional claims in published maps and institutional affiliations. Nature Communications | (2023)14:1157 15 Article https://doi.org/10.1038/s41467-023-36801-9 Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/ licenses/by/4.0/. © The Author(s) 2023 1 Translational Genomics and Targeted Therapies in Solid Tumors, August Pi i Sunyer Biomedical Research Institute (IDIBAPS), Barcelona, Spain. 2Reveal Genomics, Barcelona, Spain. 3Department of Medical Oncology, Hospital Clinic of Barcelona, Barcelona, Spain. 4Department of Medicine, University of Barcelona, Barcelona, Spain. 5Institute of Oncology (IOB)-Hospital Quirónsalud, Barcelona, Spain. 6SOLTI cooperative group, Barcelona, Spain. 7Department of Pathology, Hospital Clinic de Barcelona, Barcelona, Spain. 8Department of Medical Oncology, Vall d’Hebron University Hospital, Autonomous University of Barcelona, Barcelona, Spain. 9Breast Cancer and Melanoma Group, Vall d’Hebron Institute of Oncology, Barcelona, Spain. 10Lineberger Comprehensive Cancer Center, University of North Carolina, Chapel Hill, USA. 11Department of Medical Oncology, Institut Catalan of Oncology - Badalona, Hospital Germans Trias i Pujol, Badalona, Spain. 12AIDS Research Institute-IrsiCaixa and Health Research Institute Germans Trias i Pujol (IGTP), Hospital Germans Trias i Pujol, Badalona, Spain. 13Cancer Genomics Group, Vall d´Hebron Institute of Oncology (VHIO), Barcelona, Spain. 14Department of Medical Oncology, Hospital Universitario 12 de Octubre, Madrid, Spain. 15Department of Genetics, University of North Carolina, Chapel Hill, USA. 16These authors contributed equally: Aleix Prat, Fara Brasó-Maristany, Olga Martínez-Sáez. e-mail: alprat@clinic.cat Nature Communications | (2023)14:1157 16