Wander, Weipert, Cabel, et al. (2026) — Real-World Cell-Free ctDNA Analysis Identifies CDK4/6 Inhibitor Resistance and Tumor Evolution in HR+ Advanced Breast Cancer

Bibliographic Reference

Wander, S.A., Weipert, C.M., Cabel, L., Liao, J., Zhang, N., Safonov, A., Bardia, A., & Razavi, P. (2026). Real-world cell-free circulating tumor DNA (ctDNA) analysis identifies CDK4/6 inhibitor resistance and tumor evolution in HR+ advanced breast cancer. npj Breast Cancer. Article in press. https://doi.org/10.1038/s41523-026-00986-1

Core Argument

This study leverages the largest real-world ctDNA-based clinical-genomic database explored to date to characterize the molecular landscape of CDK4/6 inhibitor (CDK4/6i) resistance in HR+/HER2- metastatic breast cancer. Using the GuardantINFORM database — which pairs ctDNA sequencing data (Guardant360 assay) with anonymized insurance claims — the authors assess genomic alterations before and after CDK4/6i plus endocrine therapy, evaluate the impact of therapy duration on the resistance landscape, and test whether a composite panel of putative resistance genes (CDK4/6i+ET-R) predicts clinical outcomes. The central claim is that ctDNA captures acquired resistance alterations under therapeutic pressure, that pre-existing resistance alterations at baseline predict significantly worse outcomes, and that a composite ctDNA-based biomarker panel may have clinical utility for treatment selection, though prospective validation is needed.

The paper is jointly authored by academic investigators (MGH, MSKCC, UCLA) and Guardant Health scientists, reflecting a collaboration between clinical research and a commercial liquid biopsy platform.

Methods

Data source. The GuardantINFORM clinical-genomic database, comprising anonymized genomic data from patients with advanced solid tumors who underwent clinical ctDNA testing via the Guardant360 assay (Guardant Health, Palo Alto, CA). The ctDNA panel used 54-83 genes depending on version; panels assessed sequence alterations (SNVs, splice site variants, indels), fusions, and amplifications, but did not assess copy number loss. Claims data were provided via a commercial data aggregator covering ~150 payer datasets representative of US geography and demographics.

Patient population. Female and male breast cancer patients from GuardantINFORM with ctDNA testing between June 2014 and March 2024 and at least one CDK4/6i treatment claim. HER2 amplification detected via ctDNA pre-CDK4/6i was an exclusion criterion. Pre-CDK4/6i ctDNA: sample within 90 days before treatment initiation. Post-CDK4/6i ctDNA: sample after initiation and up to 90 days after discontinuation. Final cohort: 1,473 with pre-CDK4/6i ctDNA, 4,944 with post-CDK4/6i ctDNA, 207 paired. 90% of patients had detectable ctDNA pre-CDK4/6i. Median maxVAF pre-treatment: 1.8%. CDK4/6i use was palbociclib 65%, abemaciclib 20%, ribociclib 15%.

Resistance gene panel (CDK4/6i+ET-R). Curated from prior work (Brett et al., 2023), included: RB1 LOF SNVs/indels, PTEN LOF SNVs/indels, AKT1 activating SNVs/indels, ERBB2 activating SNVs/indels, FGFR1/2 amplifications, KRAS activating SNVs/indels, and CCNE1/2 amplifications. Only oncogenic alterations (by OncoKB classification) were included.

Endpoints. Real-world time to treatment discontinuation (rwTTD), real-world time to next treatment (rwTTNT), and overall survival (OS), derived from claims data. These are validated surrogates for progression-free survival in the real-world setting.

Statistical analyses. Non-adjusted Kaplan-Meier curves with two-sided log-rank p-values; Cox regression adjusted for age, sex, year of ctDNA analysis, and line of therapy with two-sided Wald p-values. Propensity score weighting (IPTW) as secondary analysis. Gene frequency comparisons: two-sided q-values (Chi-square or Fisher exact with multiple testing correction). Trend test (tertile as continuous) and ANOVA (tertile as categorical) for time-on-treatment analysis. Paired-sample VAF analysis: normalized change in log-transformed copy-number-adjusted VAF (deltaVAF) using the clonal alteration as an estimate of tumor fraction.

Key Findings

1. ESR1 and RB1 alterations are significantly enriched post-CDK4/6i therapy. The frequency of non-synonymous ESR1 alterations increased from 16% to 32%, and RB1 from 3% to 7% in post- versus pre-treatment samples. In the paired subset (n=207), ESR1 increase remained significant (12.6% to 24.2%, p=0.001); RB1 increased numerically (2.9% to 5.3%, p=0.194). No other genes showed significant frequency changes. PIK3CA hotspot mutation frequency was stable (39% pre vs 37% post), consistent with its role as an early, truncal driver.

2. ESR1 alteration frequency increases with duration of CDK4/6i treatment. When patients were divided into tertiles by time on therapy (<7 months, 7-18 months, >18 months), ESR1 frequency rose from 31% (lowest tertile) to 59% (highest tertile, trend p<0.0001, ANOVA p<0.0001). No other genes showed significant trends in the primary analysis (maxVAF >1% threshold). At maxVAF >5%, CDH1 became more frequent with longer therapy while TP53 became less frequent.

3. A composite CDK4/6i resistance biomarker panel identifies patients with significantly worse outcomes. Across all lines, 30% of patients (396/1323) had >=1 baseline CDK4/6i+ET-R alteration. These patients had significantly shorter rwTTD (adjusted HR=1.24, 95% CI 1.07-1.43, p=0.004), rwTTNT (HR=1.29, 95% CI 1.09-1.52, p=0.003), and OS (HR=1.74, 95% CI 1.35-2.25, p<0.001) compared to patients without such alterations. These differences were maintained when restricted to first-line therapy (n=613) and when assessed via propensity score weighting. The association held after stratification by visceral metastatic status for OS but lost significance for rwTTD and rwTTNT due to reduced power.

4. ctDNA detection and degree of tumor shed at baseline are independently prognostic. Patients with detectable ctDNA pre-CDK4/6i had significantly worse rwTTD, rwTTNT, and OS compared to patients with no ctDNA detected. This held when patients were stratified into ctDNA-high (maxVAF >1.8%) and ctDNA-low (maxVAF <1.8%) groups: both had worse outcomes than undetectable ctDNA, with ctDNA-high patients faring worst. Among first-line patients, the trend held but did not reach statistical significance due to the small ctDNA-negative sample.

5. Paired ctDNA analysis reveals selective pressure on specific genes. Of 66 patients with paired samples sharing a clonal alteration, 20% showed disproportionate VAF increase in ESR1, 21% in TP53, and 6% in RB1 — indicating these genes are under selective pressure during therapy. One patient acquired 13 distinct RB1 alterations, consistent with convergent evolution producing polyclonal resistance. ESR1 polyclonality increased significantly post-CDK4/6i (37% vs 29%, p=0.036), suggesting that treatment drives diversification of resistance mechanisms within the ESR1 gene.

6. The pre-treatment ctDNA genomic landscape broadly matches tissue-based NGS cohorts. Comparison to the MSK tissue cohort (n=930) and AACR GENIE DFCI tissue cohort (n=443) showed overall concordance, with some expected differences: TP53, ESR1, ATM, KRAS alterations and EGFR, PIK3CA amplifications were more frequent in ctDNA (reflecting a more heavily pre-treated, advanced population); CDH1, GATA3, AKT1, PTEN, BRCA2 alterations and MYC, CCND1 amplifications were more frequent in tissue (possibly reflecting histologic differences such as lobular subtype enrichment).

Figure Descriptions

Figure 1 presents Kaplan-Meier curves for rwTTD, rwTTNT, and OS stratified by ctDNA detection status (detected vs not detected) and by degree of tumor shed (ctDNA-high, ctDNA-low, not detected). All three outcomes show significant separation, establishing the prognostic value of baseline ctDNA detectability and quantitative tumor fraction.

Figure 2 compares frequencies of oncogenic alterations across the INFORM ctDNA cohort, the MSK tissue cohort, and the DFCI tissue cohort for 19 genes. The overall landscape is concordant, with the exceptions noted in Key Finding 6.

Figure 3 shows post- versus pre-CDK4/6i gene frequencies. Panel A presents all non-synonymous alterations across unpaired samples, with only ESR1 and RB1 reaching significance. Panels B and C break down PIK3CA and ESR1 hotspot alterations individually — PIK3CA hotspot frequencies are stable, while specific ESR1 alterations (Y537S/N, D538G, E380Q) all increase significantly post-treatment.

Figure 4 displays alteration frequency by time-on-treatment tertiles, demonstrating the monotonic increase in ESR1 alterations from 31% to 59% across tertiles, with no other gene showing a significant trend.

Figure 5 presents outcomes (rwTTD, rwTTNT, OS) for patients stratified by CDK4/6i+ET-R status, across all lines and restricted to first-line therapy. All six panels show significant separation favoring CDK4/6i+ET-R-negative patients.

Concepts Introduced or Used

• CDK4/6i+ET-R composite biomarker — A gene panel of putative resistance alterations to CDK4/6i plus endocrine therapy, comprising RB1, PTEN, AKT1, ERBB2, FGFR1/2, KRAS, and CCNE1/2. The study provides the largest real-world evidence to date that this composite panel identifies patients with worse outcomes on CDK4/6i therapy. Relevant to therapy-resistance.

• Therapy-induced clonal bottleneck — CDK4/6i therapy exerts selective pressure that prunes the tumor population, enriching for resistant clones while eliminating sensitive ones. The study demonstrates this through enrichment of ESR1 and RB1 alterations post-treatment and through disproportionate VAF changes in paired samples. Relevant to population-bottleneck and clonal-sweep.

• Convergent evolution of resistance — The documented case of 13 acquired RB1 alterations in a single patient post-CDK4/6i illustrates convergent evolution: multiple independent mutations in the same gene, suggesting strong selection pressure acting on a locus where different loss-of-function events all produce the same resistance phenotype.

• ctDNA-based clonal dynamics — The use of normalized deltaVAF (comparing individual gene VAF changes to the clonal alteration’s VAF change) to infer selective pressure during therapy. Alterations whose VAF changes deviate from the clonal trajectory are inferred to be under selection. This is a method for non-invasive tracking of clonal evolution, relevant to intratumor-heterogeneity and clonal-evolution.

• Real-world evidence (RWE) in cancer genomics — The study demonstrates the feasibility and limitations of using claims-linked genomic databases (GuardantINFORM) for large-scale analysis of treatment outcomes, acknowledging the trade-offs between sample size (N=6,210 with any CDK4/6i claim) and data granularity relative to prospective clinical trials.

Entities Referenced

• Genes: ESR1, RB1, PIK3CA, TP53, CDH1, PTEN, AKT1, ERBB2, FGFR1, FGFR2, KRAS, CCNE1, CCNE2, GATA3, BRCA2, ATM, EGFR, MYC, CCND1
• Drugs: Palbociclib, ribociclib, abemaciclib (CDK4/6 inhibitors); letrozole, fulvestrant, anastrozole (endocrine therapies); elacestrant (SERD)
• Assays/methods: Guardant360 ctDNA assay (Guardant Health); GuardantINFORM clinical-genomic database; OncoKB classification
• Cohorts/biospecimen resources: AACR GENIE consortium; MSK-IMPACT tissue cohort
• Institutions: Massachusetts General Hospital Cancer Center/Harvard Medical School; Guardant Health, Inc.; Memorial Sloan Kettering Cancer Center; UCLA Health Jonsson Comprehensive Cancer Center; Weill Cornell Medicine

Limitations (as stated by authors)

1. Retrospective, real-world design with inherent biases. CDK4/6i treatment selection is weighted toward palbociclib (65% of patients), though the authors note current evidence does not suggest meaningfully different resistance profiles between the three approved CDK4/6i. Providers may order ctDNA testing preferentially in patients who are rapidly progressing or have more complex histories, introducing selection bias.

2. Heavily pre-treated population with rare paired samples. The low number of paired pre/post samples (n=207) is noted as likely reflecting the historically later use of ctDNA testing in oncology practices. This limits the power of paired analyses.

3. Copy number loss was not assessed. The ctDNA assay versions used did not include copy number loss (LOH) detection. This means the frequency of resistance alterations in genes where oncogenicity is driven by loss of function (e.g., PTEN, RB1) may be underestimated, though the authors note that RB1 frequencies were consistent with tissue cohorts.

4. Claims data limitations. Treatment information is based on billing claims, which are subject to coding errors, incomplete clinical context capture, and challenges in accurately defining treatment lines and outcomes. The 28-day window to define combination therapy may have misclassified some treatment lines, reflected in the 12% of patients with no identified ET partner. Patients may have transitioned between insurance plans, creating missing data. Claims data structure differs from prospectively collected real-world observational datasets.

5. Lack of orthogonal tissue validation. Orthogonal validation of ctDNA findings with contemporaneous tissue testing was not possible, though the alignment of pre-CDK4/6i ctDNA results with previously reported tissue NGS profiles supports the validity of the landscape.

6. rwTTD and rwTTNT are distinct from trial PFS. While both are well-established real-world proxies for PFS, differences between these endpoints and trial-measured PFS may reflect the more diverse patient population in real-world settings, treatment outside standard recommendations, and discontinuation for reasons other than progression (adverse events, financial toxicity).

7. Histologic subtype cannot be determined. The claims-based dataset lacks information on histologic subtype (e.g., lobular vs. ductal), which may be relevant given the known association between CDH1 alterations and lobular breast cancer and emerging real-world evidence suggesting different CDK4/6i outcomes by histology.

Relevance to Clonal Evolution

Therapy as a selective bottleneck. This study provides one of the largest empirical demonstrations of CDK4/6i therapy acting as a selective bottleneck in HR+ breast cancer. The enrichment of ESR1 and RB1 alterations post-treatment, the monotonic increase in ESR1 frequency with time on therapy, and the disproportionate VAF expansion of specific alterations in paired samples all document treatment-driven clonal-evolution in action. The study is directly relevant to the population-bottleneck concept page, which describes how therapy-induced bottlenecks reshape clonal architecture — the current paper adds HR+ breast cancer on CDK4/6i as a second well-documented domain, alongside the Myeloma XI trial’s chemotherapy-induced bottlenecks.

ctDNA as a window into clonal dynamics. The study’s paired-sample analysis using normalized deltaVAF represents a non-invasive method for tracking clonal evolution during therapy. By comparing individual gene VAF trajectories to the clonal alteration’s VAF trajectory, the authors identify which genes are under selective pressure — a methodology applicable to the intratumor-heterogeneity concept page’s description of VAF-based ITH measurement. The observation that PIK3CA mutations are stable in frequency and predominantly clonal (consistent with early, truncal drivers) while ESR1 alterations become more frequent and more polyclonal post-treatment provides a concrete example of the distinction between clonal (truncal) and subclonal (branch-specific) mutations.

Convergent evolution and polyclonal resistance. The single patient with 13 acquired RB1 alterations is a striking example of convergent evolution producing polyclonal resistance — multiple independent resistance-conferring mutations arising in the same gene under strong selection pressure. This parallels the polyclonal resistance patterns documented in EGFR-mutant NSCLC and other settings of intense therapeutic selection.

Relevance to the ITH-outcome test design. The study’s ctDNA-based approach to assessing clonal architecture aligns with the wiki’s interest in intratumor-heterogeneity measurement methods. While this study does not directly test the U-shaped ITH-outcome relationship predicted by the compression-entrenchment hypothesis, it demonstrates that ctDNA can non-invasively distinguish clonal from subclonal alterations (via VAF thresholding at 50% of maxVAF), assess polyclonality within individual genes (multiple ESR1 or RB1 mutations in the same sample), and track the dynamics of subclonal expansions under therapy — all of which are relevant inputs to the ITH-outcome test design framework.

Connection to adaptive therapy. The finding that 30% of patients harbor pre-existing CDK4/6i+ET-R alterations and that these patients have worse outcomes on CDK4/6i therapy suggests a potential role for pre-treatment ctDNA profiling in treatment selection. Patients with pre-existing resistance may benefit from alternative strategies (including adaptive therapy approaches or alternative combinations), while those without detectable resistance alterations may be more likely to derive durable benefit from standard CDK4/6i treatment. This connects to the therapy-resistance concept page’s discussion of pre-existing vs. de novo resistance and to the broader principle that evolutionary biology can inform treatment sequencing.