Phase 2 study of neoadjuvant enzalutamide and paclitaxel for luminal androgen receptor-enriched TNBC: Trial results and insights into “ARness”

Summary

Luminal androgen receptor (LAR)-enriched triple-negative breast cancer (TNBC) is a distinct subtype. The efficacy of AR inhibitors and the relevant biomarkers in neoadjuvant therapy (NAT) are yet to be determined. We tested the combination of the AR inhibitor enzalutamide (120 mg daily by mouth) and paclitaxel (80 mg/m² weekly intravenously) (ZT) for 12 weeks as NAT for LAR-enriched TNBC. Eligibility criteria included a percentage of cells expressing nuclear AR by immunohistochemistry (iAR) of at least 10% and a reduction in sonographic volume of less than 70% after four cycles of doxorubicin and cyclophosphamide. Twenty-four patients were enrolled. Ten achieved a pathologic complete response or residual cancer burden-I. ZT was safe, with no unexpected side effects. An iAR of at least 70% had a positive predictive value of 0.92 and a negative predictive value of 0.97 in predicting LAR-enriched TNBC according to RNA-based assays. Our data support future trials of AR blockade in early-stage LAR-enriched TNBC.

Graphical abstract

Highlights

• •Enzalutamide plus paclitaxel (ZT) is effective against TNBC
• •Conventional AR-measuring markers are suboptimal to predict ZT efficacy
• •AR-responsive gene sets may be tested as predictive markers for ZT efficacy
• •Various ARness biomarkers of ZT response may provide biological insight

Lim et al. report that for treating triple-negative breast cancers with suboptimal response to adriamycin and cyclophosphamide before surgery, enzalutamide with paclitaxel is effective and safe to induce a response. The authors observe a higher response corresponding with androgen receptor (AR) levels measured by immunohistochemistry or RNA-seq.

Introduction

Triple-negative breast cancer (TNBC) is treated with neoadjuvant therapy (NAT) with chemotherapy alone or in combination with immune checkpoint inhibitors, but not all patients benefit from combination NAT.¹ Pathological complete response (pCR) and residual cancer burden-I (RCB-I) after NAT are associated with excellent long-term outcomes in patients with breast cancer.^{2,3,4,5,6,7,8} Many groups have used genomic analysis to determine molecular subtypes of TNBC and develop targeted therapies to improve responses and long-term survival.^9,10,11 Results of this molecular subtyping have led several groups to test androgen receptor (AR) blockade in breast cancer. For example, in a phase 2 study of bicalutamide in patients with hormone receptor-negative, AR-positive (>10% nuclear staining) metastatic breast cancer, the 6-month clinical benefit rate was 19% (95% confidence interval [CI], 7%–39%), and the median progression-free survival time was 12 weeks (95% CI, 11–22 weeks).¹² In another study of the AR signaling inhibitor enzalutamide daily in metastatic TNBC with AR staining >0%, the 16-week clinical benefit rate was 25%, and the median progression-free survival time was 2.9 months.¹³ However, the exact role of the AR phenotype (“ARness”) in TNBC biology and response to targeted therapy remains uncertain, especially in early-stage TNBC. The luminal AR (LAR)-enriched subtype of TNBC identified by RNA analysis¹⁴ has yet to be correlated with the percentage of cells expressing nuclear AR by immunohistochemistry (IHC) (iAR) used in AR-targeted clinical trials. Furthermore, the efficacy of direct AR inhibition in NAT for LAR-enriched TNBC has yet to be established. In addition, the concept of ARness within TNBC needs to be refined to guide the selection of patients who may benefit from AR-targeted therapies. Unfortunately, every trial uses its own definition of ARness, which makes a collective assessment of AR targeting in breast cancer, including TNBC, difficult.

Enzalutamide inhibits AR’s nuclear translocation and DNA binding, suppressing the recruitment of AR co-activators.¹⁵ Notably, a preclinical study using prostate cancer cells demonstrated that enzalutamide and docetaxel synergistically inhibit the growth of cancer cells and that cross-resistance can develop when these drugs are used sequentially.¹⁶ A phase 1 study of patients with prostate cancer demonstrated that enzalutamide did not have a clinically significant effect on the pharmacokinetics of docetaxel when these drugs were used concurrently, providing a rationale for combining enzalutamide with taxane-based chemotherapy.¹⁷ Therefore, we hypothesized that enzalutamide and a taxane could induce a better response than a taxane alone in patients with LAR-enriched TNBC.

In the study reported here, we tested the combination of enzalutamide with weekly paclitaxel (ZT) in patients with early-stage TNBC with a modest response, defined as less than 70% primary tumor volume reduction, to NAT with doxorubicin and cyclophosphamide (AC).¹⁸ We hypothesized that ZT produces a combined pCR and RCB-I rate of 20%, substantially higher than the historical pCR rate of 5% seen in patients with breast cancer treated with a second neoadjuvant chemotherapy regimen after a lack of response to their first chemotherapy regimen.^19,20 In addition, by multiomic analyses of tumor samples, we refined ARness in TNBCs to identify potential predictive biomarkers of ZT response. Lastly, by screening the RNA expression dataset of the TNBC tumors included in this study and the subtype-enriched genes of these TNBC tumors, we sought potential targets of inhibition for treating LAR-enriched TNBC.

Results

Patient characteristics

We enrolled 24 patients with early-stage TNBC with a suboptimal response to AC (“AC-suboptimal-response TNBC”) after confirming that tumor iAR was greater than 10% (Figure 1A). The trial patients were enrolled from 2016 to 2020, with a data lock on July 10, 2022. The median (range) patient age was 57.5 years (39–74). All patients were female; 19 were postmenopausal, 14 had T2 tumors, and 6 had T3 or larger tumors, including 1 patient with T4d disease (Table 1). More than half of the patients (13 [54%]) had an iAR greater than 75%; the median (range) iAR for the remaining 11 patients was 40% (10%–75%). Twenty patients had the LAR-enriched subtype based on RNA expression analysis. The median (range) iAR in patients with LAR-enriched TNBC was 80% (10%–100%), whereas the median iAR in patients with other TNBC subtypes was 60% (10%–95%). Patients with non-LAR-enriched TNBC were not eligible for other molecularly targeted trials based on their molecular profiling; their enrollment in our ZT cohort was approved on the basis of the ARTEMIS (ClinicalTrials.gov: NCT02276443) molecular tumor board’s decision (Figure S1A).

Figure 1.

Summary of clinical characteristics

(A) A total of 311 patients were enrolled under ARTEMIS, the principal biomarker assessment protocol, by data closure. Of these, 119 patients, who were determined to have a suboptimal response to AC, were evaluated for compatibility with the ZT protocol. Following the exclusion of 5 patients, 24 patients proceeded to undergo treatment as per the ZT protocol.

(B) The trial met its primary endpoint target by achieving the goal rate of greater than 20% of patients having a pCR or RCB-I. Four patients had pCR, and six had RCB-I, resulting in 42% of all patients accrued to this trial having a pCR or RCB-1.

(C) The 48-month EFS rate was 33.4% (95% CI, 12.6%–88.2%) for all patients, 16.9% for patients with RCB-II/III, and 67% for patients with pCR or RCB-I. The difference between these groups was statistically significant (p = 0.0043).

(D) The 48-month OS rate was 63.0% (95% CI, 40.2%–98.4%) for all patients, 36.4% for patients with RCB-II/III, and 100% for patients with pCR or RCB-I. The difference between these groups was statistically significant (p = 0.032).

Table 1.

Clinicopathologic characteristics of all patients and patients with a good response

Characteristic	No. of patients (n = 24)	No. (%) with pCR or RCB-I
Ethnicity
Hispanic or Latino	2	1 (50)
Not Hispanic or Latino	22	9 (41)
Race
Asian	3	1 (33)
Black/African American	3	1 (33)
Other race	1	1 (100)
White or Caucasian	17	7 (41)
Menopausal status
Perimenopausal	1	0 (0)
Postmenopausal	19	9 (47)
Premenopausal	4	1 (25)
Clinical stage
IB	2	1 (50)
IIA	7	6 (86)
IIB	8	3 (38)
IIIA	1	0 (0)
IIIB	1	0 (0)
IIIC	5	0 (0)
T category at diagnosis
T1c (≥1.5–2 cm)	4	2 (50)
T2 (>2–5 cm)	14	6 (43)
T3 (>5 cm)	5	2 (40)
T4d	1	0 (0)
N category at diagnosis
N0	9	6 (67)
N1	10	4 (40)
N3	2	0 (0)
N3c	3	0 (0)
Histologic type
Apocrine	4	1 (25)
Ductal	20	9 (45)
iAR, %
10–25	4	2 (50)
26–50	4	1 (25)
51–75	3	1 (33)
>75	13	6 (46)

Clinical outcomes

The primary endpoint was the combined rate of patients achieving a good response, defined as a pCR or RCB-I at the time of surgery after receipt of at least 1 week of ZT treatment, with the goal to improve the rate from the historical rate of 5%–20%. Four patients (17%) had a pCR, and six (25%) had RCB-I disease, resulting in a good response rate of 42%, exceeding the target rate of 20%. Fourteen patients had a poor response, which was defined as RCB-II (n = 8 [33%]) or RCB-III (n = 6 [25%]) (Figure 1B). Due to lower-than-anticipated participant enrollment, which did not meet the expected numbers outlined in the Simon 2-stage design, we opted to conduct an exact binomial test. This test compared the observed proportion of complete response and partial response (41.7%) against a baseline rate of 5%. The results of this test were statistically significant (p < 0.0001), confirming that the observed response rate was substantially higher than 5%. At the data lock date, the median follow-up time was 24 months. The first secondary endpoint was event-free survival (EFS). The median EFS for the whole population was 45.3 months (Figure 1C; 95% CI, 34.4–NA). At 48 months, the overall EFS rate was 33.4% (12.6%–88.2%). For the good response group (GRG; pCR + RCB-I), one patient had an event, and the median EFS was not reached at the data closure date (Figure 1C). For the poor response group (PRG; RCB-II and III), there were eight documented events, and the median EFS was 34.4 months (Figure 1C; 95% CI, 9–NA). At 48 months, the EFS rate was 66.7% (30%–100%) for the GRG and 16.9% (3.1%–92%) for the PRG. Another secondary endpoint was overall survival (OS). At 48 months, the OS rate for the whole group was 62.9% (Figure 1D; 40.2%–98.4%). The median OS time was not reached (95% CI, 38.2–NA) at the data closure date (Figure 1D). The estimated 48-month OS rate was 100% for the GRG vs. 36.4% (12.8%–100%) for the PRG (Figure 1D). The median OS time was not reached in the GRG but was reached at 38.24 months (95% CI, 36.1–NA) in the PRG. There were no deaths in the GRG, and there were 5 deaths in the PRG. One of the deaths in the PRG was not breast cancer related but due to endometrial cancer. Logistic regression did not reveal any specific clinicopathologic factors associated with good response (Table S1).

Safety

Most of the adverse events (AEs) were grade 1 or 2. Eleven patients experienced grade 3 AEs (Table 2). Neutropenia was the most common grade 3 AE (12 reports in 8 patients). One patient experienced grade 3 dizziness and syncope. This patient declined further treatment with enzalutamide while allowed re-challenge per protocol, continued receiving paclitaxel alone after two cycles of ZT, and was counted in the efficacy analysis. We observed no other unexpected toxic effects or AEs. Two patients had enzalutamide dose reductions and completed the 12 weeks of ZT. Twenty-three patients were able to complete 12 weekly doses of paclitaxel. No patients experienced grade 4 AEs during the trial, and no patient stopped either medication because of toxicity.

Table 2.

Summary of grade 2 and 3 adverse events

Adverse event	No. of reports	No. of patients
Grade 2 adverse events
Neutrophil count decreased	19	9
Anemia	15	8
Peripheral sensory neuropathy	9	7
Fatigue	7	6
Dyspnea	3	2
White blood cell count decreased	4	2
Dehydration	3	2
Alanine aminotransferase increased	2	2
Vomiting	1	1
Dizziness	1	1
Chest pain, cardiac	1	1
Sinusitis	1	1
Anorexia	1	1
Bronchial infection	1	1
Nausea	1	1
Abdominal pain	1	1
Pain other than abdominal	1	1
Nail changes	2	2
Diarrhea	1	1
Dysgeusia	1	1
Administration site condition	1	1
Vaginal infection	1	1
Lymphocyte count decreased	1	1
Urinary tract infection	1	1
Laryngitis	1	1
Pneumonia	1	1
Grade 3 adverse events
Neutrophil count decreased	12	8
White blood cell count decreased	2	2
Dehydration	2	1
Syncope	2	1
Peripheral sensory neuropathy	1	1
Dyspnea	1	1
Fatigue	1	1
Respiratory, thoracic, and mediastinal disorders	1	1
Dizziness	1	1
Infections and infestations (other)	1	1
Anemia	1	1

Association between ARness (LAR-defining) biomarkers and ZT response

To define the genomic level of ARness, we first checked the agreement of Affymetrix microarray and RNA sequencing (RNA-seq) results in the entire LAR-enriched TNBC cohort in the ARTEMIS trial. A total of 139 highly expressed genes were identified by both assays (Figure 2A). We next examined the correlation between the iAR and LAR enrichment score. RNA assay, regardless of the analysis platform, showed a high correlation between increased iAR and LAR enrichment score in the entire ARTEMIS cohort (Figure 2B). Both microarray-based and RNA-seq-based LAR-enriched tumors had a high iAR when an iAR cutoff of ≥30% was used (Figure 2C). We selected 30% as the first reasonable cutoff that could detect a significant number of LAR-enriched cases. However, an iAR cutoff of 30% yielded a positive predictive value of 0.54 and a negative predictive value of 0.98 (Table S2). An iAR cutoff of 50% showed similar positive and negative predictive values. In contrast, an iAR cutoff of 70% yielded a positive predictive value of 0.92 and a negative predictive value of 0.97 (Table S3), suggesting the use of 70% as the lowest iAR cutoff that can precisely identify the RNA-based LAR-enriched TNBC subtype, defining ARness in the TNBC tumor.

Figure 2.

LAR-enriched TNBC and correlation with IHC and clinical response to enzalutamide-based treatment

(A) We assessed two different RNA expression measuring assays (RNA microarray, RNA-seq) and IHC measurement of AR nuclear staining (iAR) as ARness biomarkers and correlated them with ZT response. The LAR-enriched TNBC subtype label is based on the RNA Affymetrix microarray, and RNA-seq had 139 overlapping genes included in each assay.

(B) The subtype calling showed 99% matching between microarray and sequencing methods, and the high iAR values were concordant with high RNA-based LAR calling (positive correlation with R=0.87).

(C) The LAR-enrichment TNBC subtype calls according to the RNA Affymetrix vs. iAR and RNA-seq vs. iAR showed similar concordance. In both correlation analyses, most LAR-enriched TNBCs fell into the iAR ≥ 70% category.

(D) Among patients with LAR-enriched TNBC subtype treated with ZT, neither LAR enrichment score nor iAR differed significantly between patients with pCR or RCB-I and those with RCB-II/III. p values were not significant (0.96 in RNAseq, 0.81 in IHC comparison.

(E) Among 14 patients who had profiling of pretreatment biopsies and complete molecular profiling, Hallmark gene set analysis showed the AR response gene pathway as the only marker that correlated with GRG. The AR response gene set expression between GRG vs. PRG showed a difference (p = 0.05).

(F) The EFS of iAR <30% vs. >30% and iAR <70% vs. >70% showed a trend favoring superior EFS in higher iAR. However, the difference was not statistically significant.

Next, we determined whether iAR or LAR enrichment determined by RNA measurement could predict ZT response in patients with AC-suboptimal-response primary TNBC enrolled in our trial. Overall, we observed a trend favoring high iAR expression and a high LAR enrichment score according to RNA-seq assay in predicting a good response of TNBC to ZT, although neither association was statistically significant (Figure 2D).

We next performed a gene set enrichment assay of the estrogen receptor (ER), progesterone receptor, and retinoblastoma pathways to determine if these gene expression pathways predicted the response to ZT. As expected, neither early nor late response ER gene pathway expression correlated with the response to ZT. The retinoblastoma pathway was downregulated in all patients regardless of their pathological response, as reported previously for TNBC.²¹ After observing the correlation of the AR response gene pathway and the ZT response, we compared the expression levels for AR response genes between the GRG and the PRG and found that expression was higher in the GRG (p = 0.05) (Figure 2E).

We next compared EFS between groups with higher and lower iAR using 30% and 70% as the cutoffs. Three of 14 (21.4%) patients with an iAR of 70% or higher had a recurrence, compared with 6 of 10 (60.0%) patients with an iAR lower than 70%, suggesting that higher iAR was associated with better EFS in the context of ZT treatment (Figure 2F; Table S4). When we tested the 30% iAR cutoff, the survival curves of the groups crossed (Figure 2F). However, there was no significant association for either iAR level, possibly due to the small sample size. OS analysis demonstrated similar relationships between iAR and EFS (data not shown).

Next, we studied gene mutations and tumor copy-number profiles using whole-exome sequencing to identify genotype changes associated with ZT response. The mutations identified included missense and nonsense mutations in known breast cancer-related genes, such as PIK3CA (n = 3) and TP53 (n = 2). Mutations in other genes were also identified, in one patient each; these other mutated genes were ACACB, ACSM2A, ATP7B, ATRIP, BCAM, BCR, C14of166B, C20of26, CACNA1H, CBR1, CCDC150, CONUL, CEP152, CLCF1, CMTM4, CMYA5, and CNTN3. Overall, the number of patients who underwent whole-exome sequencing was too small to permit a meaningful conclusion (Figure S2).

Evolution and stability of LAR-enriched TNBCs and the TME during treatment

We next examined dynamic changes in the tumor and tumor microenvironment (TME) during NAT. The LAR-enriched subtype had greater stability in RNA transcriptomic signature than the other subtypes had between the time of surgery and the end of four cycles of treatment with AC (Figure 3A). All patients except one with the LAR-enriched TNBC subtype kept their LAR-enriched phenotype during treatment with AC.

Figure 3.

Evolution and stability of AR-enriched TNBCs and TME over treatment with AC

(A) Unlike other TNBC subtypes, the LAR signature remained stable after exposure to AC in LAR TNBCs profiled longitudinally throughout T0 (baseline) and T1 (after four cycles of AC) except for one case.

(B) Among investigated TME cells, B cells, macrophages, natural killer cells, and T cells were reduced regardless of the response of tumors to ZT or AC. However, neutrophil and myeloid dendritic cell marker expressions increased throughout the treatment. Error bar in the box plot indicates standard deviation of each value (ranging from −1 to 1).

Next, we evaluated the specific subtype changes for patients in the ZT trial during AC treatment. There, we observed that all but one of the patients with LAR-enriched TNBC maintained the LAR-enriched subtype. Of note, the one patient who converted from the LAR-enriched to angio-epithelial-mesenchymal transition (angio-EMT) subtype, one of the ARTEMIS TNBC subtypes,¹⁸ after treatment with AC, had a pCR, even though the angio-EMT subtype is known to confer resistance of TNBC to chemotherapy. Unfortunately, only a few patients had sufficient surgical samples after treatment with ZT to permit assessment of the further impact of ZT on the LAR-enrichment phenotype.

Given that the US Food and Drug Administration (FDA) approved administering pembrolizumab with chemotherapy for early-stage TNBC as a standard NAT regimen after our trial was designed and conducted,¹ we investigated the TME throughout treatment in patients given ZT to see if any specific TME cell types correlated with response. As per the original protocol, an analysis of surgical stromal samples was not required for patients who had a pCR, and tumor samples from the time of surgery were not available for patients with a pCR due to a lack of residual tumor. By analyzing tumor samples collected at baseline, after four cycles of AC, and at surgery for patients with residual disease, we compared the levels of B cells; neutrophils; CD4 T cells, CD8 T cells, and regulatory T cells; macrophages; natural killer cells; and myeloid dendritic cell representative markers. For this analysis, we examined the whole group of patients, patients with a pCR, and patients with the residual disease instead of patients with good vs. poor response. We found that B cells, macrophages, natural killer cells, and T cells were reduced regardless of ZT response (Figure 3B). However, the expression of both neutrophil and myeloid dendritic cell markers increased during the treatment in all examined samples (Figure 3B).

Identifying LAR-enriched TNBC and therapeutic targets that may improve clinical response

Because we did not observe complete eradication of the tumor after treatment with ZT in more than 50% of patients with AC-refractory TNBC, we examined additional therapeutic targets using the entire ARTEMIS cohort. Specifically, we used 220 pretreatment samples with available RNA-seq data to perform Dependency Map screening (Figure 4A). We identified 52 genes with different expression in LAR-enriched TNBC compared to other groups. AR was the highest expressed gene, while others, such as FOXA1, SPDEF, AGR2, FGFR4, PIP, and ZNF552, also showed elevated expression. Next, we overlaid this group of genes to the Dependency Map that tested the effects of inhibition.²² This identified FOXA1, FGFR4, and SPDEF as the three most highly dependent genes (Figure 4B). SPDEF was the only gene that was part of the AR response genes we used in ARness analysis. LAR-enriched TNBC did not show a dependency on the other identified genes, such as CRAT, KYNU, and ALOX15B.

Figure 4.

The potential therapeutic target of LAR TNBC

(A) 220 ARTEMIS tumor cohorts with available microarray profiles were assessed for the overlap between overexpressed genes and the available drug library. LAR-enriched TNBCs show overlap with 52 genes that had available targeted inhibitors. AR was the highest expressed gene, while others like FOXA1, SPDEF, AGR2, FGFR4, PIP, and ZNF552 also showed elevated expression. Among these genes, FOXA1, SPDEF, and FGFR4 showed overlap with the Dependency Map (DepMap) screening results in genes.

(B) DepMap screening using the LAR-enriched cohort revealed FOXA1, FGFR4, and SPDEF as the top three genes that had the highest expression with high dependency among LAR-enriched tumors.

(C) Baylor College of Medicine patient-derived xenograft (PDX) breast cancer cohort showed various expression levels of AR and SPDEF, FOXA1, and FGFR4. Pearson pairwise correlation assay showed that these four genes correlate with each other with a high coefficient.

(D) Pearson pairwise correlation assay using 4 genes (AR, SPDEF, FOXA1, FGFR4) + a 150 AR response gene signature showed clustering of these genes with others (FOXA1, SPDEF, and FGFR4 clustered with FASN, AGR2, SLC44A4, TSPAN13, and XBP1 genes with strong correlation, while AR showed stronger correlation with the genes RBM25 and BRD2.

Next, we examined the Baylor College of Medicine patient-derived xenograft (PDX) breast cancer cohort²³ with known RNA-seq data to determine the distribution of FOXA1, FGFR4, and SPDEF gene expression along with AR expression to determine whether these four genes had a linear correlation with LAR-enriched TNBC. We observed various expressions of these four genes in PDX models (Figure S2C). A pairwise correlation plot using RNA-seq data from 50 TNBC PDX tumors and expression of AR, FOXA1, SPDEF, and FGFR4 showed that all four genes correlated with each other (Figure 4C). All genes showed a strong Pearson’s correlation coefficient (>0.4), with FGFR4 being the strongest. When we examined the correlation using four genes and 150 AR response genes,²⁴ FOXA1, SPDEF, and FGFR4 clustered with FASN, AGR2, SLC44A4, TSPAN13, and XBP1 with a strong correlation, while AR showed a stronger correlation with RBM25 and BRD2 (Figure 4D).

Discussion

In this study, we assessed the efficacy of the ZT regimen, which includes a direct inhibitor of AR, as a NAT option for AC-refractory LAR-enriched TNBC. To the best of our knowledge, this is unique study investigating the neoadjuvant inhibition of AR in TNBC. LAR-enriched TNBC accounts for approximately 18%–20% of all TNBC cases.^14,25 While other trials such as the I-SPY2 and German Breast Cancer Group trials have examined the pCR rate in high-risk breast cancers, specific data regarding LAR-enriched TNBC are limited. Historically, the overall rate of pCR and RCB-1 for patients who did not respond adequately to AC treatment has been generally poor, around 5%. Thus, given our finding that 42% of the patients in our trial had a pCR or RCB-1 to ZT, we conclude that the ZT regimen can effectively treat and improve outcomes in AC-refractory LAR-enriched TNBC in the neoadjuvant setting.

The pCR rate was lower in our cohort than in the overall LAR population in the ARTEMIS dataset, which included patients whose disease was not AC refractory. Our study is the first investigation of the ZT regimen as NAT specifically in an AC-suboptimal-response population. Notably, patients with AC-suboptimal-response LAR-enriched TNBC have historically not been likely to achieve a pCR, further supporting the efficacy signal observed in our trial. Regarding safety, reported serious AEs, including neutropenia, were primarily attributed to paclitaxel. One patient reported dizziness, which was possibly related to enzalutamide. The patient fully recovered, and the symptoms were temporary. Dizziness and lightheadedness have been reported as potential side effects of enzalutamide.²⁶

To evaluate ARness as a biomarker for ZT response, we examined three potential ARness measurements using different assays and found concordance between the RNA-based LAR-enrichment assays (RNA microarray and RNA-seq) and iAR (Figure 2A). The iAR with a 70% cutoff predicted LAR enrichment with high positive and negative predictive values. However, in this AC-refractory cohort, neither the RNA-based LAR score nor iAR showed clear correlations with clinical outcomes such as RCB, EFS, or OS. On the other hand, the expression of AR response gene pathways²⁴ correlated with the response to ZT therapy. Although not tested in our study due to the lack of a validation set, a combined ARness measurement based on RNA/IHC assays and AR response gene expression may serve as a predictive biomarker for ZT treatment. Future studies can validate the value of ARness measured by AR response gene pathway as a predictive biomarker for ZT.

We also observed that LAR-enriched TNBC exhibited a stable tumor transcriptome during AC treatment, suggesting biological stability during chemotherapy-induced genotoxic stress. Within the TME, we noted increased expression of neutrophil and myeloid cell markers during NAT. No other cell groups showed a clear direction of change or association with response to ZT. This finding supports the importance of LAR-targeted strategies in TNBC with ARness, irrespective of tumor exposure to standard therapy. However, more dedicated studies utilizing single-cell genomics or spatial transcriptomic analysis are required to further investigate the changes of key components of the TME after ZT exposure.

Using the ARTEMIS cohort, we aimed to identify dependent gene pathways and other characteristics of LAR-enriched TNBC. In hormone receptor-positive breast cancers, AR shares 40% of its transcriptional regions with ER, making an AR agonist act as an ER antagonist.²⁷ LAR-enriched TNBC exhibits gene ontologies heavily enriched in hormonally regulated pathways, including androgen/estrogen metabolism and genomic signatures associated with luminal/hormone receptor-positive breast cancers despite being hormone receptor negative.²⁸ AR transcriptionally regulates multiple metabolic pathways, including glycolysis, mitochondrial respiration, fatty acid metabolism, nucleotide metabolism, amino acid metabolism, and polyamine metabolism. Genes such as KLK3, AZGP1, and PIP, which are AR responsive, were also up-regulated in our LAR-enriched TNBC dataset.²⁹ Activation of these pathways may serve as surrogate response markers for AR inhibitor efficacy and guide the selection of patients who may respond well to ZT treatment. However, further validation in independent datasets is necessary. Future studies can focus on validating our ARness-defining iAR or LAR enrichment score, along with the expression of AR-responsive gene sets or selective markers within this gene set, as predictive biomarkers for ZT.

Our study demonstrated that 42% of patients achieved either pCR or RCB-I, indicating a favorable response in the neoadjuvant setting. However, the remaining patients require alternative strategies to improve the response. In our screening of LAR-enriched TNBC, we identified potential therapeutic targets such as SPDEF, FOXA1, and FGFR4. In gene expression analysis of our PDX dataset, the four genes including AR correlated with each other. SPDEF is an ETS family protein that interacts with the DNA-binding domain of AR and acts as an androgen-independent activator of the prostate-specific antigen promoter, conferring resistance to AR inhibition in AR-dependent prostate cancer.³⁰ Further investigation is needed to explore the overexpression of these genes in TNBC. Of particular significance, our findings indicate a correlation between FGFR4 and LAR-enriched TNBC. In prostate cancer, AR has been reported to regulate FGFR2 and FGFR4. FGFR4 is known to confer therapeutic resistance and aggressive behavior in cancer through cancer-associated fibroblast education via transforming growth factor β. Pooled database analysis suggests that inhibiting FGFR4 or CCL2 may be a potential therapeutic strategy for LAR-enriched TNBC. FGFR4 inhibitors, such as roblitinib,³¹ are currently being evaluated for various solid tumors, including hepatocellular carcinomas. Further efforts are required to determine the feasibility of these targets as potential therapeutic targets, considering the availability of FGFR inhibitors in preclinical and clinical settings.

AR is a known immune regulator.³² One of our two patients with the immune modulatory (IM) subtype of TNBC received ZT and had a complete pCR. Thus, not to emphasize an idiosyncrasy but rather to generate a hypothesis, in a future trial, it would be interesting to investigate whether AR inhibitors may effectively treat TNBC with IM features. Indeed, several groups have tested AR-targeted therapeutics in combination with immune checkpoint inhibitors,³³ but no consistent efficacy has been reported.

The expression of AR in TNBCs is known to be heterogeneous.¹⁵ Therefore, analyzing individual cells from tumor samples before and after treatment could provide insights into predicting response and understanding the mechanisms of ZT. Although single-cell analysis of TNBC has been performed previously,³⁴ it did not specifically focus on TNBC subtypes with ARness.

With the FDA’s endorsement of the KEYNOTE-522 protocol, the integration of pembrolizumab into a carboplatin-inclusive chemotherapy regimen has become the standard NAT for TNBC. This development necessitates a re-evaluation of traditional treatments that do not include pembrolizumab, which may result in potentially higher response rates in our study population. It is, therefore, critical to collect and analyze data from patients with LAR-enriched TNBC who have undergone the KN522 regimen to ensure the continued applicability of our observed treatment efficacy. Currently, neither the percentage of stromal tumor-infiltrating lymphocytes nor any other parameters we have can predict the potential benefit of pembrolizumab.³⁵ The acquisition of real-world data is thus paramount to further our understanding in this area. Nonetheless, the efficacy signal and preliminary biomarker data obtained from our study support the need for randomized controlled trials with a focus on biomarkers to validate the efficacy of the ZT regimen in the neoadjuvant setting and re-evaluate the nominated predictive biomarkers. Future trials can build upon our preliminary findings and provide further insights.

In summary, our study demonstrates the efficacy of the ZT regimen for treating AR-positive TNBCs that showed insufficient response to AC treatment in the neoadjuvant setting. If our findings are confirmed, then the ZT regimen could significantly contribute to reducing recurrence and mortality rates in patients with LAR-enriched TNBC. Moreover, confirmation of our findings may aid in better defining the patient population that would benefit the most from this innovative treatment approach.

Limitations of the study

We acknowledge several limitations of our study. One significant limitation of our study is the small number of enrolled patients. We faced several challenges in patient enrollment, including the relatively low proportion of LAR-enriched TNBC cases and other molecular features that led patients to other ARTEMIS subtrials. While we included 24 patients instead of the originally planned 37, we were able to demonstrate an improvement in the rate of pCR and RCB-I. However, the statistical power of our study was limited to confirm these as predictive biomarkers. When designing our study, hormone receptor expression in less than 10% of cells was broadly recognized as a criterion for TNBC.³⁶ We will consider this discrepancy between the definition in this study and the current definition according to the American Society of Clinical Oncology–College of American Pathologists guidelines carefully in our future confirmatory studies.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
clone AR441	Dako, Carpinteria, CA, USA	N/A
Critical commercial assays
RNAlater	Thermo Fisher Scientific	N/A
PicoGreen assay	Invitrogen	N/A
TapeStation	Agilent	N/A
Ultrasonicator	Covaris	N/A
KAPA Library Preparation Kit	Kapa Biosystems	N/A
AMPure PCR Purification Kit	Agencourt Bioscience	N/A
Total RNA Purification Kit	Norgen Biotek	37500
AMPure XP beads	Beckman Coulter Life Sciences	N/A
Quant-iT RiboGreen RNA Assay Kit	Thermo Fisher Scientific	N/A
RNA Nano Kit	Agilent	6000, 2100
Ovation RNAseq System V2	NuGEN	N/A
Quant-iT PicoGreen dsDNA Assay Kit	Thermo Fisher Scientific	N/A
SureSelectXT Low Input Reagent Kit	Agilent	N/A
Cycle Sequencing V3 reagents	Illumina	N/A
TruSeq Kit v3	Illumina	N/A
Software and algorithms
DESeq2	Bioconductor R package	N/A
Sciclone G3 NGSx Workstation	PerkinElmer	N/A
CIBERSORT		N/A
ConsensusTME		N/A
KEGG	MSigDB	N/A
Reactome	MSigDB	N/A
Hallmark	MSigDB	N/A
Deposited data
The DNA- and RNA-sequencing data for three of the patients	dbGaP accession number phs000001.v1.p1, 2024; https://www.ncbi.nlm.nih.gov/gap	N/A

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Bora Lim (Blim@mdanderson.org).

Materials availability

This study did not generate new unique reagents.

Data and code availability

Clinical data for the patients in this study are not publicly available to protect patient privacy. The genomic and transcriptomic data supporting this study were generated in the ARTEMIS trial (NCT02276443). A superset of this data is being made publicly available with a manuscript describing the trial and molecular profiling pending publication of the master protocol data. Not all patient-omics data are deposited to publicly available dataset due to confidential medical records/consent form agreed by each patient. Before this occurs, the data used for this trial will be made available upon reasonable request to lead contact. This study has not provided a suggested acknowledgment statement. However, Approved Users are expected to acknowledge the Submitting Investigator(s) who submitted data from the original study to an NIH-designated data repository, the primary funding organization that supported the Submitting Investigator(s), and the NIH designated data repository (e.g., dbGaP). The acknowledgment statement should include the dbGaP accession number to the specific version of the dataset(s) analyzed. Please cite/reference the use of dbGaP data by including the dbGaP accession phs003586.v1.p1.

Method details

Study population

The ZT trial (ClinicalTrials.gov identifier: NCT02689427) is a single-institution, single-arm study. The last date of data review was July 10, 2022. A copy of the most recent version of the protocol is included in the Supplemental information. To be eligible, patients had to be at least 18 years old, not be pregnant, and have an Eastern Cooperative Oncology Group performance score of 0–2 without proven metastatic disease or other cancer within 5 years of trial enrollment. Normal cardiac, renal, and hepatic organ function was required.

Patients with a history of prior therapy with doxorubicin or paclitaxel or greater than grade 1 neuropathy were excluded. Concurrent radiation therapy was not permitted. The major inclusion criteria were an evaluable primary tumor or biopsy-proven axillary node involvement with a primary tumor size of at least 1 cm in the smallest dimension based on imaging before initiation of neoadjuvant chemotherapy. TNBC was defined according to an immunohistochemical staining percentage of three receptors: less than 10% for both ER and progesterone receptor and HER2 0–1+ by IHC or 2+, fluorescence in situ hybridization nonamplified. Patients with a nuclear AR expression rate of at least 10% whose tumors did not shrink more than 70% after four cycles of AC were reviewed for further eligibility. AR scoring was done manually, not by software on scanned images. Following heat-induced epitope retrieval with citrate buffer for 25 min at 100°C, slides were incubated with mouse monoclonal antibody to AR (clone AR441, Dako, Carpinteria, CA, USA; 1:30). The percentage of any nuclear staining of any intensity in the tumor cells was recorded. Baseline measurements and evaluations were obtained within 4 weeks of registration to the study. All disease areas were recorded to assess response and uniformity of response to therapy.

Study design and endpoints

The originally planned accrual was 37 patients; however, the study completed accrual at 24 patients given the low number of patients with LAR-enriched TNBC, difficulty of continued recruitment, and a recommendation from the sponsor to proceed with early efficacy signal rather than waiting for several more years to achieve the planned accrual. The protocol (in Supplemental information) has not been modified to maintain harmony among the subtrials of the ARTEMIS master protocol (MDACC 2014-0185); however, the ARTEMIS molecular tumor board approved early closure of this specific study (Figure 1A, CONSORT diagram). The primary objective of this study was to evaluate the rate of combined pCR and RCB-I for patients with TNBC who did not respond to initial AC-based NAT and were treated with ZT in the neoadjuvant setting (Figure S1A). The pathologists evaluated the extent of residual invasive and metastatic cancer in each surgical resection specimen using the RCB calculator.³⁷ The secondary objective was to determine the EFS and OS rates. We also compared the secondary endpoints between the good and poor response groups. Exploratory objectives were to investigate the association of peripheral blood and tumor biomarkers with clinical response and survival outcomes. The most appropriate optimal regimen among phenotypically targeted second-phase trials after patients were noted to have insufficient responses to four cycles of AC was identified at the bi-weekly molecular tumor board meeting of investigators in the ARTEMIS trial. Once selected as best suited for treatment with ZT, patients were enrolled in our trial to receive enzalutamide (120 mg) by mouth daily on days 1–7 and paclitaxel intravenously over 2 h on day 1. Treatment cycles were repeated every 7 days for up to 12 cycles unless disease progression or unacceptable toxicity required a therapy modification. All AEs were documented, and AEs greater than grade 2 were reported to the Investigational New Drug Office at The University of Texas MD Anderson Cancer Center. The efficacy and safety analysis included any enrolled patient who received at least 1 week of ZT therapy.

Ethics approval and consent

All patients gave written informed consent to participate in the ZT trial. Consent for the omics analysis of collected tissue was obtained under a larger umbrella protocol, ARTEMIS (NCT02276443; MDACC 2015-0488) (Figure 1A. CONSORT diagram). The subsequent use of patients’ samples without disclosure of patient identifiers met waiver of informed consent policy criteria in accordance with the Declaration of Helsinki. The study was approved by the MD Anderson Institutional Review Board and the U.S. Food and Drug Administration for the use of drugs.

Master protocol and procedure, biomarker analysis

All molecular correlative studies involving molecular profiling data were performed using biopsy samples of the patient’s primary breast tumors and performed under an existing ARTEMIS molecular triaging protocol. Participants in ARTEMIS were included on the basis of the availability of sufficient tissue for molecular and immune analysis along with corresponding germline controls. Patients diagnosed with stage I–III TNBC had a pretreatment core-needle biopsy performed before the initiation of NAT with AC. The efficacy of AC was assessed using breast imaging techniques (ultrasound and/or MRI) after the second and fourth cycles.

Those showing disease progression or less than 70% reduction in tumor volume after four AC cycles were considered to have a suboptimal response per previously reported TNBC trials^19,20 and were given the option to participate in a clinical trial investigating novel systemic therapy combinations. Conversely, those with better responses were advised to proceed with standard-of-care NAT using paclitaxel ± carboplatin (refer to Figure S1). The ARTEMIS protocol was approved by the MD Anderson Cancer Center Institutional Review Board, and all participants gave informed consent. Only patients with molecular profiling results were included in this analysis. Upon completing their second phase of NAT with either a taxane-based regimen or targeted therapy under a molecularly targeted investigational arm, patients underwent surgical resection of tumor and lymph nodes. Board-certified breast pathologists blindly determined their RCB index. Characterization of each baseline tissue sample was used to determine eligibility for the second phase of the trial if a patient had a suboptimal response to the four cycles of AC. Not all patients had available whole exsome sequencing (WES) data. While RNA sequencing could furnish much of the information furnished by WES, the initial samples were analyzed solely using RNA microarray, which limited their utility.

Tumor biopsy collection and storage

Direct ultrasound guidance was used for all biopsies. A core needle biopsy was performed using three commercially available disposable 14-gauge biopsy needles—Marquee (Bard Biopsy Systems), MaxCore (Bard Biopsy Systems), and Achieve (CareFusion)—with similar needle throws (18 or 25 mm for Marquee, 22 mm for MaxCore, and 25 mm for Achieve). Also, fine-needle aspiration biopsy was performed using commercially available 21-gauge hypodermic needles. Biopsy samples were transferred to 2-mL cryopreservation tubes and stored in RNAlater at −80°C until extraction of DNA and RNA analytes.

Germline blood collection and DNA extraction

Whole blood was collected in a 10-mL K3-EDTA spray-dried vacutainer. Genomic DNA was extracted from whole blood using a Wizard Genomic DNA Purification Kit (A1620). Briefly, whole blood was placed in cell lysis buffer and centrifuged to separate the white blood cells. White blood cells were then incubated in nuclear lysis buffer followed by a protein precipitation solution and centrifuged, and the supernatant was separated. Genomic DNA was isolated using isopropanol and washed in 70% ethanol. After the DNA pellet was air-dried, it was rehydrated and stored at −80°C for future use. Extracted DNA was measured using a NanoDrop spectrophotometer. as per the manufacturer’s instructions.

Genomic DNA library preparation and capture

Genomic DNA was quantified using a PicoGreen assay (Invitrogen), and its quality was determined using Genomic DNA Tape with a 2200 TapeStation (Agilent). DNA from each sample used for this analysis (100–500 ng of genomic DNA) was sheared via sonication under the following conditions: peak incident power of 175, duty cycle of 20%, intensity of 5, 200 cycles per burst, and duration of 120 s using an E220 focused-ultrasonicator (Covaris). To ensure the proper fragment size, samples were checked with TapeStation using a High Sensitivity DNA Analysis kit (Agilent). Following the “with beads" manufacturer protocol, sheared DNA was subjected to library preparation using a KAPA Library Preparation Kit (Kapa Biosystems). This protocol consists of three enzymatic reactions for end-repair, A-tailing, and adaptor ligation followed by barcode insertion via polymerase chain reaction (PCR) using KAPA HiFi polymerase (six cycles). PCR primers were removed using a 1.8x provided volume of AMPure PCR Purification kit (Agencourt Bioscience). At the end of the library preparation, samples were analyzed using TapeStation to verify the correct fragment size and ensure the absence of extra bands. The amount of genome was quantified using a KAPA quantitative PCR kit. Equimolar amounts of DNA were pooled for the capture (two to six fragments per pool). Whole-exome biotin-labeled probes (Exome v3; Roche Nimblegen) were used, with the manufacturer’s protocol followed for the capture step. Briefly, DNA was pooled (two to six samples) and dried. After the addition of capture reagents and probes, samples were incubated at 47°C using a thermocycler with a heated lid (57°C) for 64–74 h. The targeted regions were recovered using streptavidin beads, the streptavidin-biotin probe target complex was washed, and another round of PCR amplification was performed according to the manufacturer’s protocol (Agencourt Bioscience). The quality of each captured sample was analyzed using TapeStation and High Sensitivity DNA Analysis kit. The enrichment was examined via quantitative PCR using specific primers designed by Roche Nimblegen. The cutoff for enrichment was a minimum of 50-fold.

RNA extraction and quality control

RNA was extracted from tumor samples using a Total RNA Purification Kit (37500; Norgen Biotek). The extracted RNA was treated with DNase I to eliminate any genomic DNA residues. The treated RNA was then purified using AMPure XP beads (Beckman Coulter Life Sciences) and eluted into a 1x TE buffer. The purified RNA was quantified using a Quant-iT RiboGreen RNA Assay Kit (Thermo Fisher Scientific), and the RNA quality was assessed using an RNA 6000 Nano Kit and 2100 Bioanalyzer (Agilent). cDNA was prepared from the extracted total RNA using an Ovation RNAseq System V2 (NuGEN). Amplification was initiated at the 3′ end as well as randomly throughout the transcriptome in each sample. The prepared cDNA was quantified using a Quant-iT PicoGreen dsDNA Assay Kit (Thermo Fisher Scientific), and its quality was assessed using Genomic DNA ScreenTape and reagents with a TapeStation 4200 (Agilent). Up to 200 ng of each cDNA sample was measured per PicoGreen quantification was sheared (mechanically fragmented) using the E220 focused-ultrasonicator. Sonication was performed under the following conditions: peak incident power of 200, duty cycle of 25%, 50 cycles per burst, and duration of 10 s for 120 iterations. To ensure the proper fragment size, samples were examined using the TapeStation 4200 with a High Sensitivity DNA Kit (Agilent). The sheared cDNA was subjected to the library preparation using a SureSelectXT Low Input Reagent Kit with 1–96 indexes (Agilent) via an automated method using a Sciclone G3 NGSx Workstation (PerkinElmer). The protocol consisted of three enzymatic reactions for end repair, A-tailing, and adaptor ligation followed by barcode insertion via PCR using Herculase II Fusion DNA Polymerase (8–14 cycles based on input DNA quality and quantity; PCR primers were removed using a 1x basic volume of AMPure PCR Purification kit (Agencourt Bioscience). The quality and quantity of the prepared libraries were evaluated using the TapeStation 4200 and High Sensitivity DNA Kit to verify the correct fragment size and ensure complete removal of primer dimers. Subsequently, the prepared libraries were individually hybridized to SureSelect Human All Exon V4 probes (Agilent). The hybridization steps were automated using the Sciclone G3 NGSx Workstation. Captures were hybridized as single-sample reactions using 500–1000 ng of prepared library as the input. All hybridization and post-hybridization captures and washes were performed according to Agilent’s protocol. Briefly, the capture reagents and probes were added to the prepared libraries, and the mixture was incubated at 65°C using a thermocycler with a heated lid for up 24 h. The targeted regions were captured using streptavidin beads, the streptavidin-biotin probe-target complex was washed, and the captured libraries were enriched via PCR amplification according to the manufacturer’s protocol (Agencourt Bioscience). The quality and quantity of each captured sample was analyzed using the TapeStation 4200 with the High Sensitivity DNA Kit. The captured libraries were sequenced on the Illumina NovaSeq 6000 platform for 2 × 150 paired-end reads with an 8-nt read for indexes using Cycle Sequencing V3 reagents (Illumina).

RNAseq analysis

Raw RNA sequence data were processed using an in-house RNAseq data analysis pipeline, which, among other tools, uses Spliced Transcripts Alignment to a Reference to align raw reads to hg19 version of the human reference genome, featureCounts to quantify aligned reads with to produce raw counts, and FastQC and Qualimap to evaluate the quality of the raw reads and feature counts. After assessment of sequencing quality control metrics, RNAseq yielded data for downstream analysis of 271 (90% of all ARTEMIS samples) tumors.

TPM counts using RSEM were used to extract expression levels for protein-coding genes only. Data were normalized using VST, and further batch correction was performed using ComBat. Batch-corrected data were used for single-set gene set enrichment analysis, which was run using GSVA separately over MSigDB gene sets, including Hallmark, Reactome, and KEGG. TME immune deconvolution was performed on a batch-corrected matrix using CIBERSORT, ConsensusTME with default parameters. The Bioconductor package DESeq2 (adjusted p < 0.05)(38) was used to identify differentially expressed genes according to pCR and non-pCR status and to perform subtype comparisons.

RNAseq TNBC subtyping

Non-negative matrix factorization was applied using two to eight factors over 1000 iterations and 30 runs to discover the optimal number of clusters in ARTEMIS TNBC cohort. The cophenetic distance and silhouette width were used to optimally select six clusters to generate the Figure 3A, and non-negative matrix factorization was run again to get the final weights and factors. To determine the association of specific gene sets with ART-Types (please see below), a t-test was used to compare single-set gene set enrichment analysis scores within samples of a subtype versus all others. Significant gene sets were derived using a false discovery rate (FDR) less than 0.05. A similar comparison was applied to immune-deconvolution estimates.

Whole exome analysis

Captured libraries were sequenced using a TruSeq Kit v3 with a paired-end flow cell (Illumina) according to the manufacturer’s instructions at a cluster density of 700–1000 K clusters/mm². Sequencing was performed using a HiSeq 4000 for 2 × 100 paired-end reads with a 7-nt read for indexes using Cycle Sequencing V3 reagents (Illumina). The resulting BCL files containing the sequence data were converted into FASTQ files, and individual libraries within the samples were demultiplexed using CASAVA 1.8.2 (Illumina) with no mismatches. More than 20 reads covered all regions. For data analysis, the WEX captured deep-sequencing data were aligned to human reference assembly hg19 using BWA, and duplicated reads were removed using Picard.

Quantification and statistical analysis

A two-stage Simon’s two-stage design was employed for the original design of this phase II study (protocol attached in supplemental information), alpha = beta = 10%, and setting the threshold for an acceptable pCR or RCB-I rate at 20%. During the enrollment, if at least 1 of the first 12 patients to undergo treatment had either a pCR or RCB-I disease, the plan was to add 25 more patients for a total of 37 patients, with planned interim analysis at every six enrolled and evaluable patients. This design had a 54% chance of termination after the first stage if the true response rate was 0.05, 28% if the true response rate was 0.10, 14% if the true response rate was 0.15, and 7% if the true response rate was 0.20.

The accrual past the first stage to ensure the true response rate greater than 20%, however the study stopped early due to lower number of patients population and changing practice pattern due to the approval of pembrolizumab added to standard therapy. The proportion of patients with a pCR or RCB-I disease as the primary endpoint and those in the remaining RCB categories were determined. We also divided patients into two groups based on the response: the GRG for those who met the primary endpoint and the PRG for RCB-II and III. EFS was defined as the time from surgery to the recurrence of breast cancer, either distant or local, or death from any cause. OS was defined as the time from surgery to death from any cause. The EFS and OS distributions were the secondary endpoints, using Kaplan and Meier to estimate the distribution from the surgery date. Patients who remained alive (OS) or alive and event free (EFS) were censored at the last contact date. Distributions were compared using the log rank test. Logistic regression was used to assess the association between ZT response and covariates of interest. A P-value threshold of <0.05 was used as a cutoff for statistical significance, and no statistical adjustment was made for the multiplicity of testing. All statistical analyses were performed using R version 4.1.1. All statistical tests used a significance level of 5%.

Additional resources

• •The trial information can also be found at https://clinicaltrials.gov/
• •The serial number for the clinical trial is NCT02689427.

Published: June 4, 2024